A genetic screen for modifiers of Deformed homeotic function identifies novel genes required for head development

The anterior-posterior axis of insects is composed of a linear array of segments that develop as variations on a basic theme. The study of the genetic basis of segmental variations in Drosophila melanogaster has led to a better understanding of axial specification of both invertebrate and vertebrate embryos (McGinnis and Krumlauf, 1992). According to the current view, each segment follows a particular developmental pathway (GarcÍa-Bellido, 1977; Lewis, 1963) under the control of proteins encoded by the homeotic selector, or Hox, genes. These homeodomain-containing transcription factors are expressed in discrete regions of the body along the anterior-posterior axis, where they differentially regulate downstream genes. In order to act as stable selectors of axial identity, HOX proteins have to be persistently expressed within their proper domain of action, and be able to recognize and regulate their targets. This process of recognition/regulation has to be modulated in different cells and tissues in order to bring about the many structures found in one segment or region.

The establishment of Hox transcription in their characteristic domains depends on regulation by particular combinations of maternal coordinate, gap, pair-rule and segment polarity genes (Akam, 1987; Harding and Levine, 1988; Irish et al., 1989; Morata, 1993). The maintenance of Hox expression depends both on auto- and cross-regulation among Hox genes themselves as well as on the negative and positive influences of the Polycomb (Pc) and trithorax (trx) group genes. Pc group genes are required to keep homeotic selectors, as well as other homeodomain con-taining genes, in a repressed state outside their normal expression limits (Jürgens, 1985; Wedeen et al., 1986). This repressive activity is thought to involve changes in chromatin structure (Zink and Paro, 1989). The trx group genes apparently work at cross-purposes to the Pc group, as trx mutations can suppress Pc loss-of-function phenotypes (Kennison and Russell, 1987; Kennison and Tamkun, 1988). In some cases, this genetic suppression is due to a direct involvement in activating the transcription of a Hox gene that Pc group functions are repressing (Breen and Harte, 1993; Tamkun et al., 1992).

The action of homeotics varies between tissues, stages of development and sub-regions within a segment, and little is known about the factors and mechanisms cooperating with the HOX proteins to govern these diverse activities. Genes proposed to be involved in such Hox activity regulation include spalt (sal; Jürgens, 1988), teashirt (tsh; de Zulueta et al., 1994), cap’n’collar (cnc; Mohler et al., 1995), extradenticle (exd; Chan and Mann, 1996; Pinsonneault et al., 1997; Rauskolb and Wieschaus, 1994), and some Hox family members (González-Reyes and Morata, 1990). At the next level of the develop-mental genetic hierarchy, a few HOX downstream targets have been identified, but little is known about how they participate in elaborating segmental structures (Morata, 1993). Until more molecules are assigned to Hox pathways, and their mechanisms understood, the specific role of HOX proteins remains little more than a developmental black box.

Many modifiers of homeotic activity have been previously identified by screening for enhancers and suppressors of Pc (Kennison and Russell, 1987; Kennison and Tamkun, 1988). Flies heterozygous for Pc mutations show a number of homeotic transformations due to ectopic expression of Hox genes. For example, Pc/+ males show a partial transformation of meso- and metathorax towards prothorax, as revealed by the presence of extra sex comb teeth on the second and third thoracic legs. This sex comb phenotype can be enhanced by mutations in other Pc group genes and, by definition, sup-pressed by mutations in trx group genes. This phenotype has been used as a marker for general Hox misregulation, and it is therefore believed that previous screens identified genes involved in the modulation of most or all of the Hox genes. In contrast, we have screened for mutations that enhance the semi-lethality of hypomorphic mutations in the homeotic selector Deformed (Dfd), attempting to focus on a single Hox pathway.

The Dfd gene is a member of the Antennapedia complex, and its function is required to specify maxillary and mandibular structures both in embryos and adults (Merrill et al., 1987; Regulski et al., 1987). A stripe of Dfd expression is initially established in the posterior head region by multiple positional cues (Chadwick et al., 1990; Jack and McGinnis, 1990; Jack et al., 1988; McGinnis et al., 1990), and its expression is maintained by autoregulatory circuits (Bergson and McGinnis, 1990; Lou et al., 1995; Zeng et al., 1994). Three genes that are downstream targets of DFD regulation have been characterized: Dfd itself, Distal-less (Dll) and paired (prd). Dll activity in the ventral portion of the maxillary segment is required for the development of cirri (O’Hara et al., 1993), and prd has been implicated in the formation of the maxillary ventral organ (O’Hara, 1995; Vanario-Alonso et al., 1995).

