Pattern formation in a secondary field: a hierarchy of regulatory genes subdivides the developing Drosophila wing disc into discrete subregions

Most work on pattern formation in Drosophila has focused on understanding the genetic regulatory hierarchy and molecular mechanisms that pattern the embryo along the anterior-posterior (A/P) and dorsal-ventral (D/V) axes. These studies have revealed much about the mechanisms of patterning within a primary field, but comparatively little is known about the mechanisms by which secondary patterning fields (e.g., insect wings and legs, vertebrate limb buds, etc.) are generated. The urodele amphibian limb, cockroach leg and Drosophila wing disc are all model systems utilized to study the patterning of secondary fields (see French et al., 1976). Traditionally, patterning of these structures was studied by observing their regulative capacities following disruption by either mechanical (ablation) or genetic (temperature-sensitive alleles or clonal analysis) means. The striking observation was that all three fields regulated in analogous manners, implying that the underlying patterning systems are similarly organized in secondary fields in organisms as far diverged as insects and vertebrates (French et al., 1976; Bryant and Gardiner, 1992). Numerous models to explain the regulative behavior of these systems have been proposed (e.g. the polar coordinate model; French et al., 1976 and Bryant et al., 1981); however, the verification of these models awaits molecular evidence for the patterning mechanisms operating in any of these systems.

Of these model systems, the Drosophila wing disc is par-ticularly well suited for the genetic and molecular dissec-tion of patterning processes in a cellular field. In previous work, we have shown that the embryonic origin and subsequent development of the wing can be followed using an antibody to the vestigial protein (Williams et al., 1991). The disc is first visualized as a cluster of cells in germ-bandretracted embryos and forms a pouch of approximately fifty cells at the time of embryo hatching. The size of the disc increases exponentially during larval life and contains approximately 50,000 cells at the end of the third larval instar (Whittle, 1990). During metamorphosis, this pouch evaginates to form most of the epidermal (cuticular) structures of the dorsal thorax, as well as the wing.

The adult wing exhibits anterior-posterior, dorsal-ventral and proximal-distal polarity, and positional information arises at some time during wing development that distinguishes the different subregions or compartments. Two compartmental restrictions that develop within the wing disc have been defined by mitotic clonal analysis and reveal the presence of boundaries that prevent clones of cells from mixing (Garcia-Bellido et al., 1976). One clear restriction is the anterior/posterior (A/P) boundary, which is established early in embryogenesis and requires the engrailed gene product in the posterior compartment for its maintenance (Lawrence and Morata, 1976). The other proposed compartmental restriction is the dorsal/ventral (D/V) boundary, which is drawn along the wing margin, separating the future dorsal and ventral wing surfaces and is detected in clones during early larval life (first or second larval instar; Garcia-Bellido et al., 1976). However, the presence of a zone of non-proliferating cells along the developing wing margin throughout the third larval instar (O’Brochta and Bryant, 1985) has led to considerable debate as to whether the detected clonal restriction is a compartment boundary or simply a region of reduced cell proliferation (Brower, 1985). In addition to these potential lineage restrictions, the late third instar disc is extensively patterned, as evidenced by the complex expression patterns identified by enhancer traps (for a review, see Whittle, 1990) and the array of cuticle structures formed when specific disc fragments are allowed to metamorphose separately (Bryant, 1978). Indeed, the latter technique has been used to make a detailed fate map of the disc such that the location of the primordia for the adult structures derived from this disc are known (Bryant, 1978).

Our approach to understanding patterning in this system is focused on the formation of the presumptive wing region of the wing imaginal disc. In terms of proximal-distal patterning, this disc forms both distal (wing) and proximal (body wall; notum) structures. The wing is a relatively recent evolutionary innovation (primitive ancestral insects lack wings) and may represent a primary division within this patterning field. Analysis of wing formation could therefore reveal basic patterning mechanisms within the larger imaginal field. A key advantage to studying this problem in Drosophila is the availability of numerous mutations that alter the adult cuticle (Lindsley and Zimm, 1992). Indeed, a large number of mutations have been identified that alter the development of the wing. However, only a few of these have been found to globally and specifically remove the entire structure. One well-studied gene, decapentaplegic (dpp), is required for proximal/distal patterning of all adult appendages (Posakony et al., 1990), while four other genes, which are now cloned, are specifically required for formation of the wing region of the disc (i.e. these genes do not appear to perform obviously analogous functions in other tissues). Strong mutations in three of these genes result in complete loss of the wing [apter ous (ap), scalloped (sd) and vestigial (vg); see Lindsley and Zimm (1992) for mutant descriptions] while one results in a transformation of wing to notum [wingless (wg)]. All of these genes are required for other functions in development (e.g. wingless is a tissue-specific component of a signaling pathway that is required for embryonic and imaginal patterning; see Peifer and Bejsovec, 1992), but the effects of loss of these genes within the wing disc (during larval development; wg is required during embryonic development for imaginal disc formation; Simcox et al., 1989) are specific to the wing region of the disc (Baker, 1988a,b; Williams and Bell, 1988; Stevens and Bryant, 1985; James and Bryant, 1981). The proteins encoded by these genes possess structural features [e.g. homeodomain (ap; Cohen et al., 1992), novel nuclear protein (vg; Williams et al., 1991), TEA domain putative transcription factor (sd; Campbell et al., 1992) and int-1 growth factor homologue (wg; Rijsewijk et al., 1987)] that are consistent with their playing primary roles in establishing the wing region of this disc.

