Head versus trunk patterning in the Drosophila embryo; collier requirement for formation of the intercalary segment

Extensive studies on the development of the Drosophila embryo have provided deep insight into the mechanisms determining the body pattern. An intricate cascade of hierarchical, cross-and autoregulatory interactions initiated by maternal morphogens, followed by activation of gap, pair rule and segment polarity genes, serially subdivides the trunk into reiterated units. Segment identity is conferred by the regionalised expression of homeotic selector genes from the Antennapedia (ANTC) and Bithorax (BXC) complexes, activated by combined action of gap and pair rule genes (reviews by Ingham, 1988, Pankratz and Jäckle, 1993). The number of segments composing the insect head has itself long been a matter of debate due to the diversity of form between, and within different taxa (Rempel, 1975). The pattern of expression of the segment polarity gene engrailed (en) during embryonic head formation in different insect orders revealed six Engrailed-accumulating segments, posterior to the labrum (pre-antennal region): from anterior to posterior, two preoral – the ocular and antennal – segments and four postoral – the intercalary and gnathal (mandibular, maxillary and labial) – segments (Rogers and Kaufman, 1997). The mutant phenotypes and patterns of expression of the head gap-like genes orthodenticle (otd), empty spiracles (ems) and buttonhead (btd) (Dalton et al., 1989; Finkelstein and Perrimon, 1990; Wimmer et al., 1993) led Cohen and Jürgens (1990) and Finkelstein and Perrimon (1991) to propose that a combinatorial input of head gap genes could be directly involved in partitioning the head anlage into a fixed number of segmental units with no need for second-level regulators, equivalent to pair rule genes in the trunk. The lack of an obvious homeotic phenotype associated with mutations in either Deformed (Dfd) or labial (lab) which are expressed in the maxillary and mandibular, and intercalary primordia respectively (Merrill et al., 1989; Mahaffey et al., 1989), led to the further suggestion that the combinatorial expression of head gap genes might also assign segmental identities (Cohen and Jürgens, 1990). This model was recently questioned, however, by the results of inducible ectopic expression of either btd (Wimmer et al., 1997) or otd (Gallinato-Mendel and Finkelstein, 1998), which affected neither the head segment identities nor, or only marginally the pattern of segment polarity gene expression, suggesting that other mechanisms and factors were involved.

We report here the isolation and phenotypic analysis of mutations in collier (col), a gene specifically activated at the blastoderm stage in a region overlapping both a (cycle 14) mitotic domain, MD2, and a parasegment (PS0: posterior region of the intercalary segment and anterior region of the mandibular segment) (Crozatier et al., 1996). We show that col activity is required for expression of the segment polarity genes hedgehog (hh), en, and wingless (wg), and the segment-identity gene cap and collar (cnc; Mohler et al., 1995) in the intercalary segment. The parasegmental register of col activation in the head at the blastoderm stage integrates inputs both from the head and trunk segmentation systems. Together, our results indicate that col is a segment-specific patterning gene acting downstream of head-gap genes and required for establishing the PS(-1)/PS0 parasegmental border and formation of the intercalary segment.

The bcd^E1, Dfd^A325, dl¹³¹⁰, ems^9Q84, eve^R13, gt^YB, lab⁴, prd^IIB, prd^6L07, otd^YHB, sal^IIA, sal^IIB, sna^II6, stg^4B51 and stg^7B69 strains were obtained from the Tübingen Stock Center (Tearle and Nüsslein-Volhard, 1987). The slp^Δ34 deficiency was from U. Grossniklaus and W. Gehring, Basel, the btd^XG81and btd^XA strains from S. Cohen, Heidelberg, the cnc^P2, cnc^VL70 and cnc^VL110 strains from J. Mohler, New York and the croc^5F59 strain from H. Jäckle, Göttingen. The D(f80850) deficiency were obtained from the Umea Stock Center.