In the present work we report the identification of eight genes whose dosage is apparently crucial for Dfd function. One is a known trx group gene: kismet (kis), previously isolated by Kennison and Tamkun (1988); another is a potential new member of the trx group, snaggletooth (snt). Four interacting genes encode potential effectors of homeotic function: two collagen IV genes, a calcium channel α₁ subunit gene, and a gene encoding muscle myosin heavy chain. Finally, two novel genes, polycephalon (poc) and apontic (apt), are required for the formation of a subset of Dfd-dependent structures. We mol-ecularly characterized the apt gene, which appears to have multiple functions in a complex spatial pattern, but a prominent site of expression is the ventral posterior head, where its function overlaps with that of Dfd.

The goal of the screen was to isolate mutations in genes whose dose is crucial for Dfd wild-type function. We screened for mutations on the second chromosome that, when heterozygous, would enhance the mortality of Dfd^rV8/Dfd^rC11 flies, reducing the survival rate from approx. 50% to approx. 0%. The genetic scheme is shown in Fig. 1. The mutagen used was EMS and the conditions and regimes used for the mutagenesis were as described in Grigliatti (1986). cn bw sp males were mutagenized and crossed en masse to Ki Dfd^rV8red/TM3 Sb virgin females; Ki non-Sb F₁ males were mated individually to cn/cn;Dfd^rC11red e/TM1 females and the progeny scored for absence of cn, red eyes. The lines selected were subjected to two rescreens. In the rescreens, the ratio of * cn bw sp/cn; Ki Dfd^rV8red/Dfd^rC11red e flies to cn/+; Ki Dfd^rV8red/Dfd^rC11red e was calculated. Ratios less than 1.0 indicate a genetic interaction, and only lines giving ratios of 0.5 or less were kept.

All crosses were performed with 5-10 virgin females and 3-5 males per vial and cultured at 25°C. We first performed inter se complementation among all mutant lines. Only groups represented by two or more lethal mutations were kept for further analysis. Meiotic map positions were determined for some of the groups (apt, vkg, snt), using a spd^fgb pr c chromosome. Selected mutations from every complementation group were crossed to a set of deficiencies covering most, but not all of the second chromosome (the Df(2) kit from Bloomington Stock Center). This analysis indicated that many mutant chromosomes carried more than one lethal mutation. We therefore recombined two alleles for each complementation group onto an appropriately marked lethal-free chro-mosome. All derived chromosomes were re-tested for complementation behavior, and two different lines for each allele were kept. All further analyses (i.e. cuticular phenotypes, interaction with Dfd and Pc⁴) were performed using these ‘clean’ lines. At the same time, the parental cn bw sp chromosome was crossed to equivalent lethal-free chromosomes and selected for the same markers; these chromosomes were used as controls in each of the genetic interaction tests.

The cytogenetic location of poc alleles was determined by their failure to complement Df(2L)TE75 (21A1-B4;21B6). The cytogenetic location of apt alleles was determined by their failure to complement Df(2R)bw-s46 (59D8-D11;60A7) and Df(2R)bw5 (59D10-E1;59E4-F1). The cytogenetic location of snt alleles was determined by their failure to complement Df(2R)nap1 (41D2-E1;42B1-B3) and Df(2R)nap9 (42A1-A2;42E6-F1).

All the vkg alleles reported here were identified as such by their failure to complement three lethal P-element insertions known to disrupt vkg function (RodrÍguez et al., 1996). Our vkg alleles fully complement DCg1^GDB, a P-element insertion allele of the adjacent collagen IV gene. All of our DCg1 alleles fail to complement the pre-viously isolated allele DCg1^GDB, but fully complement the three pre-viously identified P-element insertion alleles of vkg.

The Dfd interaction values reported in the Results section measure the interaction strength in a manner similar to that used in the initial screen, except that experimental and control chromosomes were out-crossed. All crosses were done with 15 virgin females and 3-5 males per vial and cultured at 29°C. For each line, the progeny from 35-40 vials were counted. We counted the numbers of progeny of the classes */+; Dfd^rC11/Dfd^rV8 (class a) and +/+;Dfd^rC11/Dfd^rV8 (class b). The Dfd interaction strength was calculated as the ratio of a:b. We then expressed these figures as percentages of the equivalent numbers obtained from crosses using control chromosomes. Interactions in the range of 0-40% were considered as ‘strong’, 41-60% as ‘moderate’ and 61-80% as ‘weak’.