The availability of molecular probes for these genes has allowed us to examine the dynamics of early regulatory gene expression in the wing disc, the correlations between regulatory gene expression and the wing fate map, and the regulatory interactions among these genes that are required to specify the wing region of the disc. The results indicate that the four genes fall into two classes that reflect either an early or late requirement in disc development. vg and sd are required later in development (during the third larval instar) for formation of the wing region of the disc and their expression patterns reflect the specification of the wing region of the disc during disc development. ap, in contrast, is required before wing specification to establish the proper domains of sd and vg expression and to mark the D/V boundary within the developing wing region. wg has the earliest requirement and influences both the notum/wing and dorsal/ventral boundaries. It plays a key role in positioning ap expression and promoting vg expression, which suggests that it sits at or near the top of the genetic regulatory hierarchy guiding wing formation. It is clear from the details emerging from the study of various developmental systems that wingless and its many relatives contribute to the patterning of a wide variety of invertebrate and vertebrate fields.

All gene and allele designations are as in Lindsley and Zimm (1992). vg and ap are not essential genes, and null alleles of vg (vg^83b27R and vg^AlExt; Williams et al., 1991) were used to examine vg defects. Homozygous ap enhancer trap flies show a strong ap phenotype (Fig.1, almost as extreme as ap null flies; Cohen et al., 1992) and this genotype was utilized for examination of alterations in ap mutants. wg is an essential gene, so heteroallelic combinations of two cis-acting wg hypomorphic alleles were used to generate viable wg mutant larvae (wg¹ and In(2L)wg^P; see Baker, 1988a). Likewise strong sd alleles are early larval lethals, so a pupal lethal allele (sd^31h) with severe pupal wing defects was used as a sd mutant (see Campbell et al., 1992). ap expression was detected utilizing the ap enhancer trap line. This line accurately reflects ap mRNA expression in embryos and late imaginal discs (Cohen et al., 1992) as a heterozygote or homozygote. sd expression was also detected with an enhancer trap insertion that accurately reflects wild-type sd expression in all tissues that have been examined (Campbell et al., 1992). Thus, analysis of β-gal expression patterns driven by either ap or sd is likely to reflect the true expression of these genes. The primary antibodies used to detect β-gal expression were either a β-gal-specific mouse monoclonal antibody (Boehringer Mannheim) or a rabbit anti β-gal polyclonal antibody. vg expression was detected utilizing a vgspecific rabbit polyclonal antibody, while wg protein expression was detected with either rat or rabbit polyclonal antibodies. ac was detected with a mouse monoclonal antibody. These primary antibodies were developed by two or three steps to either fluorescein or rhodamine conjugates utilizing standard methodology (see Skeath and Carroll, 1991 and Williams et al., 1991 for strategies).

Adult flies were allowed to lay eggs on yeasted molasses agar caps for defined time intervals (4 hour collections if precise staging was important). The entire cap was placed in a molasses agar 100 mm Petri plate (a hole to accommodate the cap was cut in the agar), given a generous amount of live yeast paste and incubated at 25°C. Abundant yeast and uncrowded conditions are essential for optimal timed stagings. Staging was performed by both total incubation time and optical examination of larval anterior spiracles and mouthhooks. Larval markers were utilized to facilitate identification of larvae of appropriate genotypes. ap, vg and wg stocks were balanced with the dominant Black cell (Bc) marker (associated with the In(2LR)Gla balancer) to facilitate identification of homozygous mutant larvae (absence of black cells throughout the larvae). The sd^31H stock was marked with yellow (y) and balanced with Bascn, which allowed mutant sd male larvae to be identified via the presence of yellow larval mouthhooks. The abdomen of selected early larvae were grasped with fine tweezers and one side of the cuticle (approximately mid-lateral) was pinched and held with a second set of tweezers. These tweezers were pulled rostrally until the larval head was cleanly removed. With practice this generated a dissected larval head that is either split longitudinally or inverted, either of which are satisfactory for antibody or X-gal staining. If the entire head is removed intact, grasping the inside of the mouthhooks facilitates simple eversion with the second set of tweezers. Eversion (or splitting) is essential for allowing antibody access, but is not necessary for X-gal staining. All dissections were performed in depression wells (containing PBS), and accumulated in a PBS-containing microtiter well. Fixation, antibody staining and washing were performed in microtiter wells as described previously (Skeath and Carroll, 1991). Many dissected larvae may be stained per well (at least twenty) without noticeable loss of antibody staining; however, extra care must be taken to recover tiny larvae lost from the baskets during washing! The larvae were mounted in 50 mm Tris (pH 8.8) containing 10% glycerol and 0.5 mg/ml p-phenylenediamine (to prevent quenching) as in Skeath and Carroll (1991). Individual discs were neither visible or individually dissected. Rather the larvae were fully inverted and the cuticle shredded away from the head with fine tungsten needles. The larvae were viewed under fluorescence on a Zeiss Axiophot; switching to bright field allowed identification of mouthhook and anterior spiracles to check staging of the larvae to which discs were still attached. Favorable disc preparations were viewed on a Zeiss/IM35 equipped with a Bio-Rad MRC600 Lasersharp Confocal system (LSCM). Optical projections of through focus sections were used, to ensure accurate representation of expression patterns. Late disc complexes were stained and photographed as in Williams et al. (1991) and Paddock et al. (1993).

