The location of EO053 within the large intron of pb suggested that pb itself may be the target of this enhancer (Figure 1A). Indeed, at embryonic stages when Pb protein is detected in maxillary and labial segments (Pultz et al., 1988) we find that EO053 drives GAL4 expression in these territories as well (Figure 1E-G), although not in a pattern as expansive as that driven by the previously-studied pb regulatory region 2.1 (Figure 1A; Kapoun and Kaufman, 1995a). Interestingly, in blastoderm stages EO053 also drives GAL4 expression dorsally along most of the length of the embryo (similar to the pattern seen in Figure 1B). Following gastrulation, GAL4 is detected in the amnioserosa (Figure 1C,D), a specificity derived from the earlier dorsal cell population (Hartenstein, 1995). This pattern is not representative of pb, but rather mimics the expression of zen2 (Pultz et al., 1988), the gene immediately upstream of pb (Figure 1A). This additional expression could simply represent a coincidental ectopic artifact of the precise genomic segment chosen for cloning the enhancer. However, the orthologous region from Drosophila virilis also encodes both expression specificities, reducing the likelihood of this being a chance occurrence (Figure 1—Figure Supplement 1A, C – C’’).

While numerous regulatory regions have been shown to serve more than one promoter (Choi and Engel, 1988; Deschamps, 2007; Foley et al., 1994; Lehoczky et al., 2004; Sharpe et al., 1998; Spitz et al., 2003; Tsai et al., 2016; Jones et al., 1995; Tsujimura et al., 2007; Mohrs et al., 2001), these genes typically have expression specificities in common. EO053, then, may serve as an example of a regulatory region that serves more than one promoter but with each gene utilizing the region to generate a different specificity. We thus sought to determine how linked are these specificities and whether each gene indeed requires the EO053 region for expression.

We first began analyzing EO053 under a simple model for encoding multiple specificities: Each expression pattern is dependent upon a separate subregion of the 1.4-kb EO053 sequence. We created a set of reporter constructs containing overlapping truncated portions of EO053 (trunc1, trunc2, trunc3, trunc1-2, trunc2-3; Figure 2). The central region, trunc2, drives both pb-like and zen2-like expression but not as robustly as the full EO053 construct (Figure 2E,F). This region also drives ectopic expression in the ventral embryo at stage 10 (Figure 2F) and ectopic dorsal expression (amnioserosa or dorsal vessel) in late-stage embryos (Figure 2G). The right-most region, trunc3, also drives pb-like expression, though very weakly (Figure 2I,J). A construct that encompasses both the trunc2 and trunc3 regions drives reporter expression in a robust pb-like pattern that also lacks the ectopic activities seen with trunc2 alone (Figure 2O,P), suggesting that trunc3 contains elements that repress the late dorsal expression. This model is further supported by the trunc1-2 construct (removing the right-most portion of EO053) that also drives ectopic late dorsal expression (Figure 2M).

While trunc2 retains some capacity to drive zen2-like expression (Figure 2E), it was only weakly detectable in a few early-gastrulation embryos (embryo in Figure 2E is rare; most stage 5 embryos lack GAL4 expression). A construct including trunc2 and the left-most portion of EO053, trunc1-2, restores zen2-like expression (Figure 2K), yet the left-most portion alone, trunc1, fails to drive reporter expression at any stage (Figure 2B–D). Since these initial truncation constructs failed to reveal a region in EO053 responsible for the zen2-like expression, we designed a set of 10 smaller overlapping reporter constructs to locate the zen2-like activity (Figure 3A). None of these smaller fragments drive reporter expression in a zen2-like pattern (Figure 3—figure supplement 1), while two fragments, truncF and truncG, drive reporter expression in a pb-like pattern (Figure 3C,D,F,G), consistent with their overlap with trunc2 (Figure 3A). Furthermore, we again observed late expression in the dorsal embryo driven by truncG (Figure 3G), refining the location of this ectopic activity seen neither with full-length EO053>GAL4 or endogenous pb or zen2 mRNA. These fragments allowed us to further define the domain sufficient to produce the pb-like pattern, and suggested that if the pb-like and zen2-like specificities were separable, the latter pattern would be localized to the left-most region of EO053 outside of truncF and truncG. Importantly, truncF-J, which lacks this left-most region, fails to drive zen2-like GAL4 expression (Figure 3K). However, truncA-D, a construct containing only this region, fails to drive strong zen2-like reporter expression (Figure 3H). This suggests that while the A-D region is necessary (but not sufficient) for the zen2-like pattern, the FG region likely also contains elements required for this specificity, in addition to being sufficient to drive pb-like expression. Consistent with this interpretation, a construct that deletes this region, truncΔFG, lost both pb-like and zen2-like expression patterns (Figure 3N–P).

