The control of cell fate along the dorsal-ventral axis of the Drosophila embryo

In Drosophila, embryonic pattern formation depends on contributions from both maternal and zygotic genes. Aspects of this pattern are elaborated along two embryonic axes, the anterior-posterior (A/P) and the dorsal-ventral (D/V). The establishment of the A/P pattern requires three sets of maternal genes, the anterior, posterior, and terminal systems. Each of these systems is responsible for a particular aspect of the body plan: the anterior for the head and thorax (Frohnhöfer and Nüsslein-Volhard, 1986,1987), the posterior for the abdomen (Lehmann and Nüsslein-Volhard, 1986, 1987, and the terminal for the unsegmented acron and telson (Schiipbach and Wieschaus, 1986; Klingler et al. 1988. By contrast, the D/V axis requires only one set of genes, the dorsal-ventral system, which includes the eleven genes of the dorsal group and cactus (cact) (Anderson, 1987; Roth etal. 1989; Rushlow and Arora, 1990). The molecular asymmetries that define the two axes are set down by the maternal genes in the form of gradients. The A/P axis is organized in part by a gradient of bicoid protein (Driever and Nüsslein- Volhard, 1988), and the D/V axis by a gradient of differentially compartmentalized dorsal (dl) protein (Steward etal. 1988; Rushlow etal. 1989; Steward, 1989; Roth et al. 1989). The two axes are established independently of one another, as mutations that disrupt one axis do not significantly affect the other (Nüsslein- Volhard, 1979).

The overall pattern along the A/P axis is generated by the hierarchical activities of several classes of genes that sequentially subdivide the embryo into smaller units. The maternal coordinate genes act to establish broad overlapping domains of zygotic gap gene expression. These domains, in turn, interact to specify the double segment pattern of the pair-rule genes. The interactions between pair-rule genes lead to the single segment pattern which is then further subdivided by the expression of segment polarity genes.(for reviews, see Akam, 1987; Ingham, 1988).

By contrast, the hierarchy of genes involved in D/V patterning has not been entirely established (see Ingham, 1988). While there is ample evidence demonstrating that the maternal coordinate gene for the D/V axis is dl (Steward et al. 1988; Rushlow et al. 1989; Steward, 1989; Roth et al. 1989), it is not yet clear which zygotic genes are direct targets of dl function. Mutations in several zygotic loci affecting D/V pattern have been identified that may be such targets (Nüsslein- Volhard et al. 1984; Wieschaus et al. 1984; Jurgens et al. 1984), and in general, these loci affect one of three D/V pattern elements (Fig. 1): dorsal ectoderm (Irish and Gelbart, 1987; Rushlow et al. 19876), ventral ectoderm (Mayer and Nüsslein-Volhard, 1988), and mesoderm (Simpson, 1983).

Four of these zygotic D/V patterning genes, dpp, zen, twi, and sna, have been studied in detail, and several lines of evidence suggest that these genes may be regulated by dl. In particular, it has been shown that the distribution of zen and twi protein in the early embryo responds to the state of the dl gradient. In the wild-type embryo, the distribution of twi and zen protein defines a simple pattern of three regions, zen is expressed dorsally, neither of the genes are expressed laterally, and twi is expressed ventrally (Rushlow et al. 1987b; Thisse et al. 1987; Roth et al. 1989). Genetic manipulation of the dl gradient results in shifts in the expression of these two genes: dorsalizing mutations result in a loss of twi expression and a corresponding expansion of zen expression, lateralizing mutations result in a loss of both twi and zen expression, and ventralizing mutations result in a loss of zen expression and a corresponding expansion of twi expression (Rushlow et al. 1987a; Roth et al. 1989). Furthermore, recent molecular studies have established that the dl protein has the potential to bind to specific sites in regulatory sequences upstream of the zen promoter that mediate the ventral repression of zen (Doyle et al. 1989; Ip et al. 1991). As dpp and sna are expressed in domains similar to that of zen and twi respectively, it seems reasonable to presume that these genes are also targets of dl.

We have analyzed the distribution of dpp, zen, twi and sna transcripts in whole-mount embryos mutant for a representative sample of maternal and zygotic genes affecting both D/V and A/P pattern so as to elucidate aspects of the regulatory hierarchy involved in D/V pattern formation. Consistent with the hypothesis that dpp and sna are regulated by dl, their expression patterns reflect shifts in the dl gradient. Furthermore, the shifts observed in the dpp and sna patterns are precisely those observed for twi and zen suggesting that these genes are responding to the same thresholds of dl activity. Thus, the dl gradient appears to subdivide the embryo into only three domains. Polar expression of dpp, zen, twi, and sna requires the function of the genes of the terminal system. These cues appear to act in conjunction with the cues from the dorsal-ventral system to specify zygotic gene expression at the termini.

After the initial patterns of zygotic gene expression have been established, refinement and maintenance of these patterns depends on zygotic gene activity. These events are specific to each of the three domains specified by the dl gradient. Thus, zygotic ventralizing mutations may affect the refinement and maintenance of other zygotic ventralizing genes, but not of zygotic dorsalizing genes and vice versa. In the dorsal ectoderm, interactions between dpp and zen suggest that subdivision of this domain into dorsal epidermis and amnioserosa depends on the refinement of the zen pattern by dpp and the other ventralizing genes. In contrast, the expression of dpp is unaffected by mutations in these genes. Similarly, the sna pattern depends on twi for normal expression, but the reciprocal effect is not observed. Thus, we propose that dpp and twi are likely to be the primary patterning genes for dorsal ectoderm and mesoderm, respectively.