Acridine orange staining was performed by the method of Spreij (1971), as modified by Masucci et al. (1990) and imaged using the LSCM. Xgal stainings and mountings of imaginal discs were performed as in Ghysen and O’Kane (1989) and viewed with Nomarski optics on a Zeiss Axiophot, using a Dage VE1000 video camera system and the work station of the confocal microscope for further image processing.

The aim of this study is to define how the wing is specified in the developing wing disc by the analysis of the developmental roles of and interactions between four genes (ap, sd, wg and vg) involved in wing formation. A detailed description of the alleles and gene probes utilized in this study is given in Materials and Methods. The phenotypes of adult flies mutant for ap, sd, vg or wg are shown in Fig. 1. Flies homozygous for severe hypomorphic or null ap alleles show a complete loss of the wing, retaining only a stump of wing hinge (or an amorphous mass of tissue; Fig. 1E). The phenotype of null vg flies is similar with a total elimination of wing structures and retention of only a small fragment of the wing hinge region (Fig. 1B). Certain genetic backgrounds can modify this phenotype and frequently produce flies that exhibit wing-to-notum transformations (Fig. 1C). Strong sd alleles are early larval or pupal lethal whereas weak alleles produce viable adults, which show strong wing reductions (Fig. 1F) essentially identical to those shown by vg hypomorphic flies. Strong wg alleles are embryonic lethal, whereas hypomorphic wg alleles exist that cause wing-to-notum transformations (Fig. 1D). This is likely to represent a true loss-of-function phenotype, since mitotic clonal analysis utilising an armadillo null allele (another member of the wg signaling pathway) also revealed wing-to-notum transformation (Peifer et al., 1991). While it is obvious that each gene is required globally for wing formation, one cannot deduce from these adult mutant phenotypes when and how each gene influences wing determination. To gain more direct information, we have examined the expression and regulation of each gene throughout the growth and development of the wing disc to arrive at a specific model of how the wing is patterned.

In order to determine the relationship between the domains of vg, ap and wg gene expression and the fate of cells expressing each product, we examined the expression pattern of each gene in the late third instar wing disc, a point when the growth of the disc is complete, but before the cellular rearrangements that occur during pupation have taken place. A fate map has been constructed for this stage (Fig. 2E; Bryant, 1978), which allows us to correlate spatial coordinates of gene expression with the fate of each subregion. Each gene expression pattern and certain overlapping domains of expression highlight distinct primordia. The vg protein (and sd, see Fig. 5B) demarcates those cells that will give rise to the wing proper (Fig. 2A: Williams et al., 1991). vg is not expressed uniformly throughout the presumptive wing region. Maximal levels of protein expression occur in a stripe of cells along the length of the future wing margin, and protein levels decrease in a graded fashion into the dorsal and ventral wing blade primordia (Fig. 2A). The vg expression in the most proximal portion of the disc (arrow, Fig. 2A) may not correspond solely to wing structures and could include some future notum structures (Fig. 2E). Unlike the rest of the presumptive wing, this region is not lost in vg (or sd) null mutant discs. It is possible that the vg expression in this region is a consequence of the manner in which vg gene expression is activated along the entire dorsal-ventral boundary of the disc (J. A. W., unpublished observations: see also Discussion).

The ap pattern is restricted to those cells whose fate is to form dorsal structures (both notum and dorsal wing; Fig. 2B). Superimposition of the vg and ap patterns and a comparison with the fate map demonstrates a striking correlation between the expression of these two genes and the fate of disc subregions (Fig. 2D,E): the red cells that express only vg will form the future ventral surface of the wing (and hinge region), the yellow cells that express both vg and ap will form the dorsal surface of the wing and the green cells that express only ap will form dorsal notum. Thus, it appears that two subdivisions of the wing disc are revealed by the domains of vg and ap gene expression alone: the wing region is highlighted by the pattern of ves tigial and the dorsal/ventral boundary by the edge of the ap pattern where it bisects the vg domain.

