Drosophila Paired regulates late even-skipped expression through a composite binding site for the paired domain and the homeodomain

During early Drosophila development, a cascade of regulatory interactions between the segmentation genes results in a progressive refinement of spatial information (Nusslein-Volhard and Wieschaus, 1980; Akam, 1987; Ingham, 1988). This refinement eventually gives rise to the reiterated segmental domains defined by the striped expression of the pair-rule and segment-polarity genes. In this cascade, reiterated spatial information first arises with the striped expression of the primary pair-rule genes, hairy, runt and even-skipped (eve). According to the hierarchical model of pair-rule genes, the coarse striped information of the primary pair-rule genes is then transformed into refined stripes of the secondary pair-rule and segmentpolarity genes. Understanding the molecular regulatory mechanisms of the establishment of the stripe patterns by primary pair-rule genes, and the refinement by the secondary pair-rule genes and segment-polarity genes, is key to understanding how the segmentation gene cascade defines positional information in the early embryo.

The promoters of the primary pair-rule genes, eve and hairy, have been shown to have separate cis-regulatory elements for each of the individual stripes (Howard et al., 1988; Goto et al., 1989; Harding et al., 1989; Howard and Struhl, 1990; Pankratz et al., 1990; Riddihough and Ish-Horowicz, 1991), reflecting the fact that the expression stripes of the primary pair-rule genes are each regulated by different combinations of maternal coordinate and gap gene expression. Furthermore, molecular studies of some of these stripe-specific cis elements have identified binding sites for the products of the maternal coordinate and gap genes (Stanojevic et al., 1989; Small et al., 1991). In contrast, the promoters of the secondary pair-rule genes and segment-polarity genes need not, in principle, have stripespecific elements, since the striped regulatory inputs for these genes are already repetitive in nature. Instead, expression of these genes is thought to occur through cis-regulatory elements responsible for multiple stripes. Examples of these multistripe elements include the ‘zebra’ element of the secondary pair-rule gene, fushi tarazu (ftz), responsible for all seven ftz stripes (Hiromi et al., 1985), and the even-parasegment cis element of the segment-polarity gene engrailed (en) (DiNardo et al., 1988). Regulation of multistripe cis elements is likely to involve combinatorial mechanisms. However, little is known as to how this is achieved at the molecular level.

In the past few years, it has become clear that the picture presented thus far, equating the primary pair-rule genes with establishment of stripes and secondary pair-rule genes with refinement, is over-simplified in that the refinement of stripe information also occurs within the promoters of individual genes through separate cis elements responsible for different temporal phases of expression. For example, transgenic studies of eve have shown that this gene has two regulatory programs, early and late, regulated by two distinct classes of cis-acting elements in the eve promoter (Goto et al., 1989). The ‘early elements’ (E) are stripe-specific and define the initial (early) expression of each of its stripes, which are located in the oddnumbered parasegments. In contrast, the ‘late element’ (L), responsible for later expression of eve at gastrulation, is a single multistripe element responsible for the regulation of all seven late eve stripes in the odd-numbered parasegments (eve also has secondary minor stripes in the even-numbered parasegments whose regulation is not understood and not considered here). Since the early elements (E) give rise to broad stripes whereas the late element (L) gives narrow stripes, this results in the spatial refinement of eve stripes over time: their anterior borders become sharply demarcated with concomitant loss of expression from the posterior. Other segmentation genes also exhibit refinement of their expression stripes over time suggesting that they may also have separate cis elements for different phases of their expression (Carroll and Scott, 1986; Coulter et al., 1990; Grossniklaus et al., 1992; Klingler and Gergen, 1993; Gutjahr et al., 1994; Yu and Pick, 1995)

The late element (L) of eve provides an accessible multistripe control element for molecular dissection. This late element, located more than 5 kb upstream of the transcription start site, is a cis-acting site for regulation by other pair-rule genes (Goto et al., 1989; Warrior and Levine, 1990) as well as for eve autoregulation (Goto et al., 1989; Harding et al., 1989; Jiang et al., 1991). Recent work has shown (Fujioka et al., 1995) that the pair-rule gene paired (prd) is negatively regulated by early eve stripes in a concentration-dependent manner. This study also suggested the possibility that the prd gene product (Prd) may in turn activate late eve expression, although this was not predicted by previous genetic studies that had placed prd at the bottom of the regulatory hierarchy of the pair-rule genes (Baumgartner and Noll, 1991). Here, we report that late eve expression is, indeed, positively regulated by Prd through a direct interaction with a conserved sequence in the late element. This is one of the first identified in vivo target sequences of Prd, a Pax protein with two DNA-binding domains, a paired domain and a homeodomain. Strikingly, the site is apparently a composite of a paired domain and a homeodomain binding site.

