Genetic analysis of hedgehog signalling in the Drosophila embryo

During early development of the Drosophila embryo, cells become allocated to parasegments, metameric units with the properties of secondary embryonic fields, by the activity of pair-rule genes such as fushi tarazu and even-skipped (reviewed by Ingham and Martinez Arias, 1992). This process occurs at the transition from the syncitial phase of embryonic development, when positional information is generated through intracellular gradients of transcription factors (St. Johnston and Nüsslein-Volhard, 1992), to the multicellular phase, when positional information depends upon cell-cell communication. Amongst the many genes involved in this latter phase of development, those of the segment polarity class (Nüsslein-Volhard and Wieschaus, 1980) play a fundamental role in the patterning of cells along the anteroposterior body axis of the embryo.

The establishment of parasegments is marked by the activation of the segment polarity genes wingless (wg-, Baker, 1987) and engrailed (en;DiNardo et al., 1985; Fjose et al., 1985; Kornberg et al., 1985), the interfaces of whose expression domains define the parasegmental boundaries (van den Heuvel et al., 1989). Although these expression patterns are initiated by the pair-rule genes during the blastoderm stage (DiNardo and O’Farrell, 1987; Howard and Ingham, 1986; Ingham et al., 1988), both genes subsequently come to depend upon the activity of each other for their maintenance (Martinez Arias et al., 1988). Expression of wg is maintained only in cells immediately adjacent to the en domain, whereas en expression persists in a stripe one to two cells wide immediately posterior to the wg-expressing cells. In the absence of wg activity, en expression ceases soon after gastrulation (DiNardo and O’Farrell, 1987; DiNardo et al., 1988; Martinez Arias et al., 1988). This requirement for wg activity is very transient, en expression becoming independent of wg within 5 hours of fertilisation, at stage 9 (Bejsovec and Martinez Arias, 1991 ; DiNardo and O’Farrell, 1987; DiNardo et al., 1988; Heemskerk et al., 1991; Martinez Arias et al., 1988). Since wg is the orthologue of the mammalian proto-oncogene Wnt-l (Rijsewijk et al., 1987), encoding a secreted glycoprotein, it has been postulated to act as the signalling molecule that directly mediates maintenance of en transcription. This interpretation is supported by immunolocalisation studies that reveal that wg protein is found 1-2 cells away from its source (Gonzalez et al., 1991; van den Heuvel et al., 1989), corresponding to the width of the en domain under wg control.

Once en expression becomes independent of wg activity, the parasegment boundaries and hence the metameric organisation of the embryo, can be considered to be stable. Expression of wg persists at each boundary, however, and is subsequently required for the correct specification of cell identity within each parasegment (Bejsovec and Martinez Arias, 1991; Dougan and DiNardo, 1992). The dependence of wg transcription upon en activity implies a reciprocal interaction between the two cell populations at each parasegmental boundary (Martinez Arias et al., 1988). The best candidate for a signal mediating this interaction is the product of the hedgehog (hh) gene, a protein with a single putative transmembrane domain (Lee et al., 1992; Mohler and Vani. 1992; Tabata et al., 1992; Tashiro et al., 1993). Transcription of hh is coincident with en (Tabata et al., 1992) and expression of wg in neighbouring cells decays during stage 9 in hh mutant embryos (Hidalgo and Ingham. 1990; Ingham and Hidalgo. 1993). Genetic analysis suggests that the hh signal is transduced in an unusual manner, acting in some way to antagonise the activity of the segment polarity gene patched (pic), which functions to repress the transcription of wg (Ingham and Hidalgo. 1993; Ingham. 1991).

The consequences of removing either wg or hh activity as revealed by their terminal cuticular mutant phenotypes are superficially very similar: in both cases, the precise segmental pattern of anterior denticle rows and posterior smooth or naked cuticle seen on the ventral surface of the newly hatched larva is replaced by a fairly homogeneous lawn of denticles (Baker. 1988; J ü rgens et al., 1984; Nüsslein-Volhard ct al.. 1984; Mohler. 1988; sec Fig. I for details). Given that hh mutants lack irg expression, this similarity is not unexpected; moreover, the hh mutant phenotype is almost totally suppressed by the restoration of wg expression that occurs in embryos doubly mutant for hh and pic (Ingham et al., 1991). Yet while wg is required for the maintenance of en transcription along the entire parasegment boundary, in hh mutant embryos en expression persists in certain regions of the border (DiNardo et al., 1988). This finding suggests that the regulation of wg transcription by hh commences only after the expression of en has become wg-independenl. That en expression does eventually decay in hh mutants implies a seperalc role for hh in the maintenance of en and helps explain the similarity between the terminal phenotypes of wg and hh mutants.

