Multifactorial Regulation of a Hox Target Gene

Hox proteins play fundamental roles in controlling morphogenetic diversity along the anterior–posterior body axis of animals by regulating distinct sets of target genes. Within their rather broad expression domains, individual Hox proteins control cell diversification and pattern formation and consequently target gene expression in a highly localized manner, sometimes even only in a single cell. To achieve this high-regulatory specificity, it has been postulated that Hox proteins co-operate with other transcription factors to activate or repress their target genes in a highly context-specific manner in vivo. However, only a few of these factors have been identified. Here, we analyze the regulation of the cell death gene reaper (rpr) by the Hox protein Deformed (Dfd) and suggest that local activation of rpr expression in the anterior part of the maxillary segment is achieved through a combinatorial interaction of Dfd with at least eight functionally diverse transcriptional regulators on a minimal enhancer. It follows that context-dependent combinations of Hox proteins and other transcription factors on small, modular Hox response elements (HREs) could be responsible for the proper spatio-temporal expression of Hox targets. Thus, a large number of transcription factors are likely to be directly involved in Hox target gene regulation in vivo.


Introduction
Distinct morphological structures exist along the anteriorposterior (A/P) axes of animals, and the Hox genes represent the major regulators for patterning of this body axis in organisms as diverse as fruit flies, fish and humans [1,2,3]. Since Hox genes code for transcription factors, Hox-dependent morphogenesis is driven by the differential regulation of downstream genes [4,5,6]. In line with the very diverse and many-fold effects of Hox proteins on morphogenesis, Hox transcription factors are known to regulate a large number of Hox downstream genes [7,8], including genes that themselves have broad effects on morphology, as well as genes involved in terminal differentiation [reviewed in 3].
Hox genes are expressed in broad and partially overlapping domains along the A/P axis [1,2,3], and their constant and simultaneous activity within hundreds of cells is required throughout development [1]. Despite being active in a very large number of cells, Hox proteins affect target gene expression in precisely defined sub-domains in the animal [for example 9,10,11,12,13,14,15]. In the most extreme case the regulation of a Hox target gene can be limited to a single cell [16]. In addition, some downstream genes can be activated and repressed by the same Hox protein depending on the tissue or developmental stage. Finally, this context dependency also allows a single Hox protein to affect distinct sets of target genes in the same cells during the course of development [8,11,17]. While more and more of these complex regulatory interactions are being described, the molecular mechanisms underlying the spatio-temporal precision of Hox target gene regulation is only poorly understood. This is in large part due to our limited knowledge of the design and function of Hox-dependent enhancers and promoters and their interaction with the regulatory environment.
Only a few Hox regulated enhancers have been analyzed in some detail in Drosophila [3]. For example, the activation of wingless (wg) expression in the visceral mesoderm of Drosophila embryos had been shown to be dependent on the Hox protein Abdominal-A (Abd-A) and the Dpp/TGF-b signalling pathway, with both activities functioning on a small wg enhancer [11]. Here, two transcriptional effectors of the Dpp/TGF-b pathway, Mother against dpp (Mad) and Creb, had been shown to mediate the Dpp response on the wg enhancer, and were thus assumed to work in concert with the Hox protein Abd-A [11]. Only recently, Mad and another effector of the Dpp/TGF-b pathway, Medea (Med), have been found to collaborate with the Hox protein Ultrabithorax (Ubx) to repress transcription of the Hox target gene spalt major (sal) in the haltere by independently interacting with adjacent Mad/ Med and Ubx binding sites in a small sal enhancer [18]. And finally, two transcription factors very well known for their function in the Drosophila segmentation cascade, Engrailed (En) and Sloppy paired 1 (Slp1), were shown to assist the Hox proteins Ubx and Abd-A in repressing Distal-less (Dll) expression in the abdomen of Drosophila embryos by occupying their identified binding sites in a minimal Dll enhancer [10].
Another well-studied direct Hox target gene in Drosophila is the apoptosis gene rpr, which is activated by the Hox protein Dfd in the anterior part of the maxillary segment through four binding sites located in the rpr-4S3 regulatory fragment [13]. Since Dfd is active throughout the maxillary segment, whereas rpr RNA is found only locally [19], it seemed likely that additional factors contribute to region specific Dfd-dependent rpr expression. Here, we find that eight transcriptional regulators, with diverse roles in patterning or differentiation processes, co-operate with Dfd in the regulation of rpr. Within their spatially restricted expression domains, these regulators are recruited to a minimal rpr enhancer through specific cis-regulatory DNA sequences and act together with Dfd to regulate rpr expression in the appropriate spatio-temporal pattern. Thus, our data support the idea that the combinatorial activity of Hox proteins and diverse transcriptional regulators on small regulatory elements is responsible for the spatially and temporarily restricted expression of Hox target genes in vivo. In addition, our data show that even small Hox-dependent enhancers are complex and integrate diverse regulatory inputs, which result in precise spatio-temporal expression of Hox target genes.