Flies of the genotype y w;+/+;+/+ were used as host for injection of C3D-based constructs (see below). The injection procedures are as described previously (Rubin and Spradling, 1982). The allele of prd used for mutant analysis was prd⁴ (=prd^2.45.17), which has a 1.1 kb insertion into the paired box (Frigerio et al., 1986) and shows no protein staining with a polyclonal antibody directed against the Cterminal half of the Prd protein (Bertuccioli et al., 1996). The lacZ-marked balancer chromosome CyOpthb used to distinguish mutant embryos was generated (Morrissey et al., 1991) by mobilizing a Hunchback-promoter/lacZ P-element (Driever et al., 1989) onto the CyO chromosome.

A transgene vector, called C3D, was constructed by removing restriction sites for PstI, SalI, XbaI and BamHI within the polylinker sequence of pCaSpeR3 (Thummel et al., 1988). A fragment containing the eve promoter fragment (from −275 to +166), lacZ and the poly(A) attachment site of the tubulin α subunit gene was cloned into the BglII and EcoRI sites of C3D in such a way that the directions of the transcription of the w (white) marker gene and the β-galactosidase gene are opposite. This ‘injection plasmid’, called C3DZ, contains XbaI and NotI sites just upstream of the eve-lacZ fusion gene, which provides a cloning site for the late element (L) or its deletion derivatives.

The LmLd fragment, from the EcoRV site (−5653; converted to a BglII site) to Kpn site (−4799), was cloned into a modified pSP72 vector (the SmaI site in the polylinker was converted to a BglII site and the BglII site to a NotI site). This plasmid (RVKp) therefore contains the LmLd sequence flanked by XbaI and NotI sites and was used for creating deletions shown in Fig. 2. Most endpoints were created by cleavage with restriction enzymes identified in Fig. 2. Three endpoints (−5532, −5446, −5349) were created by Bal31 exonuclease digestion. One endpoint (−5541) was created by polymerase chain reaction (PCR) using the synthesized primer EL12 (5′TCGAAGGATCCACTCACCGTGGCTAATTGC-3′), which creates a BamHI site just upstream of the PTE.

The in vitro mutagenesis of the PTE was also carried out with the RVKp plasmid and synthetic primers using standard PCR procedures (Sambrook et al., 1989). For the PTE-pd and PTE-phd mutations, 58mer oligonucleotides containing mutated PTE and the surrounding sequences were synthesized: PTE-pd primer, (5′-CTCCCCGGGCCCTTTATTGATGGGCAATTAGCCGCCTCGAGCGGCGTTTCGAGTGAG-3′); PTE-phd primer, (5′-AGGAATTCCTCACTGCCCGGGCCCTTTATTGATGGGCAGTCGACCACGGTGAGTGGC G-3′). The PCR reaction was performed with one of these primers plus the SP6 promoter primer with the RVKp template. Each product was cleaved with either XbaI and SmaI (PTE-pd) or XbaI and EcoRI (PTE-phd), and was cloned into the RVKp plasmid cut with the same set of enzymes. The LmLd sequence was then transferred to the injection plasmid C3DZ. For creating the PTE-hd mutation, two complementary primers containing the mutated PTE and the surrounding sequences were used: (5′-CCGTGGCTCGAGGCCCATCAATAAAGGGC-3′) and (5′-GATGGGCCTCGAGCCACGGTGAGTGGCG-3′). The first primer was combined with the T7 promoter primer, the second primer with the SP6 promoter primer, and PCR reactions were carried out with the RVKp template. Products from this first round were mixed and the second round PCR reaction was performed with SP6 and T7 promoter primers. The product of the second round was cleaved with XbaI and EcoRI, and cloned into the RVKp plasmid as described above. Fidelity of all PCR reactions and subsequent cloning steps were confirmed by DNA sequencing. To create internal deletion 12 (−5541 to −5503; Fig. 2A), the synthetic oligonucleotide EL22 (5′TGAGTGGATCCTTCGAGTGAGCTGCAGG-3′) was used in conjunction with EL12 (above) to create a BamHI site at −5541. The SmaI site (−5503) of this plasmid was converted to a BamHI site by ligation of BamHI linker to SmaI-cleaved plasmid. The sequence between these two BamHI sites were removed by BamHI digestion before reclosure. Plasmid DNAs for injection were purified with polyethelene glycol as described by Sambrook et al. (1989) with minor modifications. DNA was filtered through a Probind filter (Millipore) before injection.