If maintenance of wg transcription and maintenance of en in a specific sub-set of en-expressing cells are two distinct and independent functions of hh, we reasoned that mutations in genes downstream of the hh signalling pathway, responsible for maintenance of wg transcription, should abolish wg transcription without eliminating en expression. Amongst the known segment polarity mutations we have identified two in particular, fused (fu. Limbourg-Bouchon et al., 1991; Martinez-Arias. 1985) and cubitus interruptus Dominant (ci^D ; Orenic et al., 1990). that fulfil this criterion for components of the hh-wg pathway.

Since the maintenance of ptc transcription in narrow stripes of cells that Hank each en domain has previously been shown to require hh activity (Hidalgo and Ingham. 1990). we have also compared the effects of hh, fu and ci^D mutations on ptc transcription. The results suggest that transcriptional control of both ptc and wg by hh is mediated by the same signal transduction pathway.

Like its transcript, the protein product of the hh gene is distributed in a series of stripes located around the anterior boundaries of each parasegment (Fig. 2B). Whereas the hh transcription domain occupies about one quarter the width of each parasegment, coinciding with the domain of expression of en (Tabata et al., 1992). Hh protein, however, appears to be more widely distributed than the Fn protein (compare Fig. 2A and B) (Taylor ct al.. 1993). The protein is first clearly detectable some lime after the onset of gastrulation (stage 8) with levels peaking (.luring stage 10 and becoming almost undetectable again by stage 12 (Taylor et al., 1993). This expression profile is consistent with the known requirements for hh function, expression of veg disappearing in hh mutant embryos by stage 10 (Ingham and Hidalgo. 1993).

At the sub-cellular level. Hh protein displays a distinctive non-uniform distribution; antibody staining is excluded from the apical region of ectodermal cells (sec Fig. 2C) and the protein has a “capped” appearance accumulating in discrete patches which are almost invariably juxtaposed to similar accumulations in adjacent cells (Fig. 2D). In addition, the protein appears to accumulate in “dots”, similar to those previously seen with Wg-specilic antibodies (Gonzalez cl al., 1991; van den Heuvcl et al., 1989). This is reflected in the highly particulate appearance of the staining in whole embryos.

In wild-type embryos wg is activated during the blastoderm stage in dorsoventrally continuous stripes at the posterior margin of each parasegment; this expression is maintained until the beginning of stage 10 when lateral expression is lost so that each parasegment has separate dorsal and ventral stripes of veg-expressing cells (see big. 3). This later phase of veg expression in the ventral ectodermal cells has been shown to be specifically required for the differentiation of naked cuticle in the posterior part of each segment (Bejsovec and Martinez Arias, 1991; Dougan and DiNardo, 1992). Embryos mutant for the segment polarity genes ci^D and fu display cuticular phenotypes similar to that resulting from the late loss of wg activity (see Fig. 1). In ci^DR5° homozygous embryos, a reduction in the level of wg expression can first be seen at stage 8, and by the end of stage 9 ectodermal expression in the trunk region of the embryo is lost from all but 6 neuroblasts in each parasegment (see Fig. 3). As in wild-type embryos, this neuroblast expression fades as embryogenesis proceeds. During stage 10, ectodermal expression is activated in segmental patches along the dorsal edge of the embryo. Such dorsal activation is also observed in hh mutants (Ingham and Hidalgo, 1993) but in both ci^D and hh mutants this expression never reaches the levels observed in wild-type embryos and is lost during dorsal closure.

We have also re-examined the expression pattern of wg in embryos lacking wild-type fu activity, first reported by Limbourg-Bouchon et al. (1991). Complete loss of fu activity is lethal, and some alleles cause ovarian tumours and therefore female sterility. Accordingly, we used females of the adult viable genotype fu¹ /fu^v22 that have non-tumorous ovaries and crossed these to fu¹ males to generate embryos with reduced fu activity (hereafter referred to as fu⁻ embryos). The pattern of wg transcription in these embryos is essentially identical to that seen in ci^D and hh homozygotes (see Fig. 3).