To identify lethal mutations in the col chromosomal region (51C1-C3; see Fig. 1), we first generated a deficiency by removing this region. X-ray treatment was applied to the (Pw+, AN34) strain, which contains a (Pw+, lacZ) insert located at 51C and was provided by A. Nose. Irradiated males were mated to Tft/CyO virgin females. The F₁ progeny bearing white eyes were individually mated to Tft/CyO-twist-lacZ flies to establish P*/CyO-twist-lacZ stocks. A subsequent screen for loss of col expression by in situ hybridisation on whole-mount embryos with a mixture of lacZ and col probes allowed chromosomal rearrangements where col is not transcribed to be identified. From a screening of 2×10⁵ irradiated chromosomes, a single deficiency was recovered, and designated Df(2R)AN293. We then used EMS as a mutagen and selected chromosomes that were lethal over Df(2R)AN293 and viable over Df80850 (51C1-51C3 region, Fig. 1A). From 6300 individual chromosomes tested, 50 mutations were isolated, representing 7 independent complementation groups.

Genomic DNA was isolated from hand-selected homozygote col¹ mutant larvae to act as a matrix in PCR amplification experiments. The PCR amplification products were directly sequenced.

The P[col5-lacZ] and Hscol constructs have been described by Crozatier and Vincent (1999). P[col5-cDNA] was constructed by subsituting the col cDNA for lacZ in P[col5-lacZ]. Males homozygousfor the Hscol transgene were crossed with w⁻ females, with w⁻ males used as controls. The embryos resulting from these crosses were allowed to develop for 3 hours at 25°C, before heat-shock treatment for 45 minutes at 37°C. Development was continued at 25°C until the embryos formed cuticles.

Whole-mount in situ hybridisation to embryos, double labelling using several probes simultaneously, or double immunostaining and in situ hybridisation were performed as described by Crozatier et al. (1996), and references therein. RNA probes were synthesised from cDNA plasmids except for the col intronic probe (Crozatier and Vincent, 1999). Primary antibodies were used at the following dilutions: rabbit anti-phospho histone H3 from Upstate Biotechnology (2 μg/ml); home-produced rabbit anti-En (1/400), rabbit anti-Dfd (1/250), rabbit anti-Lb (1/100), mouse monoclonal anti-Wg (1/200). The antibody against Wg was kindly provided by S. Cohen, Heidelberg.

The col gene maps to the chromosomal region 51C1,2 (Fig. 1A). In order to establish its molecular organisation, we isolated approx. 45 kb of overlapping genomic DNA covering the col transcription unit and sequenced the relevant regions. The col transcription unit consists of 12 exons and 11 introns spanning a genomic region of about 30 kb (Fig. 1B). Introns separate the coding regions for each Col functional domain, defined by biochemical dissection of EBF and sequence conservation during evolution (Hagman et al., 1993; Crozatier et al., 1996). These are the DNA binding domain (aa 59 to 288), the homodimerisation domain (aa 289 to 429), and a putative transactivation domain at its carboxy-terminal end (Fig. 1C). However, additional introns split the Col DNA binding and homodimerisation domains, despite their extensive primary sequence conservation in all COE proteins identified so far, from nematode to vertebrates (Bally-Cuif et al., 1998, for ref.). Within the homodimerisation domain, the helix-loop-helix (HLH) motif is encoded by a single exon, exon 9 (Fig. 1C). Finally, the genomic structure of col indicates that the two predicted Col embryonic protein isoforms, which differ in their carboxy-terminal protein coding region (Crozatier et al., 1996), result from alternative splicing of exon 11.