Radioactive probes homologous to the D. melanogaster eve gene were made using random primers and restriction fragments from the late element and the coding sequence. A phage genomic library of D. hydei (gift of D. Maier) (Maier et al., 1990) was screened with the probes using low stringency hybridization procedures. Hybridization was carried out with 2× SSC, 0.1% SDS, 1× blocking reagent (Boehringer Mannheim) at 50°C overnight, and the filters were washed with 0.5× SSC, 0.5% SDS at 60°C for three times, 15 minutes each. Three independent positive clones were isolated for further analysis. DNA was made from these clones using plate lysates (Sambrook et al., 1989). Southern blot hybridization analysis shows that they together contain 24 kb of the putative D. hydei eve locus. EcoRI fragments of the inserts were cloned into the pSP72 vector. Southern blot hybridization was used to identify the locations of the late element and the coding sequence in the subclones. DNA sequencing was carried out using either the Sequenase sequencing system (US Biochemicals) or an automated sequencer (Applied Biosystems). Comparative analysis of DNA sequences (Fig. 1) was performed using the GCG software (Genetics Computer Group). The nucleotides are numbered from an upstream ClaI site. To test the transcriptional activity in vivo of the D. hydei DNA, two DNA fragments (9 to 1594, and 788 to 1594, corresponding to L and LmLd in D. melanogaster, respectively) were cloned into C3DZ as described above. Their activities in vivo were found to be equivalent; an example of the stripeforming activity of the shorter version (788 to 1594) is shown in Fig. 2F.

In situ hybridization to whole-mount embryos using digoxigenin(DIG-) labeled probes was performed as described (Tautz and Pfeifle, 1989). After hybridization, embryos were washed with ethanol to reduce background (Manoukian and Krause, 1992). DIG-labeled antisense mRNA and DNA probes were used for hybridization and were visualized using alkaline phosphatase-conjugated anti-DIG antibody (Boehringer Mannheim). Embryos were mounted in Fluoromount (Southern Biotechnology) or Aqua-Poly/Mount (Polysciences).

Experiments with hs-prd constructs were performed as described (Morrissey et al., 1991) using two 10 minute heat treatments. To obtain homozygous prd− embryos with one copy each of hs-prd and L-lacZ, we examined embryos from the cross: prd⁴/CyOpthb; hs-prd X prd⁴/CyOpthb; L-lacZ.

³²P-labelled probes used in EMSA were 26-mer fragments containing the wild-type or mutant PTE sequences. A histidine-tagged Prd protein fragment (amino acids 27–273) was prepared from bacteria as described (Novagen, Madison WI). 25 minute binding reactions with 0.1–1 ng of labelled DNA were performed at 20°C in 15 mM TrisHCl pH 7.5, 60 mM KCl, 0.5 mM DTT, 0.25 mg/ml BSA, 0.05% NP-40 and 7.5% glycerol in a volume of 20 μl. We included 100 ng/reaction of poly(dIdC) (Sigma) to minimize non-specific binding by the PD and HD, based on preliminary tests with single PD (Jun and Desplan, 1996) and HD (P2, Wilson et al., 1993) sites using proteins with a mutated HD or PD (Miskiewicz et al., 1996) (data not shown). Binding reactions were electrophoresed on an 8% native polyacrylamide gel (29:1 acrylamide:bis) using 0.25× TBE at 4°C and subjected to autoradiography.