Approximately 10 virgin Pc⁴ females were mated with three males for each mutant line, and cultured at 27°C. The male progeny het-erozygous for both Pc⁴ and the mutation of interest were collected and fixed in 95% ethanol. The T2 legs were mounted in Hoyer’s medium and the sex comb teeth counted under a compound microscope. For each mutant chromosome and its control, we calculated the average number of sex comb teeth per T2 leg; the strength of interaction with Pc was defined as the average number of sex comb teeth per T2 leg in the mutant divided by the average number of sex comb teeth per T2 leg in the control.

Embryos were dechorionated, devitellinized, fixed and mounted in 1:2 lactic acid:Hoyer’s medium as described by Harding et al. (1995). In all cases balancer chromosomes were removed by outcrossing the mutant stocks to wild type. Unless otherwise stated, all embryos shown are hemizygous.

Genomic DNA from the P-2360 line corresponding to l(2)09049 was extracted following standard protocols and purified in a CsCl gradient by ultracentrifugation. 7 μg of DNA were then digested with XbaI F1 restriction enzyme, the enzyme was heat inactivated, the solution diluted 50-fold in ligation mixture, and ligation performed overnight at 15°C. The product of ligation was phenol-extracted, precipitated and resuspended in 4 μl of water. Half of it was used to transform electrocompetent DH5α cells, which were plated in kanamycin-selective medium.

We used a λ iso-1 Drosophila genomic library prepared by J. Tamkun (Clark et al., 1993), and a plasmid cDNA library from 8-to 12-hour Drosophila embryonic mRNA kindly provided by N. Brown. Southern and northern hybridization were used following standard protocols; all probes were labeled by nick translation.

These were performed with modifications as described by Tautz and Pfeifle (1989). The apt antisense RNA probe was made using the T7 promoter of the vector described in Brown and Kafatos (1988).

This was done using the Sequenase Version 2.0 DNA and PCR Product Sequencing Kits from United States Biochemical according to protocols provided by the manufacturer.

Deformed null mutations are embryonic lethal; the unhatched larvae are missing structures derived from maxillary and mandibular segments, such as mouth hooks. Flies carrying the hypomorphic alleles Dfd^rV8/Dfd^rC11 and reared at 29°C reach adulthood at a rate of approximately 50% that of wild type. We reasoned that by reducing the dose of the limiting factors in the Dfd pathway, the mortality rate of these Dfd hypomorphs would increase, leaving no survivors. We therefore screened for modifier mutations on the second chromosome that in single copy decrease the viability of Dfd^rV8/Dfd^rC11 cohorts to nearly zero (see Fig. 1 and Harding et al., 1995).

Approximately 8,000 mutagenized cn bw sp chromosomes were screened for interactions. Lines that showed a decreased survival in the hypomorphic Dfd background were outcrossed, and then subjected to inter se lethal complementation. Further studies were performed on those interacting loci represented by two or more alleles. Mutant loci were mapped by using a com-bination of meiotic recombination and deficiency mapping. In some cases, we were able to assign mutant alleles to known genes using lethal complementation tests. For example, the screen generated six new alleles of the trx group gene kis, seven alleles of vkg, an α2(IV) collagen gene (RodrÍguez et al., 1996; Yasothornsrikul et al., 1997); four alleles of DCg1, an α1(IV) collagen gene (Yasothornsrikul et al., 1997); four alleles of Muscle myosin heavy chain (Mhc) and three alleles of Ca-alpha1 (Zheng et al., 1995), a calcium channel α₁ subunit gene (Fig. 2). The three other complementation groups (poc, apt, snt) represent novel genes required for head development. We obtained six alleles of poc, three alleles of apt and two alleles of snt. Fig. 2 shows the location of these genes on the second chromosome along with a summary of the results obtained previously for the third chromosome (Harding et al., 1995).

We measured the strength of the interaction with Dfd using two mutant alleles for each modifier gene. Outcrossed chromo-somes carrying the relevant mutation were placed into the Dfd hypomorphic background; from this cross we counted all the flies that carried the genotypes */+; Dfd^rV8/Dfd^rC11 (where * rep-resents any given mutation) and +/+; Dfd^rV8/Dfd^rC11. The strength of the genetic interaction between Dfd and a given modifier mutation was calculated as the ratio of these two numbers (0 = maximum interaction, 1 = no interaction). Mutant alleles in vkg scored as strong interactors (ratio of approx. 0.2). The apt, kis and poc mutations are moderate interactors (ratios of 0.4-0.6), while the Mhc and snt mutations are weak interac-tors (ratios of 0.6-0.8). Finally, one allele of Ca-alpha1 tested as a strong interactor and the other allele as a moderate interactor.