Because the FG region is necessary for both the pb-like and zen2-like patterns, we sought an alternative approach to determine if the two patterns are indeed separable. In a series of eight constructs, we created successive 47-nt non-complementary transversion mutations along the length of the FG region to identify sequences required for either the pb-like or zen2-like patterns (Figure 4). Interestingly, none of the 47-nt mutations could recapitulate the strong reduction of zen2-like activity seen in EO053ΔFG>GAL4. Two mutants, FG3 and FG7, strongly reduce the pb-like expression pattern (Figure 4F’–F’’, J’–J’’). Each of these mutants also affects the zen2-like pattern, though in opposite directions: FG3 causes an ectopic anterior expansion of the zen2-like pattern (Figure 4F), while FG7 reduces zen2-like expression (Figure 4J). FG1, FG2, FG4, FG6, and FG8 also reduce expression in the zen2-like pattern (Figure 4D–K). The reduced expression seen with the FG4 mutation results in dorsal stripes (Figure 4G), which were also observed with the insufficient truncA-D construct (Figure 3H). Together, these data suggest that while the pb-like pattern can be effectively localized to the FG region in EO053, the elements required for zen2-like expression are spread much more broadly throughout EO053 and are even linked to regions necessary for the pb-like pattern.

While the pb- and zen2-like specificities appear to be linked within the EO053 sequence, we sought to determine whether EO053 is functionally linked to either pb or zen2, or both. We thus generated via CRISPR/Cas9 a deletion of the EO053 region at the endogenous pb locus, designated pb^M2:20. A chromosomal deletion removing zen2 and null for pb, pb²³ (a.k.a. pb^map8), lacks any embryonic cuticle phenotype (Pultz et al., 1988). Therefore, we examined effects upon both pb and zen2 mRNA accumulation (Figure 5). Kapoun and Kaufman have shown that pb mini-genes lacking large sections of the intron overlapping EO053 are capable of rescuing adult mouthparts-to-leg transformations in pb null flies and that the 2.1 enhancer is required for rescue in the context of these small pb mini-genes (Kapoun and Kaufman, 1995a). Thus, because the 2.1 enhancer is unaffected in pb^M2:20 homozygous embryos, we were not surprised to observe detectable pb mRNA in these embryos (Figure 5B,C), as well as in labial discs from 3^rd-instar larvae (Figure 5—figure supplement 1). Kapoun and Kaufman also showed that a 10.6-kb fragment—apparently overlapping EO053 sequence—was able to drive LacZ expression in maxillary and, to a lesser extent, labial segments in embryos (Kapoun and Kaufman, 1995a). While this is strikingly similar to the EO053 expression pattern we observe, it is possible that additional pb enhancers outside of EO053 reside on this 10.6-kb fragment. Despite 2.1 and other potential pb enhancers remaining intact in pb^M2:20 flies, in double-blind scoring of pb expression in parallel in situ hybridization experiments, we were able to observe a statistically significant reduction in pb mRNA accumulation in pb^M2:20 embryos compared to w¹¹¹⁸ controls (Figure 5—figure supplement 2A–C). We failed to validate this difference by qPCR in staged embryos, however (Figure 5—figure supplement 2D,E), and suspected that region 2.1 may be masking the consequence of EO053 deletion. Consistent with this hypothesis, deletion of region 2.1 alone was sufficient to cause a noticeable reduction in expression area in maxillary and labial segments in mutant embryos (Figure 5E, Figure 5—figure supplements 3, 4) and in labial discs from 3^rd-instar larvae (Figure 5—figure supplement 1), and to cause a proboscis-to-leg transformation in adult flies (Figure 5—figure supplement 5) reminiscent of pb null mutants. The remaining pb mRNA detectable in Δ2.1 single mutants, however, was reduced to an even greater extent when EO053 was also deleted, as observed in maxillary and labial segments in double mutant embryos (Figure 5F, Figure 5—figure supplements 3, 4) and in labial discs (Figure 5—figure supplement 1). This strong effect upon pb mRNA expression relative to the Δ2.1 single mutants did not appear to enhance the proboscis-to-leg transformation, however (Figure 5—figure supplement 5). These data suggest that indeed EO053 serves a role as a dual enhancer of pb, operating in addition to the 2.1 and potentially other enhancers (Kapoun and Kaufman, 1995a).