The alleles, or combinations thereof, used to represent the various mutant genotypes were as follows: maternal ventralizing: Tl^l0b, Tl^9Q. ea¹/ea², cact^A2; maternal lateralizing: ea^5.13, Tl^rm9/Tl^rm10; maternal dorsalizing: snk⁰⁷³/snk²²⁹, tub’¹⁸/tub²³⁸, Tl^reQ/Df(3)Tl, dl¹; terminal: tor^PM, trk^RA; zygotic ventralizing: zen^w36, tld^10E, dpp^H37, dpp^H46, tsg^YB56, srw^B18, sew⁵¹², sog^Y506; zygotic dorsalizing: twi^ID96, Df(2R)S60, twi, sna^11GOS. In all cases we have chosen the strongest allele available, which are known to be nulls in the cases of: ea¹/ea², Tl^reQ/Df(3)Tl, dl¹, tor^PM51, Df(2R)S60, twi. All stocks were obtained from the Tübingen Stock Center.

Mutant embryos in most zygotic genotypes were not distinguishable from wild-type siblings at cellular blastoderm, so the analysis of the expression patterns at these stages was done by examining a large number of stained embryos and looking for consistent aberrations that were observable in 25% of the embryos. In some cases, mutant embryos were identified with the aid of marked balancer chromosomes. All staging was done according to Campos-Ortega and Harten- stein (1985).

Flies were cultured on standard Drosophila cornmeal yeast extract sucrose medium in 25mm × 95mm shell vials or quarter pint urine specimen bottles. Crosses were reared at 25 °C. Embryo collections were done at 25 °C, in inverted 150 ml tripour beakers covered with a 60 mm plastic Petri dish filled with apple juice or grape agar. Cuticle preparations were performed according to standard techniques (Wieschaus and Nüsslein-Volhard, 1986).

Immunological staining of whole-mount embryos was carried out as described by Macdonald and Struhl (1986) using the Avidin/Biotin ABC System (Vector). Antibodies directed against the dl protein and the Kr protein were provided by C. Rushlow. Stained embryos were mounted in methyl salicylate for photography.

Plasmids used as probes for dpp, zen and twi are: pBEhl, a 4.5 kb dpp cDNA in pNB40 (St Johnston et al. 1990); pS60–7, a 1.4 kb cDNA fragment of the zl zen transcript in pGEM-1 (Rushlow et al. 1987a); for twi, a 530 bp genomic EcoRl- BamHl fragment cloned into pSP65 was used (Thisse et al. 1987). The probe for sna was generated in a PCR (polymerase chain reaction) using primers flanking the DNA binding domain. The template for the reaction was a complete sna cDNA isolated from a 4–8h cDNA library (R. Ray and N. Brown, unpublished).

Plasmids were cut with appropriate restriction enzymes to isolate fragments containing only insert DNA, and the digestions were electrophoresed in 0.5 % Agarose gels in TBE buffer. Expected cDNA fragments were electroeluted into dialysis tubing, and the eluate was run over a NACS Pre-Pak DEAE column (BRL), then precipitated with ethanol. The pellet was resuspended in distilled water to a concentration of 15ngμl^-1, and, in general, 2μl were used per oligolabelling reaction.

Gel purified cDNA fragments were labelled with digoxigenin according to the protocol accompanying the Kit (Boehringer Mannheim, Cat. 1093–657), using 30 ng of template DNA in an overnight reaction at room temperature. After incubation, the probe was run over a standard G-50 spin column equilibrated with water, and lyophylized to dryness. The pellet was resuspended in 10μl distilled water, boiled for five minutes, and 2μl of this probe solution was used for each hybridization.

Hybridizations were done essentially as described by Tautz and Pfeifle (Tautz and Pfeifle, 1989), with some minor modifications. Stained embryos were mounted in Permount (Fisher) or in 90 % glycerol for photography.

In this report, we distinguish between the different phases of expression of the four zygotic genes. We use the term initial pattern to refer to the pattern that is evident prior to nuclear cycle 13 for dpp, zen, and twi, and nuclear cycle 12 for sna. After this initial phase, each of these patterns sharpens into a refined pattern which is a derivative of the initial pattern. For dpp and sna, we will use the term late patterns to refer to novel expression patterns that appear after the refined patterns. Although the wild-type expression patterns for zen, dpp and twi have been previously described (St. Johnston and Gelbart, 1987; Rushlow et al. 1987a;Thisse et al. 1987), we include a brief description of them for comparison with the mutant patterns.

Prior to nuclear cycle 14, dpp and zen are both expressed in the dorsalmost 40% of the blastoderm embryo with transcripts extending around both the anterior and posterior poles to label more ventral cells (Fig. 1; Fig. 2A,E; St. Johnson and Gelbart, 1987; Rushlow et al. 1987a). Expression at the anterior pole is consistently less intense than that at the posterior pole for both genes (Fig. 2E). Despite this similarity in initial pattern, the refined patterns of dpp and zen are quite different, zen expression refines during cellulariz- ation to occupy only the dorsalmost 10% of the egg circumference in a narrow band 5–6 cells wide, the mohawk, and two patches of expression anterior to the cephalic fold, the head spots (Fig, 2F; Rushlow et al. 1987a, 1987b). This refined pattern fades during fast germ band extension, and no further expression of zen is observed.