Dissection of the Dfd-Dependent rpr Enhancer
To isolate a minimal rpr regulatory element able to recapitulate endogenous expression in the maxillary segment, we divided the known 674 bp long rpr-4S3 enhancer element [13], and analyzed lacZ expression driven by the resulting sub-fragments ( Figure 1). We found that the 39 part of the enhancer termed rpr-4S3/39, which contained all previously defined Dfd binding sites [13], was sufficient to drive lacZ expression in a few cells located in the anterior part of the maxillary segment ( Figure 1C9). Doublelabelling experiments demonstrated co-localization of rpr and lacZ transcripts in the rpr-4S3/39 line in a subset of rpr expressing cells ( Figure 1C0).
We next tested the functional relevance of the Dfd binding sites in the rpr-4S3/39 enhancer by mutational analysis. Surprisingly, we observed an increase in lacZ expression in the anterior part of the maxillary segment after mutating all Dfd binding sites ( Figure 1D9), rather than a reduction, as observed with the same mutations in the context of the larger rpr-4S3 reporter [13]. Additionally, weak lacZ expression was observed in the anterior part of all other segments in the rpr-4S3/39-Dfdmt line ( Figure 1D). These findings showed that the Dfd binding sites are not exclusively used for activation, but also for repression of rpr transcription. Additionally, these results indicated that the overall binding site composition of the rpr-4S3 enhancer determines its regulatory output and that most information for repression is located in the 39 part of the enhancer. Consistently, we found ectopic reporter gene expression in the posterior part of the maxillary segment when using the remaining 59 part of the rpr-4S3 enhancer (rpr-4S3/59) (Figures 1E and 1E9). While the rpr-4S3/39 enhancer harbours most of the binding sites for transcriptional repression, it still has the capacity to direct region-specific activation of rpr, since lacZ expression is maintained in a few cells in the anterior part of the maxillary segment in the rpr-4S3/39 line ( Figure 1C9). Taken together, these results showed that the Dfddependent regulation of rpr is highly complex and that Dfd has activating and repressing activity even when acting on a small regulatory element. Thus, we decided to study the rpr-4S3/39 enhancer in detail, because its reduced complexity provided a sensitive background to uncover the mechanisms of Hox target regulation in vivo.

Identification of Transcription Factors Necessary for Proper rpr Expression
To test the effect of Dfd on the minimal rpr-4S3/39 enhancer fragment, we ubiquitously expressed Dfd in the rpr-4S3/39 reporter strain using the armadillo (arm)-GAL4 driver [20]. We observed specific reporter gene activation in the anterior part of every segment ( Figure 2F), a result we had observed before when using the rpr-4S3 reporter line (data not shown). lacZ RNA never extended into the anterior-dorsal or anterior-ventral zone (Figures 2A and 2F). This led us to hypothesize that essential factors for the Dfd-dependent rpr expression are locally expressed in sub-domains of every segment, either in the anterior part if they act as activators or in the posterior, dorsal or ventral part if they act as repressors on the rpr-4S3/39 enhancer.