A regulatory element from the NdeI site of the eve gene at −6415 (measured from the transcriptional start site) to the KpnI site at −4799 was shown previously to be fully functional in driving the expression of a linked reporter gene in seven narrow stripes, mimicking late eve expression (Goto et al., 1989). We refer to this region as the late element (L). In order to identify potentially important sites within L, we undertook a cross-species comparison of sequences based on the assumption that regulatory regions crucial for gene function would diverge more slowly than other non-coding regions in the course of evolution (Wilson et al., 1987). The eve gene from D. hydei was cloned using the coding and L sequences of the D. melanogaster gene as probe. The L sequences of D. hydei and D. melanogaster share a number of islands of conservation. These conserved islands form a cluster from −6080 to −4875, as shown in Fig. 1. Based on the pattern of the conservation, we divided the D. melanogaster L into three regions, Lup (Lu), Lmiddle (Lm) and Ldown (Ld), with the EcoRV site (−5653) marking the Lu/Lm boundary, and the PstI site (−5306) the Lm/Ld boundary. Lm contains the Minimal Autoregulatory Sequence (MAS), consisting of two in vitro Eve-binding regions (EVE-D and EVE-P) and two nuclear factor-binding sites (DENF-1 and DENF-2), which have previously been shown to be important for eve autoregulation (Jiang et al., 1991). The EVE-D and EVE-P sequences are well conserved. The DENF-1 and DENF-2 sites were previously found to be essential for eve autoregulatory activity, since deletion of either site eliminated the stripe-forming activity of an MAS multimer (Jiang et al., 1991). It has been proposed that the proteins that bind to the DENF-1 and DENF-2 sites are GAGA factor and Tramtrack, respectively (Jiang et al., 1991). However, the consensus binding sequence for GAGA factor within the DENF-1 site is not conserved, and the DENF-2 site is absent, apparently deleted, in D. hydei. Similar findings have been made in a parallel study of sequence conservation between D. melanogaster and the Hawaiian Drosophila, D. picticornis (Sackerson, 1995). No function is known for the other observed islands of conservation in Fig. 1. The importance of the conserved nucleotides in vivo is underscored by the finding that the D. hydei L apparently has full late activity in the D. melanogaster embryo (see below and Fig. 2F)

In addition to this evolutionary comparison, we initiated a dissection of the late element using transgenes in which altered late elements were placed upstream of the eve promoterproximal region (−275 to +166) contiguous to the lacZ reporter. The in vivo activities of altered L elements were tested by staining whole-mount embryos of transgenic lines for β-galactosidase (β-gal) mRNA. We found that the EcoRVKpnI fragment (LmLd) is fully active as a late element, as judged by the staining intensity of the embryos (see Fig. 2B), as previously reported (Goto et al., 1989). We also found that Lu+Lm is active, whereas Lu+Ld is not (data not shown). Thus, Lu and Ld are redundant in that either one can augment Lm for activity. Despite this redundancy, a comparison of the conserved sequences in Lu and Ld (Fig. 1) does not reveal obvious homology, although short stretches of sequence similarity can be noted.

To further dissect late function, we created deletions in the LmLd fragment and tested their activities in vivo (Fig. 2A). Examples of the activities of several constructs are shown in Fig. 2B-E. The presence or absence of activity was scored by the intensity of staining. We note that some of the constructs scored as ‘negative’ (marked with an asterisk in Fig. 2A) do express extremely weak stripes. These weak stripes are sometimes incomplete and are typically detectable at only later stages due in part to accumulation of the stable β-gal product. However, as illustrated in Fig. 2, the levels of expression by these ‘negative’ constructs (e.g. embryos Fig. 2D,E) were extremely weak compared to those of transgenes scored as active (embryos Fig. 2B,C). Moreover, these scoring criteria were consistent with the results of a second, independent ‘localized rescue’ assay. In this assay, eve regulatory regions were used to drive the eve coding sequence and the ability of late elements to augment activity of early expression elements in rescuing the eve deficiency phenotype were assayed (see Fig. 2 legend and Fujioka et al., 1995). Analysis of a subset of the constructs in Fig. 2A revealed that elements scored as positive in the first assay (e.g. LmLd) had full rescue activity in the second assay, whereas elements with very weak β-gal stripes (e.g. Lm) showed only partial rescue. We note that the very weak β-gal stripes of Lm were previously scored as positive by less stringent criteria (Harding et al., 1989).

In our deletion analysis (Fig. 2), the absence of activity by deletion 1 confirms that the MAS, and possibly surrounding sequences, is critical for late activity. The activities of the next three deletions (deletion 2-4) indicate that at least two subfragments of Ld (−5349 to −5206, or −5018 to −4799) can independently augment Lm for activity. Deletion 5 shows that the upstream-most 97 bp of Lm can be deleted with minimal loss of activity. A further deletion of 24 bp (deletion 6) drastically reduces activity, identifying the location of a critical sequence between −5556 and −5532. The next two deletions (deletions 7 and 8) show that this conclusion is not dependent on the choice of Ld subfragments (Fig. 2C,D). The activity of deletions 9 and 10 further narrows the upstream limit of the newly identified critical sequence to −5541. The importance and uniqueness of the sequence in question are also underscored by the elimination of late activity by the small deletions 11 and 12 (deletion 12 is shown in Fig. 2E). Deletion 13 sets the downstream limit of the critical sequence at −5503, and shows further that the interval between the newly identified critical sequence and the MAS can be deleted without effect. Thus, this deletion analysis identifies three separate regions within LmLd to be critical for the cis-activity: the 38 bp upstream element between −5541 and −5503, the MAS, and either of the redundant Ld subfragments.