About the time zen expression is fading, the dpp pattern refines. As the germ band extends, dpp transcripts are progressively excluded from the dorsally located amnioserosa, and by full germ band extension are located exclusively in the dorsal epidermis (Fig. 1; Fig. 2B). This pattern is maintained until midway through slow germ band extension, when it is succeeded by the late pattern which consists of two stripes of spots running parallel to the A/P axis, the first of which appears dorsally, followed by the second which appears laterally (Fig. 2C). This pattern is maintained throughout stages 10 – 14, though during late germ band extension and germ band retraction, these stripes become continuous lines of expression, elaborated with segmental modulations (Fig. 2D).

Transcripts of twi and sna are first detected during nuclear cycle 11 – 12 in a single continuous stripe, comprising the ventral most 20% of the embryo, that extends up to and around both poles (Fig. 1; Fig. 2G, and 21, respectively; Thisse et al. 1987, 1988). Shortly after this initial pattern appears, sna transcripts are excluded from the polar regions, particularly from the posterior pole (Fig. 2J; this refinement is complete by nuclear cycle 13). The boundaries of the sna pattern are sharply delineated and a distinct sinus appears at the site of the presumptive cephalic furrow in late stage 4 embryos (see Fig, 5A). Cells expressing sna at this stage invaginate with the ventral furrow, but this expression in the presumptive mesoderm is lost after fast germ band extension (stage 8). However, as expression is lost in the mesoderm, sna transcripts are observed to accumulate in subsets of cells in the ventral ectoderm and in cells of the procephalic neurogenic ectoderm (Fig. 2K). Based on their segregation pattern, these cells appear to be neuroblasts (Campos-Ortega and Hartenstein, 1985). Later, during germ band retraction, this pattern is supplanted by a new round of expression in the dorsal ectoderm consisting of a single row of segmentally repeated spots from T2 to A8, the thoracic spots being collinear with the lateral dpp stripe, described above, and the abdominal spots being slightly more dorsal. The thoracic spots persist until well after dorsal closure (data not shown).

The initial pattern of twi expression becomes more restricted at the posterior pole from a position just above the pole cells in stage 4 embryos to a position just beneath them in stage 5 embryos (Fig. 2H; Thisse et al. 1988). Unlike the sna expression pattern, the boundaries of the twi domain are less sharp; cells at the edges express lower levels of transcript, relative to those along the ventral midline, suggesting that the twi pattern tapers off in a graded fashion. Most of the twi cells invaginate with the ventral furrow during stage 6, and twi expression persists in the mesodermal derivatives until late in embryogenesis (Thisse et al. 1988).

In the following sections, information on mutant phenotypes precedes the description of expression patterns. Cuticular phenotypes engendered by representative maternal loci are shown in Fig. 3. We will use the term mutant embryos to refer to embryos derived from mothers mutant for a particular locus. AU staging is according to Campos-Ortega and Hartenstein (1985). In general, specific alleles will not be referred to in the text, but all genotypes tested in this study are listed in the Materials and methods.

Ventralized embryos are produced by mothers bearing certain alleles of three maternally acting genes: Toll (Tl), easier (ea), and cact. This phenotype is characteristic of loss-of-function alleles of cact and select dominant gain-of-function alleles of Tl and ea. Unlike the cuticle of wild-type embryos, which consists of ventral denticle belts and dorsal hairs (Fig. 3A), ventralized mutant embryos (that produce cuticle) have rings or patches of ventral denticles along the entire D/V axis (Fig. 3C; Anderson et al. 1985a). Strongly ventralized embryos do not lay down cuticle (Fig. 3B).

Predicted fate shifts in the mesodermal anlage vary depending on the particular allele. The dominant allele Tl^Ob shows the most severe ventralization, in which all blastomeres behave as mesodermal cells (Fig. 3B; Leptin and Grunewald, 1990). A weaker fate shift is observed in cact mutant embryos, which show expansion of the mesoderm at the expense of the dorsal ectoderm (Fig. 3C; Roth et al. 1989). Phenotypic analysis of the dominant ventralizing allele Tl^9Q clearly shows a global shift primarily in the ectoderm (Anderson et al. 1985a).

Expression of dpp and zen in ventralized embryos is initiated at the normal time, but the pattern is altered. Transcripts for both genes accumulate at the termini, but not over the dorsal surface (Fig, 4A,B,E,F). zen expression at the anterior pole is transient and appears distinctly later than that at the posterior pole (Fig. 4E,F). dpp transcripts show a similar asymmetry in the initiation of expression at the two poles. Neither dpp nor zen are expressed in embryos that have entered stage 5, and the late patterns of dpp expression are not observed (data not shown).

The expression patterns of twi and sna in the ventralizing alleles correlate well with the effect a particular allele has on the expansion of the mesodermal anlagen. Mutant embryos derived from Tl^10b/+ mothers show uniform expression of twi and sna over the entire D/V axis (Fig. 41,M). cact mutants show an expansion of the twi and sna domain (Fig. 5D), which is consistent with the gastrulation phenotype (Roth et al. 1989; Roth, 1990). We observe no expansion of the mesoderm in embryos derived from TP®/ + mothers (Fig. 4J,N; Fig. 5B). Regardless of the maternal genotype, the sna pattern refines normally during late stage 4, and thereafter is not expressed at the poles (Fig. 4N). sna expression appears to be initiated in late stage embryos, but no specific pattern is evident (data not shown).