To test this hypothesis mechanistically, we assayed 16 transcription factors, which meet the expression criteria outlined above, along with two known Dfd interactors, Apontic (Apt) and Disconnected (Disco) [21,22] for their ability to modulate Dfddependent rpr expression (Table 1). We ubiquitously mis-expressed all factors in embryos harbouring the rpr-4S3/39 reporter, and categorized them dependent on their capacity to affect reporter gene expression ( Figure 2). Seven transcription factors were able to elicit the response predicted by their expression patterns ( Figure  S1): Apt and Glial cells missing (Gcm) activated reporter gene expression ( Figures 2C and 2D), whereas Brinker (Brk), Disco, Dorsocross 1 (Doc1), En and Slp1 repressed lacZ expression ( Figures 2K to 2O). Two of the factors identified in our screen, En and Slp1, have recently been shown to assist Hox proteins in target gene regulation [10], supporting the validity of our approach. Additionally, Disco and Apt were known to genetically and/or biochemically interact with Dfd [21,22]. While over-expression of the activating transcription factors alone had a modest effect ( Figures 2B to 2D), simultaneous over-expression with Dfd strongly enhanced reporter gene expression ( Figures 2G to 2I). Similar effects were observed when we analyzed endogenous rpr RNA expression in these embryos ( Figures S2E to S2H), suggesting that these factors are likely to function in concert with Dfd in the induction of rpr. Using this co-expression strategy, we identified another factor modulating rpr expression: Empty spiracles (Ems) enhanced the ability of Dfd to activate reporter gene expression ( Figure 2G), although Ems mis-expression alone had no effect ( Figure 2B). Since other candidates tested had no effect on reporter gene expression either alone or in combination with Dfd

Author Summary
Bilateral animals share a common genetic mechanism to control development along the anterior-posterior body axis, and transcription factors of the Hox class are key regulators of this conserved process. It is thought that Hox proteins drive morphological diversification of body segments by differentially controlling the expression of downstream genes. However, due to their highly conserved DNA binding domain, the homeodomain, Hox proteins alone bind to very similar and frequently occurring sequences in the genome. This implies that Hox proteins alone are likely insufficient to activate or repress their target genes in a proper spatio-temporal fashion. In contrast to this observation, Hox proteins have very specific effects during development and execute their function with high precision. A solution to this paradox could lie in the context-specific interaction of Hox proteins with other transcriptional regulators; however, only a few examples are known. By analysing the mechanism underlying the regulation of the Hox target gene reaper, we identified a set of eight transcription factors to be important for the precise spatio-temporal regulation of this gene. Based on our findings, we suggest that Hox proteins functionally interact with a plethora of unrelated transcription factors on small, yet complex enhancer elements to execute their specific functions throughout development of diverse organisms.

Role of Co-Regulatory Factors in Dfd-Dependent rpr Expression
To test whether the factors identified are necessary for proper rpr transcription in the maxillary segment, we analyzed rpr transcripts in embryos mutant for the individual transcription factor genes ( Figure 3). Additionally, we studied the morphology of the gnathal lobes, in particular the boundary between the maxillary and mandibular segments, since it is known that maintenance of this boundary depends on proper rpr activity [13]. In mutant embryos of two activators, gcm and apt, we observed a reduction of rpr expression in the anterior part of the maxillary segment ( Figures 3C9 and 3E9) and a slight fusion of the maxillary and mandibular segments ( Figures 3C0 and 3E0). The latter phenotype was not as pronounced as in Dfd mutants ( Figure 3B0), which exhibit a strongly reduced rpr expression in the anterior part of the maxillary segment ( Figure 3B9; [13]). In ems mutants, the maxillary-mandibular boundary developed normally ( Figure 3F0). Here, rpr transcript levels were only reduced in the middle part of the anterior rpr expression domain, whereas dorsally and ventrally to this area rpr transcript levels were elevated ( Figure 3F9). As reported previously [23], a loss of the mandibular segment was observed in ems mutant embryos ( Figure 3F0). To test the interactions between Dfd and the activating transcription factors genetically, we extended our studies to Dfd gcm double mutants. In these embryos, rpr expression and the formation of the maxillary-mandibular boundary were completely lost ( Figures 3D9  and 3D0). This result not only confirmed an important role of the activating factors for rpr expression and the maintenance of the segment boundary, but also suggested that Dfd and Gcm act independently.
In embryos mutant for the repressing transcription factor genes, we observed ectopic rpr expression in the maxillary segment, primarily in the central or posterior part ( Figures 3G9 to 3K9), showing that the factors are involved in repressing rpr transcription. The de-repression of rpr transcription in only a few cells suggested that repression of rpr transcription is redundant. Thus, we aimed to analyze rpr expression in embryos mutant for multiple transcription factor genes with repressive function. However, due to lethality, we were not able to generate any double mutant combination. To circumvent this problem, we made use of the fact that the activity of Slp1, Brk and En is dependent on the transcriptional co-repressor Groucho (Gro) [24,25,26,27]. We hypothesized that gro mutant embryos should behave similarly to a triple knock-out of the repressor genes with regards to rpr regulation. Consistently, the number of cells ectopically expressing rpr was further increased in gro mutants when compared to the single mutants. The effect was most pronounced in the posterior part of the maxillary segment ( Figures 3L and 3L9). Phenotypic analysis of the repressor mutants revealed that the maxillarymandibular boundary was not affected ( Figures 3F0 to 3K0). This was consistent with largely unchanged rpr transcription in the anterior part of the maxillary segment in all mutants for repressive transcription factors ( Figures 3F9 to 3K9). Nevertheless, the overall morphology of the gnathal lobes in these mutants was abnormal ( Figure 3).
