Caudal counter-represses Hunchback to regulate even-skipped stripe 2 expression in Drosophila embryos

Hunchback is a bifunctional transcription factor that can activate and repress gene expression in Drosophila development. We investigated the regulatory DNA sequence features that control Hunchback function by perturbing enhancers for one of its target genes, even-skipped. While Hunchback directly represses the eve stripe 3+7 enhancer, we found that in the eve stripe 2+7 enhancer, Hunchback repression is prevented by Caudal binding—this relationship is called counter-repression. We found evidence that this relationship is conserved by comparing predicted binding sites for Hunchback and Caudal across orthologous eve stripe 2 enhancers. These results alter the textbook view of eve stripe 2 regulation wherein Hb is depicted as a direct activator. Instead, to generate stripe 2, Hunchback repression must be counteracted by Caudal binding. We discuss the implications of this interaction for eve stripe 2 regulation and evolution.


Introduction
Bifunctional transcription factors (TFs) that can activate or repress their target genes are critical in animal development (1,2) and are associated with human disease (3,4). The function of these TFs can depend on the context of the enhancer sequences they bind (5)(6)(7)(8)(9). To infer accurate regulatory networks in development from regulatory DNA, we must define the sequence features that control the activity of bifunctional TFs (10).
Other studies examined endogenous eve expression in embryos by misexpressing hb (sna::hb embryo) along the ventral surface of the embryo and found that Hb represses eve stripes 3 and 7 (24). Qualitative measurements of mutated versions of the eve stripe 3+7 enhancer (eve3+7) suggested that Hb repression of eve3+7 is direct (15,17,25,26).
Even after decades of study, the DNA sequence features that control Hb bifunctionality are not known. One hypothesis is that DNA-bound Hb dimers act as repressors while Hb monomers act as activators (dimerization hypothesis). The dimerization hypothesis is supported by computational work that predicts expression of eve3+7 and the gap gene Krüppel (27,28), as well as in vitro experiments identifying zinc finger domains in Hb and Ikaros that allow for dimerization (29). In another hypothesis, binding of a different TF converts Hb from a repressor to an activator (co-activation hypothesis). In vivo measurements of synthetic binding site arrays for Bicoid (Bcd) and Hb support the co-activation hypothesis (30), and Hb co-activation by Bcd has been incorporated into computational models of eve enhancer function (31)(32)(33). Recent models have also incorporated Hb co-activation by the TF Caudal (Cad) (34). Cad activates gap and pair-rule gene expression in the posterior of both long and short-band insects (35)(36)(37)(38), and Cad homologs are critical in vertebrate development and human disease (39,40). Though both hypotheses have been included in computational models of enhancer function (28,34), neither has been experimentally tested.
Here we experimentally test the co-activation hypothesis by perturbing two enhancers that are active in the same cells but respond differently to Hb. eve stripe 7 is generated by two shadow enhancers; Hb represses eve3+7 (24,26) and activates eve2+7, an extended version of eve2min that also generates stripe 7 (41). Because both enhancers are active in the same nuclei, Hb function must be partially controlled by enhancer sequence. In this system, we can test the co-activation hypothesis by measuring the effects of perturbing both enhancers quantitatively and at cellular resolution ( Fig. 1) (42)(43)(44).
We find that Hb directly represses eve3+7 and indirectly activates eve2+7. Indirect Hb activation occurs because Cad binding sites prevent Hb repression in eve2+7; this interaction appears to be evolutionarily conserved. This result alters the textbook description of eve2min, where Hb is depicted as a direct activator that synergizes with Bcd (45). Instead, Hb represses eve2, but this activity is masked by Cad, a direct regulator that counter-represses Hb. We discuss the possibility that additional activators may function as counter-repressors in development.