Lateralized embryos are produced by mothers bearing certain alleles of Tl or ea. The cuticle of the mutant embryos is an elongated tube covered with rings of denticles that appear to be derivatives of the lateral regions of the wild-type denticles (Fig. 3D). Lateralized embryos fail to form a ventral furrow, and the cephalic folds are prominent both dorsally and ventrally. Thus, these alleles are expected to show an expansion of lateral fates at the expense of both more dorsal and more ventral fates (Anderson et al. 1985a).

The initial expression patterns of dpp and zen in lateralized embryos are similar to those seen in ventralized embryos; dpp and zen are expressed only at the poles (Fig. 4C,G). The initial expression of dpp and zen at the poles is lost by the onset of stage 5, after which no further expression of zen is observed. However, initiation of the late patterns of dpp expression is observed (data not shown).

Lateralized embryos lack the initial pattern of twi transcripts ventrally, but expression is observed at the poles (Fig. 4K). The expression of twi at the posterior pole fades by the onset of gastrulation, while that at the anterior pole persists (data not shown). This persistence of twi expression is consistent with the observation that, in wild type, the expression at the anterior pole is retained until germ band extension, while that at the posterior pole fades earlier (our observations, and Thisse et al. 1988). sna transcripts only appear transiently at the poles if at all (Fig. 40). This is probably a reflection of the rapid refinement of the sna pattern observed in wild type. Expression of sna later in embryogenesis is observed (data not shown), consistent with the onset of the late patterns.

Null alleles of each of the maternal effect genes in the dorsal group produce a characteristic embryonic lethal phenotype that is entirely dorsalized. The cuticle of mutant embryos reflects a fate shift in the D/V axis such that all cells along the axis behave like dorsal cells of the wild-type embryo. Mutant embryos differentiate as long, thin tubes of cuticle covered with fine dorsal hairs (Fig. 3E). Ventral denticle belts are entirely lacking. At the time of gastrulation, these embryos do not form a ventral furrow, and the cells normally recruited for this purpose fold to mirror the characteristic dorsal and transverse folds (Nüsslein-Volhard, 1979).

In dorsalized embryos, dpp and zen are expressed uniformly in all cells of the blastoderm stage embryo (Fig. 4D,H). The zen pattern sharpens along the A/P axis, but not along the D/V axis: in stage 5 mutant embryos, zen is expressed in two narrow stripes anteriorly, and a single broad stripe in the middle of the embryo, reminiscent of the head spots and mohawk seen in wild type (see Fig. 6C; cf. Fig. 6A). By contrast, the initial pattern of dpp does not refine at all. Yet, in late stage embryos, new rounds of dpp expression are observed, consistent with the normal onset of the late patterns (data not shown).

twi is not expressed in dorsalized mutant embryos during the early stages of embryonic development (Fig. 4L). While sna transcripts cannot be detected prior to the onset of gastrulation (Fig. 4P), later in development, mutant embryos show hybridization in the segmented region of the embryo, which can be correlated with the onset of the late patterns (data not shown).

The uniform expression of zen in embryos derived from dl mothers is not entirely consistent with the cuticular phenotype: while the expression of zen in a dl mutant embryo predicts that the blastomeres should be fated as amnioserosa, the cuticular phenotype of these mutant embryos implies that some of the cells expressing zen in fact secrete dorsal cuticle. This can be accounted for, in part, by the fact that in the wild-type embryo, it is the mohawk that is coincident with the prospective amnioserosa, and not the initial expression pattern or the refined expression in the cephalic segments. Thus, since the zen pattern refines along the A/P axis in dl⁻ embryos (see above), it is only the centrally located region, corresponding to the mohawk, that should produce amnioserosa fates. These results suggest that a broad central domain of amnioserosa cells should appear in dorsalized embryos. To determine if this is the case, we have stained dorsalized embryos with an antibody directed against the Kriippel (Kr) protein. Analysis of the wild-type distribution of Kr protein has shown this antigen to be present in all amnioserosa cells from the onset of germ band extension until the completion of germ band retraction. Although other cells express Kr protein at this time, the amnioserosa cells are morphologically distinct in that they have particularly large nuclei (Fig. 6B, Gaul et al. 1987).

In late stage dl embryos, we observe a subset of cells producing the Kr antigen that are morphologically similar to wild-type amnioserosa cells. These cells form a broad ring in the middle of the embryo that is located in approximately the same position where the expanded mohawk appeared earlier in development (Fig. 6D; cf. Fig. 6B). While the position of these cells is relatively constant, the number of cells is not; some dorsalized embryos show as many as 200 amnioserosa cells (Fig. 6D), others as few as 40 (Fig. 6E). We suspect that this variation reflects the extinction of Kr expression in the amnioserosa that is observed in wild type embryos during stage 13 (Gaul et al. 1987). Nevertheless, our results do indicate that, consistent with the refined expression pattern of zen, some cells in dl embryos behave as amnioserosa cells and that these cells are expressed in the region of the embryo predicted by the refined expression pattern of zen.

Based on the preceding results, it might be expected that the positional value of cells at the poles of the embryo would be ambiguous or undefined with respect to their position along the D/V axis, insofar as genes that are exclusively expressed either dorsally or ventrally throughout the segmented region of the embryo (i.e. dpp, zen, twi, and sna), share a common domain of expression at the poles. This hypothesis is substantiated by the expression patterns of dpp and zen in the maternal ventralizing genotypes, and by the expression patterns of twi in the maternal lateralizing genotypes. In all of these cases, expression at the poles persists despite dramatic shifts of positional values along the D/V axis, suggesting that control of gene expression at the poles requires functions that are not specified by the D/V patterning genes. A similar singularity of the poles is observed in the regulatory network controlling A/P pattern formation, and the genes of the terminal system have been shown to mediate this effect (Nüsslein-Volhard et al. 1987). Two representatives of this group, trunk (trk) and torso (tor) (Schijpbach and Wieschaus, 1986), are analyzed here.

