The role of binding site cluster strength in Bicoid-dependent patterning in Drosophila

Abstract

The maternal morphogen Bicoid (Bcd) is distributed in an embryonic gradient that is critical for patterning the anterior–posterior (AP) body plan in Drosophila. Previous work identified several target genes that respond directly to Bcd-dependent activation. Positioning of these targets along the AP axis is thought to be controlled by cis-regulatory modules (CRMs) that contain clusters of Bcd-binding sites of different “strengths.” Here we use a combination of Bcd-site cluster analysis and evolutionary conservation to predict Bcd-dependent CRMs. We tested 14 predicted CRMs by in vivo reporter gene assays; 11 show Bcd-dependent activation, which brings the total number of known Bcd target elements to 21. Some CRMs drive expression patterns that are restricted to the most anterior part of the embryo, whereas others extend into middle and posterior regions. However, we do not detect a strong correlation between AP position of target gene expression and the strength of Bcd site clusters alone. Rather, we find that binding sites for other activators, including Hunchback and Caudal correlate with CRM expression in middle and posterior body regions. Also, many Bcd-dependent CRMs contain clusters of sites for the gap protein Kruppel, which may limit the posterior extent of activation by the Bcd gradient. We propose that the key design principle in AP patterning is the differential integration of positive and negative transcriptional information at the level of individual CRMs for each target gene.

Gradients of two transcription factors, Bicoid (Bcd) and Dorsal (Dl), are critical for patterning the major body plan axes in the Drosophila embryo. This studied focused on Bcd, a maternal effect gene whose mRNA is initially localized at the anterior pole of the oocyte (Fig. 1A) (1). Upon egg deposition, the trapped bcd RNA is translated, and a nuclear gradient forms, with highest concentrations near the anterior pole and progressively lower concentrations in more posterior regions (Fig. 1B) (2). Bcd protein distribution matches its biological activity, with high levels required for the positioning of the most anterior structures, intermediate levels for head structures, and low levels for the thoracic and anterior abdominal segments. Varying the shape of the Bcd gradient has a direct effect on positional identity in the blastoderm (3). Increasing the maternal input of bcd can shift morphological landmarks posteriorly, whereas decreasing it shifts them anteriorly. These experiments suggest that the concentration of Bcd protein present at each position along the length of the embryo (EL) determines the destiny of cells that will occupy that region.

Fig. 1.

Regulation of otd and hb by the Bcd morphogen. (A–D) Lateral views of blastoderm-stage embryos showing bcd mRNA (A), Bcd protein expression (B), and the mRNA expression patterns of the Bcd target genes otd (C) and hb (D). All embryos are oriented with anterior to the left and dorsal up. (E) Schematic summary of the data shown in B–D. Thresholds for Bcd-mediated activation of otd (T1) and hb (T2) and the percentage of EL on the x axis are shown. (F–I) Computational analysis of the Bcd-binding site clusters in the otd and hb loci. (F and G) Graphical representations of Bcd clusters in 4-kb sequences from the otd (F) and hb P2 (G) genomic regions. (Lower) Each predicted Bcd site is shown as an individual dot in the scatter plots, with PWM scores on the y axis. (Upper) Colinear cluster view representations of the data in the scatter plots (see Materials and Methods). Cluster significance is represented by a color code (far right): cool colors (light blue and green) indicate low-significance clusters; hot colors (orange and red) represent highsignificance clusters. (H and I) Results of mutating one (H) or two (I) highaffinity sites (red asterisks) in the hb P2 region. [The image in B is reproduced with permission from the embryo tu9 entry of the FlyEx database (Copyright 1998, David Kosman and John Reinitz).]

Bcd protein contains a homeodomain and functions as a morphogen by activating the transcription of multiple target genes in different positions along the anterior–posterior (AP) axis (reviewed in refs. 4 and 5). Driever, Thoma, and Nüsslein-Volhard (6) first proposed that differential positioning could occur by means of gene-specific regulatory elements that respond to different concentration thresholds of Bcd protein. In their model, genes containing high-affinity sites might respond to lower Bcd concentrations than those with low-affinity sites. It was also proposed that the number of binding sites might contribute to the level of transcription.

