Multi-step control of muscle diversity by Hox proteins in the Drosophila embryo

Anatomy drawings illustrate the stereotyped patterns of skeletal muscles that are essential for coordinated movements. Each muscle has its own name/identity, reflecting its specific properties and function. The genetic and molecular bases of muscle identity remain, however, largely unknown. The rather simple pattern of Drosophila embryonic/larval skeletal muscles makes it an ideal model with which to study this process (Bate 1990; Bate 1993). Every (hemi)segment of the Drosophila larva contains ~30 different somatic muscles, each composed of a single multinucleate fiber. Whereas all muscles express the same myogenic program (e.g. of sarcomeric proteins), each has its own identity characterized by its specific position and orientation with respect to the dorsoventral (D/V) and anteroposterior (A/P) axes, size, sites of attachment on the epidermis and innervation (Bate 1993; Baylies et al., 1998; Knirr et al., 1999). Each syncytial muscle fiber is seeded by a ‘founder’ myoblast (founder cell, FC). FCs possess the unique property of being able to undergo multiple rounds of fusion with fusion-competent myoblasts (FCMs). The current view is that muscle identity reflects the expression by each FC of a specific combination of ‘identity’ transcription factors (iTFs), with Apterous, Collier (Col; Knot), Even-skipped (Eve), Krüppel, Ladybird, Nautilus (Nau) and Slouch (Slou; S59) being among the best characterized (Bourgouin et al., 1992; Ruiz-Gomez et al., 1997; Jagla et al., 1998; Crozatier and Vincent, 1999; Knirr et al., 1999; Balagopalan et al., 2001; Fujioka et al., 2005; Dubois et al., 2007). FCs originate from the asymmetric division of progenitor cells, which are themselves singled out from promuscular clusters (equivalence groups) by Notch (N)-mediated lateral inhibition (Carmena et al., 1995; Ruiz Gomez and Bate, 1997). Each muscle progenitor/FC is specified at a specific A/P and D/V position within the somatic mesoderm. This position determines the final location of the muscle(s) issued from this progenitor.

The abdominal A2-A7 segments present the same final muscle pattern. The patterns of the thoracic T2-T3 and abdominal A1 segments are variations on this pattern, whereas the first thoracic segment (T1) and eighth abdominal segment (A8) present fewer and more diversified muscles (Bate and Rushton, 1993). Although it is well established that the Hox transcription factors are major regulators for patterning of the animal body, our understanding of how each Hox protein specifies distinct morphological features and cellular identities within each body part is still fragmentary (Hueber and Lohmann, 2008; Mann et al., 2009). The segment-specific aspects of the Drosophila musculature represent an interesting paradigm with which to address this question (Greig and Akam, 1993; Michelson, 1994). Based on the expression pattern of Nau, which is the Drosophila ortholog of the mammalian bHLH myogenic factor MyoD, in different Hox conditions, Michelson (Michelson, 1994) suggested that segmental differences in the somatic muscle pattern reflect the regulation of muscle iTFs by Hox proteins. Subsequent studies of apterous expression in thoracic lateral muscles supported this notion and suggested that Hox activity could control the segment-specific variation in the number of myoblasts allocated to a specific group of muscles (Capovilla et al., 2001). As a whole, however, when and how homeotic genes act during the muscle-specification program remain to be established.

Here, we addressed this question, focusing on the dorsal acute DA3 muscle lineage, which depends upon the combinatorial activity of Nau and Col, which is the Drosophila ortholog of mammalian early B-cell factors (EBFs) (Michelson et al., 1990; Keller et al., 1997; Balagopalan et al., 2001). Our data show that Hox activity is superimposed on mesoderm-specific and positional information at the progenitor stage to establish the pattern of somatic muscles and acts at several independent steps during the muscle-specification process. Very few studies have addressed the role of Hox genes in controlling the fate of skeletal muscle precursors in vertebrates (Alvares et al., 2003). Our data provide a new framework for studying the integration of Hox information into the generic process of myogenesis and the generation of muscle diversity.