The wg pattern also highlights the dorsal/ventral boundary, as a brilliant stripe of wg expression runs along the presumptive wing margin where the future dorsal and ventral surfaces of the wing meet (Fig. 2C; and Baker, 1988b). As we shall see later, however, this late pattern of wg is misleading with respect to the time and place where wg has its most dramatic impact on wing formation.

The vg and sd genes are expressed in identical patterns (Fig. 5A,B) and label the presumptive wing region of the imaginal disc (Fig. 2A). These expression patterns and the complete removal of the wing observed in vg and sd mutant discs (see below) suggest that both genes encode components essential for formation of the wing region of this disc. These figures also demonstrate that the wing region of the disc exists as a domain of gene expression in the late larval disc.

ap is expressed in all cells of the dorsal region of the disc (Fig. 2B). Double labeling with achaete (which marks bristle precursor cells that flank the wing margin) indicates ap expression extends precisely to the wing margin (data not shown), and therefore extends to the border of the putative dorsal compartment of the disc and is excluded from cells of the ventral compartment. Mitotic clonal analysis will have to be utilized rigorously to determine if the border of ap expression is exactly coincident with the entire length of the D/V boundary in a cell-by-cell manner. Since ap is not expressed in the ventral wing pouch, it is not clear why this region is lost along with the rest of the wing in ap mutants, since this region is lost from the disc as early as the third instar (Fig. 5E).

The late wg expression pattern is highlighted by a stripe along the presumptive wing margin, and a patch of expression in the presumptive notum region (Fig. 2C). Although wg clearly prefigures the future wing margin, the global wing-notum transformations seen in wg mutant discs are difficult to reconcile with the observed limited pattern of expression in late third instar discs.

As we show below, a more complete picture of wing formation comes from understanding the dynamics of gene expression and the regulatory interactions between the genes. In the case of wingless, the wing-to-notum phenotype can better be explained by the pattern of wingless expression in very early discs and by the effects of mutations in wingless on the expression of the other three wingdetermining genes.

vg and sd mark the developing wing in early discs Since the vg protein is expressed in the embryonic primordia of the wing disc (Williams et al., 1991), it is possible that early larval discs also express vg. Indeed, it turns out that staining with the vg antibody, which in embryos and third instar larvae specifically stains wing and haltere discs, has allowed us to identify wing discs in as early as newly molted second instar larvae (data not shown, and Fig. 3). This has greatly facilitated the identification of the early disc expression patterns of ap, sd and wg, using a double fluorescence labeling strategy to identify the otherwise nearly invisible young wing discs. From correlation of the vg patterns with those of the other regulatory genes, we have determined that various subregions and boundaries are specified progressively and depend upon interactions among some of the four regulatory genes.

The mature vg expression pattern develops in roughly two stages. We first detect low level ubiquitous expression in early second instar discs at the time when the discs contain approximately 100 cells. The expression eventually becomes elevated at approximately the early-mid second instar (Fig. 3A; disc size is less than 200 cells when elevation of vg expression is first detected) to give strong staining throughout the disc. This expression is then repressed distally through the remainder of the second instar (Fig. 3D) and finally resolves to form a well-defined stripe by the early third instar (Fig. 3G). This stripe clearly is non-uniform and drops off in intensity both proximally and distally (Fig. 3G), similar to the graded distribution seen in late third instar discs (Fig. 2A, 5A). Double labeling with antibody to vg and the sd enhancer trap indicates that sd-driven lacZ expression closely matches vg pattern in early discs (data not shown). The sd pattern resolves into a stripe more slowly than vg, but this probably reflects the slow kinetics of β-galactosidase degradation rather than differential regulation of the two genes. Thus, sd and vg expression appears identical, in both early and late imaginal discs. However, double labeling experiments with a sd-specific antibody (when available) will be necessary to determine if the sd enhancer trap does accurately reflect sd gene expression in early discs. Examination of vg expression throughout the third instar, and later, indicates that the graded vg stripe (Fig. 3G) largely corresponds to the early wing region of the disc. This suggests that the wing region exists as a specific domain of gene expression by the early third instar discs, but is not detected as a clearly localized domain at earlier stages (as assayed by sd and vg expression).

ap expression is not detected in early second instar discs, but is activated in its apparently mature pattern in earlymid second instar discs (Figs 3B, 4E). This initial activation of ap is likely to reflect accurately the initial pattern of ap expression, and certainly cannot be due to perdurance of β-galactosidase. While the number of cells subsequently increases enormously in the wing disc, the ap pattern remains relatively unaltered with respect to the ventral notum and wing boundaries throughout later larval development (Fig. 3E,H). The early expression of ap in a dorsally restricted pattern indicates that a D/V boundary exists in early second instar discs, even before the developing disc field contains 200 cells. Interestingly, ap expression never outlines the developing wing (as revealed by vg and sd expression), which implies that the loss of the wing in ap mutants is caused by indirect effects on cells not expressing ap. The expression pattern indicates that ap specifies a dorsal-specific component which is indirectly necessary for wing formation. Comparison of ap and vg expression (Fig. 3I) demonstrates that the early third instar vg stripe (i.e. the early wing region) and the resolving vg stripe in the second instar (Fig. 3F) is centered precisely on the D/V boundary, which is also the border of ap expression. This is the same spatial relationship between ap and vg that is observed in late discs.