The cuticular phenotype of embryos derived from mutant trk or tor mothers indicates the role these genes play in the specification of the unsegmented acron (anterior pole) and telson (posterior pole). The differentiated mutant embryos lack the labrum, the head skeleton is reduced in size, and all structures posterior to the seventh abdominal segment are deleted (Fig. 3F, Klingler et al. 1988). The first manifestation of the mutant phenotype is observed at the time of gastrulation when the ventral furrow can be seen to invaginate over the entire length of the embryo rather than between 20 and 65% egg length, as seen in wild type (Schiipbach and Wieschaus, 1986).

Expression of dpp and zen in mutant embryos derived from tor or trk mothers is initiated at the proper time, but not in the proper pattern. Rather than extending over the dorsal surface and around both poles, expression is confined to the dorsal surface. Neither dpp nor zen transcripts are detected at the poles (Fig. 7A,C). Notably, this pattern of expression does not correspond to the fate shifts predicted from the cuticular phenotype, which predicts an expansion of the segmented region of the embryo (Klingler et al. 1988). The lack of terminal gene function does not affect the refinement of these patterns along the D/V axis or the initiation of late patterns. In stage 5 embryos, refinement of the zen pattern is observed along both axes, though the refined pattern is condensed along the A/P axis, and the anterior-most head spot is absent (Fig. 7D, cf. Fig. 2F). In late stage embryos, dpp is expressed in the characteristic double stripe pattern (Fig. 7B, cf. Fig. 2C).

As seen for dpp and zen, the expression of twi and sna is also initiated at the proper time, but not in the proper pattern. Both twi and sna transcripts are detected in early stage 4 embryos in the characteristic ventral domain. The width of the domain remains 18 cells wide, as observed in wild type, and the sinus in the sna pattern at the cephalic furrow is observed (Fig. 5D). However, the pattern along the A/P axis is aberrant. Transcripts are detectable in the ventral cells up to the middle of the poles, but no dorsal cells are labelled (Fig. 7E,G). This alteration is consistent with the A/P fate shifts predicted from the cuticular phenotype and the expression patterns of pair rule genes in these mutant embryos (Klingler, 1989; Casanova, 1990). Notably, the sna pattern does not refine (Fig. 7G). After stage 5, both twi and sna transcripts can be detected in the derivatives of the mesoderm (Fig. 7F, for twi), and later in development sna transcripts accumulate in the cells of the ventral and dorsal ectoderm, characteristic of the wild-type pattern (Fig. 7H). In this last respect, dpp, zen, twi, andsna are similar: improper expression at the poles does not affect the ontogeny of the refined and late patterns in the segmented region of the embryo.

Taken together, the above results indicate that the polar expression of both dorsally expressed genes like dpp and zen, and ventrally expressed genes like twi and sna, show a requirement for the genes of the terminal group.

In the following two sections, we report on the initial and refined expression patterns of dpp, zen, twi, and sna in embryos mutant for zygotic genes affecting D/V patterning. Representative cuticle preparations for these zygotic genotypes are shown in Fig, 8. We will use the term decapentaplegic group (dpp group) to refer to those genes that are required for the specification of dorsal structures, and the term twist group (twi group) to refer to those genes required for the specification of the ventral structures.

In addition to dpp and zen, the dpp group includes five other zygotic loci: screw (sew), tolloid (dd), shrew (srw), twisted gastrulation (tsg), and short gastrulation (sog). Mutations in these loci are associated with a general loss of amnioserosa, dorsal ectoderm, and dorsolaterally derived structures of the acron and telson (Anderson, 1987; Rushlow and Arora, 1990; K. Arora and C. Nüsslein-Volhard, in prep). Accompanying this loss of dorsal structures is an expansion of ventrolateral pattern elements (Fig. 8C-E), The seven genes can be subdivided into three classes based on a comparison of the phenotype of the most severe loss-of-function allele of each (R. Ray, K. Arora, and W. Gelbart, in prep): weak loci (sog, tsg, and zen;Fig. 8C), moderate loci (tld, sew, and srw;Fig. 8D). and a single strong locus (dpp;Fig. 8E). Loss-of-function mutations in these loci result in a characteristic disruption of germ band extension that leads to the invagination of the posterior segments into the interior of the embryo (our observations, see Fig. 8).

The initial expression patterns of ail four zygotic genes are unaffected by mutations in any of the dpp group genes (Table 1). Thus, the expression of each gene appears to be independently initiated by dl. In the case of twi and sna, proper refinement, and initiation of late patterns is also observed (Table 1). For dpp and zen, however, maintenance and refinement of the initial patterns is aberrant (see also Rushlow and Levine, 1990). A comparison of the effects of weak, moderate and strong dpp group mutations on the dpp and zen expression patterns indicate that these two genes have reciprocal effects on their refinement. For instance, in a wild-type gastrulating embryo, expression of zen is confined to the dorsal-most region of the embryo, the presumptive amnioserosa, while dpp expression is excluded from this domain (Fig. 9A,D). In embryos mutant for weak loci of the dpp group, zen expression is maintained until germ band extension, but the pattern fails to refine along the D/V axis, and expression is observed over 40% of the dorsal surface (Fig. 9B). dpp also fails to refine in these embryos, and expression is observed in the amnioserosa (Fig. 9E). In embryos mutant for moderate and strong loci of the dpp group, zen expression is lost throughout the dorsal ectoderm prior to cellularization, and thus is never seen to refine (Fig. 9C). Again, dpp expression is normal but does not refine and is expressed throughout the dorsal ectoderm (Fig. 9F).