This model was primarily based on studies of the target gene hunchback (hb), which is expressed throughout the anterior half of the embryo (Fig. 1D) (7). The 5′ regulatory region of hb contains several high-affinity Bcd-binding sites, consistent with the idea that it responds to a low-concentration threshold (Fig. 1E, T2) (8, 9). A second target gene, orthodenticle (otd), is expressed only in the anterior 30% of the embryo (Fig. 1C) (10). The otd regulatory region contains lower-affinity sites than those detected in the hb promoter and, thus, may require higher levels of Bcd (Fig. 1E, T1) for activation (11).

However, several lines of evidence suggest that the simple threshold model for positioning target genes may be incomplete. First, a recent study showed that the posterior border of the hb expression domain does not always correspond to a specific Bcd concentration threshold (12). Second, Bcd-dependent activation of several target genes requires synergistic interactions with the gap protein Hb (13, 14). Finally, the posterior borders of expression of two Bcd targets, the even-skipped (eve) stripe 2 enhancer and the early expression domain of sloppy-paired 1 (slp1) are established by repression mediated by the gap protein Kruppel (Kr) (14–16). Kr is expressed in a central domain and forms an opposing gradient formed by diffusion toward the anterior. Thus, it is not clear whether the differential sensitivity hypothesis can explain the positioning of many different Bcd target genes.

Here we use a combination of bioinformatics predictions and evolutionary conservation to expand the list of known cis-regulatory modules (CRMs) that respond to Bcd-dependent activation. Fourteen candidates were analyzed by reporter gene analysis. Of these candidates, 11 show expression patterns that resemble those of their associated endogenous genes. Several computational methods were used to estimate the relative Bcd binding strengths of all known Bcd-dependent CRMs, but no correlation between Bcd binding strength and AP position of CRM reporter expression was found. We also scanned each CRM for clusters of Hb and Kr sites. These experiments, which suggest that most Bcd-dependent CRMs are regulated by combinations of binding factors, are discussed in light of the differential sensitivity hypothesis.

Materials and Methods

Computer Predictions of Bcd-Dependent CRMs. The position weight matrices (PWM) used to identify Bcd-, Hb-, and Kr-binding sites were described in ref. 17. We used standard techniques (17) to evaluate the statistical significance of finding a given number of binding-site matches within a DNA sequence interval (base pairs). However, because binding-site matches have different PWM scores, the number of identified sites in a given interval will depend on the PWM cutoff value. To generate an adequate recognition model, we scanned a training set consisting of the known Bcd-dependent enhancers at each site cutoff value and identified common parameters shared by all. All known clusters contain (i) at least six Bcd sites scoring at least 4.2 in a window of 550 bp, (ii) at least two sites scoring 5.6 within a 200-bp window, and (iii) at least one site scoring 6.9. Simultaneous scans with these parameters were used to discover new transcriptional signals in genomic regions. Here, we searched the regions of genomic DNA that surround 10 putative target genes [giant (gt), hairy, paired, slp1, slp2, tailless (tll), Dichaete (D), bowl, CG9571, and goosecoid]. The region searched for each gene extends from 20 kb upstream to 20 kb downstream of each gene's coding region.

In Vivo Tests of Predicted CRMs. We chose 13 significant clusters from the above list of genes for reporter gene analysis. We also analyzed a strong cluster located in the sixth intron of the bancal (bl) gene. Fragment sizes to be tested were determined by using an alignment of Drosophila melanogaster and Drosophila pseudobscura genomic sequences in the vista browser (18). The chromosomal coordinates of each identified CRM in the D. melanogaster genome were used to retrieve the corresponding alignment with the D. pseudoscura genome. The curve parameters used were as follows: calculation window, 200 bp; minimal conservation width, 100 bp; and conservation identity, 50%. The boundaries of fragments to be tested were established by picking the region where the conservation identity dropped to <50%. In the case of the cluster in the bl gene, the Bcd cluster was located in a large conserved region, so we selected a fragment containing only the cluster. The cluster near the CG9571 gene was not well conserved, so conservation was not used in choosing the fragment to be tested. Primers used for fragment amplification are shown in Table 1.