The strains used were w¹¹⁸ as wild-type (wt) reference (Bloomington Stock Center, IN, USA), Antp¹ (Abbott and Kaufman, 1986), Ubx¹ (Bloomington Stock Center), Antp²⁵; Ubx¹ (Ernesto Sanchez-Herrero, Madrid, Spain), rP298-lacZ (Nose et al., 1998), UAS-Antp (Heuer et al., 1995), UAS-Ubx, UAS-abdA (Michelson, 1994), UAS-col (Vervoort et al., 1999), UAS-nau (Keller et al., 1997), 24B-Gal4 (Brand and Perrimon, 1993), rP298-Gal4 (Menon and Chia, 2001) and sns-Gal4 (Kocherlakota et al., 2008). The mutant strains were balanced over marked (TM3 twist-lacZ) chromosomes. All Gal4-UAS crosses were performed at 25°C.

A BamHI-NaeI DNA fragment containing the attB site from the pUASTattB vector (Bischof et al., 2007) was substituted for the P-element-containing NsiI-AvrII fragment in the H-Pelican lacZ transformation vector (Barolo et al., 2000) (GenBank AF242361.2) to generate attB-inslacZ, in which lacZ is under the control of the minimal hsp70 promoter and flanked by gypsy insulator sequences. The 4_0.9, 2.6_0.9 and CRM276 col genomic fragments were inserted into the attB-inslacZ and/or attB-inslacZⁱ vectors. Each attB construct was inserted at position 49D on the second chromosome by injection into nosC31^NLS; Zh8 embryos (Bischof et al., 2007).

The unique intron of the D. virilis β-tubulin56D (βtub56D) gene (Dvir\GJ20466; FlyBase ID FBgn0207606) was inserted into the lacZ coding region between Asp119 and Val120 by standard PCR-based cloning. The 5′ to 3′ sequences of the created splice junctions are as follows: 5′-CGTGAGGT — AGGTCTCG-3′; the intron sequence is underlined. This resulted in a single C-to-G nucleotide change (bold) in the lacZ coding sequence, resulting in an Asp235-to-Glu substitution. The resulting intron-containing lacZ gene, denoted lacZⁱ, was used to generate attB-inslacZⁱ from attB-inslacZ. Mutagenesis of the consensus AbdA/Ubx binding site TAATTA (Ekker et al., 1994) to TGGGGA was by PCR.

Embryos were fixed and processed for antibody staining and/or in situ hybridization as described (Crozatier et al., 1996). Primary antibodies were: mouse anti-Col (1/100) (Dubois et al., 2007); rabbit anti-Mef2 and anti-Nau (1/100) (provided by Eileen Furlong, Heidelberg, Germany and Bruce Paterson, Bethesda, MD, USA, respectively); mouse anti-β-galactosidase (Promega, 1/1000); and mouse anti-Ubx and anti-Antp (1/100, Developmental Studies Hybridoma Bank, IA, USA). Secondary antibodies were: Alexa Fluor 488-conjugated goat anti-rabbit and goat anti-mouse; Alexa Fluor 555-conjugated goat anti-rabbit and goat anti-mouse; Alexa Fluor 647-conjugated goat anti-mouse (all Molecular Probes, 1/300); and biotinylated goat anti-mouse (Vector Laboratories, 1/1000). Double fluorescence in situ hybridization and immunostaining using intronic probes and Col antibodies were as described (Dubois, 2007).

Pairwise sequence alignments of col upstream and CRM276 sequences from various Drosophila species (http://flybase.org/cgi-bin/gbrowse/) were performed using NCBI-BLAST (bl2seq), Genome Browser (University of California, Santa Cruz, CA, USA) and Evoprinter (NINDS, NIH, Bethesda, MD, USA) and manually edited by eye. The search for individual binding sites for transcription factors made use of Cis Analyst (http://rana.lbl.gov/cis-analyst/) and FlyEnhancer (http://opengenomics.org/) and manual inspection based on the matrices for binding sites of Twi, Tin, dTCF, Mad and Pnt (Philippakis et al., 2006). Access to the Tin and Twi in vivo binding sites (Sandmann et al., 2006; Sandmann et al., 2007; Zinzen et al., 2009) was via the E. Furlong lab site (http://furlonglab.embl.de/data/).