To exclude the possibility that changes in Dfd expression cause the modifications in rpr activity and boundary formation observed in the transcription factor mutants, we analyzed Dfd protein localization in these embryos. Since Dfd expression was always unaffected (Figures 4B0 to 4E0), we concluded that the factors identified do not act upstream, but in parallel to Dfd in the regulation of rpr. In addition, we could rule out cross-regulatory effects between gcm and Slp [28] in the maxillary segment (data not shown).
After having shown a functional relevance for the identified transcription factors, we studied their contribution to rpr expression in the context of the rpr-4S3/39 enhancer element, since gene expression is often resistant to the modulation of individual trans-acting factors acting on large and redundant enhancers. We observed a strong reduction of lacZ RNA in embryos mutant for the activating transcription factors ( Figures 4B9  to 4D9). This result suggests that all three factors play important roles in rpr activation and that they act on regulatory elements contained within the rpr-4S3/39 enhancer. In embryos mutant for repressing transcription factors, ectopic lacZ activation was observed only in some maxillary cells, as shown for the Doc mutant ( Figure 4E9), suggesting that transcriptional repression is redundant even at the level of the rpr-4S3/39 enhancer. Consistent with the binding site mutations ( Figure 1D9), lacZ was strongly activated in the anterior part (but also in other parts) of the maxillary segment in Dfd mutants ( Figures 4F and 4F9). These results confirmed that in the rpr-4S3/39 context Dfd acts primarily as a repressor, and suggested that full repression is achieved by the combined action of Dfd and additional transcription factors.

Direct Interaction of Co-Regulatory Factors with Minimal rpr Enhancer
We next addressed whether the identified factors act directly in the regulation of rpr in the maxillary segment. Thus, we studied the expression of rpr and all factors with cellular resolution using double-labelling experiments. rpr transcripts always co-localized with the activating transcription factors, whereas they were mostly excluded from cells positive for the repressing transcription factors ( Figures 5F9 to 5M9). We obtained the same result when lacZ transcript distribution and expression of the co-regulatory factors in the rpr-4S3 reporter line were analyzed ( Figures 5N to 5U). rpr transcripts and the activators co-localized in distinct sub-domains of the rpr expression zone: Ems in the dorsal most, Gcm in the middle and Apt in the ventral most part (Figures 5F9 to 5H9). This suggests that individual activating factors are responsible for rpr transcription in distinct cells in the anterior part of the maxillary segment and that their combined activity is required for the expression of rpr in its complete domain.
To further test whether the identified transcription factors are directly involved in the expression of rpr on the mechanistic level, we mapped transcription factor binding sites in the rpr-4S3/39 enhancer using phylogenetic footprinting [29,30,31]. Using species-specific rpr RNA probes, we could show that rpr was expressed specifically in the anterior part of the maxillary segment in all seven Drosophila species chosen ( Figure S3). We then isolated the rpr-4S3/39 enhancer fragment from all species and after aligning the sequences using the TCoffee algorithm [32,33], we identified three highly conserved boxes, which contained all four Dfd binding sites previously characterized ( Figure S3). Furthermore, we found known consensus binding motifs for three of the eight factors within the conserved regions ( Figure 5A, Figure S3) [10,34,35,36].
To molecularly test direct binding of all factors identified in the rpr enhancer, we performed electrophoretic mobility shift assays (EMSA). We found that all eight transcription factors interact with conserved regions in the rpr-4S3/39 enhancer in vitro ( Figure 5, Figure S4). To define the DNA sequences necessary for this interaction, systematic competition experiments using overlapping and mutated oligonucleotides for each conserved box were performed. This analysis allowed us to confirm the published Gcm consensus sequence -ACCCGCAT- [37] (Figure 5D), which is located directly adjacent to Dfd binding site 1 in the rpr-4S3/39 fragment ( Figure 5A, Table 2). Similarly, En and Slp1 binding sites are found in close proximity to Dfd binding sites 2 and 3 ( Figure 5A). Our EMSA analysis uncovered that the assisting factors Apt, En, Slp1 and Brk interact with binding sites slightly divergent from published consensus sequences ( Figure 5, Figure  S4, Table 2) [10,38,39,40]. Finally, we identified unknown DNA binding sequences for the assisting factors Doc1 and Disco ( Figure 5B, C, Figure S4, Table 2). Our competition experiments revealed that Disco protein interacts with two binding sites in the rpr-4S3/39 enhancer ( Figure 5B, Figure S4), which share an invariant five nucleotide core motif, -TGACA-( Figure 5A, Table 2).
