The secreted proteins of the Hedgehog family have been implicated in many different processes in vertebrate development including cartilage differentiation, myotome and sclerotome specification, hair follicle development, limb morphogenesis and the specification of different neuronal cell types. In addition, the aberrant activation of the Hedgehog pathway has been identified as the likely cause of a number of tumours in humans including basal cell carcinomas (BCCs) and primitive neurectodermal tumours (PNETs). Elucidating the mechanisms by which Hedgehog signals are transduced will thus have widespread implications for our understanding of both normal development and disease.
Members of the hedgehog (hh) gene family encode a novel class of secreted proteins that act as intercellular signals. The active signalling form of the protein (HH‐N) (Lee et al., 1994; Bumcrot et al., 1995; Porter et al., 1995, 1996b) is covalently coupled to cholesterol during autoproteolysis of full‐length Hh (Porter et al., 1996a), a modification that appears to influence the range of its signalling activity. Now widely recognized as being of major importance in vertebrate development (Hammerschmidt et al., 1997), Hh signalling was originally described in Drosophila. Unlike vertebrates, where multiple Hh family members have been identified, flies have only a single hh gene, but this nevertheless controls a number of key developmental processes. Though most recent studies have focused on its role in regulating the growth and patterning of the wing and other appendages in the adult fly (Basler and Struhl, 1994; Diaz‐Benjumea and Cohen, 1994; Zecca et al., 1995; Blair and Ralston, 1997; Rodriguez and Basler, 1997), hh was first identified as one of the Drosophila segment polarity genes, so called because their mutations disrupt the pattern and polarity of the larval segments (Nüsslein‐Volhard and Wieschaus, 1980). In hh homozygotes, the larval cuticle becomes covered in spiky processes called denticles (Mohler, 1988), hence the name ‘hedgehog’. This phenotype is shared by many members of the segment polarity class, suggesting that all of the genes identified by these mutations may function in a common pathway. Genetic analysis has revealed that most segment polarity genes are in fact involved in the transduction of either the Hh signal (Forbes et al., 1993) or of the signal encoded by the wingless (wg) gene, a member of the Wnt family of signalling proteins (van den Heuvel et al., 1993). The segment polarity genes have thus provided a key resource in the dissection of these two signalling pathways, and their molecular characterization is now shedding light on the detailed biochemical mechanisms that underlie transduction of Hh signals.
A major function of hh in the Drosophila embryo is the maintenance of wg transcription at the boundary of each segmental unit (Hidalgo and Ingham, 1990; Ingham, 1993; Ingham and Hidalgo, 1993); from here, Wg protein diffuses across the segment to specify the character of the ectodermal cells that secrete the larval cuticle (Lawrence et al., 1996). Like hh, mutations in three other segment polarity genes smoothened (smo), fused (fu) and cubitus interruptus (ci) eliminate wg transcription at parasegmental borders (Forbes et al., 1993; Ingham, 1993; Préat et al., 1993; van den Heuvel et al., 1993; Motzny and Holmgren, 1995; Alcedo et al., 1996; van den Heuvel and Ingham, 1996); by contrast, mutation of a fourth gene, patched (ptc), leads to the derepression of wg (Ingham et al., 1991; Martinez Arias et al., 1988). By making double mutant combinations between ptc and the other genes, it was established that smo, fu and ci all act downstream of ptc to activate wg transcription (Forbes et al., 1993; Hooper, 1994) whilst, on the other hand, transcription of wg becomes independent of hh in the absence of ptc (Ingham and Hidalgo, 1993). These findings suggest a simple pathway whereby hh acts to antagonize the activity of ptc which in turn antagonizes the activity of smo, fu and ci. The universality of this pathway subsequently has been established both in Drosophila, where ptc, smo, fu and ci mediate the activity of Hh in all processes studied to date (Ma et al., 1993; Chen and Struhl, 1996; Forbes et al., 1996; Sanchez‐Herrero et al., 1996; van den Heuvel and Ingham, 1996; Strutt and Mlodzik, 1997), and in vertebrates, where homologues of ptc, smo and ci (but not as yet fu) have been identified and implicated in processes mediated by one or other of the Hh family proteins (Concordet et al., 1996; Goodrich et al., 1996, 1997; Marigo et al., 1996b,c; Stone et al., 1996; Hynes et al., 1997; Lee et al., 1997; Quirk et al., 1997).