Col expression in stage 15 embryos showed that the DA3 muscle forms in the thoracic (T) T2-T3 segments and abdominal (A) A1-A7 segments, but not in the T1 segment and was thinner in T2-T3 than in A1-A7 (Fig. 1A). Since the size of Drosophila larval muscles correlates with their number of nuclei (Demontis and Perrimon, 2009), we counted the number of nuclei in the DA3 muscle. Fewer nuclei were present in T2 and T3 (six on average) than in A1-A7 (eight on average) (Fig. 1B), showing that this number is segment specific. Col expression in the somatic mesoderm is first detected at embryonic stage 10, in a cluster of cells at the same dorsal position in all trunk segments, including T1 (Fig. 1C). This cluster gives rise to the DA3/DO5 progenitor in T2 and T3 and to the DA3/DO5 and DO4/DT1 progenitors in A1-A7 (Fig. 1D). Following asymmetric division of the DA3/DO5 progenitor, col transcription is maintained in the DA3 FC but is repressed in the sibling DO5 FC, a repression mediated by N; it is also not maintained in the DO4 and DT1 FCs (Crozatier and Vincent, 1999). Nau is expressed in the same progenitors as Col (Fig. 1C,D). However, Nau and Col co-expression is only transient. col transcription is maintained in the DA3 FC and is activated in the nucleus of each FCM incorporated into the growing DA3 myofiber, whereas nau is transiently transcribed in the DA3 lineage and is activated in FCM nuclei incorporated into the DO5 myofiber (Dubois et al., 2007). This leads to a specific accumulation of Col and Nau in the DA3 and DO5 muscles, respectively (Fig. 1E). Thus, the progenitor/FC stage is the specific step in the DA3/D05 lineage at which Nau and Col are expressed together. One Col- and Nau-expressing progenitor is found in T2 and T3, two progenitors in A1-A7 and none in T1, correlating with the final muscle pattern (Bate and Rushton, 1993; Crozatier and Vincent, 1999). Transient co-transcription of col and nau was nevertheless observed in a cell issued from the Col-expressing cluster in T1, although at a very low level compared with other segments (Fig. 1F). This indicates that a positive input required to upregulate the expression of these two iTFs in progenitor cells is missing in T1. In summary, transient expression of Nau and Col at the progenitor stage is segment specific and foreshadows the segment-specific formation of DA3/DO5 and DO4/DT1 muscles (Fig. 1G).

A lacZ reporter gene containing 4 kb of the col upstream cis-regulatory region (4_0.9-lacZ; Fig. 2A) reproduces col expression in the DA3 FC and muscle, but not in promuscular clusters (Dubois et al., 2007). An enhancer that drives Eve expression in dorsal promuscular clusters at the origin of the DA1 and DO2 muscles (see Fig. S1 in the supplementary material) has been described (Halfon et al., 2000; Knirr and Frasch, 2001; Speicher et al., 2008). This enhancer integrates positional information issued from the ectoderm with mesoderm-intrinsic information via the binding of five different transcription factors: Mothers against dpp (Mad), dTCF (Pangolin), Pointed (Pnt), Twist (Twi) and Tinman (Tin) (Carmena et al., 1998; Halfon et al., 2000; Knirr and Frasch, 2001). Mad and dTCF are the downstream effectors of Decapentaplegic (Dpp) and Wingless (Wg) signaling, respectively, with these ectodermal signals defining a positional grid within each segment. Pnt is an effector of the Ras/MAPK pathway that contributes to the specification of equivalence groups within the mesoderm. Twi and Tin are mesoderm-specific transcription factors (Halfon et al., 2000).