Hb directly represses eve3+7.
Mutating Hb binding sites in eve3+7 previously showed that Hb defines the anterior border of stripe 3, but the effect of Hb on stripe 7 was unclear (26). The stripe 7 pattern driven by eve3+7 retreats in response to ectopic Hb, but this effect could be indirect (41). To test whether Hb directly represses stripe 7 in the context of eve3+7, we removed predicted Hb binding sites from eve3+7 using the most current PWMs available, and measured expression using quantitative in situ hybridization in wild-type and sna::hb embryos (eve3+7mutHb, Fig.   2A). Some predicted Hb sites remain in eve3+7mutHb because they overlap with sites for other factors. We compared expression driven by this construct to expression driven by a wild-type version ( Consistent with previous results (26), eve3+7mutHb drove an expression pattern where stripe 3 expanded anteriorly ( Fig. 2B and E). In addition, the stripe 7 expression pattern driven by the eve3+7mutHb construct no longer retreats from ventral Hb in sna::hb embryos ( Fig. 2B and F). These results confirm that Hb directly represses eve3+7.
Previous experimental and computational work suggested that Hb directly activates eve stripe 2 (23,31). We previously found that the stripe 7 pattern driven by eve2+7 bulges in sna::hb embryos, and thus hypothesized that Hb directly activates eve2+7 (41). Here, we tested whether Hb directly activates stripe 7 in eve2+7 by mutating predicted Hb binding sites in eve2+7 (eve2+7mutHb, Fig. 3A) and measuring gene expression in wild-type and mutant embryos.
We were surprised to find that mutating Hb binding sites had no effect on the spatial pattern driven by eve2+7 in either genetic background (Fig. 3). We tested for quantitative effects of removing Hb binding sites by measuring expression levels using a co-stain method (44). We found that eve2+7mutHb drove slightly lower expression levels in stripe 2 (p-value = 0.022, To quantify the effect of ventral Hb in individual embryos, we take average measurements from lateral strips, extract the local maxima corresponding to stripes 3 and 7, and subtract those values from the corresponding peaks from the ventral strip. We perform the same analysis in wild-type embryos to account for modulation along the dorsal ventral axis. A decrease in the ventral/lateral difference between wild-type and sna::hb embryos indicates Hb repression, while an increase indicates Hb activation (see F). (E) Lateral line traces from individual wild-type embryos containing eve3+7 reporter constructs are shown (wild-type: grey, n= 26; mutant: blue, n = 9). Embryos are from all six timepoints in stage 5. (F) Differences between ventral and lateral stripe peaks are plotted for individual wild-type and sna::hb embryos in all 6 timepoints in stage 5. Top: wild-type eve3+7 (wt: n = 26; sna::hb: n = 19); Bottom: eve3+7mutHb (wt: n = 9; sna::hb: n = 13). Asterisks indicate significant differences between wild-type and sna::hb embryos (p-value < 0.001, Mann-Whitney U test). Differences between wild-type and sna::hb embryos containing eve3+7mutHb were not significant (p-value > 0.4 for both stripes, Mann-Whitney U test).
Mann-Whitney U test), but stripe 7 levels were indistinguishable between eve2+7 and eve2+7mutHb (p-value = 0.304, Mann-Whitney U test) (Fig 3C). Given the subtle effects that Hb binding site mutations had in eve2+7, we examined the behavior of other eve stripe 2 enhancer fragments in the literature. When we mutated all three footprinted binding sites in a fragment used in comparative studies (46), we observed a significant decrease in stripe 2 expression level ( Fig S2). However, we observed no significant decrease in expression level as a result of mutating the footprinted Hb binding site in eve2min (Fig. S2), despite contradictory evidence from previous qualitative work (23,47). These data indicate that Hb binding site mutations in eve stripe 2 enhancers have small, context dependent effects on gene expression. They further suggest that while Hb may be a weak activator of stripe 2, it does not influence stripe 7 expression at endogenous Hb levels in either eve3+7 or eve2+7. were measured by normalizing lacZ expression levels using a huckebein co-stain (44) and extracting local maxima from lateral line traces. Asterisks indicate significant differences (p-value < 0.05, Mann-Whitney U test). (D) Differences between ventral and lateral stripe peaks are plotted for individual wild-type and sna::hb embryos in all 6 timepoints in stage 5. Top: wild-type eve2+7 (wt: n = 22; sna::hb: n = 26); Bottom: eve2+7 mut Hb (wt: n = 9; sna::hb: n = 11). Asterisks indicate significant differences between wild-type and sna::hb embryos (p-value < 0.001, Mann-Whitney U test).