In the absence of any known deficiency removing the col gene, we generated one by X-ray irradiation, using as a marker a P[w+, lacZ] insertion located at 51C (see Material and Methods). Approximately 2×10⁵ irradiated chromosomes were first screened for loss of the w⁺ marker. By searching for associated loss of col expression by in situ hybridization to col mRNA on whole-mount embryos, we were able to recover a deficiency removing the col gene, deficiency Df(2R)AN293 (Fig. 1A). A subsequent screen for EMS-induced lethal mutations in the 51C1-C3 interval defined by Df(2R)AN293 and Df(2R)80850 (Fig. 1A) led to the isolation of 50 mutations representing 7 different complementation groups. As a first step to identify col mutations, cuticles from hemizygous mutant embryos were examined for head skeleton defects. All three independent alleles of one complementation group (designated as col¹, col² and col³) displayed lethality at the late embryonic/first instar larval stage with specific defects in the head skeleton similar to those observed upon heat-shock induced expression of col antisense RNA in early gastrula embryos (Fig. 2A,B; Crozatier et al., 1996). These defects are a complete lack of the ventral arms (VA) and a strong reduction of the lateral gräten (LG), two structures thought to be derived from the intercalary/mandibular segment anlagen (Jürgens et al., 1986). The T-ribs in the floor of the pharynx and the antennal, maxillary or hypopharyngeal sensory organs (see Jürgens et al., 1986) are present, and a normal pattern of internal sensory structures, detected by 22C10 antibody staining (Schmidt-Ott et al., 1994), is observed (data not shown). The mutant embryos that hatch give rise to larvae that do not grow and tend to crawl out of the medium, suggesting that they are unable to feed. That these defects result from col mutations was further substantiated by the lack of a somatic dorsal muscle in which col is specifically expressed (Crozatier and Vincent, 1999). The same head and muscle phenotypes are observed for the 3 alleles in homozygous or hemizygous combinations, suggesting that they are strong hypomorphic or null alleles. Further support was provided by sequence analysis of the col¹ allele, where a G to A transition (amino-acid position 228) eliminates a splice acceptor site. This lesion should result in the non-removal of intron 6 and the production of a truncated Col protein (Fig. 1C); indeed, no Col protein can be detected in col¹ mutant embryos using anti-Col polyclonal antibodies directed against the divergent carboxy-terminal end of the protein (data not shown, and Crozatier and Vincent, 1999).

In order to confirm that our 3 putative col mutations were indeed affecting the col gene, we undertook to rescue this phenotype using a col transgene. The cis-acting regulatory region responsible for col expression in the head was identified by generating reporter transgenes expressing a Col/β-galactosidase fusion protein under the control of various col genomic DNA fragments. A reporter gene containing 5 kb of col upstream DNA (P[col5-lacZ], see Fig. 1B) showed a pattern of lacZ transcription that faithfully reproduced col transcription in MD2/PS0 from early cycle 14 (stage 5) up to stage 11 (see Fig. 7A and data not shown), and in the embryonic mesoderm (Crozatier and Vincent, 1999). We used this 5 kb upstream region to drive expression of a col cDNA in phenotypic rescue experiments (P[col5-cDNA] construct). A single transgenic copy of this construct rescued the embryonic/ larval lethality of col¹, col² or col³ hemizygous embryos: about 80% of the rescued embryos hatched into wandering larvae and reached adulthood. Complete rescue of the head skeleton defects was observed in the cuticules of such embryos, confirming that only col function was affected in these embryos.

The embryonic head phenotype of col¹ hemizygous mutant embryos indicates a loss of skeletal structures derived from the intercalary, and possibly mandibular, segments (Fig. 2A-B; Jürgens et al., 1986) without transformation towards another segment identity. To investigate this segmentation phenotype in more detail, we first compared col expression with that of the segment polarity genes hh and wg. At the blastoderm stage, the posterior limit of col expression is parasegmental (PS0/PS1), as it precisely abuts the mandibular stripe of hh-expressing cells (Fig. 3A). Whether its anterior limit is also parasegmental cannot be answered at this stage because the expression of segment polarity genes in pre-gnathal segments is not yet established. Examination of early stage 11 embryos shows that col expression overlaps the intercalary hh stripe and abuts the intercalary Wg spot, indicating a parasegmental anterior border for col expression. At this stage however, col expression has been lost from the posterior part of PS0, as it does not overlap mandibular Wg expression (Fig. 3C,D). The cnc gene, which codes for a b-ZIP transcription factor, has been postulated to act as a segment identity gene in the mandibular segment (Mohler, 1993; McGinnis et al., 1998). Consistent with col being expressed in PS0 col and cnc expression only partly overlap, in the region corresponding to the anterior mandibular segment (Fig. 3B). Together, these data indicate a parasegmental register of col expression at the blastoderm stage, which is subsequently restricted to anterior PS0.