The wg expression pattern was assayed in both a wg enhancer trap line and with a wg-specific antibody. Both patterns of expression are essentially identical in late discs (Fig. 4A,B) except for the slightly broader domain of wg protein expression, which is expected since wg encodes a secreted product (Van den Heuvel et al., 1989). The spatial coordinates of wg enhancer trap expression in the second instar are radically different from the later pattern, and is localized ventrally throughout the second instar (Fig. 4C,D). Expression is first detected in a cap comprising less than 10 cells in early-mid second instar discs (Fig. 4C). This is approximately the time vg expression is elevated and ap expression is activated (Fig. 4E). The wg-specific antibody reveals that protein expression is regulated more dynamically than lacZ expression. The sequential development of the late wg protein pattern from the early ventral cap is shown in Fig. 4F,I,L (and with respect to ap expression in Fig. 4E,H,K). In mid-early third instar larvae, the wg expression pattern is similar to the late pattern (compare Fig. 4B and L); indeed, the future wg margin stripe is first apparent at this stage. wg expression in the early disc roughly complements that of ap, and is highest in the ventral region of the disc (Fig. 4G,J,M). This may indicate a requirement for wg in the ventral region of the disc, and suggests that both ap and wg may be required for D/V polarity in the early disc. As is the case with ap, at none of the stages examined does wg prefigure the developing wing, which argues that wg does not directly specify the wing region of the disc and that loss of the entire wing in wg mutants may be a secondary effect of its requirement in establishing D/V polarity. As we shall see below, the dynamics of wg expression are directly related to the progressive specification of restrictions within the wing.

In order to approximate the time of action of the sd, vg, ap and wg genes, the morphology of imaginal discs from each of these mutants was examined. By the late third instar, wing discs from all of the mutants are morphologically abnormal and exhibit a complete loss of the wing region (Fig. 5C-F). In order to examine earlier stages, the ap enhancer trap was introduced into sd, vg and wg mutant backgrounds and the morphology, ap expression patterns and regions of cell death in second and early third instar discs were viewed by Nomarski images of the X-gal-stained tissues and by staining with acridine orange. Second instar and early third instar vg and sd discs appear morphologically normal; the first detectable defect in these mutants is a jaggedness at the border of ap expression (the future wing margin) which is observed in late second instar discs and throughout the third instar (data not shown). In contrast, wg and ap discs are clearly smaller and morphologically abnormal in the late second instar, significantly earlier than sd or vg morphological defects become apparent (Fig. 6A-D). In addition, ap expression is abnormal in ap and wg wing discs as early as the middle second instar (Fig. 6C,D and see below). These observations indicate that ap and wg may be required earlier than sd and vg.

The timing of cell death in mutant discs also suggests an early requirement for wg and ap. Previous analyses of cell death in vg, sd and wg mutant wing discs has revealed the presence of cell death in the wing region of vg and sd discs throughout third instar development, but no cell death was detected in wg mutant discs (James and Bryant, 1981). In wild-type discs, sd mutants and vg mutants, we observed very little cell death in the wing disc during second instar development until a ‘burst’ of cell death occurs at the second to third instar molt. Thus, no significantly increased levels of cell death are detected in vg and sd discs, relative to wild type, until the third instar. In wg discs, no cell death is detected throughout the second instar; however, elevated cell death is detected at the edges of the presumptive wing region at the second to third instar molt (Fig. 6B acridinestained cells; and Fig. 6D, blebbing cells above the focal plane of ap expression). Cell death is not increased in the central wing region of wg mutant discs throughout second or early third instar development (or later; James and Bryant, 1981). Thus, the abnormal morphology in newly hatched third instar wg discs cannot be due to cell death, and must be the result of some other patterning defect. In contrast, extensive cell death was observed throughout the second half of the second instar in ap mutant discs (Fig. 6A,C) and is clearly visible even in ventral cells that do not express ap (Fig. 6C, the blebbing cells outside the region stained with B-gal). In addition, the D/V border is jagged in late second instar ap mutant discs, perhaps the result of the earlier cell death that we have observed. Since ap discs undergo increased levels of cell death in the second instar and ap and wg discs are clearly morphologically abnormal, well before vg or sd mutant defects (or cell death) become apparent, it is likely that wg and ap are required and act before the vg and sd genes.