We similarly anticipated that if EO053 regulates zen2 it may not act alone on this target either. Endogenous zen2 mRNA accumulation expands to the anterior and posterior poles of blastoderm-stage embryos (Figure 5I), a pattern that differs from expression driven by EO053, which is absent from the poles (Figure 1B). While the regulation of zen2 expression has not yet been pursued, an investigation of the paralogous zen gene found that a reporter driven by the promoter-proximal region recapitulates endogenous zen gene expression (Doyle et al., 1989). Guided by the zen2 inversion between D. melanogaster and D. virilis (Figure 1—figure Supplement 1D), we cloned the zen2 promoter-proximal region (zen2US) and found that it is indeed capable of driving reporter expression in a pattern that recapitulates the polar expansion of endogenous zen2 mRNA (Figure 5G). Intriguingly, the EO053 and zen2US patterns differ not only spatially but temporally, with zen2US driving reporter expression only until the completion of cellularization at stage 5 (Figure 5G,H), while EO053 is strongly active at this stage and continues to gastrulation (Figure 1B–D). Such a transition is also observed with endogenous zen2 mRNA: Early expression includes expression in polar regions as well as dorsally at stage 4 (Figure 5I), but following cellularization the mRNA is largely detectable only in dorsal-most cells and is absent from the anterior and posterior poles (Figure 5J). We find that only this later and not the earlier accumulation of endogenous zen2 mRNA requires EO053, as pb^M2:20 embryos largely fail to express zen2 beyond cellularization (Figure 5L,L’). Thus, EO053 appears to have dual roles in Drosophila embryogenesis, assisting other enhancers in early stages with zen2 expression and then with pb expression during later morphogenetic events (Figure 7).

The zen, zen2, and bicoid (bcd) genes in Drosophila are derivatives of the ancestral Hox3 ortholog in basal arthropods, and have diverged in expression and function from the ancient homeotic role. Why, then, do they remain at their ancestral genomic location within the Hox complex? Splits and inversions within the Hox complex are common in Schizophoran flies, suggesting loosened constraints on colinearity (Negre and Ruiz, 2007; Von Allmen et al., 1996). The sharing of regulatory elements among members of the Hox complex has been a model to explain the persistent linkage of Hox genes in metazoan genomes (Spitz et al., 2003; Sharpe et al., 1998), and the function of EO053 provides direct support for maintenance of an ancestral regulatory linkage as a contributing factor to persistence of a pb-zen linkage. While it is challenging to trace EO053 itself across evolution, patterns of synteny between pb and zen following the functional transition offer an opportunity to test such a model. Specifically, any translocation of a zen ortholog away from the pb ortholog would presumably not be favored if one or more regulatory elements are shared between the two genes. Indeed, examining available genomic scaffolds across 80 different Arthropods, we were unable to detect any translocation event that breaks the synteny between pb/Hox2 and zen/Hox3 (Figure 6—figure supplements 1–9). Furthermore, among the 66 species examined that evolved following the Hox3/zen divergence, only three species—all members of the Formicoidea—exhibit a change in synteny: only via the loss of the zen coding sequence (Figure 6—figure supplements 1 and 5). In contrast, 3/14 species examined that predate the Hox3/zen divergence exhibit loss of Hox3 or pb—representing Crustacea, Myriapoda, and Chelicerata (Figure 6—figure supplements 1, 8 and 9; Chipman et al., 2014; Grbić et al., 2011; Pace et al., 2016; Kim et al., 2016; Kenny et al., 2014).

Examining patterns of regulatory sequence conservation is a complementary approach to exploring the model of ancient, shared regulation as an explanation of the persistent pb/zen linkage. Sequence conservation makes possible the identification of EO053 throughout the Schizophora (Figure 6A,B). In particular, 33/36 nt of the region containing the 5' 12nt of FG4 and the 3' half of FG3, the latter of which we have shown to be required for the proper expression of both pb- and zen2-like specificities (Figure 4F–F’’), are identical between D. melanogaster and Ceratitis capitata (Figure 6C). Outside of the Brachycera it is challenging to identify orthologous regulatory regions. Comparing D. melanogaster EO053 and pb intronic sequence from the mosquito Anopheles gambiae identified a 12-nt sequence from within the Schizophora 36-nt span (ATCATTAATCAT, henceforth referred to as ‘the EO053 motif’, in green in Figure 6B,C) that is also found in the Anopheles intron (Figure 6—figure supplement 2). This sequence is similar to others that have been shown to be bound by Exd/Hox dimers (Bergson and McGinnis, 1990; Chan et al., 1997; Regulski et al., 1991; Rusch and Kaufman, 2000; Zeng et al., 1994), and thus represents a plausible candidate for an ancient motif with regulatory function. We find that this motif is important for EO053 function, as a TTAA>GGCC mutation to abrogate Hox binding dramatically reduces GAL4 expression in both the zen2-like and pb-like specificities (Figure 6D–F’). The deepest we are able to identify a region orthologous to EO053 outside of Schizophora is in the assassin fly Proctacanthus coquilletti (Brachycera; Orthorrapha). This species contains a variant EO053 motif (ATCATAAATCAT) that could still mediate an Exd/Hox interaction (Slattery et al., 2011). Given the functional importance of this motif for both aspects of EO053 expression and its conservation within Brachycera, we chose to examine the 80 Arthropod Hox2/3 regions for patterns consistent with an ancient regulatory function for this or similar sequences.