We then determined whether col mutations affect the expression of wg and En, which mark the anterior and posterior compartments of each segment, respectively (Schmidt-Ott and Technau, 1992). In col¹ hemizygous embryos, both the intercalary stripe of En and the spot of wg expression are missing (Fig. 4A). Since col expression does not overlap the intercalary Wg spot (Fig. 3D), the loss of this spot in col mutant embryos suggested that col does not regulate wg expression directly but possibly by an hh-dependent mechanism (Gallinato-Mendel and Finkelstein, 1997). We indeed found that in col mutant embryos, the intercalary stripe of hh is also absent, or much reduced (Fig. 4A). Together, these results show that col controls hh, en and wg expression in the intercalary segment and is required for establishing the PS(-1)/PS0 parasegmental border. The VA and LG, which are missing and reduced, respectively, in col mutant embryos, are also affected in two other head mutants, crocodile (croc), which codes for a forkhead-domain protein (Häcker et al., 1995) and cnc. These structures are also affected in embryos mutant for the homeotic genes Dfd and lab, which are expressed in the mandibular and maxillary segments, and in the intercalary segment, respectively (Diederich et al., 1991; review by Rogers and Kaufman, 1997). We first looked at col expression in embryos mutant for croc, cnc, Dfd or lab. In none of these embryos did we observe a change in col transcription. Conversely, no changes could be detected for croc, Dfd or Lb expression in col¹ hemizygous embryos, indicating that expression of each of these three genes is independent of col (not shown). In contrast, col is required for cnc transcription in the posterior intercalary segment at stage 9-10 (Fig. 4B). Because this region is anterior to the region of overlap between col and cnc expression at the blastoderm stage (Fig. 3B), we conclude that it corresponds to a secondary site of cnc expression initiated at stage 9, under control of col activity. In cnc mutant embryos, intercalary hh expression is normal (data not shown), indicating that hh and cnc are regulated by col independently of each other.

The normal pattern of Lab expression and morphology of the intercalary lobe (sometimes referred to as hypopharyngeal lobe; Rogers and Kaufman, 1997) at stage 11 indicated that lack of hh and cnc expression in col¹ hemizygous embryos was probably not linked to cell death. In order to address this question more directly, we looked at the distribution of reaper (rpr) mRNA, which marks cells fated to undergo apoptosis (White et al., 1993). In wild-type embryos at stage 11, there are two invariant sites of rpr expression: inside the epidermal layer of the gnathal region where the primordium of the salivary gland invaginates, and near the caudal tip of the extended germ band (Fig. 4C; Abrams et al., 1993). In col¹ embryos, an additional site of rpr expression is observed, which does not correspond to cells of the intercalary lobe but rather to cells of the hypopharyngial epithelium (floor of the forming pharynx). These cells have been previously shown to derive from the ventral side of the intercalary segment primordium (Jürgens and Hartenstein, 1993) and lineage tracing analysis has linked them to MD2 (Cambridge et al., 1997). This specific cell death is consistent both with the domain of col expression at the onset of gastrulation and the head skeleton phenotype of col mutant larvae.

col expression is first detected during the interphase of mitotic cycle 14 (stage 6), when expression of head-gap genes has already resolved from initial broad domains into defined stripes. The stripe of col expression is included in that of btd, overlaps that of ems, and is restricted both dorsally and ventrally to neuroectodermal cells (Crozatier et al., 1996 and Fig. 5A). Examination of dorsal (dl) mutant embryos showed that Dl is required for col repression in the mesodermal plate. The ectopic expression of col observed in twist (twi) and snail (sna) mutant embryos suggests that Dl target genes (review by Rushlow and Arora, 1990), rather than Dl itself, are involved (Fig. 5B and data not shown). Embryos lacking ems function also show a ventral derepression of col expression. Further, at stage 10, ems mutant embryos show an abnormal pattern of col mRNA accumulation, with a mandibular in addition to intercalary stripe of col-expressing cells (Fig. 5C). This suggests a second role for ems in regulating col. In btd mutant embryos, there is a complete loss of col expression, whereas there is no change in embryos lacking both slp (slp1 and slp2) genes (data not shown), consistent with previous data establishing that btd but not slp is required for intercalary en and wg expression (Wimmer et al., 1993; Grossniklaus et al., 1994). Together, these results confirm that col acts downstream of head gap genes in the transcriptional cascade patterning the head. The fact that col is expressed in a parasegment immediately anterior to the trunk, raised the possibility of a regulation by components of the trunk segmentation system such as pair rule genes. In paired (prd) mutant embryos, col expression starts normally but, soon after gastrulation, becomes restricted to a few cells located anterior to the abnormal cephalic furrow which forms in these embryos (Sander, 1980; Fig. 5C). col mis-expression might thus be a rather indirect effect of the prd mutation. In addition to prd, the only pair rule gene whose mutations appear to affect col expression is eve.