In order to determine whether the temporal and spatial relationships between regulatory gene expression and the morphological abnormalities in discs and adult wings reflected a temporal and spatial hierarchy of interactions between these genes, we examined the expression of each gene in mutant discs of each genotype. We showed earlier that the wg expression pattern in the second instar wing disc roughly opposes the ap pattern (Fig. 4) and thus may help to define the ventral region of the disc and/or to set the position of ap expression and the D/V boundary. Indeed, we find that ap expression expands into the ventral region in wg mutant discs (Fig. 7B,D). Thus, wg appears to be required early to set the limit of ap expression and the location of the D/V boundary properly, which also expands ventrally in wg mutants. The variability of the observed expansion of ap expression is probably due to the fact that the wg alleles are hypomorphic; in fact only reduced levels of wg protein are detected in these wg mutant discs (data not shown) indicating that a complete removal of wg expression could result in more extreme defects.

Proper expression of vg protein requires both wg and ap function and the dynamics of vg expression are similar in early discs of both wg and ap mutants. In the second instar, the early elevated ubiquitous phase of vg expression is normal, and vg is initially restricted from the distal tip of the disc. However, rather than becoming restricted to a stripe defining the presumptive wing in the early third instar, vg expression is lost in both mutants (Fig. 7A,C). Thus, both ap and wg are required for retention of vg expression flanking the D/V boundary and for formation of the wing region.

It is important to stress that the effect of wg on ap or vg expression cannot be due to cell death. This is because, as indicated above, cell death is only detected in wg discs at the second to third instar molt and is limited to the lateral edges of the disc (Fig. 6B,D). In addition, since no cell death has been observed in third instar wg discs (data not shown and James and Bryant, 1981), the observed expansion of apterous expression in the central region of the disc (or loss of vg expression from this region) cannot be due to cell death. Since the notum is a dorsal derivative of the disc, the observed expansion of the dorsal region in wg mutant discs may account for the wing-to-notum transformations seen in wg adults.

The vg and sd wing expression patterns are essentially identical during all of the stages of wing development that we examined. The possibility of cross-regulation between these genes was investigated by examination of vg and sd expression in sd and vg mutants, respectively. Although low level vg expression is retained, all elevated vg expression is eliminated in both early and late sd discs (Fig. 5D, and legend). Since embryonic vg expression is unaltered in this mutant, and in another strong early larval lethal sd allele (sd^47m; data not shown), it appears that the observed dependence of vg expression upon sd is a wing-specific requirement. The vg and sd embryonic expression patterns are quite different (see Campbell et al., 1992 and Williams et al., 1991), so sd appears to be a specific activator of vg expression only in the wing disc. Conversely, sd expression is not detected in second or early third instar discs from a vg null mutant. Since early sd expression is low (even in wild-type discs) and the vg mutant discs cannot be easily identified (double labeling with vg was not possible), it is difficult to determine if early sd expression is just reduced or totally absent. However, late sd wing disc expression is clearly eliminated in a vg null mutant (data not shown). These results argue that sd and vg may be spatially regu-lated in a similar manner and probably act to maintain the expression of each other in the developing wing disc.

We have examined the developmental roles of and interactions between four genes that are globally required for wing formation. By determining their temporal and spatial patterns of expression throughout the growth and development of the wing disc, correlating domains of gene expression with the wing fate map, identifying the stage when mutant discs first exhibit defects and demonstrating the dependence of some genes on the proper function of others, an initial picture of the genetic hierarchy guiding the subdivision of the wing disc has emerged. In addition, we have demonstrated two key and surprising functions of the wg protein in establishing dorsal/ventral polarity and distinguishing the presumptive wing from notum. By opposing ap expression, wg acts to distinguish the ventral from dorsal regions of the wing surface. And, as a positive regulator of vg and sd expression, wg promotes the development of the wing from the surrounding notum.

Genetic analysis has identified only a few genes that are globally and specifically required for wing development (Lindsley and Zimm, 1992). It is likely that additional genes exist that affect the global development of the wing; however, the paucity of adult viable alleles of essential genes and the difficulty of performing systematic mitotic clonal analyses have thus far prevented their identification. Of the four genes that we have concentrated on, two (vg and sd) have striking expression patterns that demarcate the wing region of the mature wing disc, and each gene is required to maintain the expression of the other. Their expression in early third instar discs indicates that the wing region exists as a specific domain of vg and sd gene expression at this time and is subsequently lost completely in vg or sd mutants. These results indicate that these two genes are ‘pro-wing’ genes, required specifically to form the wing region of the disc. The identification of sd as a putative transcription factor (TEA domain protein; Campbell et al., 1992) and vg as a novel nuclear protein (Williams et al., 1991) is consistent with the idea that these genes are transcription control genes required for the differentiation and morphogenesis of the wing region of the wing disc in the third larval instar. Loss of either gene leads to extensive cell death and loss of the wing region of the disc during third instar development (James and Bryant, 1981), presumably due to an inability to pattern this region properly.

Inspection of the overlapping vg (or sd) and ap late third instar expression patterns clearly demonstrates that these two genes differentially mark the dorsal and ventral surfaces of the wing; vg expression specifically labels the ventral wing surface while vg and ap expression mark the dorsal wing blade (Fig. 2). This raises the intriguing possibility that these genes regulate dorsal/ventral identity in the wing by controlling the expression of structural genes in cells of each wing surface. Certain position-specific (PS) proteins are known that are expressed in either the dorsal (PS1) or ventral (PS2) wing surfaces (Brower et al., 1984). It is possible that genes such as those encoding PS1 or PS2 are differentially regulated by vg and ap.