Previous studies have shown that in ptc mutant embryos wg transcription is hh independent, leading to the notion that hh acts by antagonising ptc activity in some way (Ingham and Hidalgo, 1993; Ingham et al., 1991). A ptc mutation can therefore be thought of as being equivalent to a hh gain of function mutation, at least with respect to its effect on the regulation of wg transcription. Accordingly, if fu and ci^D are required downstream of hh they should both be epistatic to ptc. To investigate this possibility, we constructed fu⁻ptc and ptc ci^D double mutants and assayed them for wg expression. In both ptc^G12; ci^DR50 homozygous embryos and in fu⁻ embryos homozygous for ptc^G12 the expression of wg is indistinguishable from that observed in fu⁻ or ci^DR50 single mutant embryos (Fig. 4); similarly, the cuticular phenotypes of the double mutants are identical to those of the single mutants alone (data not shown). Thus both genes are required for the maintenance of wg expression after stage 9, irrespective of the presence or absence of ptc, indicating that they act downstream of ptc and, by extension, hh.

Although the spatial distribution of ptc transcript is initially quite distinct from that of wg (Hooper and Scott, 1989; Nakano et al., 1989), by stage 10 of embryogenesis the two genes share the characteristic of being expressed in cells adjacent to those expressing en, and hence hh. Whereas expression of wg is confined to cells anterior to each hh domain, ptc is transcribed in stripes of cells flanking each hh stripe (see Fig. 5). This results from repression of transcription in the middle of each broad stripe of pic-expressing cells, and the maintenance and further activation of transcription in cells that are adjacent to those expressing en (Fig. 5)

Using promoter deletion analysis, we have separated regions of the ptc 5 ′ DNA that are required for this later expression, from those required for the initial pattern of broad stripes (Y. N., A. J. F. and P. W. L, unpublished observations). Reporter genes containing 12 kb of ptc upstream sequence give an expression pattern which is indistinguishable from that of the endogenous gene from gastrulation onwards. Broad segmental stripes appear during stage 8 which then resolve into two narrow stripes showing dorsoventral modulation typical of ptc (Fig. 5B,C). By contrast, smaller constructs containing 3.2 kb or 2.5 kb of upstream sequences show only the later part of this expression pattern. In wild-type embryos these small constructs are first activated in segmental pairs of narrow stripes from early stage 10 onwards (Fig. 5D). These narrow stripes are maintained through germband extension although they become less well defined during stage 11 due to dorsoventral modulation of expression within each stripe. Thus 2.5 kb of cis-acting sequences are sufficient to activate transcription in cells flanking the hh domain.

In the absence of hh activity ptc transcription fails to be maintained or activated in the cells bordering those expressing en and hh (Hidalgo and Ingham, 1990). In embryos homozygous for the strong loss of function allele, hh^iJ, this is first obvious at stage 10. In wild-type embryos at this stage, expression within the broad stripe has faded sufficiently dorsally for the maintenance of pairs of narrow stripes to be observed at the dorsal edge of the embryo (Fig. 6A). These narrow stripes do not however appear in hh mutants (Fig. 6B). Similarly, expression of the 3.2 and 2.5 kb reporter constructs fails to be activated in hh mutants (data not shown).

The disappearance of ptc transcription in hh mutants occurs with the same time course, and in the same pattern as the disappearance of midstripe expression in wild-type embryos. Expression is lost in a dorsal to ventral direction, leaving only a ventral posterior triangle of transcription in stage 11 embryos. This corresponds to the region of strongest expression within the broad stripes of stage 9 embryos, and is the last region of midstripe expression to be lost in wild-type embryos.

During stage 11 in hh mutants, ptc transcription is reactivated transiently in the tracheal placodes. This seems to correspond to the strong dorsal patch of activation, observed in the anterior narrow stripe in wild-type embryos. In hh mutants this activity and the remaining ventral posterior expression is completely repressed by stage 12 when all ectodermal staining has disappeared. Mesodermal expression of ptc appears to be independent of hh activity as transcription is activated normally during germband contraction in hh mutant embryos.