Table 1. Primers used for fragment amplification.

Gene	Sequence
bowl	CCGGAATTCGTACCAAGATTTGTGTGC
	TTGGCGCGCCGAATTGAAACCTTCTTAGG
gscA	CCGGAATTCGTGTACCTCTGGGATCCGCTG
	TTGGCGCGCCCCCAGAAGCGGTAAAACCGGG
gscB	GAAGATCTAGGGATTCCGCCTTCTGCGC
	TTGGCGCGCCATGGAGCGGCTGGGGTCT
prd	CCGGAATTCTGCCGATCCCCATCACTAGG
	TTGGCGCGCCCGATCCCAGTAACTCAATC
slpA	GAAGATCTTCCACGCCAGCCAACACTG
	TTGGCGCGCCCTGCGAACCACTGCACTAGCG
slpB	CCGGAATTCGCAGCTCCCAAAGAAGCAC
	TTGGCGCGCCGCTGGGTAATCAGTGCTCGGCATC
slpC	GAAGATCTGTCTAGGGAGCTTCAAGCG
	TTGGCGCGCCGCTAAATAGCAGCCAAATCC
tll	CCGGAATTCATGGATGTGTGTGTTGGCC
	TTGGCGCGCCGGAGCTGCGGTGCGCAGACC
hairy	CCGGAATTCGATTTGCACACGGCGTGAAGG
	TTGGCGCGCCCCTGCGGCTTACTGGCCAAC
D/fsh	CCGGAATTCGTCCCGCCAGCAAGTTAATG
	TTGGCGCGCCGCCGAACCCAAACGGAAGTG
bl	CCGGAATTCATCAGGCACTAGGTGTCCCG
	TTGGCGCGCCGAGAGAACAAGAGGGCGAGCG
CG9571	CCGGAATTCCCCAGAGCATGTGTATAGTACTCG
	TTGGCGCGCCCAGTCCACCGGCCGCCGATCT
gt23	GGGAATTCGGCGACTTGGATCGTGAG
	AAAACTGCAGCTGCCCTGCCCTGCTCTG
gt1	GGGAATTCGATTCCCCTGCATTACG
	AAAACTGCAGGAACGGATGCGCTGCCC

Primers are oriented 5′ (left) to 3′ (right).

PCR-generated fragments contained either EcoRI or BglII sites at the 5′ end and AscI at the 3′ end. They were then cloned into pAOE1, a modified CaSpeR vector that contains the eve basal promoter fused to the lacZ gene and the α-tubulin 3′ UTR (19). Reporter constructs were introduced into the Drosophila germ line by standard injection techniques (20, 21). Between three and five independent lines were obtained for each construct.

Reporter gene expression was examined by using an antisense lacZ probe in at least three independent lines for each construct. In all cases, expression patterns were consistent among the lines, and only one line was further characterized. Endogenous gene expression patterns were assayed in yw embryos by using antisense RNA probes. The bl and D/fsh expression patterns were assayed by using antisense RNA probes corresponding to an ≈900-bp fragment from the largest bl exon and an ≈1.3-kb fragment from the D/fsh coding region, respectively, which were amplified from genomic DNA.

Posterior border positions (PBPs) of expression patterns were measured from digital images of blastoderm-stage embryos acquired at ×200 on a Zeiss Axioskop. Percentage EL was calculated as the distance between the posterior expression boundary and the posterior tip divided by the total EL. EL measurements were performed on five blastoderm embryos from a single transgenic line for each CRM analyzed, and the percentages shown in this paper are averages of the five measurements. We detected no significant variation among embryos containing the same construct. Posterior borders for previously identified CRMs were obtained in the same manner from single embryos in literature images.