Caudal binding prevents Hb repression of eve2+7.
Our data indicate that Hb directly represses eve3+7 but has little effect in eve2+7. In this system, the co-activation hypothesis makes a strong prediction: mutating Cad sites in eve2+7 should convert Hb from an activator to a repressor of that sequence. We tested this hypothesis by mutating predicted Cad binding sites in eve2+7 (eve2+7mutCad, Fig. 4A), and measuring gene expression in wild-type and sna::hb embryos. These mutations abolished stripe 2 expression completely and caused an anterior expansion of stripe 7 (Fig. 4B). We hypothesize that this anterior expansion of stripe 7 was caused by unintended mutations in binding sites for the repressor Giant (Gt) (see Discussion). In sna::hb embryos, removing Cad binding sites causes the stripe 7 pattern to retreat (Fig. 4B and D), confirming that without Cad, Hb behaves as a repressor in this sequence. To confirm that this effect was due to direct repression by Hb, we mutated both Hb and Cad binding sites in eve2+7 (eve2+7mutCadmutHb).
eve2+7mutCadmutHb drives a stripe 2 pattern that expanded to the anterior compared to WT eve2+7 ( Fig. 4B and C); we hypothesize this effect is also due to unintended mutations in Gt binding sites (see Discussion). Most importantly, these additional Hb mutations abolished repression by ventral Hb (Fig. 4). These results therefore support a variation of the co-activation hypothesis: Cad prevents Hb from directly repressing eve2+7. . Asterisks indicate significant differences between wild-type and sna::hb embryos (pvalue < 0.001, Mann-Whitney U test).

Caudal binding sites co-evolve with Hunchback binding sites in eve2+7.
Our results reveal that Cad sites in eve2+7 ensure that Hb does not turn off eve stripe 2.
We therefore hypothesized that the balance between Hb and Cad binding sites is under strong selective constraint, which would result in Hb and Cad binding sites co-evolving in eve stripe 2 orthologs. To test this hypothesis, we calculated binding site enrichment scores for Cad and Hb (48) in orthologous eve stripe 2 enhancers previously identified in Drosophila and Sepsid genomes (49). We found that Cad and Hb enrichment scores were significantly correlated in orthologous eve stripe 2 enhancers (Fig. 5A). In contrast, we did not observe any correlation between Hb and Cad binding sites in orthologous eve3+7 enhancers when using a similar threshold for scoring predicted binding sites (Fig. 5A). This result is notable since mutating Cad binding sites in eve3+7 decreases expression of both stripes (Fig. S3), suggesting that Cad also directly activates eve3+7. We also used the same method to also test for co-evolution of Hb and Bcd in eve2 and eve3+7 orthologs, but we observed no significant correlation in either case ( Fig. S4). Although we cannot rule out the possibility that this interaction also occurs in eve3+7 (see Discussion), these results suggest that Cad counter-repression of Hb is a conserved feature of eve stripe 2 enhancers.