During early extended germband (stage 9-10) the pattern of ptc expression in fu⁻ embryos is very similar lo that observed in hh^i,j mutants. Narrow stripes fail to be activated dorsally as the broad stripes of expression start to disappear (Fig. 6C). Fading of (he broad stripes in fu⁻ embryos continues during stages 10-11 and. as in hh^i,j. the strong ventral-posterior patch of expression remains and activation lakes place around the forming tracheal pits. However, in contrast to hh^IJ mutants, during stage 11 expression also begins to be activated dorsally and anterioventrally. in a pattern that resembles wild-type embryos at this stage. Although this activation is weaker than in wild-type, by the time the germband starts lo contract narrow stripes of expression can be distinguished bordering the en domains. As the germband contracts these narrow stripes of low level expression are maintained with even weaker expression throughout the broad domain between them.

In ci^DR50 homozygous embryos, the pattern of ptc expression in early extended germ bands similarly resembles that observed in hh mutants. As the broad stripes of ptc expression fade, transcription fails to be maintained or activated in the cells bordering the hh/en domain (Fig. 6D). Expression is lost completely dorsally but remains at low levels throughout the broad stripe ventrally (see Fig. 6). During stage II. pie transcription is activated around the tracheal placodes but in contrast to hh mutants, it is also reactivated al low levels in broad segmental stripes that exclude only the en domain. These stripes are modulated dorsoventrally such that expression is lowest laterally; however, no signs of increased levels of transcription in cells adjacent to the en domain arc observed. These broad stripes of weak expression are maintained throughout germ band contraction. Loss of ci^D activity also has an effect on mesodermal expression of ptc. the transcript being present at abnormally high levels from stage 11 through germband contraction.

To confirm that the changes in pic and mg transcript accumulation seen in ci^D mutant embryos reflect changes at the level of transcriptional control, we analysed the effects of the CI^DR50 mutation on the expression of IacZ in two enhancer trap lines in which the reporter gene is under the control of enhancers of cither ptc or ug. Expression of the wp-lacZ enhancer trap fades from the ectoderm during stage 8–9 leaving only neuroblast expression, which in turn disappears (Fig. 7); expression is then activated at the dorsal edge of the embryo in the same pattern as the endogenous gene at stage 10. Similarly, expression of the ptc-lacZ reporter line I184 fails to be maintained in the cells bordering the cn domain as the broad stripes of expression fade in stage 9–10 embryos (big. 7) and during stage 11 low level activation occurs throughout the non-rzz-expressing region of the segment. Like the endogenous gene, the ptc-lacZ reporter is also expressed at elevated levels in the mesoderm during germband contraction.

Embryos lacking both maternal and zygotic wild-type activity of the costal-2 (cos-2) gene develop a larval cuticular phenotype that resembles that of ptc mutant larvae (Grau and Simpson. 1987). In both cases the denticle rows in the posterior part of each ventral bell are eliminated and replaced with denticles of more anterior character. In contrast to ptc mutants, however, there is no duplication of the segment boundary or of the anterior row of small denticles in cos-2 mutant larvae and no reversal of polarity of the remaining large denticles.

To investigate whether the phenotypic similarity between cos-2 and ptc mutations is based on similar effects at the cellular level we analysed the expression of ptc and wg in cos-2 mutant embryos. Females with cos-2 mutant germ lines were generated by pole cell transplantation and crossed to males heterozygous for the cos-2⁻ allele. By germband extension (stage 9) the level of ptc transcript in embryos as judged by in situ hybridisation is much higher than in their heterozygous siblings; each broad stripe of ptc expression persists through stage 12. failing to resolve into the two narrow stripes that characterise wildtype embryos at this stage (Fig. 8). Both these effects resemble the changes in the pattern of ptc transcription seen in embryos that lack a functional copy of the ptc gene itself (Hidalgo and Ingham. 1990). There are. however, some subtle differences between ptc and cos-2 embryos. In the absence of ptc activity the broad stripes of ptc expression do eventually split lo give a narrow stripe of nonexpression in the middle of the segment. It has been suggested that this is due to the repression of ptc in cells expressing en eclopically (Hidalgo and Ingham. 1990). In cos-2 embryos in which there is no ectopic en expression (Forbes, 1992). such splitting of the ptc expression domain is not observed.