Several points can be made from these results. First, all of the dpp group genes appear to be required for the normal ontogeny of the zen pattern and the fating of the amnioserosa. Second, the weak and moderate ventral- izing loci have different effects on the expression pattern of zen-. dpp, tld, sew, and are required for the maintenance of the zen expression pattern, while tsg and sog are required for its refinement. While these functions are distinct, the terminal phenotype is the same: loss of the amnioserosa. This failure to fate the amnioserosa undoubtedly accounts for the failure of the dpp expression pattern to refine, since the cells normally fated to the amnioserosa would be expected to behave like cells of the dorsal epidermis in the mutant embryos.

Thus, the subdivision of the dorsal ectoderm into amnioserosa and dorsal epidermis depends only on the dpp group genes and is subordinate to the initial specification of dorsal ectoderm by dl. Furthermore, there appears to be a regulatory loop involved that has the following organization. Moderate and strong dpp group genes are required for the proper ontogeny of zen expression, which determines the fate of the amnioserosa. Once this domain has been established, genes specific to the dorsal epidermis, like dpp, are excluded from it. Notably, zen has no role in its own refinement, as the ontogeny of zen expression is normal in zen mutant embryos. Similarly, null mutations of dpp do not affect the expression of dpp, excepting that since the amnioserosa is not fated, no refinement of the dpp pattern is observed (Table 1).

Loss-of-function alleles of twi and sna produce embryos that lack all derivatives of the mesoderm. In such mutant embryos, the cells on the ventral side do not invaginate and no ventral furrow is formed (Simpson, 1983; Leptin and Grunewald, 1990). In spite of this defect, twi and sna mutant embryos produce fairly norma) cuticles with only minor truncations of ventrally derived pattern elements (Fig. 8B).

The initial expression patterns of all four zygotic genes are not affected by mutations in twi or sna (Table 1). Thus, in conjunction with the results obtained for the dpp group genes, it seems clear that the initiation of expression of dpp, zen, twi, and sna is entirely controlled by the maternal morphogen dl. In particular, we note the initiation of twi expression is essentially normal in sna mutant embryos, and the initiation of sna expression is normal in twi mutant embryos (Fig. 10A,C,E,G). Thus, these two genes are not sequential elements in a hierarchial series, and neither depends entirely on the other. As was true for twi and sna expression in dpp group genes, refinement and initiation of late patterns is observed for both dpp and zen in twi and sna mutant embryos. In contrast, the ontogeny of twi and sna expression in twi group mutant embryos is aberrant, as discussed below.

twi expression is aberrant in both twi and sna mutant embryos. In twi mutant embryos, no effect is observed until the pattern begins to deteriorate during cellulariz- ation, at which time the level of expression drops and the pattern begins to develop gaps in a segmental fashion (Fig. 10B). By the beginning of germ band extension, twi expression is no longer detectable. Notably, twi mutations do not affect the width or refinement of the twi pattern, only its perdurance (Table 2).

In sna mutant embryos, twi expression appears in the normal pattern, but throughout stages 4 and 5, the mutant embryos can be distinguished from their wildtype siblings by the fact that the level of expression is reduced (Fig. 10C, Table 2). As was observed in twi mutant embryos, twi expression is extinguished in the presumptive mesoderm by the beginning of germ band extension. Notably, the expression of twi in these mutant embryos is extinguished only within the domain of the presumptive mesoderm; expression in the cephalic segments and telson is not affected and must be under separate control.

The expression of sna in twi mutant embryos reveals that sna function is at least partially subordinate to twi. In such embryos initiation of sna expression is normal, but the width of the pattern never reaches the characteristic 18 cells, and the edges, which are sharp and distinct in wild type, are diffuse and irregular (Fig. 10E, Table 2). Furthermore, stage 5 mutant embryos still show weak expression of sna at the poles suggesting that the refinement of the sna pattern is also aberrant. Thus, twi appears to be required for the proper ontogeny of the sna expression pattern, while the reciprocal, that sna is required for the ontogeny of twi, was not observed. The sna expression patterns described here have two features in common with the twi patterns: sna expression is extinguished by the beginning of germ band extension, and the expression in the cephalic segments and telson are under a separate control (Fig. 10F).

Consistent with the hypothesis that the width of the sna domain depends on twi, sna expression in sna mutant embryos is as broad as in wild type (Fig. 10G). However, the edge of the domain is rough, as was observed in the twi mutant embryos. This latter effect suggests that sna is required for the refinement of its own expression, at least as far as the sharpness of the edges is concerned. As was true for sna expression in twi mutant embryos, the expression of sna in sna embryos is not detectable after the beginning of germ band extension (Fig. 10H). Thus, regardless of probe and genotype, the expression patterns are extinguished at approximately the same time, suggesting that the failure of the patterns to be maintained in the mutant embryos may not be an effect of the mutations themselves, but rather a programmed point in development after which other fates are established in the mesodermal domain if twi and sna are not both expressed.