Estimating Cluster Binding Strength. The graphical analyses of Bcd CRMs were generated by the cluster draw visualization tool (17) by using the following parameters: minimal score cutoff, 4; score increment, 0.1; window size, 500 bp; and minimum shift of overlapping matches, 2. cluster draw generates a graphical depiction of the statistical significance of a site cluster at different PWM cutoffs within a given window length. The analyses of Hb and Kr sites were performed by using the same program with the same parameters as those used for the Bcd analysis. All graphical depictions of regulatory elements are presented in the same orientation as their associated transcripts.

The power of cluster draw is demonstrated by the in silico mutational analysis of the hb P2 promoter. Two high-affinity binding sites, TCTAATCCC (PWM score, 7.86) and CGTAATCCC (PWM score, 7.97) were mutated to a low PWM score binding site, CGTGATCCT (PWM score, 4.14). We then analyzed 4-kb regions containing these mutations exactly as described above for the wild-type hb P2 sequence.

The structural classifications of Bcd-site clusters were performed as follows: 2-kb regions centered over the defined CRMs were scanned for Bcd site matches by using PWMs (17) and the m-match 2.73 program (22). Three different properties of CRM structure were evaluated: site number, average site score above a cutoff of 5.0, and the PWM score of the highest scoring site. Scores from these analyses were plotted versus PBPs of the patterns driven by each CRM. More details on these procedures are described by Papatsenko and Levine (23).

Results

Identification of Bcd-Dependent CRMs. Previous studies of segmentation gene regulation led to the identification of several CRMs that are directly activated by Bcd. All Bcd-dependent CRMs identified so far contain clusters of Bcd sites. We used simultaneous scans trained on the general structures of the known CRMs to search for more Bcd target elements (see Materials and Methods). This method correctly identified many previously known Bcd target elements and previously uncharacterized Bcd site clusters located throughout the genome. Thirteen previously uncharacterized clusters were found within 15 kb of genes known to be expressed in the early embryo, and all of these were tested for regulatory activity in vivo. We also tested one strong cluster located within the sixth intron of the bl gene. The sizes of tested fragments were determined in part by cross-species comparisons (see Materials and Methods).

In reporter gene assays, 11 of the 14 tested fragments directed expression patterns in wild-type embryos that recapitulate all or part of the endogenous patterns of the associated genes (Fig. 2 and Table 2). These experiments identified several elements that control segmentation genes, including three new gap gene CRMs. Two CRMs were found in the genomic region that lies 5′ of the gap gene gt. One CRM (gt23) is initially expressed in a broad anterior domain and then refines into two stripes (Fig. 2B). A second CRM (gt1) is expressed later in a small dorsal domain very near the anterior tip (Fig. 2D). Double stain experiments (data not shown) indicated that the timing and spatial regulation of both patterns are indistinguishable from the anterior expression domains of the endogenous gt gene (Fig. 2 A and C) (24). We also identified a CRM 3′ of the gap gene tll (Fig. 2F) that drives expression similar to the anterior tll domain (Fig. 2E) (25).

Fig. 2.

Expression patterns directed by Bcd-regulated CRMs. Wild-type mRNA expression patterns (Left) are shown for the indicated genes, as well as lacZ RNA patterns in transgenic embryos carrying reporter constructs driven by computationally predicted CRMs (Right). All embryos are oriented with anterior to the left and dorsal up. (F, J, L, P, and T) Several CRMs direct patterns that do not perfectly coincide with the patterns of their respective endogenous genes, suggesting that other sequences are required for the wild-type expression pattern.

Table 2. Experimentally validated Bcd-dependent CRMs.