By comparing gene expression profiles for myoblasts of different genotypes representing perturbations of the Wg, Dpp, Ras and N pathways, Estrada et al. identified ~160 genes that were possibly regulated similarly to eve in the mesoderm, including col (Estrada et al., 2006). Using an in silico search, with the ModuleFinder protocol (Philippakis et al., 2006), we identified five non-coding sequences that were selectively enriched for combinations of Tin, Twi, Pnt, Mad and dTCF binding sites within the col gene. Each was individually tested by transgenic reporter analysis (data not shown). The intron-located sequence with the highest prediction score, despite the absence of a conserved Mad binding site (see Fig. S2 in the supplementary material), which is referred to below as CRM276 (Fig. 2A), drove lacZ expression in promuscular clusters. Double staining of stage 10-11 embryos showed a precise overlap between CRM276-lacZ and Col expression (data not shown).

The stability of lacZ mRNA and β-gal protein prevents, however, a precise determination of temporal aspects of the activity of CRMs, a central aspect of our study. To circumvent this problem, we introduced the Drosophila virilis βtub56D intron into the lacZ coding region (lacZⁱ, see Materials and methods) and generated reporters in which lacZⁱ is placed under control of the CRM276 and 4_0.9 col fragments (4_0.9-lacZⁱ and CRM276_lacZⁱ, respectively; Fig. 2B,C). The patterns of lacZ expression were identical when driven by intron-devoid (not shown) and intron-containing transgenes (Fig. 2B,C), demonstrating that the intron is efficiently spliced. In situ hybridization with a mixture of col and lacZⁱ intronic probes confirmed at the primary transcript level that CRM276 precisely reproduces col activation and integrates the same A/P and D/V positional information (Fig. 2D). CRM276-lacZⁱ transcription was subsequently restricted to the DA3/DO5 progenitor, identical to endogenous col, indicating that it is subject to repression by N (Fig. 2E). Unlike col, however, CRM276-lacZⁱ transcription was very weak in progenitors and was not detected beyond that stage (Fig. 2F,G). Thus, CRM276 is an early mesodermal enhancer that imparts positional and mesodermal information to col activation in a specific promuscular cluster. Conversely, the 4_0.9 CRM was only active from the progenitor stage and was activated in the nuclei of FCMs that have fused with the DA3 FC (Fig. 2H-K). CRM276 and 4_0.9 CRM together account for all aspects of col transcription, including its repression by N signaling, during the successive steps of selection and asymmetric division of the DA3/D05 progenitor (Fig. 2I).

The precise determination of temporal windows of activity established that the activity of the position-specific CRM276 is transient and is relayed by another 4_0.9 CRM in progenitor cells. As it does not operate in T1, this relay is segment specific, providing evidence that it is at the progenitor stage that segment-specific information superimposes on positional information provided by segmentation and D/V patterning genes (Fig. 2L).

The register of mesodermal expression of the thoracic Hox proteins Sex combs reduced (Scr), Antennapedia (Antp), Ultrabithorax (Ubx) and Abdominal-A (AbdA) is schematized in Fig. 3A (Bate, 1993) (see Fig. S3 in the supplementary material; data not shown) (Bate and Rushton, 1993). Comparison between 4_0.9-lacZ and either Scr, Antp or Ubx expression in stage 11 embryos (see Fig. S3A,B in the supplementary material; data not shown) confirmed that the Col-expressing progenitor in T2 and T3 expresses Antp and that the two col-positive progenitors in A1-A7 express Ubx. Overlapping expression with the FC-specific marker, rP298-lacZ (Nose et al., 1998), in late stage 11 embryos showed that Antp and Ubx expression is maintained in T2-T3 and A1-A7 FCs, respectively (see Fig. S3C,D in the supplementary material). We looked at the effect of overexpressing either Antp or Ubx (Fig. 3B,C) or AbdA (not shown) throughout the mesoderm on the specification of the DA3 muscle. Pan-mesodermal expression of either Hox protein (Brand and Perrimon, 1993; Michelson, 1994) resulted in the formation of an ectopic Col-expressing muscle in T1, at the same position as in other segments (Fig. 3B,C). Thus, providing Antp, Ubx or AbdA activity is sufficient to convert the Col-expressing promuscular cluster in T1 into a DA3 muscle. We observed, however, that the size of this ectopic DA3 muscle varied depending on which Hox protein was expressed. When Antp was overexpressed, the number of nuclei in this muscle was identical to that in wild-type (wt) T2 and T3 (Fig. 3D). However, when either Ubx or AbdA was overexpressed, this number was converted to that in wt A segments (Fig. 3E and data not shown), a clear example of posterior prevalence (Duboule and Morata, 1994). These data show that Antp, Ubx and AbdA activities in the mesoderm control the formation of the DA3 muscle and regulate its number of nuclei.