Patched and Smoothened: opposing partners in the Hedgehog receptor
Since ptc encodes a transmembrane protein (Hooper and Scott, 1989; Nakano et al., 1989), the simplest interpretation of the genetic data from Drosophila is that it acts as a receptor for Hh (Ingham et al., 1991). The predicted topology of the Ptc protein—two large extracellular loops and 12 transmembrane domains—is not, however, typical of receptors for secreted polypeptides and its proposed role accordingly remained controversial until relatively recently. In the meantime, an alternative candidate for a receptor emerged with the discovery that smo also encodes a transmembrane protein that bears at least a passing resemblance to the large family of G protein‐coupled receptors (Alcedo et al., 1996; van den Heuvel and Ingham, 1996). Moreover, Smo exhibits sequence similarity to members of the Frizzled family of serpentine proteins, some of which have been shown to act as receptors for Wnt proteins (Bhanot et al., 1996), making the case for Smo as the Hh receptor seem more compelling. In vitro binding assays have, however, failed to detect any physical interaction between vertebrate Smo and Hh proteins (Stone et al., 1996) whereas, under the same conditions, vertebrate Ptc binds the Sonic hedgehog (Shh) protein with relatively high affinity (Marigo et al., 1996a; Stone et al., 1996). As might be expected, the two large extracellular domains of Ptc are required for this binding (Marigo et al., 1996a), though the binding domains have yet to be mapped precisely. These biochemical properties are supported by genetic data from Drosophila implicating Ptc in binding Hh. In the appendage primordia of the adult fly, the diffusion of Hh away from expressing cells is strictly limited, but can be significantly increased by eliminating Ptc expression from the Hh‐receiving cells. The Hh protein diffuses across clones of cells lacking Ptc expression until it reaches genetically wild‐type Ptc‐expressing cells at the edge of the clone. The implication of this behaviour is that Hh is normally sequestered by Ptc (Chen and Struhl, 1996).
Taken together, these findings suggest that Hh acts by binding to Ptc, thereby releasing an inhibitory effect of Ptc on Smo (Figure 1). Since Ptc and Smo are both transmembrane proteins, a likely scenario is that they physically associate to form a receptor complex. This interpretation is supported by the fact that Ptc can be co‐immunoprecipitated with Smo from cultured cells expressing both proteins (Stone et al., 1996). The release of Smo from Ptc inhibition most likely involves a conformational change rather than the dissociation of the complex, because the association of the two proteins, as assayed by co‐immunoprecipitation, is unaffected by the presence or absence of Hh (Stone et al., 1996). It is, however, important to remember that binding of Smo to Hh‐bound Ptc is not essential for its activity, since Smo becomes constitutively activated in the complete absence of Ptc protein (Alcedo et al., 1996; Quirk et al., 1997). This does of course beg the question as to whether some as yet unidentified protein may act as a ligand for Smo, a possibility suggested by the long extracellular N‐terminal domain in the Smo protein. Indeed, although the in vitro binding assays cannot detect an interaction between Hh and Smo, a direct interaction between the two proteins cannot be ruled out entirely; in this regard, it is notable that the phenotype of an animal lacking both hh and ptc activity is not identical to that of an animal mutant for ptc alone, implying that Hh can signal even in the absence of Ptc (Bejsovec and Wieschaus, 1993). Whether or not this residual activity is mediated by Smo remains to be established.
It seems clear, however, that the principal regulation of Smo is via its interaction with Ptc, and it follows from the model that at least some loss‐of‐function mutations in ptc should act by disrupting binding to Smo. The discovery that mutations in the human ptc homologue, Ptch, are widespread in basal cell carcinomas (BCCs) (Hahn et al., 1996; Johnson et al., 1996) has provided a major stimulation for the analysis of Ptc/Smo function as well as an abundant source of loss‐of‐function mutations. Many tumour‐derived alleles of Ptch have now been sequenced, but the results of these studies to date have been relatively unrevealing, the majority of the mutations characterized being due to premature termination of the coding region (Chidambaram et al., 1996; Wicking et al., 1997).