Although the screen was designed to identify genes required for Dfd wild-type function, many of these genes might also be required more generally for the function or transcriptional regulation of other Hox genes. Mutations in some general regulators would be expected to fall into the trx group class of genes and therefore suppress the T2 to T1 transformation seen in Pc heterozygotes. Consequently we tested two alleles of each gene identified in this screen for interaction with Pc⁴.

We crossed males carrying either an interacting mutation or the relevant control chromosome to Pc⁴/TM3 females and counted the number of sex comb teeth in the T2 legs of the male progeny with the genotype */+; Pc⁴/+. In wild-type males, sex combs are confined to T1 legs and have an average of ten sex comb teeth per leg. In the conditions used in our assay, +/+; Pc⁴/+ flies (control flies) show ectopic sex combs in T2 with approximately 90% penetrance and an average of three sex comb teeth per leg. We determined the average number of teeth per leg for each of the mutants and controls. The numbers obtained for the mutants was expressed as a fraction of the appropriate controls. Thus, a ratio of 1.0 corresponds to no deviation from the control; ratios between 0 and 1.0 represent suppression; ratios greater than 1.0, enhancement.

The two kis alleles that we tested show a level of Pc suppression comparable to that previously reported for other alleles (approx. 0.1 of controls; Kennison and Tamkun, 1988). In addition, mutations in the novel gene snt appear to weakly suppress Pc (0.45 of controls). The rest of the mutations exhibited no significant interaction with Pc.

Two of the interacting complementation groups on the second chromosome are adjacent genes (viking and DCg1; RodrÍguez et al., 1996; Yasothornsrikul et al., 1997), both coding for collagen IV family proteins. We found that some of the mutant alleles in these genes are lethal when heterozygous in combi-nation with mutant alleles of another Dfd modifier gene, LanA, which maps to the third chromosome and encodes a Drosophila laminin α-5 isoform (Henchcliffe et al., 1993); our alleles of LanA were formerly designated as headline in Harding et al. (1995). Based on their complementation behavior, our vkg alleles can be divided into three classes (Table 1). Class 1 alleles (e.g. vkg³⁰) complement all DCg1 alleles, but fail to complement LanA alleles. Class 2 alleles (e.g. vkg³¹⁹) com-plement LanA but fail to complement most DCg1 alleles. Finally, vkg¹⁷⁷ and vkg²³² fail to complement both DCg1 and LanA mutations. Our alleles of DCg1 can be classified in a similar way. DCg1⁴¹² fails to complement all of our LanA mutants and most of the vkg ones. DCg1²³⁴ fails to comple-ment vkg but not LanA. DCg1⁴¹¹ and DCg1⁴⁴ complement LanA and fail to complement a few vkg alleles. Both collagen IV and laminin are interacting components of the basal lamina, a specialization of the extracellular matrix that lines most tissues and has been implicated in a variety of developmental processes (see Discussion).

Embryos mutant for any of the three novel genes identified in the screen display head defects that share similarities to the defects observed in Dfd mutant embryos. In the head of poc mutants, the anterior portions of the lateral bars of the H-piece are missing or severely distorted (Fig. 3D). These sclerites are of maxillary origin and depend on Dfd function (Merrill et al., 1987; Regulski et al., 1987); they probably serve the role of mechanically bridging the cephalopharyn-geal plates with the chewing apparatus. poc mutants also develop ectopic sclerotic patches of apparent head cuticle in ventral and ventro-lateral thoracic regions. The shape and distribution of the ectopic sclerotic patches vary somewhat among embryos, but they are often highly refractory and dense. In wild-type embryos, sclerites with this appearance are found only in the head, so we believe that the patches in poc mutants correspond to ectopic head tissue. The ectopic sclerites are most pronounced in T2 and T3, where they are located in the ventral naked cuticle near or overlapping the Keilin’s organs. Ectopic sclerites also can be seen in the head (Fig. 3D shows one such patch near the median tooth).

snaggletooth mutants have cuticular defects restricted to derivatives of the clypeolabrum in the anterior embryonic head (Fig. 3E). The median tooth is reduced in size, often misshapen, and points ventrally instead of anteriorly. The epistomal sclerite is malformed and reduced. In addition, supernumerary sclerotic fragments are frequently associated with each of these anterior head structures. The lining of the dorsal pouch is also sometimes abnormally sclerotized, resulting in threads of cuticle connecting the median tooth with the dorsal bridge.