As shown in Fig. 1, the functionally critical 38 bp element between −5541 and 5503 contains a stretch of 32 nucleotides of which 29 are conserved between D. melanogaster and D. hydei. (These nucleotides are also conserved between D. melanogaster and D. piticornis; Sackerson, 1995) As shown in Fig. 3, the conserved region (labeled as PTE in the figure) is also found to be homologous to e5, a site within the eve promoter that has been shown to be bound by Prd and Eve in vitro (Hoey and Levine, 1988; Treisman et al., 1991). The e5 sequence is also conserved between D. melanogaster and D. hydei. Significantly, many of the bases shared by the 32 bp region and e5 are among those conserved. Prd contains two DNA-binding domains, a Paired-class homeodomain (HD) and a paired domain (PD) (Bopp et al., 1986; Frigerio et al., 1986), and footprinting analysis has shown that the e5 site has two adjacent half-sites that can be bound either simultaneously or individually by the two domains (Treisman et al., 1991). These half-sites can be aligned with optimized binding sites, obtained by PCR selection, for the isolated Prd HD (HD opt in Fig. 3) (Wilson et al., 1993) and the isolated PD (PD opt) (Xu et al. 1995; Jun and Desplan, 1996). Comparison of the 32 bp conserved region with these PCR-optimized sites suggests that it also contains adjacent half-sites for both domains.

To test whether Prd can bind to the conserved sequence, we performed an electrophoretic mobility shift analysis (EMSA) using DNA fragments containing the sequence and a histidinetagged fragment of the Prd protein containing both the PD and the HD (amino acids 27-276). As illustrated in Fig. 4, the Prd protein shifted the conserved sequence (lanes 1-5). This binding was effectively competed by adding cold probe, but not random salmon sperm DNA sequences (not shown). This indicates that the conserved sequence is indeed a Paired binding site. Interestingly, with higher concentrations of Prd protein, lower mobility band shifts of PTE and e5 are observed (not shown), consistent with the fact that the protein has two DNA-binding activities and that the DNA probes have two binding sites that can be bound by the HD and PD of separate Prd molecules (Treisman et al., 1991).

In a parallel study, Jun and Desplan (1996) recently determined a PCR-optimized Prd-binding sequence using a Prd peptide containing both the PD and HD. As shown in Fig. 3, the conserved sequence, PH0, matches this optimal Prd sequence remarkably well (12 out of 15 bases). The sequences match at all 6 base pairs predicted to contact the HD (Wilson et al., 1995), and at 6 of the 11 base pairs thought to contact the PD (Xu et al., 1995). In the following sections, this 32 bp conserved sequence will be referred to as the PTE (the Paired Target Element).

Prd is an activator of transcription in cell culture assays (Han et al. 1989; Ananthan et al., 1993; Cai et al., 1994), a property expected for a factor interacting with the PTE, a positive regulatory site. In addition, prd stripes overlap with the anterior portion of early eve stripes, where late eve stripes are activated. Therefore, Prd may directly interact with the PTE to activate late eve expression. However, previous studies have found that prd is at the bottom of the pair-rule gene hierarchy (Baumgartner and Noll, 1991), and no alterations in the eve expression pattern were previously observed in prd mutants (Frasch and Levine, 1987). However, this and other early studies of segmentation gene interactions did not make use of lacZ-tagged balancer chromosomes, which allow unequivocal identification of mutant embryos. We therefore re-examined eve expression in prd mutants making use of a CyO balancer chromosome with a hunchback/lacZ P-element insertion (see Fig. 5 legend). The patterns of eve expression in prd mutant and in wild-type embryos are shown in Fig. 5. At the blastoderm stage, there is no appreciable difference between the two (Fig. 5A,C), but during gastrulation, eve expression starts to fade in the prd mutant (Fig. 5D), particularly in the ventral ectoderm, and becomes undetectable as germband elongation proceeds (not shown). In contrast, the wild-type sibling embryos at the same morphological stages (and stained simultaneously) express strong, refined eve RNA stripes (Fig. 5B). Thus, early eve stripes are normal in prd mutants, but late stripes are absent. To test whether the prd mutation exerts its effect through the late element, we examined the expression of late element reporter constructs in prd mutants. The L-lacZ and the LmLd-lacZ transgenes were crossed into a prd mutant line and their expression patterns examined. Homozygous prd⁻ embryos showed no L-lacZ (Fig. 6B) or LmLd-lacZ (not shown) expression stripes, unlike prd⁺ siblings which showed the characteristic stripe expression (Fig. 6A). Thus, prd is required for L to activate late stripes.