An interesting question is ‘does the wing exist as a domain of gene expression before the resolution of vg and sd expression? If a more primary tissue-specific pro-wing gene than sd or vg exists, mutational analysis has yet to identify it. Indeed, both ap and wg alter the disc earlier than vg or sd, but this appears to be the consequence of D/V polarity defects rather than reflecting earlier specification of the wing region. It seems likely that accumulation of vg protein accurately reflects the initial specification of the wing, and thus the wing is first identifiable as a domain of gene expression in the early third instar. The graded expression of vg protein (Figs 3, 4) may be required to specify coordinates within the wing, with the margin region expressing maximal vg protein, and the interior of the wing expressing lower levels. The vg expression patterns can be correlated with the fact that hypomorphic vg alleles progressively delete wing tissue from the margin, implying that higher levels of vg are required to make the wing margin than are required to make the wing blade (Williams and Bell, 1988). Analysis of the effects of ectopic expression of vg (and sd) in the wing blade should help to address this possibility.

The D/V compartmental boundary, which is drawn along the wing margin, separates the dorsal and ventral wing surfaces (Brower, 1985). This boundary is first clonally detected during early larval life in first or second instar development (Garcia-Bellido et al., 1976). However, the presence of a zone of non-proliferating cells along the developing wing margin throughout the third larval instar has led to considerable debate regarding whether the detected clonal restriction is a compartment boundary or simply a region of reduced cell proliferation (Brower, 1985), and speculation that the actual in vivo timing of the formation of mitotic clones in the clonal analysis studies is considerably later than initially assumed. The visualization of this boundary in the early second instar via the ap expression pattern argues that it is a true compartmental restriction, since ap is compartmentally localized well before the zone of non-proliferation is detected (O’Brochta and Bryant, 1985). In addition, the timing of the onset of ap expression in the early-mid second instar disc correlates well with initial predictions from clonal analysis about the timing of the restriction, arguing that ap may mark the dorsal compartment. Although further analysis is required to confirm that the D/V boundary is a true clonal restriction, it is clear that the dorsal compartment exists as a domain of gene expression in mid-early second instar discs (before the disc size is 200 cells) and may be considered a compartment in this context (Brower, 1985). Analysis of ap and wg mutants indicates that proper compartmentalization is essential for subsequent disc patterning. Loss of either ap or wg is associated with D/V pattern abnormalities in late second instar discs; the D/V boundary is irregular in both ap and wg mutants, and is expanded ventrally in wg mutants. These results and the early ap and wg expression patterns indicate that ap and wg are primarily required for establishment of the D/V boundary; wg appears to help set the proper location of this boundary and may provide an important ventral region component, while ap presumably provides an essential dorsal-specific component. It is clear that wg cannot affect ap expression directly since wg is an extracellular component of a signalling pathway. Indeed, since wing-to-notum transformations have been observed in clones of cells mutant for other identified members of this signalling pathway (armadillo, dishevelled and porcupine; see Peifer and Bejsovec, 1992), it is likely that some downstream target of the wg signal regulates ap directly.

In the early third instar larva, vg and sd are expressed at discernably higher levels in stripes that are centered along the entire length of the D/V boundary (Fig. 3G). Earlier vg expression is ubiquitous (in the wing disc) and is progressively restricted to this stripe via repression of vg expression in the remainder of the disc (compare Fig. 3D,G). This is consistent with a model that the D/V restriction is a necessary prerequisite to wing formation, and that the wing is ‘specified’ from the D/V boundary. The loss of vg expression in ap and wg mutants which show D/V defects is also consistent with this model, and argues that wing formation is a sequential event, requiring the establishment of the D/V boundary (ap and wg functions) and subsequent wing specification (vg and sd functions). Indeed, a vg region control element (which is essential for normal wing patterning) has been identified that directs reporter gene expression specifically in a strip of cells directly spanning the D/V boundary in third instar wing and haltere discs, but not in the remainder of the wing region of these discs. This expression is independent of vg or sd function (J. A. W., unpublished results) and implies that this regulatory element interprets information along the D/V boundary of developing wing and haltere discs. These recent results may also explain why small regions of potentially notal identity express vg (see arrow, Fig. 2A); since these cells flank the base of the D/V boundary, they presumably reflect the fact that vg is initially activated along the entire length of the D/V boundary, not all of which may form wing tissue. The mechanism by which vg and sd (or the vg control element) expression is retained in cells flanking the D/V boundary is unknown; however, the analysis of this process may yield valuable information about how pattern may be generated by local discontinuities between adjacent cell populations.