To test if the identified target sequences are directly bound by the factors in vivo, we analyzed the ability of the transcription factors to bind to the rpr-4S3/39 enhancer in the context of chromatin. To this end, we performed chromatin immunoprecipitation assays (ChIPs) for all factors for which functional antibodies were available. A significant enrichment of the rpr-4S3/ 39 locus was observed using Dfd, Gcm and En antibodies ( Figure 6). These results demonstrated that Dfd, Gcm and En directly interact with specific target sequences in the rpr-4S3/39 enhancer in vivo.
Our EMSA experiments also revealed that all factors are able to bind independently of Dfd to the rpr-4S3/39 enhancer, since formation of protein complexes between Dfd and the coregulatory transcription factors was not observed (Figures 5B to 5E, Figure S4). Consistently, GST pulldown and yeast-two hybrid assays did not provide any evidence for direct interactions of the identified transcription factors with Dfd (data not shown). To exclude the possibility that more than two factors are required for complex formation on DNA, which has been shown before for other HREs [41], we performed EMSA experiments using a mixture of three transcription factors and conserved box 1 as probe. In this region the binding sites for Disco, Dfd and Gcm lie in close proximity, which is considered a requirement for cooperative binding [41]. However, we did not observe a higher-order complex when incubating conserved box 1 with extracts containing all three transcription factor proteins (data not show). Thus, we conclude that these regulators do not bind the rpr-4S3/39 enhancer in a cooperative manner.

Contribution of Co-Regulatory Factors to Dfd-Dependent rpr Expression
We next tested the importance of the identified DNA binding sites for Dfd-dependent rpr expression in the embryo. To this end, we mutated all sites for the activating or repressing transcription factors in the rpr-4S3/39 element and analyzed reporter gene expression. We found that lacZ expression was abolished, when binding sites for all three activating factors, Gcm, Apt and Ems, either alone or in combination with the Dfd binding sites, were mutated ( Figures 7C and 7D). These results show that these factors, independently of Dfd, are responsible for activation of the rpr-4S3/39 enhancer element in the anterior part of the maxillary segment. A reduction of lacZ expression was even observed when a single binding site for an activating factor was mutated (Figures 7G and 7G9), further suggesting that the combined action of all three factors is required for activation of the rpr-4S3/39 element. In embryos carrying the rpr-4S3/39 enhancer fragment with all sites for the repressing transcription factors mutated, we observed ectopic reporter gene expression in some, but not all maxillary cells ( Figure 7E9). Additionally, lacZ was expressed throughout the embryo ( Figure 7E). Finally, when all Dfd and repressor sites were mutated, lacZ transcription was activated in even more cells of the maxillary segment ( Figure 7F and 7F9), confirming that Dfd acts in parallel to the repressing factors. Here, reporter gene expression  To test specificity of binding of the proteins to the DNA fragments, competitor oligonucleotides for the mapped binding sites were used either in their wild-type (c wt ) or mutant (c mt ) sequence versions. Red arrowheads indicate specific DNA-protein complexes containing Dfd protein, turquoise, green, orange or light-yellow arrowheads indicate specific DNA-protein complexes containing Disco, Doc1, Gcm or Ems proteins, respectively. Note that in all competitor oligonucleotides only binding site sequences for co-regulatory transcription factors are mutated, but not for Dfd binding sites. (F to U9) Protein or RNA co-localization of co-regulatory was strongly increased in the anterior and posterior part of the maxillary segment ( Figure 7F9). However, we never observed lacZ expression throughout the whole maxillary segment, suggesting that there are additional, unidentified repressors of rpr expression. When single binding sites for repressing transcription factors were mutated, as shown for the Doc1 binding site ( Figure 7H9), ectopic reporter gene expression in the maxillary segment was observed, confirming the importance for the direct interaction of repressors with rpr-4S3/39 element.