apontic mutants have severe defects in structures derived from the gnathal segments. The lateral bars of the H-piece (maxillary origin) and hypostomal sclerites (labial origin) are missing and the lateralgräten (mandibular origin) are truncated (Fig. 3F). All of these sclerites arise from the ventral portion of the gnathal region; other structures mapping nearby in the embryo anlage are still present (e.g. ectostomal sclerite of maxillary origin). The heads of apt mutant embryos are open anteriorly and the dorsal pouch is severely reduced, replaced in part by sclerotic rubble. The dorsal bridge, which is secreted by the cells lining the dorsal pouch, is broken or missing. The rest of the cephalic sclerites can be identified in the mutant; any slight defects are probably due to the overall damage to the head architecture. The open head phenotype (also referred to as the dorsal pouch syndrome; Jürgens et al., 1986) is common to a number of mutations affecting head development, including mutations in the homeotic genes Scr (Pattatucci et al., 1991; Sato et al., 1985; Wakimoto and Kaufman, 1981) and, to some degree, Dfd. Since the apt mutant phenotype has a drastic effect on Dfd-dependent head structures, we decided to clone and characterize apontic coding sequences.

Seven lethal P-element insertions map near the cytological location (59F) of apontic. These were tested for their ability to complement the lethality of two of our apt alleles (apt⁴¹ and apt¹⁶⁷). One of the P insertions (P-2360) failed to complement both alleles. We isolated chromosomes from which the P-2360 element had precisely excised, and found that these did complement apt mutations, indicating that the P-2360 insertion had interrupted the function of the apt locus. We will hereafter refer to P-2360 as the apt^PZ mutant allele.

We extracted total genomic DNA from apt^PZ flies and isolated a fragment of approx. 500 bp adjacent to the P insertion by plasmid rescue. The 500 bp fragment was used as a starting point to isolate genomic clones from an iso-1 Drosophila library. A 1.2 kb NdeI-BamHI genomic fragment immediately adjacent to the insertion site of the P element hybridized to a 2.8 kb band on a poly(A)+ northern blot. A cDNA library was screened, and eight cloned copies of this RNA were isolated. All had identical restriction maps containing inserts of the expected size of 2.8 kb. When the entire cDNA was used as a probe, it hybridized to a single band in a developmental northern blot from embryonic mRNA (Fig. 4B).

The genomic organization of the putative apt transcription unit is shown in Fig. 4A. The first and second exons are separated by an intron of approximately 14 kb; all other introns are each less than 100 bp. The first exon contains the initiation codon of the longest putative open reading frame; the remainder of the coding sequence spans the other four exons. In order to confirm that this transcript encodes the apt function, we PCR-amplified coding sequences corresponding to the 2.8 kb RNA from the apt⁴¹ and apt¹⁶⁷ mutant chromosomes as well as from the parental wild-type chromosome. After obtaining and comparing the sequences of these DNAs, we found that apt⁴¹ and apt¹⁶⁷both have missense mutations in the third exon, shown in Fig. 4A. The mutation in apt⁴¹ DNA is a G→A transition, which would change Arginine 55 to Cysteine (R55C). The mutation in apt¹⁶⁷ DNA is a G→A transition, which would change Arginine 110 to Histidine (R110H). These results confirm that the 484-amino-acid ORF in the 2.8 kb mRNA encodes the apt function. The predicted APT protein sequence is shown in Fig. 4C. No significant similarities to known proteins were found in database searches, although it contains clusters of basic amino acid sequence that are often found in transcription factors. Both mutational changes occur within a short repeat of amino acid sequence of Acidic-Arg-Thr-X-Asn.

We performed in situ hybridizations to whole mount embryos using an RNA antisense probe from the full length apt cDNA. Transcripts from apt are expressed in a complex and dynamic pattern that includes, but is not limited to, the head regions affected in apt mutants. At the syncytial blastoderm stage (Fig. 5A) maternal apt transcripts are distributed throughout the embryo. At cellular blastoderm, apt transcripts can be detected at both poles of the embryos (Fig. 5B). During germ band elongation (Fig. 5C,D) most of the posterior expression disappears and a segmentally repeated pattern of expression arises in the trunk. apt transcripts in the trunk region are detected in cells along the ventral midline, in dorsal cells abutting the amnioserosa, and in the tracheal placodes (cells that will invaginate to form the respiratory tree). In the head of stage 9-10 embryos, apt is expressed in the dorsal part of the acron, the entire clypeolabrum and the ventral gnathal region (see Fig. 5D for a ventral view). Transcripts can also be detected in the anterior-lateral regions of the mandibular, maxillary and labial lobes. During stage 11, apt transcripts accumulate at high levels in the dorsal ridge (small arrow in Fig. 5E), and at this stage, transcript levels gradually disappear from the vicinity of the tracheal pits (Fig. 5D). As the amnioserosa is absorbed and dorsal closure ensues, dorsal mesodermal cells arranged in a single row on either side of the amnioserosa accumulate apt transcripts (Fig. 5E, large arrow); later, these cells, still expressing apt transcripts, will contribute to the dorsal vessel. The patterned expression of apt also persists to late stages of embryogenesis in head epidermis and CNS (Fig. 5F).