Fig. 5. Caudal and Hunchback binding sites co-evolve in orthologous eve stripe 2 enhancers. (A) Enrichment scores for predicted
Hb and Cad binding sites in orthologous eve stripe 2 and eve3+7 sequences from different Drosophila and Sepsid species (49). Scores were calculated by comparing the number of predicted sites to an expected value calculated from the genomic background (48). Scores for Drosophila melanogaster enhancers are indicated in magenta. Spearman correlation and p-value are displayed for each set of enhancers. (B) Summary of current findings. Top: levels of Hb protein (blue), Cad protein (red) and Eve protein (grey) are plotted as a function of anterior-posterior position. Data was taken from the FlyEx database (67). Bottom: cartoon indicating Hb function in different parts of the embryo. In the anterior, Cad levels are low, so Hb represses eve2+7. In the trunk, Cad binding to eve2+7 prevents Hb repression.

Discussion
The eve enhancers are often used to teach the fundamental principles of patterning (45,50). Hunchback (Hb) is a key regulator of eve; it is thought to activate the eve stripe 2 enhancer in concert with Bicoid (Bcd), while it acts as a repressor at other enhancers, such as eve3+7 (24,26,51). Here, we couple quantitative imaging and systematic perturbations of regulatory DNA and TF expression to uncover the DNA sequence features that control Hb activity. We show that Hb activity is controlled by a second TF, Caudal (Cad), and revise the classic description of eve stripe 2 regulation.

Revising our picture of eve stripe 2 regulation
In the textbook picture of eve stripe 2 (45), Bcd and Hb directly activate the enhancer in the anterior of the embryo, while Gt and Kr directly repress it to carve out a single stripe of expression (14,22,23). However, attempts to reconstitute the minimal eve stripe 2 enhancer from its component binding sites have failed, which suggests that there is still more to learn about its regulatory logic (52). For example, these interactions alone cannot explain how eve stripe 2 is repressed in the anterior tip of the embryo. Two other mechanisms have been proposed to account for this: direct repression by Slp1 and downregulation of Bcd activity by Torso (53). Our results suggest an additional mechanism: Hb represses eve2+7 in the anterior due to the absence of Cad (Fig 5B). In our data, an eve stripe 2+7 enhancer without Hb or Cad binding sites generates stripes 2 and 7 in the correct place with an anterior expansion of both stripes, which we hypothesize is due to mutations in Gt binding sites (see below). However, while an eve2+7 enhancer lacking Hb and Cad sites could generate the correct pattern, its function could be easily compromised by the creation of Hb binding sites over evolutionary time.
Regulation of eve stripe 2 by Hb and Cad therefores provide two different buffers against binding site mutations: Hb mediated repression in the anterior may allow for loss of other repressor binding sites (such as Slp1), and counter-repression between Cad and Hb guards against intrusion of Hb binding sites, which would turn off eve stripe 2 without Cad.
The correlation we observe in eve stripe 2 enhancers may reflect counter-repression in that enhancer specifically. However, this bioinformatic signature is not enough to rule out the possibility that Cad also counter-represses Hb in eve3+7. Sensitivity analyses indicate that the correlation of Hb and Cad binding sites in eve stripe 2 orthologs is highest when sites with lower PATSER scores are included, while in eve3+7 orthologs we find that correlation values generally increase with PATSER threshold (Fig S5). These results may reflect differences in the constraints on the regulatory logic of eve2+7 and eve3+7. Hb repression in eve2+7 must be prevented or counteracted by Cad to allow for stripe 2 expression, whereas Hb repression in eve3+7 is necessary for correct positioning of the stripes but may be modulated by Cad binding.
Nonetheless, the co-evolutionary signature we observe in eve stripe 2 orthologs suggests that the functional interaction between Cad and Hb sites may be a conserved feature of eve stripe 2 regulation.
Bcd and Hb have been proposed to synergistically activate eve stripe 2 (14,30), but our results do not support this hypothesis. Computational models that include Bcd/Hb synergy predict that Hb binding site mutations cause large decreases in expression (31). However, our data suggest only modest effects of Hb binding site mutations in accordance with qualitative data in the literature (23,47). In addition, our results do not support the previous co-activation model for Bcd/Hb synergy in eve2+7; without Cad, Hb represses eve2+7 even though Bcd binding sites are the same.