Discussion
We have shown that eight transcriptional regulators, Apt, Gcm, Ems, En, Slp1, Brk, Doc1 and Disco, are required in addition to Table 2. Published and mapped binding sites for all transcription factors within the rpr-4S3/39 Hox response element.    the Hox protein Dfd to properly regulate the expression of the apoptosis gene reaper in a specific subset of cells of the maxillary segment. The finding that such a large number of structurally unrelated transcription factors with important and diverse functions during differentiation and cell-type specification processes assist Dfd in the regulation of the rpr4S3/39 enhancer element was surprising. For example, Gcm is one of the major regulators of glial cell differentiation, consistent with its expression in glial precursor cells during embryogenesis [42,43]. Two other factors, Brk and Doc1, are both known to play important roles in the Dpp/TGF-b signalling pathway: Brk, a genuine transcriptional repressor [24], acts as a negative regulator of Dpp-dependent genes [44,45,46], whereas Doc1, one of the three genetically redundant Dorsocross transcription factors required for amnioserosa development in Drosophila [47], is a direct target of the Dpp pathway [48]. Furthermore, Apt and Disco have previously been found in genetic screens designed to identify modifiers/interactors of Dfd [21,22]. Mutations in these genes, which are expressed in the gnathal segments [21,22], result in severe malformations of structures derived from these segments, similarly to the defects observed in Dfd mutants. And finally, we have identified three factors known to be critically involved in patterning the A/P axis as important regulators of rpr expression: the gap-like segmentation gene product Ems, which is important for patterning embryonic head structures [49,50], and two segment polarity factors, En and Slp1. Taken together, our findings suggest that proper spatiotemporal Hox target gene regulation is achieved by the combined action of multiple transcriptional regulators: the Hox proteins themselves and a large number of structurally diverse transcription factors. Although the interaction with additional transcription factors has been reported before [11,14,51], the finding that a multitude of diverse factors is required to regulate the activity of a small HRE adds a new layer of complexity to the mechanisms of Hox target gene regulation.
Our results not only demonstrate that the newly identified factors are functionally involved in rpr regulation, but also show mechanistically that they contribute to localized rpr activation by direct interactions with specific DNA sequences located in the rpr4S3/39 enhancer element both in vitro and in vivo. Additionally, all factors very likely bind to their target sequences independently of Dfd. Thus, our findings support and significantly extend recent observations: the repression of the Hox target sal in the haltere requires the direct interaction of the Hox protein Ubx and two Dpp downstream effectors, Mad and Med, with adjacent binding sites in the sal1.1 CRE [18]. As in our case, no evidence for a direct cooperative interaction of the assisting transcription factors with the Hox protein was detected. Thus, we postulate that the Hox-dependent regulation of rpr expression is, as in the case of sal repression by Ubx, achieved through combinatorial regulation, in which two or more regulatory proteins bind to nearby sites, but not necessarily to each other [18].
While combinatorial regulation of gene expression has been extensively studied for diverse transcription factors [52,53], our results shed new light on the mechanisms of Hox target gene regulation. Previously, much attention focused on the Hox cofactors Extradenticle (Exd) and Homothorax (Hth), which allow Hox proteins to differentially recognize and select some of their target genes through cooperative complex formation [54,55,56,57]. Although Hox cofactors like Exd can explain why different Hox proteins have different DNA binding specificities, the interaction with these factors was not able to explain how broadly expressed Hox proteins are able to affect target gene expression in only a subset of cells. One of the major reasons for that is that Exd and Hth, the only well-known Hox cofactors, are expressed throughout the embryo and interact promiscuously with most Hox proteins. In addition, studies on Exd and Hth revealed that Hox target gene regulation more or less inevitably includes complex formation between Hox proteins and assisting coregulatory factors. We and others have now shown that the ability of a broadly expressed Hox protein to regulate a target gene in a proper spatial and temporal context is achieved by the Hoxindependent recruitment of context-specific transcription factors to cis-regulatory sequences present in compact HREs [18]. Based on our findings, we now suggest that a large number of transcription factors could dictate the transcriptional output in combination with the respective Hox protein by binding selectively and independently to cis-regulatory sequences within HREs of target genes ( Figure 7I). Since every cell has a unique combination of transcription factors, the combinatorial interactions for the broadly expressed Hox proteins are almost limitless in such a scenario, accounting for the precise modulation and fine-tuning of Hox target gene regulation, even on the level of the individual cell. Additionally, this model can explain how Hox proteins can act as repressors in one context and as activators in another, because the combined transcriptional output is dependent on the regulatory activity of all transcription factors assembled on a HRE ( Figure 7I). There are several lines of evidence that support this model: first, the invariable ectopic activation of Hox downstream genes in spatially restricted domains of every segment when upstream Hox proteins are ubiquitously mis-expressed [8], and second, the accumulation of binding sites for additional transcription factors in enhancers predicted to be controlled by Hox proteins [8]. Alternatively, it seems also possible that Dfd regulates the expression of its target gene rpr only in some maxillary cells, while the novel co-regulatory factors mediate regulation in other maxillary cells. However, since Dfd protein is present in all cells of the segment, the first model seems more likely. Irrespective of the mechanism used by Dfd, it will be essential to study the architecture of HREs, with a special focus on the binding site composition of these enhancers and the diverse factors binding to them to further advance our understanding of Hox target gene regulation in vivo.