apontic is required for the formation of some but not all Dfd- and Scr-dependent structures (see Fig. 6). The lateral arms of the H-piece are missing and the lateralgräten are shortened in Dfd mutants (Merrill et al., 1987; Regulski et al., 1987); the hypostomal sclerites and dorsal pouch are missing in Scr mutants (Pattatucci et al., 1991; Sato et al., 1985; Wakimoto and Kaufman, 1981). Other Dfd- and Scr-dependent structures are intact in apt mutants, such as the ectostomal sclerites (Dfd-dependent), mouth hooks (Dfd-dependent) and cross bar of the H-piece (Scr-dependent). Thus, the apt phenotype suggests that apt might be contributing to the diversification of Dfd and Scr function in a specific cell population within each selector’s domain.

Such overlap in phenotypes could occur in principle by three different mechanisms. apt could mediate Dfd and Scr functions by regulating their transcription in a particular region (i.e. apt acts upstream), apt could be a target of Dfd and Scr in a discrete population of cells (apt acts downstream), or apt could act in conjunction with Dfd and Scr (or with Dfd and Scr targets) to produce a distinct biological effect in a subpopulation of cells (apt acts in parallel).

We performed whole mount in situ hybridizations of apt mutant embryos with Dfd and Scr RNA probes and found their patterns of transcription to be indistinguishable from wild type (data not shown). Therefore, apt is not required to establish or maintain Dfd and Scr transcription. Conversely, we looked at apt transcription in Dfd and Scr mutants by in situ hybridiz-ation and detected no changes (data not shown). Thus, neither Dfd nor Scr is required to establish or maintain apt transcrip-tion. We conclude that apt acts in parallel with Dfd and Scr proteins to produce ventral gnathal structures.

These studies identify eight genes on the second chromosome whose dose is crucial for the function of the homeotic gene Dfd. One of them, apt, is required for structures dependent on both Dfd and Scr selectors during embryogenesis (see Fig. 6). We focused our attention on its characterization.

Head defects in apt mutants can be ascribed to the loss of apt function in two distinct cell populations, one dorsal and one ventral. Dorsally, apt mutants display an open head and a malformation or loss of dorsal pouch derivatives, such as the dorsal bridge. The dorsal pouch forms when the acron is enveloped by the dorsal ridge cells during head involution. Within the dorsal pouch, the acron forms the ventral wall of the pouch and the cells derived from the dorsal ridge form the dorsal wall (Schoeller, 1964; Turner and Mahowald, 1979). Both the acron and the dorsal ridge express apt transcripts. Dfd is expressed in the anterior part of the dorsal ridge and Scr in the posterior part (Gorman and Kaufman, 1995; McGinnis et al., 1990; Sato et al., 1985). Scr null mutations are also asso-ciated with open head and disruption of the dorsal pouch. In Dfd mutants, these structures are not consistently abnormal, suggesting either that Dfd is not required for dorsal ridge migration, or that the anterior part of the dorsal ridge is not necessary for head involution and dorsal pouch formation. In any case, the dorsal pouch phenotype in both Scr and apt mutations might be due to their loss of function in the dorsal ridge.

In the ventral head, apt mutants lack three sclerites derived from the most ventral aspect of mandibular (Dfd-dependent), maxillary (Dfd-dependent) and labial (Scr-dependent) lobes (see Fig. 6). Other structures derived from these lobes are intact, e.g. ectostomal sclerites (maxillary) and the cross bar of the H-piece (labial). The prothorax, where Scr is also required, appears normal in apt mutants.

These results are consistent with apt being part of Dfd and Scr pathways only in certain cells.

Within these cells, apt function appears to act in parallel to the Hox genes, since apt seems neither to regulate nor to be regulated by Hox gene functions. There are many mechanisms by which this parallel role in potentiating Hox function might be accomplished. The most direct mechanism invokes apt as a cofactor required for Hox target element activity in the ventral- and dorsal-most portions of the gnathal segments. One group of Hox target elements that apt does not detectably influence is Dfd autoactivation elements, since Dfd transcription in apt mutants is normal. Thus even in the region of overlap, apt is not required for all of Hox functions since the activity of Dfd autoregulatory enhancers remains intact.