Disruption of Smo–Ptc binding could also be caused by mutations in smo; in contrast to ptc mutations, these should be dominantly acting (since they would lead to constitutive activity of the mutant protein) and would not, therefore, be recovered in conventional Drosophila screens (since they would be dominant lethal). Recent studies of human BCCs however, have identified two such activating mutations in Smo, both of which appear to be responsible for the transformation of basal keratinocytes (Xie et al., 1998). Despite the predictions, neither of these appear to act by disrupting the association of the protein with Ptc, since both mutant forms can still be co‐immunoprecipitated with Ptc. While it is possible that these mutations reduce the affinity of Smo for Ptc, an alternative explanation is that they uncouple the intracellular signalling activity of Smo from regulatory domains within the protein. Intriguingly, one of these alleles has been identified as a Trp to Leu substitution at a position in the seventh transmembrane domain which, when mutated in the G‐protein coupled a‐1B adrenergic receptor, causes its constitutive activation (Xie et al., 1998). This raises the possibility that Smo may itself couple to heterotrimeric G proteins, a possibility supported by recent evidence suggesting that related Frizzled family members stimulate the phosphatidylinositol signalling pathway via G proteins (Slusarski et al., 1997).
Inside the cell: a complex process
If events at the cell surface seem unconventional, the manner in which the signal is transduced inside the cell is no less unusual. Although genetic analysis identified Fu and Ci as downstream components of the pathway, their epistatic relationships with Ptc could not determine precisely where in the signalling pathway each protein acts. A high degree of sequence similarity with members of the vertebrate Gli family of transcription factors (Orenic et al., 1990), however, gave a strong indication that ci encodes the ultimate component of the pathway. A considerable body of evidence suggesting that Ci does indeed directly mediate the transcriptional activation of target genes in response to Hh signalling has now been adduced both from in vitro and in vivo studies: binding sites for Ci have been identified upstream of the promoters both of wg and of ptc, which is itself transcriptionally regulated by Hh, and these sites have been shown to be sufficient to mediate activation of reporter genes by Ci in tissue culture assays (Von Ohlen and Hooper, 1997; Von Ohlen et al., 1997) and, in the case of the ptc promoter, to mediate Hh‐dependent transcription in transgenic flies (Alexandre et al., 1996).
So how does Hh signalling regulate the activity of ci? In vertebrates, there is good evidence that the transcription of Gli genes is under the control of Hh proteins (Marigo et al., 1996b; Borycki et al., 1998), the consequent increase in levels of Gli proteins induced by Hh signalling resulting in the activation of different targets. However, in Drosophila, transcription of ci is clearly independent of Hh (Slusarski et al., 1995); in both the embryonic segments and the adult appendage primordia (the imaginal discs), ci is transcribed uniformly in all cells that are responsive to Hh. Thus, if Hh does control ci activity, it must do so post‐transcriptionally. Since overexpression of ci is sufficient to activate target genes independently of Hh activity (Alexandre et al., 1996; Dominguez et al., 1996; Hepker et al., 1997), a simple model is that Hh acts by increasing the levels of Ci protein, perhaps by regulating its rate of turnover. A key insight into this process came last year with the discovery by Aza‐Blanc et al. (1997) that Ci is subject to proteolytic cleavage. In cells not exposed to the Hh signal, the predominant form of Ci is a 75 kDa species that comprises the N‐terminus and the central DNA‐binding zinc finger domain, but lacks the C‐terminal half of the full‐length protein (Aza‐Blanc et al., 1997). The missing portion contains not only an acidic region identified as a transcriptional activation domain (Alexandre et al., 1996) but also a binding site for the CBP, which itself is essential for the activation of Hh target genes (Akimaru et al., 1997). Thus, in the absence of Hh induction, it seems that the form of Ci that can activate transcription is specifically eliminated from cells by this proteolytic cleavage. When cells receive the Hh signal, however, the cleavage of Ci is blocked, resulting in the accumulation of the full‐length 155 kDa form of the protein. While the simplest interpretation of these data would seem to be that Hh acts by promoting the accumulation of the active form of Ci, the picture is complicated by the fact that the smaller 75 kDa form retains DNA‐binding activity and can itself act as a transcriptional repressor (Aza‐Blanc et al., 1997). This property, together with its accumulation in non‐Hh‐induced cells, correlates well with the finding that some genes—including another Hh target decapentaplegic (dpp), as well as hh itself—are ectopically transcribed in cells that lack expression of Ci. This paradoxical finding suggests the possibility that activation of dpp by hh might actually be achieved by the elimination of the 75 kDa repressor form of the protein from hh‐responding cells rather than by accumulation of the activating form of the protein (Aza‐Blanc et al., 1997). Against this, however, genetic analysis has shown that ci is absolutely required for the activation of other Hh targets such as ptc and wg (Forbes et al., 1993); moreover, ectopic expression of ci is sufficient to activate dpp as well as ptc and wg transcription (Alexandre et al., 1996; Dominguez et al., 1996; Hepker et al., 1997). Thus, the principal role of Ci seems to be to activate transcription of Hh targets.