We have shown that two sources of maternal information are required to generate the initial expression patterns of dpp, zen, twi, and sna. This information is provided by genes of the dorsal and terminal systems. In the segmented region of the embryo, the expression of all four zygotic genes is controlled exclusively by the dorsal group genes and cact. In the polar regions, the situation is more complex and requires the activity of the terminal system. In particular, dpp and zen require the activity of the terminal system alone, while twi and sna require the coordinate action of both the dorsal and the terminal systems, as will be discussed below. These two maternal controls are all that is required for the specification of the initial expression of all four zygotic genes.

The dl gradient appears to be interpreted on three levels. Ventrally, dl activates twi and sna expression and represses dpp and zen expression, so only twi and sna are expressed. Laterally, the reduced activity of the dl protein is not sufficient to activate twi and sna, but can still repress dpp and zen, so none of the four genes are expressed. Dorsally, the activity of dl is so low that it can neither activate twi and sna nor repress dpp and zen, so only dpp and zen are expressed (Fig. 10). Thus, there are two discernible thresholds of dl activity along the D/V axis, twi and sna respond to the higher threshold. If dl activity is above this threshold, twi and sna are expressed; if it is below, they are not. dpp and zen respond to the lower threshold. If dl activity is above this threshold, dpp and zen are repressed; if it is below, they are not.

In the absence of double staining experiments or alternate sections, we cannot say that twi and sna, on the one hand, and dpp and zen on the other are responding to precisely the same level of dl activity. In fact, careful analysis of the twi and sna patterns suggests that the graded expression of twi may extend more dorsally to label cells that are not expressing sna (unpublished). Similarly, the ventralmost limit of dpp and zen expression may not coincide precisely. Nevertheless, within the resolution of the experiments presented here, the two thresholds are respected. The fact that all four of the genes analysed here appear to be responding to only two thresholds of dl activity suggests that the dl gradient may only be involved in defining these three broad domains of the embryo. Further subdivision of the embryo within each of these domains undoubtedly depends on the interactions between the zygotic gene products that are specifically required for each domain. Our results on the interactions between zygotic genes suggest that this is in fact the case. In particular the relationship between dpp and zen in the subdivision of the dorsal ectoderm into dorsal epidermis and amnioserosa suggests that only the zygotic genes of the dpp group are responsible for this partitioning event.

Although the initial expression pattern of zen suggests a primary role for this gene in the organization of the D/V pattern, our data strongly suggest that this gene plays a subordinate role to dpp. The most notable of these is that the cuticular phenotype associated with zen mutations, which consists of a loss of amnioserosa and deletions in the optic lobes (Wakimoto et al. 1984), does not reflect a general requirement for zen in all cells of the dorsal ectoderm, as predicted by the initial expression pattern. Instead, the deleted structures correspond precisely to those regions of the fate map delimited by the refined expression pattern. Since the establishment of this refined pattern depends on the function of all other dpp group genes, it must be concluded that zen acts at a level further down the hierarchy. By contrast, the failure of other zygotic genes to affect the initial expression pattern of dpp, and the severity and singularity of the dpp phenotype, are strong indicators of the primary role this gene plays in the specification of the dorsal ectoderm.

The specification of the ventral domain requires both twi and sna. Our results indicate that these two genes are independently specified by the dl gradient, and the interactions between the two genes suggest that, despite their respective homologies to DNA binding proteins (Murre et al. 1989; Boulay et al. 1987), they are not sequential elements in a hierarchical series. Nevertheless, the expression of sna depends on the normal expression of twi, and in this respect, sna is subordinate to twi. Thus, we propose that twi is the primary patterning gene for mesoderm. However, further studies on the roles these genes play in mesoderm development suggest that they carry out complementary roles. In particular, twi and sna have different effects on the expression of genes in mesodermal and surrounding cells. For instance, twi has been shown to be responsible for the activation of mesoderm specific genes like msh-2 (Bodmer et al. 1990) and PS-2 integrin (unpublished). In contrast, sna has been shown to be involved in repression of genes that are expressed in the mesectoderm and ventral ectoderm like single-minded (Nambu et al. 1990) and rhomboid (unpublished). Thus, it seems clear that twi and sna carry out complementary, not redundant, functions in the specification of the mesoderm.

Although we have only analyzed a small number of zygotic genes which might be targets of dl activity, we speculate that the dl gradient may only be responsible for coarse subdivision of the embryo into the three domains described above. While it seems clear the determination of the dorsal and ventral fates are specified directly by dl, there is some evidence to suggest that lateral positional values appear to arise as a default state in the absence of the zygotic gene expression characteristic of the dorsal and ventral positional values. The support for this hypothesis comes from the cuticular phenotype of double mutant embryos lacking the maternal function of dl and the zygotic function of dpp. Such double mutant embryos do not express twi or sna due to the loss of dl function, and also lack the function of the primarily determinant in the dorsal ectoderm, dpp. Phenotypically, the embryos are lateralized and manifest zygotic gene expression consistent with this fate (Irish and Gelbart, 1987; our observations), suggesting that these positional values are fated in the absence of dl activity. A similar type of intercalation is observed in the regulation of the gap gene Kr which is expressed in embryos derived from mothers lacking the function of the two A/P maternal coordinate genes, nanos and bed (Gaul and Jackie, 1989; Lehmann and Frohnhbfer, 1989).