CRM	Size, bp	Position	Source
hb P2	243	0 kb 5′	6
Kr CD1	730	3 kb 5′	46
kni	1,019	1.2 kb 5′	39
tll	1,036	1.5 kb 3′	This study
gt1	787	6 kb 5′	This study, 34
gt23	1,213	10 kb 5′	This study, 34
btd	1,080	3 kb 5′	47
otd early	925	3.3 kb 5′	11
salBE	421	10 kb 5′	48
bowl	388	2nd intron	This study
hairy0	470	8 kb 3′	This study
hairy2	1,080	8.5 kb 5′	49
hairy7	932	9.5 kb 5′	40
eve1	788	5.5 kb 5′	50
eve2	488	1 kb 5′	14
slpA	372	1 kb 5′ of slp1	This study, 34
slpB	793	2 kb 3′ of slp1	This study
		7 kb 3′ of slp2
prd	1,400	1 kb 3′	This study, 34
D/fsh	748	2 kb 3′	This study
bl/Mir7	363	Sixth intron of bl	This study
		7 kb 5′ of Mir7
CG9571	757	5 kb 5′	This study

Sizes of minimal fragments sufficient for Bcd-mediated activation are shown. 5′ and 3′ positions are calculated with respect to the start and stop of the coding region, respectively.

Four novel CRMs were identified near known pair rule genes. One CRM was detected in the 3′ region of hairy and drives expression of a small anterior dorsal domain (Fig. 2H) similartothe hairy 0 stripe of the endogenous gene (Fig. 2G) (26). Another CRM is located 3′ of the paired gene and directs expression of an early broad domain (Fig. 2J) that coincides with the later position of the native paired stripes 1 and 2 (Fig. 2I) (27). Two more CRMs (slpA and slpB) (Fig. 2 L and N) were identified in the slp locus, which contains the two related genes, slp1 and slp2. Both slpA and slpB faithfully reproduce parts of the early slp1 and slp2 expression patterns (Fig. 2 K–M) (28).

Four other CRMs were identified near the genes bowl, CG9571, D/fsh, and bl/Mir7 (Fig. 2 O–V). In three cases (bowl, CG9571, and D/fsh), the newly identified CRMs direct patterns similar to their associated endogenous genes (29–32). The final CRM (bl/Mir7) is located in the sixth intron of the bl gene and directs a strong anterior domain of expression (Fig. 2V). However, the endogenous bl gene is expressed nearly ubiquitously (Fig. 2U), which makes it an unlikely target of regulation by this CRM. One potential target of this element is the microRNA gene (Mir7), which is located 7 kb downstream in the eighth intron of bl (33). Four of the CRMs reported here (gt1, gt23, slpA, and D/fsh) were also identified in a recent genome-wide search for new patterning elements based on clusters of combinations of different binding sites including Bcd (34). The fragments used in that study were significantly larger in size but show very similar patterns to those in Fig. 2.

Three fragments that were predicted to be CRMs did not show any embryonic expression in reporter gene assays. One fragment contains a third Bcd site cluster in the slp locus; the other two fragments are located near the goosecoid gene, which is expressed in a Bcd-dependent anterior domain (35). It is not clear why these elements failed to direct expression in the embryo, but the chosen fragments may lack critical sequences for activation.

To confirm that the 11 CRMs are bona fide Bcd target elements, males carrying each reporter gene were crossed to bcd^E1 mutant females (36). In all cases, the lacZ reporter expression patterns were abolished in embryos from these crosses. These results suggest that Bcd activity is required for activation of each identified CRM.

Embryonic Position Versus Bcd-Binding Strength. The discovery of the elements described above brings the total number of experimentally validated Bcd-dependent CRMs to 21 (Table 2). This set of CRMs permitted us to begin an examination of the mechanisms involved in AP patterning by the Bcd morphogen. Specifically, we wanted to correlate the PBPs of the patterns driven by each CRM (see Materials and Methods) with the relative binding “strength” of each cluster. First, we attempted to correlate specific characteristics of the CRMs with their PBPs. Three features were tested: site number, average PWM score above a cutoff of 5.0, and the single highest PWM score (see Materials and Methods). There was no correlation between any of these features and the PBPs of reporter gene expression (Fig. 3 A–C). However, individual CRMs showed very different rankings among themselves when different features were tested. This inconsistency suggests that no single characteristic tested so far accurately reflects the in vivo binding strength of the CRM.