At the beginning of cycle 14, the stripe of col expression is located immediately anterior to Even-skipped (Eve) stripe 1, these two domains becoming separated by a single row of cells during the process of cellularisation (Fig. 6A). We found that col is ectopically expressed in eve mutant embryos, in a region roughly corresponding to PS1, indicating that Eve acts as a repressor of col in this parasegment (Fig. 6B). During the interphase of cycle 14, the broad band of ectopic col activation resolves into a distinctive stripe, separated from the normal PS0 stripe by one to three cells going from ventral to dorsal (Fig. 6B). col expression precisely overlaps the expression of string (stg), in the region prefiguring MD2 (Foe, 1989; Edgar et al., 1994; Crozatier et al., 1996). Expression of stg, which triggers the G₂/M transition, is unchanged in embryos deficient for col and vice versa, arguing that MD2 cells undergo a concerted mitotic and differentiation programme, set upstream of both col and stg (data not shown). This led us to examine whether eve was also involved in defining the position of MD2, using antibodies against the phosphorylated form of histone H3 as a marker of mitosis (Fig. 6C). Like col transcription, MD2 expands posteriorly in eve mutant embryos at early cycle 14 to form a second, ectopic, stripe of mitotic cells at the beginning of gastrulation. Together, our results show that col expression and MD2 position integrate inputs from both the head and trunk segmentation systems, which were previously considered as being essentially independent.

At the onset of gastrulation (stage 7), col is expressed in the entire PS0 (Fig. 3A). At stage 10, the two separate patches of cells which keep expressing col correspond to lateral cells of the intercalary lobe and ventral cells invaginating within the atriopharyngeal cavity, respectively, indicating a restriction of col expression to anterior PS0 (Fig. 5C). To determine when this restriction occurs, we made use of the greater stability of the β-gal protein compared to lacZ (or col) mRNA and compared the two patterns in P[col5-lacZ] embryos. In order to follow col transcription rather than transcript accumulation, we used a col intron probe which labels nascent nuclear transcripts (Crozatier and Vincent, 1999). Before mitosis 14, the lacZ and col mRNA and β-gal protein patterns completely overlap (Figs 5B and 7A, and data not shown). After completion of mitosis 14, all the ectodermal cells derived from the MD2 domain can be visualised by β-galactosidase immunostaining. Only the most anterior cells continue to transcribe col-lacZ (or col) (Fig. 7A). This observation suggests that, at cycle 14, an asymmetric cell division occurs with respect to the maintenance of col transcription. After stage 11, col transcription is only maintained in a subset of the cells of the intercalary lobe. In col mutant embryos, this transcription is lost (Fig. 7B), indicating a direct or indirect, positive auto-regulation.

To investigate further the role of col transcriptional regulation in head morphogenesis, we examined the functional consequences of ubiquitous expression of the Col protein, using hs-col transgenic lines. One 45 minutes heat shock was applied at different times during embryonic development. Whereas heat-shock treatment of hs-col embryos later than 5 hours after egg laying (AEL) has no apparent effect on embryonic development and viability (Crozatier and Vincent, 1999), heat-shock treatment between 3 and 5 hours AEL (stages 6 to 9) causes embryonic death. Cuticle preparations of dead embryos display a normal segmental array of thoracic and abdominal denticles, but a disrupted head skeleton (Fig. 2C and data not shown). The H piece, which originates in part from the maxillary, and partly from the labial segment anlagen (Jürgens et al., 1986), is consistently missing and the LG are also affected. Because col activity is required for maintaining its own expression in the head ectoderm (Fig. 7B), we asked whether ubiquitous expression of the Col protein was altering endogenous col expression. In situ hybridization to hs-col embryos with a col intron probe which allows hs-col and endogenous col transcripts to be distinguished, revealed that col transcription is ectopically activated in mandibular cells (Fig. 7C). Whether this ectopic col transcription results from activation and/or maintenance in all MD2-derived cells after mitosis 14 remains to be established. No specific change in the expression pattern of En, cnc, Dfd and Lab could be detected, however, suggesting that ectopic activation of other Col targets is responsible for the induced phenotype.