A picture of how Ci processing is controlled has emerged recently, based to a large extent on the characterization of three genes identified by mutations that cause wing duplications (Jiang and Struhl, 1995, 1998; Lepage et al., 1995; Li et al., 1995; Robbins et al., 1997; Sisson et al., 1997). This phenotype is characteristic of the inappropriate activation of the Hh pathway in the developing wing primordium and, consistent with this, high levels of full‐length Ci protein accumulate independently of Hh signalling in the wing cells of each mutation. Thus, each mutant identifies a gene required in some way for the normal processing of Ci, any one of which could therefore mediate the control of Ci by Hh. The product of the slmb gene (Jiang and Struhl, 1998) is an F box/WD40 repeat protein homologous to the Saccharomyces cerevisiae cdc 4 protein required for the ubiquitination of cell cycle regulators. By analogy, Slmb may target Ci for ubiquitination and hence for cleavage by the proteasome pathway (Jiang and Struhl, 1998), a possibility supported by the finding that proteasome inhibitors block the processing of Ci in tissue culture cells (D.Stark and P.W.Ingham, unpublished observations). Although the partial degradation of a protein via the proteasome is unusual, analogous proteasome‐dependent processing of the NF‐κB and cystic fibrosis transmembrane regulator (CFTR) proteins has been reported previously (Palombella et al., 1994; Jensen et al., 1995). Since targeting of proteins to the proteasome is commonly regulated by their phosphorylation, a plausible scenario would be that Hh acts by modulating Ci phosphorylation. While the phosphorylation status of Ci has yet to be established, sequence analysis has revealed the presence of a number of consensus protein kinase A (PKA) phosphorylation sites that have been conserved in the vertebrate Gli proteins (Akimaru et al., 1997; Chen et al., 1998). The implication that PKA may thereby regulate Ci processing is supported by the fact that another of the wing duplication mutations inactivates the catalytic subunit of PKA (Jiang and Struhl, 1995; Lepage et al., 1995; Li et al., 1995). Moreover, mutating each of the putative phosphorylation sites in Ci results in increased stability of the full‐length protein and a concomitant increase in its transcriptional activation activity (assayed in transfected tissue culture cells) (Chen et al., 1998). Although this does not constitute definitive proof, the strong implication is that PKA phosphorylates Ci directly, raising the possibility that Hh might regulate Ci cleavage by regulating the activity of PKA (Chen et al., 1998). While such a model would be consistent with Smo coupling to a G protein‐regulated signalling pathway, the finding that a constitutively activated PKA catalytic subunit does not suppress Hh signalling suggests rather that Hh acts in parallel with PKA (Jiang and Struhl, 1995). In fact, it transpires that the processing of Ci is also regulated by its sequestration within the cytoplasm and this process depends upon the product of the costal‐2 (cos‐2) gene, mutation of which, like slmb and PKA, causes wing duplication (Whittle, 1976; Grau and Simpson, 1987). Unlike Slmb and PKA, however, the cos‐2 product is a novel protein, though it does possess an N‐terminal domain that shares significant sequence similarity with the motor domains of kinesin superfamily proteins (Sisson et al., 1997). Through this domain, the Cos‐2 protein binds to microtubules, taking with it both the novel serine‐threonine kinase encoded by fused and Ci with which it forms a high molecular weight multimeric complex (Robbins et al., 1997; Sisson et al., 1997). Intriguingly, the association of this Cos‐2–Fu–Ci complex with microtubules is itself regulated by Hh signalling, the entire complex dissociating intact from the microtubules when cells are exposed to Hh (Robbins et al., 1997). Since the full‐length form of the Ci protein accumulates in cells lacking Cos‐2 activity (Sisson et al., 1997) as well as in cells induced by Hh, it follows that the association of Ci with microtubules is essential for its processing (Figure 1). In this view, the critical step in the regulation of Ci activity by Hh is the modulation of its sequestration rather than its PKA‐dependent phosphorylation. The latter process may be necessary to target Ci to the complex or, as suggested above, for its ubiquitination.