Fig. 3.

Cluster analyses of all known Bcd-dependent CRMs. (A–C) Correlations between PBPs (expressed as the percentage of EL) for each CRM (x axes) and three different CRM features: site number (A), average PWM score at >5.0 (B), and highest PWM score for a single site (C). (D) cluster view representations of the 4-kb genomic regions containing known Bcd-dependent CRMs (see Fig. 1 legend). CRMs are classified into groups according to ranges of PBPs. The average PBP for each CRM is shown above each image.

We also used cluster view software to generate a graphical depiction that simultaneously accounts for several binding characteristics (see Materials and Methods). This program was first tested on 4-kb genomic regions surrounding the previously characterized Bcd-dependent CRMs, otd early (11) and hb P2 (6). The otd region contains more sites than the hb P2 region, and most sites have intermediate or low PWM scores, whereas two sites in the hb P2 region have very high PWM scores (Fig. 1 F and G). Also, the sites in the otd region are dispersed over a relatively large genomic region compared with the hb P2 region. These differences are accurately reflected in the associated color graphs. Furthermore, in silico mutations of single sites in the hb P2 sequence cause dramatic changes in the graphical appearance of the cluster (Fig. 1 H and I).

We next used cluster view to analyze all known Bcd-dependent CRMs and attempted to correlate the resulting graphs with the PBPs of reporter gene expression (Fig. 3D). In general, we did not detect any correlation. For example, there are seven CRMs whose PBPs lie between 79% and 70% EL. Within this group, three Bcd clusters (CG9571, bl/Mir7, and slpA) appear very strong, one (otd early) appears very weak, and three (slpB, bowl, and tll) lie somewhere in between. The other four groups contain fewer CRMs but also show significant differences in the depictions of individual clusters. Together, these results strongly suggest that differential binding alone does not play a major role in determining the limits of expression of most Bcd target genes.

AP Positioning by Combinations of Activators and Repressors. An alternative to the simple gradient mechanism is a combinatorial mechanism, which is supported by several previous studies of Bcd-dependent CRMs (13, 14, 16, 37). To determine whether regulatory inputs by Hb and/or Kr are general characteristics of Bcd-dependent CRMs, we scanned all known elements for clusters of Hb and/or Kr sites and grouped them into distinct classes according to their binding site composition (Fig. 4). Interestingly, only five CRMs (Class 1) seem to be regulated primarily by Bcd alone. The expression patterns driven by all five are restricted to the most anterior 30% of the embryo (average PBP, 75.4% EL; PBP range, 85% to 72% EL). Four CRMs (Class 2) also contain strong Hb clusters but very few, if any, Kr sites. Expression patterns driven by this class extend more posteriorly (average PBP, 52.5% EL; PBP range, 88% to 30% EL), consistent with the hypothesis that synergistic interactions between Bcd and Hb increases CRM sensitivity to the Bcd gradient (13, 14). Six Bcd-dependent CRMs (Class 3) also contain relatively strong clusters of Kr sites but very little, if any, Hb binding. Compared with class 1 CRMs (regulated by Bcd alone), these elements direct transcription in more posterior regions (average PBP, 62.6% EL; PBP range, 70% to 56% EL). This finding is quite interesting in light of the fact that Class 3 elements do not contain significant Hb-binding clusters. Finally, six CRMs (Class 4) showed strong binding clusters for Bcd, Hb, and Kr. PBPs of the patterns driven by these elements varied considerably, ranging from 70% to 25% EL.

Fig. 4.

Analysis of Bcd, Hb, and Kr binding-site clusters in Bcd-dependent CRMs. Each graph depicts individual scans (Bcd, left; Hb, center; Kr, right) of experimentally verified CRMs, which are grouped according to their major transcriptional inputs (shown on left). The average PBP for each CRM is shown above each panel. Color codes representing cluster significance for each regulatory factor are also shown.