A number of genes have been implicated in aiding Hox fate determination in the gnathal region. Dfd is expressed in both maxillary and mandibular segments; the cnc gene is expressed in mandibular and labral segments. It has been suggested that Dfd requires the action of the cnc gene product to elicit a mandibular fate whereas without cnc, it determines maxillary structures (Mohler et al., 1995; Harding et al., 1995). Scr is expressed in and required for the development of labial and prothoracic segments. In mutants for the sal gene, Jürgens (1988) observed defects consistent with a transformation from labium to prothorax, suggesting that sal may act with Scr to specify labial identity in a combinatorial manner similar to that of Dfd and cnc. It is possible that apt acts in combination with Hox and other homeotic genes to specify structures on the dorsal-ventral axis of the gnathal segments. For instance, Dfd, cnc and apt may act together to give rise to lateralgräten, Dfd and apt to specify lateral bars of the H-piece, and Scr, sal and apt to specify the hypostomal sclerites.

Variations in pathways controlled by a given Hox gene may be ascribed to differences in target choice, to a differential action on targets, or to differential inputs from other pattern-ing genes further downstream in the pathway. HOX functional specificity (or that of a target) can in principle be changed in different regions or cells by exposing HOX (or target) proteins to different sets of specificity cofactors. HOX specificity could also be regulated by chemical modification of ubiquitous factors through transduction of a localized signal. In principle, apt could be acting at any of these levels. To date, the only protein shown to act as a direct cofactor of HOX proteins is EXD, which has been implicated in aiding HOX proteins at two levels: discrimination between target and non-target genes, and whether a target gene is activated or repressed (Mann, 1995; Pinsonneault et al., 1997). We are currently testing the subcellular localization of APT protein by immunolocalization, and attempting to identify common targets of apt and Dfd or Scr.

Although the biochemical function encoded by apt is still unknown, the gene has an intriguing expression pattern. Most or all of the cells where the gene is expressed undergo migra-tions in a coherent manner: the dorsal ridge extends over the acron, ventral gnathal cells migrate into the stomodeum to form the cibarial plate, tracheal pits invaginate and form the tracheal tree, during dorsal closure the dorsal mesoderm migrates to form the heart primordia. There is no evidence that apt is directly required for the migration process, but apt function might be required for many of these tissues to properly differ-entiate once they reach their final destinations. More analysis of the apt mutant phenotypes in internal structures will allow us to address this interesting possibility.

Both apt and dfc (isolated as a Dfd interactor on the third chromosome, see Fig. 2) appear to act in parallel to one or a few Hox genes, but their mutant phenotypes do not involve homeotic transformations. Another group of Dfd interacting genes, which includes cnc, exd and poc, also act in parallel but do mutate to homeotic phenotypes. It is not clear whether these two groups of genes are functionally distinct, as exd could be placed in either the first or second group, depending on whether its zygotic or maternal/zygotic phenotype is used for classification.

Two of the mutations isolated in the screen, kis and snt, are strong and weak suppressors, respectively, of Pc in genetic tests. Since trx and brm mutants suppress Pc haploinsufficient adult phenotypes, it has been suspected that many of the other genes that suppress Pc might also encode proteins that directly affect transcription, possibly through regulation of chromatin architecture. However, Pc group functions may also be regulated by other mechanisms besides direct antagonism of their repressive capabilities. The trx group of genes is likely to be a collection of disparate factors able to modulate homeotic transcription and function at many different levels. For example, mutations in hh (which encodes a secreted signal) act as Pc suppressors in genetic tests, but it is highly unlikely that hh would affect transcription directly. snt zygotic loss of function leads to mild defects restricted to labral structures; this gene may prove to be a part of the poorly understood web of interactions leading to determination of the anterior head.

The dosage of Mhc, Ca-alpha1, viking and DCg1 is also critical for Dfd function. The known biological functions of the molecules encoded by these genes is consistent with activities that are either downstream of homeotic selectors or required to mediate Hox function as crucial components of signaling pathways. None of these mutations cause cuticular defects in first instar larvae that are reminiscent of those seen in Hox mutants. We therefore believe that their interactive relationship with the Hox genes is likely to take place in soft tissues (particularly the mesodermal derivatives), and/or at developmental stages after embryogenesis. It is also possible that some of these genes may have a maternal contribution that masks their embryonic loss-of-function phenotypes.