Hb 'bifunctionality' is due to counter-repression
Our previous work indicated that Hb activates eve2+7 and represses eve3+7. Because Hb was "known" to be an activator of eve2min and repressor of eve3+7, we hypothesized that these interactions were direct. Here, we find that Hb activation of eve stripe 7 is indirect. We now hypothesize that sna::hb causes retreat of Gt and subsequent anterior expansion of the stripe 7 pattern (41). Knirps forms the anterior boundary of stripe 7 through the eve3+7 enhancer (26), while Gt has been proposed to form the anterior boundary of stripe 7 through eve2+7 (31). Indeed, we believe that the anterior stripe 7 expansion we observe after mutating Cad sites in eve2+7 is caused by unintended mutations in Gt binding sites. Predicted Gt sites overlap with predicted Cad sites in this sequence, and our Cad mutations disrupt a predicted Gt site that overlaps a footprinted Gt site in eve2min (Fig S6) (54). While these effects make it difficult to determine whether Cad directly activates eve2+7, they do not affect our conclusions concerning the influence of Cad on Hb function. Which TFs activate stripe 7 expression is less clear, though Cad and Zelda are obvious candidates shared by eve2+7 and eve3+7. Overall, the conclusion of our previous paper-that eve stripe 7 shadow enhancers use different regulatory logic-still holds. Hb represses eve3+7 but has little effect on eve2+7 because of counter-repression. Furthermore, we hypothesize that the two enhancers use different repressors to set the anterior border, though this remains to be tested directly. We also previously showed that Hb activates and represses separate Kr shadow enhancers using the same sna::hb misexpression assay (48); it is thus possible that Cad also counter-represses Hb in the proximal Kr enhancer.
Cad may counter-repress Hb through any number of molecular mechanisms. For example, Cad may interfere with Hb via direct protein-protein interactions. The simplest idea is that when Cad is present, Hb does not bind. This isn't supported by genome-wide chromatin immunoprecipitation (ChIP) data for both Hb and Cad; both are bound to eve2+7 (55,56). We note two caveats of this data. First, it is not spatially resolved, so it remains possible that Hb binding is affected in a subset of cells. Second, Cad may disrupt Hb binding quantitatively, enough to affect function but not to detect changes by ChIP. Another possibility is that direct protein-protein interactions between Cad and Hb interfere with Hb protein domains that execute its repressive function (21). Finally, interactions could occur indirectly through co-regulators, or mechanisms analogous to the effects of short-range repressors on nearby activators (57).
Importantly, Cad is not the only counter-repressor in Drosophila development. Stat92E functions as a counter-repressor in the formation of the posterior spiracle, but its target repressor remains unknown (58). Different counter repressors may exert their effects through different molecular mechanisms.
We note that endogenous eve stripe 2 is expressed in caudal mutant embryos (38), which conflicts with our result that mutations in Cad binding sites in eve2+7 abolish stripe 2 expression. This discrepancy may be explained in a number of ways. First, hb expression may be affected in a Cad mutant background. While hb expression is detectable in this background by qualitative in situ hybridization (38), hb expression levels may be lower, which may affect its capacity to repress stripe 2. Second, the piece of DNA that we tested in reporters may not contain all relevant DNA that contributes to eve stripe 2 expression. Indeed, in the endogenous eve locus, the DNA between eve2+7 and eve3+7 contains binding sites for many eve regulators, but is notably devoid of predicted Hb binding sites (41). It may therefore not be affected by Hb repression in the absence of Cad. Finally, the placement of the eve2+7 enhancer adjacent to the promoter in our reporter constructs may allow Hb to exert a stronger repressive effect than in the endogenous locus. As discrepancies between eve3+7 and the endogenous stripes revealed the influence eve2+7 as a stripe 7 shadow enhancer (41), further dissection of stripe behavior in cad mutant embryos may yield additional insights into eve regulatory logic over the entire locus.
A long-term goal of studying gene regulation is predicting the output of a specific enhancer (10). Context-dependent function of TFs [e.g. (59)] is a huge hurdle to meeting this goal. Here we further elucidate counter-repression, one type of context dependence, and offer some strategies for uncovering it in regulatory DNA. First, single mutations of annotated activators or bifunctional TFs may be misleading. Analyzing the effects of single and double mutations together are necessary to classify activators, repressors, counter-repressors and the targets of counter-repression. For example, let's say that you mutate binding sites for TF #1 and expression decreases. TF #1 may either be an activator or a counter-repressor. You now combine mutations in TF #1 with mutations in TF #2, and expression is restored. TF #2 may be a repressor or the target of counter-repression. Finally, you examine the the effect of mutating TF #2 alone. If TF #2 is a repressor, expression will increase, but if TF #2 is the target of counter-repression, there will be no effect. All three experiments are necessary to correctly classify both TF #1 and TF #2. High-throughput mutational studies both in reporters (9) and in vivo (60) may be able to systematically gather this type of data on combinatorial effects.