In cos-2 embryos there is also a dorsovenlral modulation in the intensity of ptc expression not observed in ptc embryos. This modulation is first apparent in the broad stripes in stage 9 extended germbands. Expression fades laterally during stages IO and II. then as the germband contracts two regions of strong expression, both al the anterior of the broad stripe become clear in the ventral-lateral. and dorsal-lateral regions of each stripe. As in the wild-type ptc pattern, expression from the late extended germband (stage I I) to the end of development is strongest al the anterior of the segment i.e.: adjacent to the anterior segment boundary.

In contrast to (he early effects on ptc transcription, expression of ug initially appears normal in cos-2⁻ embryos. Al stage 11. however, the ug domain becomes significantly broader than in wild type (Fig. 8). ‘This ectopic expression of wg persists ventrally until the end of embryogenesis. By contrast, expression of wg fades dorsally during dorsal closure (stage 13–14).

The broadened expression of wg in cos-2 embryos from stage I I is similar to the pattern of wg observed in ptc mutants; in both cases the ventral expression domain of wg expands lo fill about half of each segment however, in cos-2 embryos (his change in wg expression occurs much later than in ptc mutants in which the wg domains broaden in stage 9 embryos. The irregularity of some of the abdominal wg stripes in ptc mutants compared lo cos-2 may partly be due lo the ectopic segment boundaries which form in the former, but not in the latter mutant embryos. A further difference between these mutants is in the late maintenance of the dorsal domain of teg expression; dorsal expression fades in cos-2 embryos, while in ptc mutants, as in wild type, wg continues to be expressed dorsally until the end of embryogenesis.

The wg gene has at least two temporally distinct functions in the development of each parasegment, acting first to consolidate the parasegment boundaries by maintaining en expression (DiNardo cl al., 1988; Martinez Arias et al., 1988) and subsequently regulating the differentiation of individual cells such that they secrete naked cuticle rather than denticles (Bejsovec and Martinez Arias. 1991 ; Dougan and DiNardo. 1992). A particular level of wg activity appears to be required lo specify naked cuticle: in its absence, cells of the ventral ectoderm produce a lawn of denticles, whereas if wg is eclopically expressed either under the control of a heal shock promoter (Nordcmecr et al., 1992). or through the effects of other mutations such as naked and ptc (Dougan and DiNardo. 1992; Martinez Arias et al., 1988), denticle differentiation is supressed. The maintenance of wg in a single narrow stripe therefore seems essential for determining the proportion of each parasegment that will secrete naked cuticle. Here, we have presented evidence that the ci^D and fu genes are specifically required to maintain wg expression after en has become independent of the wg signal, and suggest that they do so by transducing the hh-encoded signal.

The transcription pattern of wg in fu and cl^D mutant embryos is essentially indistinguishable from that seen in hh mutants. In each case wg expression is lost from the ectoderm during stage 9. Since the maintenance of wg at this stage is known to depend upon wg activity, it could be that both fu and ci^D act downstream of the wg signal, perhaps regulating the transcription or activity of en, which in turn feeds back on wg transcription via the activity of hh. Two considerations argue against this possibility: first, the phenotypic consequences of fu and ci^D mutations are significantly less severe than those caused by loss of wg function; second, the expression of En protein in the stage 10 embryo is relatively unaffected by the absence of activity of either fu (Limbourg-Bouchon et ah, 1991) or ci^D (A. J. F. and P. W, I., unpublished observations).

Of course the possibility remains that either gene might post-translationally modulate the activity of en, thus influencing its ability to regulate the expression of the putative signal (Limbourg-Bouchon et al., 1991). Although we have not analysed the expression of hh in either mutant it is clear from the results of our epistasis analysis that both genes act downstream of the hh signal. It has been proposed previously that hh acts by antagonising the activity of ptc (Ingham et al., 1991) and since ptc encodes an integral membrane protein (Hooper and Scott, 1989; Nakano et al., 1989), one implication being that it acts as a receptor for the hh signal. In embryos doubly mutant for ptc and hh, wg is ectopically expressed (Ingham and Hidalgo, 1993), indicating that in the absence of ptc, wg transcription becomes independent of the signal. By contrast, wg transcription ceases at late stage 9 in both ptc;ci^D and fu\ptc embryos. Thus fu and ci^D are required for wg expression irrespective of the activity of either ptc or hh, implying that both genes act as downstream components of the hh signalling pathway.