The col gene was first suggested to be a second level regulator of head patterning, in the light of its expression pattern and the phenocopy produced by expressing antisense RNA (Crozatier et al., 1996). We report here the analysis of col mutations, which shows that col is a segment-specific patterning gene required for establishing the PS(-1)/PS0 parasegmental border and formation of the intercalary segment.

The extensive amino acid sequence similarity between Col and rodent EBF/Olf-1 defined a new family of transcription factors highly conserved among metazoans (Crozatier et al., 1996 and references therein). The subsequent characterisation of mouse EBF paralogs and closely related orthologs in other vertebrates (see Bally-Cuif et al., 1998) and C. elegans (Prasad et al., 1998) confirmed the high degree of evolutionary conservation of these transcription factors, designated as COE factors. A helix-loop-helix (HLH) motif is the only region of significant homology to other families of transcription factors. The structure of the col gene shows that this motif is encoded by one of the 12 exons; it also reveals that the DNA-binding domain, although sharing between 74% and 92% sequence identity between all COE proteins (Bally-Cuif et al., 1998), is encoded by 6 exons scattered over 15 kb of genomic DNA. The same organisation in 12 exons is found in the orthologous nematode unc3 gene and at least four introns are located at the same position in the DNA binding domain of col and Olf-1 (Prasad et al., 1998). Whether the highly splintered structure of the COE genes is linked to their diversity of functions in different phyla (Lin and Groschedl, 1995; Dubois et al., 1998; Prasad et al., 1998; Crozatier and Vincent, 1999; this report), remains unknown. So far, alternatively spliced exons have been found in Olf-1 (Wang et al., 1997) and col (Crozatier et al., 1996) but it is not clear whether these various splicing forms perform different functions.

We found that a 5 kb col upstream region reproduces expression of a reporter gene in PS0 at the blastoderm stage and a single somatic muscle precursor (muscle DA3; Crozatier and Vincent, 1999), but not in the embryonic PNS and CNS or the wing disc (see below). Correspondingly, P[col5-cDNA] fully rescues the col embryonic head and muscle phenotypes (this report and Crozatier and Vincent, 1999), but it does not rescue other col developmental functions. First, about 20% of rescued embryos do not reach adulthood and the adults that do emerge are weak and have a short life span. Second, these adults have wings lacking the central intervein region (Vervoort et al., 1999). col expression in the wing disc, in a stripe of cells located anterior to the A/P compartment boundary is not reproduced by the P[col5-lacZ] transgene, providing an explanation for this wing phenotype. The isolation of col mutants has, so far, led us to identify three independent Col functions during development, each correlating with a tightly regulated site of expression: establishment of the intercalary segment in the embryonic head, patterning of the central part of the wing, and formation of a specific somatic muscle. col is also expressed in specific subsets of post-mitotic sensory neurons in the head and trunk, and in CNS neurons both in the brain and ventral nerve cord (Crozatier et al., 1996). Mutations of the C. elegans COE gene unc-3 result in defects in the axonal outgrowth of some motor neurons and in dauer formation, a process requiring inputs from specific sensory neurons (Prasad et al., 1998). While a key role of COE proteins in vertebrate neuronal differentiation is also suggested by data obtained in mouse, Xenopus and zebrafish (Garel et al., 1977; Wang et al., 1997; Bally-Cuif et al., 1998; Dubois et al., 1998), the potential functions col expression in the nervous system remain to be explored.