If prd is a regulator of late eve expression, then ectopic expression of an introduced prd transgene would also be expected to affect eve expression. If this were the case, it would provide an independent way to examine interaction of Prd with the late element. It was previously shown that induction of ectopic Prd driven by a heat-shock promoter leads to a posterior expansion of odd-numbered en stripes (Morrissey et al., 1991). Moreover, the loss of odd-numbered en stripes in prd⁻ embryos can be rescued by heat treatment of prd⁻hs-prd embryos (Miskiewicz et al., 1996). We therefore examined the expression patterns of endogenous eve and L-lacZ following ectopic Prd induction. In heat-treated prd⁻ embryos containing a hs-prd transgene, we observed rescue of L-lacZ striped expression (Fig. 6C) and of late eve RNA expression (not shown), confirming that Prd is an activator of late eve expression. Moreover, in both prd⁻ and prd⁺ embryos, ectopic Prd caused the posterior borders of both L-lacZ (Fig. 6C) and late eve RNA (not shown) stripes to be shifted posteriorly. This suggests that Prd is at least partly responsible for determining the posterior borders of late eve stripes.

The results described above show that Prd interacts with L in vivo to activate late eve stripes. We next asked whether this interaction is a result of Prd binding to the PTE. As noted above, the PTE contains juxtaposed binding sites for a pairedclass HD and a PD. If binding of Prd to the PTE regulates the activity of the late element, then we would expect that a targeted mutation in either of these half-sites, which reduces Prd-binding in vitro, would weaken the in vivo activity of the late element. We created specific mutations within the PTE to weaken or eliminate Prd binding and tested their effects on maintenance of late stripes in vivo. These experiments were carried out in the context of LmLd which is fully active as a late element.

The first mutation, PTE-pd, changes four nucleotides in the apparent core of the PD-binding sequence (Fig. 3). In the second mutation, PTE-hd, the putative HD-binding sequence is changed from TAATTG to TCGAGG, significantly altering the core TAAT sequence. The third mutation, PTE-phd, was designed to interfere with the binding of both the HD and the PD of Prd to the PTE. It changes the HD-binding sequence, TAATTG, to CGACTG, and the PD-binding sequence ACCGTGGC to ACCGTGGT (Fig. 3). This eliminates both the core TAAT sequence of the HD-binding site, and changes a single nucleotide in the PD-binding site that is expected to reduce the affinity of the PD, since a change in this position of the site-selected PD site significantly reduces the affinity of Prd (Jun and Desplan, 1996). As shown in Fig. 4, the affinity of Prd for PTE-pd was reduced more than 4-fold (lanes 6 –10) when compared to wild-type PTE (lanes 1–5), and the affinity for PTE-hd was more than 16-fold lower (lanes 11–15). The affinity for PTE-phd was even lower (by at least 60-fold; data not shown). We note that PTE-pd retains limited PD DNAbinding activity (data not shown), consistent with the fact that only 4 of the 11 base pairs predicted to be contacted by the PD are mutated (Xu et al., 1995). When tested in vivo, the mutations PTE-pd, -hd and -phd all caused reduced activity of LmLd-lacZ. The effects of all three mutations on stripe activation and maintenance are roughly comparable (Fig. 7) and approach that of the deletion 12 construct (RVΔ41S), which lacks the PTE entirely (Fig. 2E). Moreover, like deletion 12, all three mutant constructs show only limited in vivo rescue activity when placed upstream of eve coding sequences, in contrast to wild-type LmLd (results not shown; see Fig. 2 legend). Thus, both putative PD and HD-binding sites are required for the activity of the PTE. From this observation, in conjunction with the results of the deletion analysis and the fact that the PTE sequence closely matches the highest affinity site selected in vitro, we conclude that the PTE is a direct target site of Prd in vivo.