Since en expression is largely unaffected in stage 10 ci^D and fu embryos, the loss of en expression observed in hh embryos at the same stage (DiNardo et al., 1988) implies an additional role for hh that must be mediated by a different pathway. Several other lines of evidence have suggested that regulation of wg expression is not the only role of hh’, for instance, the hh cuticular phenotype is stronger than that of wg (Fig. 1); the cuticular phenotype of ptc;hh more closely resembles that of ptc en than of ptc alone (Hidalgo, 1991); and ubiquitous expression of wg, driven by a hs-wg construct, while able to partially rescue wg mutants fails to rescue the hh mutant phenotype (Sampedro et al., 1993).

Previous studies have implicated hh in the maintenance of ptc transcription, in addition to its roles in regulating wg and en expression, in the stage 10 embryo. In normal development, the pattern of ptc transcription resolves from a broad stripe, occupying the posterior three quarters of each parasegment, into two narrow stripes of cells that flank the AA-expression domain (Hooper and Scott, 1989; Nakano et al., 1989). Deletion analysis of the ptc 5′ region has allowed us to separate regions of DNA responsible for later Ci^D-dependent expression in segmental pairs of narrow stripes from those necessary for early expression in broad stripes. While a fragment containing 12 kb 5′ of the ptc transcription start site gives the complete post-blastoderm pattern of expression, smaller fragments of 3.2 kb and 2.5 kb direct expression only in pairs of narrow stripes flanking the hh domain in extended germband embryos. That this symmetrical expression of ptc depends upon the activity of hh suggests that the hh signal is not polarised, but rather that the anterior and posterior cells have differing competences to respond to the same signal. Since the activity of ptc represses its own transcription as well as that of wg (Hidalgo and Ingham, 1990), the maintenance of ptc transcription by hh is probably mediated in a similar manner, hh antagonising the activity of ptc in responding cells. Thus, the difference in competence between the two cells expressing wg and those expressing ptc seems unlikely to be at the level of signal reception. Another possibility is the existence of distinct signal transduction pathways such that the hh signal can elicit different outcomes in different cells; for instance one such pathway might operate in cells anterior to each hh domain, transducing the signal to activate wg transcription, whilst a distinct pathway could mediate the activation of ptc transcription in cells on either side of M-expressing cells. Our finding that both/it and ci^D are required for the activation of ptc transcription in cells flanking the hh domain, strongly suggests, however, that the same factors transduce the hh signal to.regulate both wg and ptc transcription.

The molecular structures of both/u and ci^D fit well with their proposed roles as intracellular transducers of the hh signal. ci^D encodes a zinc finger protein homologous to the mammalian proto-oncogene GLI (Orenic et al., 1990) suggesting that it may regulate ptc and wg expression at the level of transcriptional control. This interpretation is supported by our finding that the effects of the ci^D mutation on expression of wg-lacZ and ptc-lacZ reporter genes mirror those on the endogenous genes. Thus ci^D is exerting its control via the cfs-acting regulatory regions of both genes, suggesting that the Ci^D protein binds directly upstream of the wg and ptc promoters to activate transcription. At present nothing is known about the distribution of the Ci^D protein during embryogenesis; however transcription of ci^D is not limited to those cells in which it activates wg and ptc transcription but occurs throughout most of each parasegment, in all cells except those expressing en. If the Ci^D protein is similarly distributed, it follows that its activity must be regulated at the post-translational level. In this regard it is interesting that the product of the/w gene is a serine threonine kinase (Preat et al., 1990), raising the possibility that ci^D activity may be modulated by phosphorylation.

In embryos lacking wild-type activity of the cos-2 gene we have found an effect on wg and ptc expression reciprocal to that seen in hh, fu and ci^D mutants. As in ptc mutant embryos, wg and ptc are both ectopically expressed when cos-2 activity is reduced; in contrast to ptc mutants, however, expression of en is unaffected. Thus cos-2 most likely acts downstream of ptc to repress the transcription of both wg and ptc. If, as we have suggested, fu activates Ci^D by phosphorylation, one possible role for cos-2 could be to inhibit this activation, by interacting either with Fu or with Ci^D itself. In the former case, cos-2 might simply inactivate Fu. Alternatively, cos-2 might encode a phosphatase whose activity could reverse the phosphorylation of the Ci^D protein. It is, of course, equally possible that cos-2 is itself the substrate for fu activity in a linear pathway of negative regulation leading from the cell surface to the nucleus (Fig. 9). According to this scheme the function of ptc would be to inhibit the activity of Fu, whose activity in turn inhibits the activity of Cos-2. The latter might repress ptc and wg transcription by inactivating the Ci^D protein or perhaps sequestering it, preventing its entry into the nucleus.