In summary, most Bcd-dependent CRMs appear to contain significant contributions from Hb and/or Kr. Several CRMs (including kni and hairy 7) also contain clusters of binding sites for Caudal (Cad), which is expressed in a posterior gradient (38). Previous work suggests that Cad functions with Bcd to activate expression of these CRMs (39, 40).

Discussion

Cis-Regulatory Design Features of Bcd-Dependent CRMs. The hypothesis that positioning of Bcd target genes is controlled by the affinity of Bcd sites was derived from reporter gene analysis of the hb promoter (6). Here we have tested whether Bcd binding to CRMs is a critical factor in the differential positioning of 21 target expression patterns. By using several different scoring methods and the cluster view graphical analysis, we find no correlation between the Bcd binding strength of a given CRM and the posterior border of its expression pattern. Significantly, only five CRMs seem to be regulated primarily by Bcd. Expression driven by these elements is restricted to the most anterior part of the embryo and is reminiscent of patterns directed by synthetic reporter genes containing only Bcd sites (13). Thus, Bcd activity alone cannot account for the expression of Bcd target genes activated in middle and posterior regions.

Ten of the 21 Bcd targets also contain binding activities for Hb, and several contain clusters of Cad sites. Hb and Cad are expressed maternally. Where their inputs have been examined experimentally, they seem to be involved in increasing CRM sensitivity to Bcd-dependent activation (13, 14, 38, 40), which may permit activation in middle and posterior embryonic regions. Embryos lacking both maternal and zygotic hb also lack all anterior structures (41). Also, artificially increasing Hb levels in the anterior part of the embryo can rescue many of the patterning defects in bcd mutants, including the thoracic segments (42). However, some head structures cannot be rescued by ectopic Hb. Perhaps these structures are regulated by a few target genes that are activated by Bcd alone.

At least 12 Bcd-dependent CRMs also contain clusters of Kr sites. Kr has been shown to act as a repressor to set the posterior expression borders of eve stripe 2 and slp1 (14–16). Furthermore, Kr can act as a repressive gradient in posterior regions, where it differentially positions the anterior borders of hairy stripes 5 and 6 (43). These stripes are controlled by CRMs that contain Kr sites of different affinities. The presence of Kr clusters in many Bcd-dependent CRMs (Fig. 4) suggests that Kr-mediated repression may be an important general mechanism for setting posterior borders. The Kr gradient directly opposes the Bcd gradient, and differential repression by this gradient provides an attractive alternative but not mutually exclusive mechanism for setting multiple expression boundaries in the anterior half of the embryo. Several CRMs containing Bcd plus Kr inputs are activated more posteriorly than those containing Bcd alone, even in the absence of a significant contribution from Hb. These expanded patterns is unexpected if Kr acts solely as a repressor, and the mechanism involved here is not clear. One possibility is that low levels of Kr could potentiate the binding of Bcd. Some Kr sites overlap extensively with Bcd sites (14), which is consistent with this idea.

Our results suggest that combinatorial input is a key design feature in the majority of Bcd-dependent regulatory elements. How can we reconcile this with the strong experimental evidence supporting Bcd as a concentration-dependent morphogen? We propose that Bcd plays multiple roles in AP patterning, which may be reflected in the structure of the Bcd-dependent CRMs that have been characterized so far. (i) Bcd may activate a small number of genes in a strictly concentration-dependent manner. These genes would be expressed near the anterior tip. (ii) Bcd may combine with Hb and/or Cad to activate a second set of genes in middle body regions. Obvious candidates are Kr and the other gap genes, zygotic hb, knirps, and gt, all of which function as concentration-dependent repressive gradients (43–45). (iii) For other Bcd target genes (perhaps the majority), the function of the Bcd gradient may be to act in combination with Hb and/or Cad to provide large embryonic domains where CRM activation can occur. These broad expression domains would be refined and positioned by gradients of gap represser proteins, leading to the ultimate establishment of the number and spatial order of expression patterns required for the AP body plan.