The early (E) and late (L) control elements of the eve promoter participate in distinct phases of the refinement of spatial information during segmentation of the early Drosophila embryo. Early elements direct expression in broad stripes, each of which spans the entire odd-numbered parasegmental primordium. This gives way, under the control of the late element, to expression in sharply defined narrow stripes in the anterior cell row of each primordium. Both phases are important for correct regulation of downstream genes in the cascade. The early expression is critical for defining expression boundaries of several pair-rule genes including prd, runt, sloppy-paired (slp) and odd-skipped (Fujioka et al., 1995; see below). Late eve expression regulates a later phase of expression of these same pair-rule genes, thereby providing for the initiation of strong, continuous en stripes in odd-numbered parasegments (Fujioka et al., 1995).

We have analysed the cis-regulatory element for late eve expression and have found that this multistripe element (see Introduction) is regulated by the Prd protein through an evolutionarily conserved sequence (PTE). This observation supports our model for the formation of eve late stripes (Fig. 8; see also Fujioka et al., 1995). The model proposes that bellshaped early eve stripes act as morphogenetic gradients that differentially regulate prd, runt and slp. Although all three genes are repressed by eve, it is postulated that runt and slp are sensitive to lower levels of Eve protein than is prd. This creates a row of cells at the anterior edge of early eve stripes that express prd but not runt or slp. Since prd is an activator of late eve (this study), and runt and slp are repressors (Cadigan et al., 1994a, b; Manoukian and Krause, 1993), activation of the late eve is restricted to this row of cells.

In its regulation of late eve, the function of Prd may be augmented by a structural homolog of Prd, the Gooseberry (Gsb) protein, which also contains a paired-class HD and a PD. Recent studies (Li and Noll, 1994) have shown that, when appropriately expressed, the Prd protein can substitute for Gsb, suggesting that the two proteins may be functionally equivalent. Hence, since Gsb is turned on at about gastrulation in the same cell rows in which late prd is expressed, it is possible that Gsb contributes in part to later maintenance of late eve expression in collaboration with Prd protein, which continues to be expressed through much of germ band elongation. However, since gsb⁻ embryos have normal en expression which depends on late eve for full activation in odd-numbered parasegments (Hidalgo, 1991; Li and Noll, 1993; Fujioka et al., 1995), it appears that Prd protein alone is sufficient to activate late eve in the absence of gsb function. In contrast, in prd⁻ embryos in which odd-numbered gsb stripes are deleted (Bopp et al., 1989), there is no activation of late eve (this study), consistent with Prd and Gsb functions both being absent.

Central to our model (Fig. 8) is the idea that Prd functions combinatorially on the late element. Strong support for this comes from the effects of ectopic Prd on L-lacZ expression. Uniform expression of Prd does not cause uniform L-lacZ expression. Rather, L-lacZ stripes are expanded posteriorly by only one or two cells (Fig. 6), implying that other regulators act in combination with Prd. Indeed, ectopic Prd has the equivalent effect on en expression: the odd-parasegment en stripes, which coincide with L-lacZ, are expanded 1-2 cells posteriorly, suggesting that Prd normally specifies the posterior borders of these stripes (Morrissey et al., 1991). Like L-lacZ, the anterior borders of these stripes are thought to be specified by the repressors slp and runt. The question remains how these repressors might function in combination with Prd at the molecular level. With the identification of a functional Prd site within L, this question can now be addressed.

The PTE element is a conserved sequence that deletion analysis indicates is critical for function of the late element. Our present study shows that Prd can bind to the PTE in vitro and that Prd functions through this sequence in vivo. The significant loss of activity caused by mutations in the PTE (Figs 2, 7) suggests that the interaction of the site with Prd plays a critical role that cannot be substituted for by any other sequences within our minimal single-copy L element (LmLd). Nevertheless, although sequence analysis of LmLd does not reveal any conspicuous sequence similarities to previously identified Prd-binding sites, the possibility that there may be secondary Prd site(s) within L is suggested by our observation that several PTE deletion constructs have extremely weak stripes that appear to be maintained through germ band elongation (see Fig. 2 legend). This contrasts with the complete absence of L activity in prd⁻ embryos.

The e5 site (Fig. 3), located just upstream of the eve TATAA box, is a previously identified in vitro target of Prd that has been used to study the interaction of Prd, and its mammalian homologues, with DNA (Chalepakis et al., 1994; Hoey and Levine, 1988; Treisman et al., 1991). However, the function of e5 in vivo has not been addressed. We have recently shown that removal of the e5-binding site, along with some adjacent sequence, from the promoter-proximal sequence (−275 to +166) normally included in our reporter constructs, causes somewhat weakened expression of L-lacZ stripes (result not shown). A similar reduction was observed previously when the promoter-proximal region was replaced with a heat-shock promoter (Goto et al., 1989). This observation, together with the presence of a similar sequence in the D. hydei promoter (Fig. 3), suggests that the e5 site may function in vivo in collaboration with the PTE. However, our observation that PTE deletions have much more severe consequences in vivo than the e5 deletion, and that Prd has a much higher affinity for the PTE than for e5 (data not shown), suggests that the PTE plays a more critical role for Prd mediated activation of late eve expression. We should also note that the eve promoter contains another site, the e4 sequence, which is bound by Prd in vitro (Hoey and Levine, 1988; Treisman et al., 1991). However, unlike the PTE or e5, this sequence is not conserved in D. hydei (data not shown; see below).