Activation of col expression in PS0 is dependent upon activity of the head-gap gene btd. btd is, however, expressed in a much larger domain than col, both posterior-and anteriorwards, indicating that col expression is restricted by other factors. The factor defining the anterior parasegmental border of col remains unknown. It is probably a target of Bicoid (Bcd), however, since col expression is progressively shifted anteriorwards by gradually decreasing the amount of maternal Bcd by means of sry δ mutations (Payre et al., 1994; data not shown). Posteriorly, the limit of col expression abuts Eve stripe 1. Each of the 7 pair rule stripes of Eve protein foreshadows the position of odd-numbered parasegments (PS1 to 13) (Frasch and Levine, 1987; Lawrence et al., 1987). We have recently shown that btd is required for Eve expression in PS1 (Vincent et al., 1997). We show here that eve, in turn, negatively regulates col expression in this parasegment and defines the posterior limit of the mitotic domain MD2. This cross-talk between the head and trunk segmentation systems thus represents a key component of the transcriptional cascade patterning the head.

Early col expression is repressed ventrally by the mesoderm-determining genes snail and twist, and by ems. Mutations in each of these three genes, causes a premature activation of col in the mesodermal plate. The D/V regulation of col expression differs from that of stg, – MD2 is either narrowed or lacking in ems mutant embryos, and fused in snail but not twist mutant embryos (Edgar et al., 1994) and cnc, whose expression is affected in neither ems nor twist mutant embryos (Mohler, 1993; McGinnis et al., 1998). Thus, different combinations of transacting factors appear to be involved in precisely defining the spatial expression of each of these genes in the head, along both the D/V and A/P axes.

The maintenance of col expression in intercalary cells after stage 11 requires the activity of col itself, suggesting a positive autoregulation. A similar loop operates in the formation of the embryonic muscle DA3 (Crozatier and Vincent, 1999). Further support for a positive autoregulation comes from the activation of endogenous col transcription, following ubiquitous expression of the Col protein. The observation that this activation is restricted to mandibular cells indicates the involvement or another factor cooperating with Col, whose expression is itself limited to these cells.

Recent experiments have questioned the existence of a simple combinatorial code of head-gap genes responsible both for direct activation of segment polarity genes and for assigning segment identity in the anterior head region. First, misexpression of btd in the anterior half of the blastoderm indicated that the spatial limits of btd expression are not instructive (Wimmer et al., 1997). Second, ectopic expression of otd results in variable changes in En and wg expression in different head segments, but does not change segment identity (Gallinato-Mendel and Finkelstein, 1998). The absence of intercalary En and wg expression in col mutant embryos is the first demonstration that activation of segment polarity genes does not result solely from the direct action of cephalic gap genes but involves intermediate regulators (Fig. 8). Contrasting with the trunk, the interactions between segment polarity genes in the head seem to be segment-specific (Gallinato-Mendel and Finkelstein, 1997). col requirement for intercalary wg expression, although the two genes are not expressed in the same cells, indicates that this requirement is non cell-autonomous but probably relayed by Hh signalling (Fig. 8). cnc is another target of Col in the intercalary segment, correlating with the partly overlapping col and cnc head phenotypes (Mohler et al., 1995 and this report). Together, our results enable us to draw up a detailed cascade of transcriptional regulation specifically controlling the formation of the intercalary segment. How to integrate lab into this model and which of these interactions are direct at the molecular level are now issues to be addressed.

We are grateful to the Bloomington, Tübingen and Umea Drosophila Stock Centers for mutant strains. We thank Tom Kaufman and Steve Cohen for gifts of antibodies, Maryvonne Mevel-Ninio for performing X-ray irradiations of flies, Jym Mohler for communication of results prior to publication, and Laure Bally-Cuif, David Cribbs, Marc Haenlin and Julian Smith for their comments on the manuscript. We also wish to thank Claude Ardourel for excellent technical assistance and Anaid Chahinian for some of the in situs. This work was supported by CNRS and Human Science Frontier Organisation. D. V. was supported by a fellowship from CNRS, S. I. by a fellowship from ARC and L. D. by a fellowship from la Fondation pour la Recherche Médicale.