Little is known about Ca-alpha1, other than it is expressed in the nervous system and is homologous to the rat brain type D calcium channel α₁ subunit (Zheng et al., 1995). The action of Hox genes in the patterning of the nervous system is well documented. Dfd is expressed in embryonic and larval sub-esophageal ganglia, and adult mutants for hypomorphic Dfd alleles often show motor defects: they have difficulties extending their proboscides and feeding (Merrill et al., 1987) or moving their maxillary palps (Restifo and Merrill, 1994). It is possible that Ca-alpha1 is a tissue-specific target gene, one of the terminal products of a developmental event in the CNS, or that it acts in a signalling pathway required for Dfd expression or function in the nervous system.

Mhc is expressed only in muscle cells; it is a large gene containing numerous alternative exons; the expression of particu-lar isoforms is under developmental regulation (Bernstein et al., 1993). Muscle myosin is required for muscle function and to date is not known to mediate any developmental process. It seems, therefore, that Mhc lies near the end of developmental pathways. It is known that Hox genes are responsible for imparting segmental identity to muscle fibers (Bate, 1993). They do so by regulating many steps in muscle development, including determination of myoblasts, migration and fusion (Fernandes et al., 1994). It is plausible that different Hox genes might promote structurally different muscle types in different regions by indirectly regulating the alternative splicing of Mhc transcripts. We have not examined the integrity of muscle fibers in Dfd⁻ embryos, but Restifo and Merrill (1994) have reported that Dfd^rV8 (one of the alleles used in this screen) hemizygotes show a reduction or absence of the retractors of the rostrum, a pair of head-specific adult muscles. The motor defects seen in Dfd mutants could be due to muscular malfunction or lack of differentiation. It is also possible that the genetic interaction could be highly indirect, as there are examples of HOX proteins having an effect on muscle morphology in a non-cell-autonomous way. Fernandes et al. (1994) demonstrated that the segment-specific determination of myoblasts in the wing discs and their subsequent segment-specific migratory patterns in T3 are determined by the action of Ubx in the epidermis, whereas their fusion and innervation requires Ubx in a cell-autonomous manner.

Induction phenomena across germ layers involve several signalling modes, and cell populations often separated by the basal lamina. The basal lamina is a complex sheet-like assemblage lining many tissues (it covers muscle fibers, separates epithelia from connective tissue and surrounds most organs). Both vkg and DCg1 encode collagen type IV molecules, found exclusively in basal lamina. Our screen for Dfd interactors on the third chromosome (Harding et al., 1995) identified four mutant alleles in LanA, a gene encoding one of the subunits of laminin (Henchcliffe et al., 1993), another major component of the basal lamina. In this report we have shown that the EMS-induced mutations that we isolated in each of the collagen genes can be classified as interacting with mutations in either the other collagen IV gene, in the laminin α-5 gene, or both. P-element-induced mutations in the same collagen genes do not display such interactions. Such extragenic lethal non-complementation is consistent with combinations of truncated proteins being capable of ‘poisoning’ the higher order basal lamina assembly. Alternatively, it is possible that the alleles isolated in this screen affect regions within the basal lamina components that are devoted to specific interactions, perhaps ones required for Hox-dependent morphogenetic signals.

There is evidence that the basal lamina plays a crucial role in muscle differentiation or morphogenesis. Drosophila LanA mutants have defects in heart and somatic musculature; in tissue culture, laminin has been shown to promote myoblast proliferation, fusion and myotube formation (Gullberg et al., 1994; Vachon et al., 1996). A detailed analysis of collagen IV mutant phenotypes has not yet been done, but overexpression of truncated proteins and antisense RNA leads to defects in muscle structure and attachment (Borchiellini et al., 1996). The basal lamina has been implicated in other developmental phenomena as well. For instance, it was recently shown that the ocellar pioneer axons of LanA mutants have pathfinding defects, while bristle axons do not seem to need laminin to find their targets (GarcÍa-Alonso et al., 1996). Thus, there are many possible ways in which basal lamina could contribute to mor-phogenetic processes under Hox regulation, but the current data does not allow us to determine which is the most likely scenario.

We thank Diane Inglis, Aarron Willingham and Tobey Tam for help with complementation crosses and cuticular preparations; Elizabeth Wiellette and Alexey Veraksa provided useful comments on the man-uscript. We are also indebted to the Bloomington Stock Center for providing us with innumerable lines and Fly Base for much useful information. This research was supported by a grant from the NIH (HD30368).