We have shown that the initial expression pattern of twi at the poles requires the activity of both the dorsal- ventral system and the terminal system: twi expression is entirely lacking in dorsalized embryos, and aberrant at the poles in tor⁻ embryos. The failure for twi expression to appear in dorsalized embryos clearly indicates that these two control mechanisms are not independent. As discussed below, it appears that the ability for the terminal system to act on twi depends of the activity of dl. We have therefore proposed that the terminal system acts through or with dl to affect expression of twi at the poles.

Expression of twi is lacking in all genotypes tested that produce dorsalized embryos. However, it is not the presence or absence of dl protein, per se, that is necessary for the activation of twi by the terminal system, since dorsalized embryos derived from Tl, snk, or tub mothers produce dl protein but nevertheless lack expression of twi at the poles. Thus, the effect of the terminal system on twi depends on the activity of the dl protein. This point accounts for the polar expression of twi in lateralized embryos, as dl is at least partially active (Steward et al. 1989; RothetaZ. 1989; Roth, 1990) in these embryos, and thus twi can be expressed. This conclusion presumes that the expression of twi in lateralized embryos depends on terminal system function. In fact, this is the case, as we find that a lateralized embryo that is also lacking the function of one of the maternal terminal system genes does not express twi (unpublished). Furthermore, we can rule out the possibility that the terminal system is acting to alter the distribution of dl protein at the poles (i.e., causing the nuclear localization of dl at the poles) as no accumulation of dl protein is observed in the terminal nuclei of lateralized embryos (unpublished).

Further evidence to suggest that dl and the terminal system act in conjunction comes from analyses of twi expression in double mutants of loss-of-function alleles of the dorsal group genes and cact. Such double mutant mothers produce lateralized embryos (Roth et al. 1989; Roth 1990), and, like the lateralized embryos produced by mothers bearing lateralizing alleles of dorsal group genes, these embryos express twi at the poles. However, in the double mutant combinations with cact, the alleles of the dorsal group genes are nulls, and thus the dorsal pathway is not active. Thus, the lack of twi expression at the poles in a dorsalized embryo appears to be due to interference by cact. As it has been proposed that cact interacts with the dl gene product, we can account for our data by assuming that the interaction between the cact and dl gene products interferes with the action of the terminal system.

Two mechanisms could account for these data. First, two positive regulators may be required to activate twi: dl and another transcription factor, either maternal or zygotic, downstream of tor. We must presume that neither of these factors are sufficient to activate twi expression alone, in order to account for the fact the dorsalized embryos do not express twi at all. This simple model is appealing, but at present, there is no candidate for the second transcription factor. An alternative explanation is that dl is the only transcription factor required, and that dl activity is modified by the action of tor or downstream genes. While we know that this does not occur by redistribution of dl protein, it could occur by modulation of dl activity (i.e., directly acting on dl protein), thus making the lower concentration of dl protein normally found in polar nuclei sufficient to activate twi. At present, these two hypotheses cannot be distinguished.

With regard to the control of sna expression, the initial pattern is clearly regulated by the same network as described for twi. However, since the refinement of this pattern also depends on the terminal system, we must conclude that the genes of this system play a role in the repression as well as activation of sna transcription. Thus, there are two superimposed controls: the terminal system and dl activate sna expression initially, and then later in development other terminal genes act to refine this pattern. As the latter regulatory feature appears to override the effect of the former, we suspect that the transient polar expression of sna is not a reflection of a specific requirement for sna gene activity at the poles, but rather a consequence of these overlapping control mechanisms.

Finally, as we have noted earlier, the problem that arises from expressing twi and sna at the poles by affecting the activity of dl is how dpp and zen can be expressed in the same domain when these two genes are repressed by dl. We have shown that this effect is also mediated by the terminal system, and that this activity does not depend on dl. The latter fact is clearly illustrated by the expression patterns of dpp and zen in embryos derived from Tl^10bI + mothers. Although such ventralized embryos show uniform localization of dl protein to all blastoderm nuclei, the expression of dpp and zen persists at the poles. Thus, the activation of dpp and zen by the terminal system overrides the repression due to dl. This assertion is supported by the expression patterns we observe in tor⁻ embryos. These embryos show a retraction of the dpp and zen expression patterns away from the poles (Fig. 11), an effect which is not consistent with the fate changes predicted by the cuticular phenotype (Klingler et al. 1987; Casanova, 1990). However, though inconsistent with the tor phenotype, this pattern of expression is consistent with the repression of dpp and zen by the virtually normal gradient of dl protein in the mutant embryos (unpublished).

We conclude that the terminal system is required for two layered functions that are necessary to achieve expression of the four zygotic genes at the poles. First, the terminal system is required to express twi and sna at the poles by modulating the activity of dl. Then, other functions of the terminal system activate dpp and zen, and repress sna at the poles. Both of these latter effects override aspects of dl regulation that are required for normal pattern formation in the segmented region of the embryo. The combination of these two functions can account for the modulations observed in the wild type patterns.

We wish to thank the following for their contributions to this work. C. Rushlow, for providing the antibodies used in this study. N. Brown, for his contribution to the cloning of the sna cDNA. R. Warrior, N. Brown, C. Rushlow, J. Birchler, and D. Hursh, for discussions and critical reading of the manuscript. S. Findley, for help in preparing the figures. This work was funded in part by an Office of Naval Research Predoctora] Fellowship and a National Institutes of Health Training Grant to R.P.R. and by a National Institutes of Health grant to W.M.G. K.A. was supported by a Max Planck Fellowship.