We then looked at when during the muscle-specification process expression of Col and Nau was modified by pan-mesodermal expression of Antp or Ubx. In stage 11 embryos, residual Col expression was detected in all segments in cells that do not become progenitors; high levels of Col and Nau co-localized in the DA3/DO5 progenitor in T2-T3 and in the DA3/DO5 and DO4/DT1 progenitors in A1-A7 (Fig. 1D,G; Fig. 4A). Pan-mesodermal expression of Antp resulted in high-level expression of Col and Nau in one dorsal progenitor in T1 (Fig. 4B), a pattern normally restricted to T2 and T3 (Fig. 4A). Pan-mesodermal expression of Ubx resulted in high-level expression of Col and Nau in two progenitors in all three T segments (Fig. 4C), a pattern typical of wt A1-A7 segments (see Fig. 1D). We conclude that Hox activity in the mesoderm controls the segment-specific number of progenitors expressing Col and Nau that emerge from a pre-defined promuscular cluster.

In order to complement the Hox gain-of-function data, we followed Col expression and the DA3 muscle lineage in Hox loss-of-function mutant embryos, using null alleles (Fig. 4D-I). No DA3 muscle formed in T2 and T3 in Antp mutant embryos (Fig. 4E). In Ubx mutant embryos, the DA3 muscle pattern was similar to that of the wt (not shown), presumably owing to the derepression of Antp expression in the A1 and A2 segments (Hooper, 1986). Accordingly, we observed that Antp, Ubx double-mutant embryos lacked a DA3 muscle in all T and A1-A2 segments (Fig. 4F). Close examination of stage 11 Antp mutant embryos showed that one cell in each thoracic segment transiently accumulated more Col protein than the other cells of the cluster, similar to T1 in wt embryos. This transient, low-level Col accumulation confirms that the process of progenitor selection is initiated and prematurely aborts in the absence of Antp input (Fig. 4H). In Antp, Ubx double mutants, Col upregulation in progenitors was only observed in A3-A7 (Fig. 4I), further confirming that Hox-dependent upregulation and the final muscle pattern are linked (Fig. 4J).

The loss of col expression in Hox mutant embryos and the pattern of 4_0.9-lacZ expression together indicated that the 4_0.9 CRM mediates col regulation by Hox proteins (Figs 2 and 4; Fig. 5A). The observation that a reporter construct containing only 2.6 kb of col upstream DNA (2.6_0.9-lacZ) was only active in A progenitors (Fig. 5B,C) further raised the possibility that differential control by the Antp and Ubx/AbdA paralogs could involve different cis elements. To investigate this, we compared 2.6_0.9-lacZ and 4_0.9-lacZ expression in stage 11 embryos when either Antp or Ubx was expressed throughout the entire mesoderm. Unlike col, 2.6_0.9-lacZ was not activated by Antp in T1, consistent with the fact that it is not expressed in T2 and T3 in wt embryos (Fig. 5E; see also Fig. 4). By contrast, 2.6_0.9-lacZ was activated by Ubx (or AbdA, not shown) in two progenitors in each segment, including all three T segments, thereby reproducing the pattern of col expression under these conditions (Fig. 5F). These results showed that upregulation of col expression by Ubx/AbdA, and not Antp, involves cis elements present within the 2.6_0.9 col upstream region. Conversely, 4_0.9-lacZ expression was upregulated by Antp in one progenitor in T1, therefore reproducing Col expression under the same conditions (Fig. 5D; Fig. 4B). These data show that Antp and Ubx/AbdA control col expression in muscle progenitors via distinct cis-regulatory elements (Fig. 5C).