It has been argued before that context-specific transcription factors assisting Hox proteins in target gene regulation are not likely to act as transcriptional repressors or activators themselves, but rather recruit co-repressors and/or co-activators, and thereby dictate the transcriptional output imprinted in HREs [18,58]. Our finding of the co-repressor Groucho playing a role in the Dfddependent repression of rpr transcription now substantiates this hypothesis, since three of the factors identified in our work, En, Slp1 and Brk, are known to require interactions with the Groucho co-repressor for the transcriptional repression of some of their downstream genes [24,25,59]. Interestingly, at least two other coregulatory transcription factors identified in this work are also known to interact with co-activators/co-repressors: Apt is able to recruit the transcriptional co-activator Multiprotein bridging factor 1 (MBF1), thereby mediating Apt-dependent transcriptional activation [35]. Disco has been found in a yeast-two-hybrid screen to interact with the well-known co-repressor C-terminal Binding Protein (CtBP) [60], which is also recruited by Brk to repress some Dpp-responsive genes [24]. Since there is accumulating evidence that co-repressors, like CtBP, execute their function on transcriptional regulation through chromatin modification [61,62], it is tempting to speculate that Hox proteins regulate their target genes also by epigenetic control mechanisms. In summary, the multitude of potential regulatory mechanisms used by Hox proteins might be the reason why it has been impossible to fully elucidate how Hox proteins mediate their function with high specificity and precision in vivo. In the light of recent advances, it now seems likely that the mechanisms leading to functional specificity of Hox proteins are dependent on the cellular context, the composition of the target enhancer element and the identity of the individual Hox protein.
Thus, one could argue that during evolution Hox proteins have undergone individualization in trans, as well as sequence diversification in cis.

Histology and Scanning Electron Microscopy
In situ hybridization and immunochemistry were performed as described [74,75]. Fluorescent RNA / protein double labelling and fluorescent duplex in situ hybridizations were done as described previously [13,76]. Probe detection was done using the TMR and Fluorescein TSA Amplification kits from PerkinElmer

Electrophoretic Mobility Shift Assays
EMSA was performed as described previously (Lohmann et al., 2002). For the zinc finger transcription factor Disco, 1 mM ZnSO 4 was included in the binding reaction. For the mapping of binding sites, the ability of all eight transcription factors to interact with conserved boxes 1 to 3 was tested, to define the interaction domains on the rpr-4S3/39 fragment. Subsequently, if known binding sites were present within the conserved boxes, competition experiments were performed to test if these sites are necessary for binding. For all factors with unknown binding sites, systematic competition experiments using overlapping and mutated oligonucleotides covering the binding region were performed. All oligonucleotide sequences used for these experiments can be obtained upon request.

Chromatin Immuno-Precipitation (ChIP)
ChIP experiments were performed as described previously at www.flychip.org. Four independently staged wild-type embryo populations were collected, chromatin samples were prepared from 5 to 9.5 hr embryo collections. Antibodies used were the following: guinea pig anti-Dfd, guinea pig anti-IgGs (gift from H. Schwarz, MPI Tuebingen), mouse anti-En, mouse anti-LacZ (Invitrogen), rat anti-Gcm (gift from M. Wegner, University Erlangen) and rat anti-GFP (Invitrogen). A dilution of 1:500 was used for the anti-Dfd, anti-En and anti-Gcm antibodies, the mock antibodies were used at equivalent protein concentrations. Amplification of the rpr-4S3/39 and an unrelated, non-coding control locus were analyzed by quantitative real-time PCR in technical triplicates using at least two biological replicates. Precipitates were normalized to input DNA (i.e., sonicated, pre-ChIP DNA) and compared to the non-coding negative control region. A PCR efficiency of 1.8-fold amplifications per cycle was assumed. PCR primer sequences can be obtained upon request. whereas Gcm is able to ectopically induce rpr expression alone. To select identical stages, three characteristic spots of rpr expression in the thoracic segments normally seen in stage 11 wild-type embryos (marked by three asterisks) were used. In (B to H) one spot of ectopic rpr expression at the very posterior end in the prd-GAL4 overexpression embryos is marked by a red arrow, the blue box outlines an additional stripe of rpr RNA expression in the T3 primordium.  