An analagous pathway of negative interactions between the genes that regulate sex determination in C. elegans has previously been proposed on the basis of the results of genetic analysis (Kuwabara and Kimble, 1992). Intrigu-ingly, not only are there similarities between the two pathways at the genetic level, but also in the molecular structures of some of the gene products involved. Thus the nematode her 1 gene, like hh, encodes a product with the characteristics of a secreted protein (Kuwabara and Kimble, 1992); tra-2 encodes an integral membrane protein with multiple transmembrane domains showing gross topological similarity, though only marginal sequence similarity, to ptc (Kuwabara et al., 1992). Finally, tra-1, the most downstream component of the nematode pathway, encodes a zinc finger protein with significant homology to ci^D (Zarkower and Hodgkin, 1992).

Although the different signalling pathways outlined in Fig. 9 are consistent with the results of our epistasis analysis, such analysis cannot distinguish between linear and parallel pathways. It is equally possible that hh acts via an entirely separate pathway to activate Ci^D_t independent of the ptdcos-2/fu pathway. In this case the balance between activation and inactivation of Ci^D by these two parallel pathways would regulate the transcription of wg and ptc.

While the expression of both wg and ptc in embryos lacking wild-type cos-2 activity resembles their expression in ptc mutants by stage 12, the ectopic activation of wg in the former is significantly delayed. Thus while ptc is expressed at elevated levels from stage 8 onwards, the domains of wg expression only broaden after stage 11. Since the cos-2 allele that we have analysed is not a complete loss of function mutation, it may be that residual cos-2 activity is able to maintain repression of wg during the early stages of embryogenesis. In terms of our models, this could be taken to mean that activation of wg transcription requires a higher threshold level of ci^D activity than does activation of ptc. In this regard, it is interesting that transcription of ptc is reinitiated in stage 12/13 embryos lacking wild-type fu activity. Like the cos-2⁵ allele, the fu alleles that we have analysed necessarily cause only the partial inactivation of the fu gene. Thus it is possible that such mutant embryos retain sufficient activity to activate transcription of ptc, but not of wg. Again, in terms of our model, this would be consistent with different threshold levels of ci^D being necessary for the activation of the two genes. Why these threshold requirements should differ with developmental stage is unclear; however, it is almost certain that additional transcription factors co-operate with ci^D to regulate the expression of ptc and wg at different stages. The segment polarity gene gooseberry, for instance, is known to be required for transcription of wg, though not of ptc (Hidalgo and Ingham, 1990), in the stage 11 embryo. Clearly, the transcriptional control of both genes is likely to be complex; a detailed analysis of their cw-acting regulatory regions will be needed to provide a clearer understanding of this complexity.

Our analysis has focussed on the mechanisms controlling the expression of two segment polarity genes, ptc and wg, at a particular stage of embryogenesis. Both genes are subject to complex transcriptional regulation, a complexity reflected in their rapidly evolving patterns of expression in the developing embryo. In the case of wg, we know that these different controls underpin the multiple roles of wg as development proceeds. The functional significance of this later phase of ptc transcription is unclear but it suggests that ptc may also have late functions independent of its early role in regulating wg and ptc transcription. Differences between the patterns of en expression in ptc and cos-2 mutants (Forbes, 1992) and in embryos in which hh is ectopically expressed (P.W.I., unpublished results) imply that ptc may have a role in limiting the extent of the en domain. The activty of ptc may have a repressive effect on the activation of en in response to wg, possibly by inhibiting the reception of the wg signal, or by altering the competence of cells to respond. Alternatively, ptc may act in some way to prevent en-expressing cells from mixing with their (pre-expressing) neighbours during the cell movements of germband contraction, thus helping to maintain the integrity of the en domain. The expression of ptc on both sides of the en domain means it is in the correct intrasegmental location for both these suggested functions.

We are grateful to P. Simpson, D. Gubb, B. Limbourg, S. Pinchin, C. Niisslein-Volhard and R. Holmgren for providing us with mutant Drosophila stocks, G. Riddihough and K. Howard for plasmids, and A. Hidalgo for helpful discussions.