Since 2.6_0.9-lacZ is only expressed in A segments, we asked whether its regulation by Ubx/AbdA is direct. An in silico search for high-affinity Hox binding sites [TAATTA (Ekker et al., 1994; Affolter et al., 2008)] within the 2.6_0.9 CRM identified two sites that are conserved at the same relative position in several Drosophila species (Fig. 5C and see Fig. S4 in the supplementary material). Each of these sites, named Hox1 and Hox2, was individually mutated (TAAT to GGGG) within the 2.6_0.9-lacZ construct, giving rise to the 2.6_0.9^Hox1-lacZ and 2.6_0.9^Hox2-lacZ transgenes. 2.6_0.9^Hox2-lacZ expression was no longer detected in progenitors (Fig. 5G), whereas the Hox1 mutation had little if any effect. Mutation of both Hox1 and Hox2 sites had no additional effect over the Hox2 mutation alone (data not shown), indicating that Hox2 is required for col regulation by Ubx/AbdA in abdominal muscle progenitors. Expression of 2.6_0.9^Hox2-lacZ recovered later during DA3 muscle development (Fig. 5H), consistent with our previous finding that cis elements responsible for col activation in the DA3-recruited FCM are all contained within the 2.3_1.6 interval, which does not cover the Hox2 site (Dubois et al., 2007). We conclude that a single Hox binding site mediates the specific upregulation of col expression by Ubx/AbdA in A progenitors (Fig. 5C).

Our finding that Hox activity is required in progenitors for prolonged expression of Nau and Col and for implementation of the muscle differentiation process suggested that forced expression of these iTFs in progenitor cells might bypass the need for Hox activity. We therefore repeated the 24B-Gal4×UAS-overexpression experiments, but using both Col and Nau in place of a Hox protein. To visualize the DA3 muscle, we used the 2.6_0.9-lacZ reporter (Fig. 6A). As previously shown (Dubois et al., 2007), 24B-Gal4-driven expression of Nau alone does not affect 2.6_0.9-lacZ expression, whereas expression of Col results in strong activation in the DA2 and VL1-2 muscles in T2 to A7, in DT1 in A segments, plus, more sporadically, in a few other muscles (Fig. 6B,C) (Dubois et al., 2007). The same pattern was observed upon expression of Col plus Nau, with the notable exception of additional activation in a muscle in T1. This ectopic muscle is at the same position and shows the same orientation and attachment sites as the ectopic DA3 muscle induced by Hox proteins (see Fig. 3). We infer from this observation that high levels of Nau plus Col can bypass the need for a Hox protein in implementing the DA3 muscle formation process. It strengthens our conclusion that the upregulation of expression of a combination of iTFs by Hox proteins plays a decisive role in converting positional information within the mesoderm into a specific muscle pattern.

However, the ectopic DA3 muscle that formed in T1 upon forced expression of Col and Nau displayed very few nuclei (two to three), indicating that Hox activity is essential for allocating this muscle a normal number of nuclei. To better define in which myoblasts Hox was acting, we repeated the Ubx-overexpression experiments using two different drivers, rP298-Gal4 and sns-Gal4, which are specific for FCs and muscle precursors and for FCMs, respectively (Menon and Chia, 2001; Kocherlakota et al., 2008). We found that the number of nuclei in the T2 and T3 segments was converted to that in A segments upon Ubx expression in FCs (Fig. 6E,F) but not in FCMs. These results show that Hox factors act to regulate the size of each muscle. Ubx expression in FCs did not, however, promote the formation of an ectopic DA3 muscle in T1 (Fig. 6E), strengthening our conclusion that Hox activity is required at the progenitor stage to implement the muscle differentiation process.