Figure S3) as shift probe. EMSA was performed using no protein (P), translation lysate only (L) and lysate with Slp1 protein (S). c 30 to c 36 represent competitor oligonucleotides with consecutive base-pairs mutated. Competition experiments revealed that sequences mutated in the oligonucleotides c 33 include binding site for the Slp1 protein. The purple arrowheads indicate specific DNA-protein complexes containing Slp1 protein. The asterisk indicates a complex with lysate protein seen also in the control. (A9) EMSA using box 3 (as shown in Figure S3) and no protein (P), translation lysate (L), lysate with Slp1 protein (S) and lysate with Dfd protein (D). To test specificity of binding of Slp1 protein to the DNA fragment, competitor oligonucleotides for the mapped Slp1 binding site were used either in their wild-type (c wt ) or mutant (c mt ) sequence versions. The purple arrowheads indicate specific DNA-protein complexes containing Slp1 protein. Note that in the competitor oligonucleotides only the binding site sequence for the Slp1 protein is mutated, but not for the Dfd binding site sequence. (B) EMSA for mapping of Disco binding site 1 in the rpr-4S3/39 enhancer using box 1 (as shown in Figure S3) as shift probe. EMSA was performed using no protein (P), translation lysate only (L) and lysate with Disco protein (C). c 1 to c 7 represent competitor oligonucleotides with consecutive base-pairs mutated. Competition experiments revealed that sequences mutated in the oligonucleotides c 3 and c 4 include binding site for the Disco protein. The turquoise arrowheads indicate the DNA-protein complexes containing Disco protein. The asterisk indicates a complex with lysate protein seen also in the control. (B9) EMSA using box 1 (as shown in Figure S3) and no protein (P), translation lysate (L), lysate with Disco protein (C) and lysate with Dfd protein (D). To test specificity of binding of Disco protein to the DNA fragment, competitor oligonucleotides for the mapped Disco binding site were used either in their wildtype (c wt ) or mutant (c mt ) sequence versions. The red and turquoise arrowheads indicate the specific DNA-protein complexes containing either Dfd or Disco protein, respectively. The asterisk indicates a complex with lysate protein seen also in the control. Note that in the competitor oligonucleotides only the binding site sequence for the Disco protein is mutated, but not for the Dfd binding site sequence. (C) EMSA for mapping of Brk binding site in the rpr-4S3/39 enhancer using box 2 (as shown in Figure S3) as shift probe. EMSA was performed using no protein (P), translation lysate only (L) and lysate with Brk protein (B). c 8 to c 13 represent competitor oligonucleotides with consecutive base-pairs mutated.

Supporting Information
Competition experiments revealed that sequences mutated in the oligonucleotide c 10 and c 11 include binding site for the Brk protein.
The blue arrowheads indicate specific DNA-protein complexes containing Brk protein. The asterisks indicate complexes with lysate protein seen also in the control. (C9) EMSA using box 2 (as shown in Figure S3) and no protein (P), translation lysate (L), lysate with Brk protein (B) and lysate with Dfd protein (D). To test specificity of binding of Brk protein to the DNA fragment, competitor oligonucleotides for the mapped Brk binding site were used either in their wild-type (c wt ) or mutant (c mt ) sequence versions. The red and blue arrowheads indicate the specific DNAprotein complexes containing either Dfd or Brk protein, respectively. The asterisks indicate complexes with lysate protein seen also in the control. Note that in the competitor oligonucleotides only the binding site sequence for the Brk protein is mutated, but not for the Dfd binding site sequence. (D) EMSA using box 2 (as shown in Figure S3) and no protein (P), translation lysate (L), lysate with En protein (N) and lysate with Dfd protein (D). To test specificity of binding of En protein to the DNA fragment, competitor oligonucleotides for En binding site were used either in their wild-type (c wt ) or mutant (c mt ) sequence versions. The red and light green arrowheads indicate the specific DNA-protein complexes containing either Dfd or En protein, respectively. Note that in the competitor oligonucleotides only the binding site sequence for the En protein is mutated, but not for the Dfd binding site sequence. (E) EMSA using box 2 (as shown in Figure S3) and no protein (P), translation lysate (L), lysate with Apt protein (A) and lysate with Dfd protein (D). To test specificity of binding of Apt protein to the DNA fragment, competitor oligonucleotides for Apt binding site were used either in their wild-type (c wt ) or mutant (c mt ) sequence versions. The red and light purple arrowheads indicate the specific DNA-protein complexes containing either Dfd or Apt protein, respectively. Note that in the competitor oligonucleotides only the binding site sequence for the Apt protein is mutated, but not for the Dfd binding site sequence.