How Hh induces the dissociation of the Cos‐2–Fu–Ci complex from microtubules remains a mystery for the present. One possibility is that Smo activates the Fu serine‐threonine kinase which in turn could act by inhibiting microtubule binding by Cos‐2, a possibility consistent with genetic epistasis analysis that places cos‐2 downstream of fu (Préat et al., 1993). However, although Cos‐2 does become hyperphosphorylated in response to Hh activity, this response is delayed relative to the release of the complex from the microtubules, and in any event it is independent of the activity of the Fu kinase domain (Robbins et al., 1997). Moreover, the activity of Fu is entirely dispensable for normal Hh signalling when a fourth component of the complex, the so‐called Suppressor of fused [Su(fu)] protein is also eliminated (Préat, 1992). In in vitro assays, Su(fu) can bind directly to Fu and to Ci; since there is no evidence for a direct interaction between the latter proteins or indeed between them and Cos‐2, it may be that Su(fu) forms a bridge between them, stabilizing their association with one another and with Cos‐2 (Monnier et al., 1998) (Figure 1). Other as yet unidentified proteins almost certainly participate in complex formation, however, since Su(fu) is not essential for Hh signalling; until these are identified, a complete picture of how the complex is regulated will remain elusive.
Genetic analysis in Drosophila has provided an outline of the Hh signalling pathway, the details of which are now beginning to be filled in at the biochemical level. Whilst a relatively clear picture is starting to emerge, many issues remain to be addressed. A major gap in our understanding lies in the events between activation of Smo and the dissociation of the Ci–Fu–Cos2–Su(fu) complex from microtubules, while the failure to detect significant levels of the full‐length form of Ci in the nucleus—the high levels of this species that accumulate in response to Hh are located almost exclusively in the cytoplasm (Aza‐Blanc et al., 1997)—remains a long‐standing puzzle. Another paradox concerns the role of PKA; although, as discussed above, inactivation of PKA results in activation of the pathway, recent studies in the Drosophila embryo have shown that hyperactivity of PKA can have the same effect (Ohlmeyer and Kalderon, 1997). Consistent with a role for PKA in promoting Ci cleavage, this PKA‐induced activation occurs in the absence of an increase in the levels of Ci155; one possibility is that PKA activates Ci155 through phosphorylation of some other cryptic site or sites or, alternatively, it may act on some other co‐factor. Whatever the target, this effect of PKA appears to be dependent upon the activity of Smo (Ohlmeyer and Kalderon, 1997).
Finally, it is reasonable to ask just how far the lessons learned from Drosophila can be extrapolated to vertebrates. At first sight, the evidence that the pathway has been highly conserved seems quite compelling. The results of the expression and functional analyses of vertebrate Ptc homologues largely fulfil the predictions based on analogy with Drosophila studies (Goodrich et al., 1996, 1997), while the biochemical characterization of the Hh–Ptc–Smo interactions has actually been performed using the vertebrate gene products (Marigo et al., 1996a; Stone et al., 1996). It is notable, however, that vertebrates have two homologues of Ptc (Concordet et al., 1996; Takabatake et al., 1997), and recent expression analyses of Ptc2 in mice reveal a pattern not predicted by analysis in Drosophila (Motoyama et al., 1998). Similarly, where Drosophila has just one ci gene, in vertebrates there are at least three members of the Gli family, each of which exhibits its own distinctive pattern of expression (Hui et al., 1994). One possibility is that in Drosophila the functions of all the Gli gene products have been subsumed by Ci; Gli3, for instance, appears to act as a repressor of Shh expression whereas Gli1 and 2 may act only as activators of Hh family target genes (Marigo et al., 1996b). This begs the question as to whether the Gli proteins are subject to the same processing as Ci and, if not, is their regulation by Hh signalling exclusively at the level of transcription? Significantly, recent studies of the COUP TFII gene, a target of Shh signalling in the neural tube, have identified a Shh response element that does not bind Gli proteins, but does bind an unidentified nuclear factor that is susceptible to modulation by protein phosphatase IIA (Krishnan et al., 1997). No parallel for such a factor has yet been described in Drosophila; however, evidence for Ci‐independent regulation of a Hh target has been obtained recently with the identification of a cis‐acting element upstream of wg that lacks Ci consensus binding sites yet drives transcription in a hh‐dependent manner (Lessing and Nusse, 1998). Thus, it seems clear that transcription factors other than members of the Ci/Gli family are regulated directly by Hh signalling; the next challenge will be to identify these factors and the pathways that control them.
I am indebted to Jeremy Quirk for his help in the preparation of Figure 1 and to Linda Parsons and Marysia Placzek for their comments on earlier drafts of this review.
- Copyright © 1998 European Molecular Biology Organization