Segment-specific upregulation of col and nau expression by Hox proteins suggests the following model for the transcriptional history of the DA3 muscle (Fig. 7). The first step is the activation of col transcription in a cluster of mesodermal cells in all T and A segments. This is controlled by the early-acting CRM276 enhancer, which integrates positional information issued from the ectoderm, mesoderm-intrinsic cues and repression by N signaling in non-progenitor cells. In a second step, maintenance of Col and Nau expression in the DA3/DO5 progenitor relies upon Hox factors and is mediated by another CRM, the 4_0.9 CRM, which contains separate elements for regulation by Antp and Ubx/AbdA. The 4_0.9 CRM is also subject to repression by N, leading to restriction of col transcription to the DA3 FC. In a third step, implementation of the DA3 muscle differentiation process requires positive and direct Col autoregulation, which converts all of the DA3 muscle nuclei to the same transcriptional program (Dubois et al., 2007). In essence, Col accumulation in the DA3/DO5 progenitor is required to maintain its own expression in the DA3 FC and muscle precursor, thus representing a case of forward autoregulation. Finally, Hox proteins collaborate with iTFs to control the number of myoblasts assigned to the DA3 muscle. Central to our model is the switch in regulation from the early to late muscle CRM that occurs at the progenitor stage and requires Hox activity, thereby linking positional information along the A/P axis to muscle diversity.

Eve expression in the DA1 muscle lineage provided the first paradigm for studying the early steps of muscle specification. Detailed characterization of an eve muscle CRM showed that positional and tissue-specific information were directly integrated at the level of CRMs via the binding of multiple transcription factors, including dTCF, Mad, Pnt, Tin and Twi (Carmena et al., 1998; Halfon et al., 2000; Knirr and Frasch, 2001). Based on this transcription factor code and using the ModuleFinder computational approach (Philippakis et al., 2006), we have identified a CRM, CRM276, that precisely reproduces the early phase of col transcription. This CRM also drove expression in cells of the lymph gland, another organ that is issued from the dorsal mesoderm where col is expressed (Crozatier et al., 2004). Parallel to our study, two col genomic fragments were selectively retrieved in chromatin immunoprecipitation (ChIP-on-chip) experiments designed to identify in vivo binding sites for Twi, Tin or Mef2 in early embryos (Sandmann et al., 2007; Liu et al., 2009). One fragment overlaps with CRM276. Based on this overlap and interspecies sequence conservation, we tested a 1.4 kb subfragment of CRM276 that retained most of the transcription factor binding sites identified by ModuleFinder and found that it specifically reproduced promuscular col expression (M. de Taffin, personal communication). This in vivo validation shows that intersecting computational predictions and ChIP-on-chip data should provide a very efficient approach to identify functional CRMs on a genome-wide scale (Zinzen et al., 2009).

The eve and col early mesodermal CRMs are activated at distinct A/P and D/V positions. We are now in a position to undertake a comparison of these two CRMs, in terms of the number and relative spacing of common activator and repressor sites and their expanded combinatorial code, in order to understand how different mesodermal cis elements perform a specific interpretation of positional information.

A progenitor is selected from the Col promuscular cluster in T2 and T3 but not T1. One cell issued from the Col-expressing promuscular cluster in T1 nevertheless shows transiently enhanced Col expression, suggesting that the generic process of progenitor selection is correctly initiated in T1. This process aborts, however, in the absence of a Hox input, as shown by the loss of progenitor Col expression and DA3 muscle in specific segments in Hox mutants. The similar changes in Nau and Col expression observed under Hox gain-of-function conditions allow us to conclude that the expression of iTFs is regulated by Hox factors at the progenitor stage. The superimposition of Hox information onto the intrasegmental information thereby implements the iTF code in a segment-specific manner and establishes the final muscle pattern. Unlike DA3, a number of specific muscles are found in both T1 and T2-A7 (Bate and Rushton, 1993), such as the Eve-expressing DA1 muscle; other muscles form in either abdominal or thoracic segments, as illustrated by the pattern of Nau expression in stage 16 embryos (Michelson, 1994) (Fig. 1E). This diversity in segment-specific patterns indicates that Hox regulation of iTF expression is iTF and/or progenitor specific.