Following mechanical damage or genetically induced cell death, discs or disc fragments undergo epimorphic pattern regulation. This regulation may duplicate existing struc-tures, or regenerate missing structures by intercalation, depending on the particular fragment that undergoes regeneration. Analysis of the regenerative capacity of a variety of wing disc fragments and analogous results with other systems has led to a number of loose rules of epimorphic regulation (French et al., 1976), and most models of secondary field patterning attempt to predict underlying pattern systems that could account for the observed regulative behavior of different disc fragments. Several such models predict multiple radial and angular positional determinants to exist within the field, e.g. the polar coordinate model (French et al., 1976, Bryant et al., 1981). However, these models require several rules to explain certain regulatory events, such as the full circle rule of distal regeneration. Moreover, candidate positional morphogens have not been clearly identified. In addition, the observation that distal regeneration in wing discs requires that the removed region spans the A/P boundary (Schubiger and Schubiger, 1978) is difficult to reconcile in terms of these models, which predict multiple angular determinants. In general, these models work well to explain the regulatory behavior of mature fields, but do not clearly predict the underlying mechanisms through which a mature pattern is set up.

Meinhart (1982) and Karlsson (1984) have proposed models that represent feasible molecular mechanisms by which pattern may be sequentially assigned to developing fields of cells. These models predict that cell determination boundaries (e.g. compartment boundaries) act as organizing regions for secondary fields. They propose that pattern is highest along compartment boundaries. Distal regeneration requires the presence of compartment boundaries in the healing tissue, while pattern duplications occur when the damaged tissue does not contain compartment boundaries. These models predict that normal proximal-distal patterning within the disc would also require these compartmental restrictions. The expression patterns of two wing disc patterning genes (dpp; Masucci et al., 1990 and patched; Phillips et al., 1990) define stripes, which are aligned with the A/P boundary; presumably the A/P restriction is required at some stage for establishment of these expression patterns. We have shown the presence of an early D/V boundary at least in terms of a domain of gene expression, which appears to set up the specific wing patterns of sd and vg. These results are consistent with the idea that pattern may be generated from compartment boundaries, and that both A/P and D/V oriented patterns are necessary for normal proximal-distal patterning. For example, the absence of either dpp (A/P oriented), sd or vg (D/V oriented) result in phenotypically similar losses of distal wing structures.

Larval wg function is not restricted to the wing disc and is required for normal patterning in most imaginal discs (Baker, 1988b). We did note in the course of these studies that the halteres, the other Drosophila flight appendages, appear to be patterned in the same way as the wing (Fig. 3G-I), which is not too surprising since mutations in all four genes also affect the halteres and this structure is a developmental homolog of the wing. We also note that the wingless transcript is localized to a small apically localized quadrant (in the anterior-ventral region; Baker, 1988b) of late third instar leg discs; although no relationship between this pattern and the late wing disc pattern is obvious, there is a striking similarity to the wingless expression pattern in the second instar wing disc. Thus, it is possible that the wg function in the leg disc is analogous to the early wing disc function, and may be required to define the ventral region of the leg discs. Indeed a D/V clonal restriction has been detected in the leg discs. The restriction is established in the second instar, and has been speculated to be analogous to the wing disc D/V restriction (Steiner, 1976). This restriction is only clearly identified in the large anterior compartment of the leg, and the resulting ventral region is the zone in which wg is maximally expressed. Loss of distal structures is observed in adult legs from some wg mutants (Baker, 1988b; J. A. W., unpublished observations); this is analogous to the loss of distal (wing) structures from the wing disc. Interestingly, loss of wg function in the leg results in removal of ventral parts of the pattern and their replacement with mirror-image duplications of dorsal elements (Baker, 1988b). This is consistent with the idea that the function of wg is to specify the ventral region of the leg disc, and that the dorsal domain of the leg may expand in a wg mutant. Thus, wg function in the leg and early wing may be analogous, and involved in D/V compartmentalization and ventral specification in both patterning fields. It is startling that wg (and other members of the Wnt gene family) are involved in patterning so many structures within both vertebrate (Nusse and Varmus, 1992) and invertebrate organisms. The observation that the Wnt-3 and Wnt-5A genes are expressed in the developing mouse limb bud (Nusse and Varmus, 1992) may imply that the Wnt family is also required to pattern vertebrate limbs, perhaps in a manner analogous to its patterning function in the Drosophila wing and leg.

We thank Steve Cohen for providing ap enhancer trap and ap null stocks prior to publication, and Shelagh Campbell for sd enhancer trap flies. All other sd alleles were kindly provided by Arthur Chovnick. We appreciate the gifts of both rabbit anti-wgspecific antibody (provided by Roel Nusse), and rat anti-wgspecific antibody (provided by Alphonso Martinez-Arias). We especially thank Allen Laughon, Teresa Orenic, Lisa Nagy and Seth Blair for critical reviews of the manuscript, and Leanne Olds and Jamie Wilson for help with figure and text preparation. This work was supported by a Medical Research Council postdoctoral fellowship to J. A. W. and the Howard Hughes Medical Institute.