Conclusion
By combining targeted genetic perturbations with quantitative expression measurements, we uncovered counter-repression as a key feature of eve stripe 2 regulation.
This type of TF interaction, which converts Hb from a repressor to a "nonfunctional" regulator, may be useful for preventing activation in the anterior of the embryo, or for buffering against the effect of binding site introgression. In both cases, counter-repression increases the flexibility of eve stripe 2 regulatory logic, which may impact the plasticity of the enhancer over evolutionary time.

Materials and Methods
Binding site predictions and construct design. We used PATSER (61) to predict binding sites in eve3+7 and eve2+7 for blastoderm TFs. We used position weight matrices (PWMs) derived from bacterial 1-hybrid experiments for the following factors: Bicoid, Caudal, Dichaete, Stat92E, Hunchback, Krüppel, and Nubbin (62). We used other published PWMs for Giant, Knirps, and Tailless (63). Finally, we used a Zelda PWM from a personal communication with Christine Rushlow. Count matrices were converted into frequency matrices for use in PATSER using a pseudocount of 1. In designing binding site mutations, we predicted the effects on predicted sites above a p-value cutoff of 0.003.
Fly work. All reporter constructs were cloned into the NotI/BglII insertion site in the pBΦY reporter plasmid (49) and integrated into the attP2 landing site using injection services provided by BestGene Inc (64). Successful transformants were homozygosed using the mini-white marker. We generated sna::hb embryos as described previously (24) using two copies of the sna::hb transgene, a generous gift from Steve Small.
In situ hybridization. Embryos were collected, fixed in heptane and paraformaldehyde, stained using quantitative in situ hybridization, and imaged using 2-photon confocal microscopy as described previously (42,48). For each embryo, we measure fushi-tarazu mRNA using an antidigoxigenin horseradish peroxidase (HRP) antibody (Roche) and a coumarin-tyramide color reaction (PerkinElmer), as well as lacZ mRNA using an anti-2,4-dinitrophenyl HRP antibody (PerkinElmer) and a Cy3-tyramide color reaction (PerkinElmer). The sna::hb crossing scheme generates wild-type and mutant embryos in approximately equal proportions; we identified embryos misexpressing hb using the ftz expression pattern. For experiments involving a hkb costain, we stained embryos in the same batch, discarded outliers, and ensured that hkb levels and lacZ levels were significantly correlated as described previously (44,48) Binding site overrepresentation analysis. Binding site enrichment was performed as described previously (48). To perform this analysis, we used PWMs from FlyFactorSurvey (65) with sequencing data collected on the SOLEXA platform. For our sensitivity analysis, we systematically varied the PATSER score cutoff used to identify relevant sites and calculated the effect on the Spearman correlation coefficient.