As early as 1994, Hox proteins were proposed to regulate the segment-specific expression of iTFs. Seven years later, Capovilla et al. characterized an apterous mesodermal enhancer (apME680) active in the LT1-4 muscles and proposed that regulation by Antp was direct (Capovilla et al., 2001). However, mutation of the predicted Antp binding sites present in apME680 abolished its activity also in A segments, suggesting that some of the same sites were bound by Ubx/AbdA. We now have evidence that the regulation of col expression by Ubx/AbdA in muscle progenitors is direct and involves a single Hox binding site. However, regulation by Antp does require other cis elements. It remains to be seen whether regulation by Antp is also direct. Since Antp, Ubx and AbdA display indistinguishable DNA-binding preferences in vitro (Ekker, 1994), the modular regulation of col expression by different Hox paralogs suggests that other cis elements and/or Hox collaborators contribute to Hox specificity (Mann et al., 2009). Direct regulation of col by Ubx has previously been documented in another cellular context, that of the larval imaginal haltere disc, via a wing-specific enhancer (Hersh and Carroll, 2005). In this case, Ubx directly represses col expression by binding to several sites, contrasting with col-positive regulation via a single site in muscle progenitors. This is the second example, in addition to CG13222 regulation in the haltere disc, of direct positive regulation by Ubx via a single binding site (Hersh et al., 2007). Hox ‘selector’ proteins collaborate on some cis elements with ‘effector’ transcription factors that are downstream of cell-cell signaling pathways (Grienenberger et al., 2003; Walsh and Carroll, 2007; Mann et al., 2009). In the DA3 lineage, it seems that Dpp, Wg and Ras signaling act on one col cis element and the Hox proteins on others. The regulation of col expression by Hox proteins in different tissues via different CRMs provides a new paradigm to decipher how different Hox paralogs cooperate and/or collaborate with tissue- and lineage-specific factors to specify cellular identity (Brodu et al., 2004; Gebelein et al., 2004; Walsh and Carroll, 2007; Stobe et al., 2009).

The DA3 muscle displays fewer nuclei in T2 and T3 than in A1-A7, an opposite situation to that described for an aggregate of the four LT1-4 muscles. Capovilla et al. proposed that the variation in the number of LT1-4 nuclei was controlled by Hox proteins (Capovilla et al., 2001). Our studies of the DA3 muscle extend this conclusion by showing that the variations due to Hox control are specific to each muscle and are exerted at the level of FCs. Since the number of nuclei is both muscle- and segment-specific, Hox proteins must cooperate and/or collaborate with various iTFs to differentially regulate the nucleus-counting process. As such, Hox proteins contribute to the combinatorial code of muscle identity. Identifying the nature of the cellular events and genes that act downstream of the iTF/Hox combinatorial code and that are involved in the nucleus-counting process represents a new challenge.

We thank the Bloomington Stock Center, S. Abmayr, S. Menon, E. Sanchez-Herrero and F. Karch for fly stocks; B. Patterson and E. Furlong for antibodies; F. Karch and members of our laboratory for critical reading of the manuscript; and E. Furlong for communicating unpublished data. We acknowledge the help of the Toulouse RIO Imaging Platform and B. Ronsin and A. Leru for confocal microscopy. This work was supported by CNRS and Ministère de la Recherche et de la Technologie (MRT), Université Paul Sabatier, Association Française contre les Myopathies (AFM) and the US National Institutes of Health. J.E. was supported by fellowships from MRT and AFM and H.B. by a fellowship from MRT. Deposited in PMC for release after 12 months.