The Prd protein, like a number of other Pax proteins, has both a HD and a PD (Walther et al., 1991). Although both have DNA-binding activities, the functional relationship between these two domains in vivo has remained elusive. PCR-selection experiments have revealed optimal target sequences for each of these domains when tested alone (Jun and Desplan, 1996; Wilson et al., 1993; Xu et al., 1995). These optimal sequences can be simultaneously aligned (see Fig. 3) with both the PTE and e5, forming adjacent half-sites. As was shown for e5 (Treisman et al., 1991), the PTE can be bound by Prd molecules lacking either one of the two binding activities (Miskiewicz et al., 1996). Furthermore, the PTE matches PH0, a PCR-selected site for simultaneous binding by the PD and HD (Jun and Desplan, 1996), at many of the bases thought to contact the Prd protein based on structural studies (Wilson et al., 1995; Xu et al., 1995). The order and relative orientation of the PD and HD sites in both the PTE and e5 is the same as in PH0. Moreover, the spacing between the two sites is the same in the PTE and PH0, but is one base closer in e5 (Fig. 3). In contrast, the order in the PTE is the opposite to that in e4, which is a relatively poor binding site for Prd (Treisman et al., 1991).

The PTE is the first identified in vivo target of Prd that contains both a PD and a HD target sequence. Previous analysis of the gooseberry promoter has identified a putative A/T-rich Prd-binding site (Li and Noll, 1994) which, based on comparison with PCR-selected sites, may be bound by the Prd HD (Wilson et al., 1995). Paired domain sites are probably also important for the regulation of gooseberry, since both PDand HD-binding activities of the Prd protein are required for in vivo regulation of gooseberry, as they are for the regulation of en and hedgehog (Bertuccioli et al., 1996; Miskiewicz et al., 1996). Similarly, we have found that both the PD and HD target sites of the PTE are required for Prd regulation of LlacZ. Mutation in each of these target sites leads to reduced binding of Prd to the PTE in vitro (Fig. 4), and to significantly weakened L-lacZ stripes in vivo (Fig. 7), consistent with the idea that Prd function requires both of its DNA-binding domains. Whether both binding activities need be present in the same Prd molecule for L-lacZ regulation has yet to be determined. Ectopic expression studies of Prd regulation of segment-polarity genes have shown that the two binding activities can come from separate molecules: co-expression of a PD mutant protein with a HD mutant protein restores Prd function in embryos, whereas neither mutant alone is sufficient (Miskiewicz et al., 1996). However, it remains possible that the wild-type Prd protein uses both binding activities in the same molecule (see also Bertuccioli et al. 1996; Jun and Desplan 1996). With the identification of the PTE, it is now possible to investigate these questions, as well as structural constraints on the target sequence. For example, how do spacing and orientation of the HD and PD sites affect Prd binding and function in vivo?

Prd is a member of the Pax family of transcription factors, all of which contain a PD, and some of which also contain a HD (Walther et al., 1991). Some of the mammalian members of this family have been associated with defined mutations and birth defects. For example, mutation of Pax-3, the closest mammalian homolog of Prd, which contains both a PD and a HD, is associated with Waardenburg Syndrome in humans, affecting neural crest cells and other derivatives of the dorsal neural tube (Baldwin et al., 1992; Burri et al., 1989; Tassabehji et al., 1992). Our observation that Prd functions in vivo through the PTE raises the possibility that Pax-3 and its homologs also function through composite binding sites, with adjacent sequences for HD and PD binding.

We thank Claude Desplan, Claudio Bertuccioli and Susie Jun for discussion of unpublished data and technical advice. We are grateful to Dieter Maier for providing us with the D. hydei library. We also thank James B. Jaynes for discussions and careful reading of the manuscript. This work was supported by NSF grant IBN-9507406 to T. G. and NIH grant GM42752 and ACS grant JFRA-430 to M. W.