65/66 species that arose following the Hox3/zen divergence contain instances of the EO053 motif within the large introns between the YPWM- and homeodomain-encoding exons of pb (Figure 6—figure supplements 2–9). The outlier Locusta migratoria scaffold containing the homeodomain coding sequence lacks upstream motif instances but also lacks the upstream exon necessary to confirm motif absence (Figure 6—figure supplement 7). In species that arose prior to the Hox3/zen divergence, the Ixodes scapularis (Chelicerata) (Figure 6—figure supplement 9), Daphnia pulex (Crustacea) (Figure 6—figure supplement 8), and Orchesella cincta (Hexapoda) (Figure 6—figure supplement 7) pb orthologs lack intron motif instances. Despite a lack of direct evidence for which, if any, motifs in the Hox2/3 regions are functional outside of Schizophora, we nevertheless examined these genomic intervals for patterns of conservation.

We identified in several major Arthropod clades conserved instances of the EO053 motif, based upon relative location and flanking sequences (Figure 6—figure supplements 2–9, Supplementary file 1). Lepidoptera contain a conserved motif instance between pb and zen2 (Motif 2 in Figure 6—figure supplement 3, Supplementary file 1). Coleoptera contain a conserved mismatch upstream of the pb promoter (Motif 3 in Figure 6—figure supplement 4, Supplementary file 1), and several species contain conserved motifs within the pb intron (Motif 4) and upstream of the duplicated zen genes, including on an isolated scaffold harboring a fourth zen paralog in the Asian long-horned beetle A. glabripennis (Motif 5 in Figure 6—figure supplement 4, Supplementary file 1). Within Hymenoptera, the two basal Tenthredinoidea species have five conserved motif mismatches (Motifs 6–10 in Figure 6—figure supplement 5, Supplementary file 1). Outside the Tenthredinoidea, all remaining species conserve a mismatch upstream of zen (Motif 11 in Figure 6—figure supplement 5, Supplementary file 1). Of particular note, this sequence is still present in the three Formicoidea that have lost the zen coding sequence. The Formicoidea also contain an intron motif specific to this clade (Motif 13 in Figure 6—figure supplement 5, Supplementary file 1), as well as another intron motif also conserved throughout the Aculeata (Motif 12 in Figure 6—figure supplement 5, Supplementary file 1). Several of the Hemiptera examined (excluding the Sternorrhyncha species A. pisum, D. citri, and B. tabaci) contain a conserved motif mismatch downstream of zen (Motif 14 in Figure 6—figure supplement 6, Supplementary file 1), and the two Dictyoptera species contain a conserved motif mismatch upstream of zen (Motif 15 in Figure 6—figure supplement 7, Supplementary file 1). Within the Chelicerata, gene duplication and loss present interesting opportunities to examine motif instances. The multiple rounds of whole-genome duplication in the basal Limulus polyphemus led to three copies each of pb and Hox3, two of which have confirmed synteny. All three Hox3 paralogs have a motif mismatch downstream (Motif 16 in Figure 6—figure supplement 9, Supplementary file 1). At a similar position downstream of one of the pb paralogs is the same 12-nt motif, with a variant site at the same position downstream of a second pb. We found another curious parallel between one adjacent Limulus pb and Hox3: the same 12-mer mismatch motif is found in the introns of each gene, though upstream of the putative Hox3 coding sequence, at a similar distance from a separate motif instance (Motif 17 in Figure 6—figure supplement 9, Supplementary file 1). This same motif was also found upstream of the coding sequence of the third Limulus Hox3, again at a similar distance from another motif instance. We also found instances of these motifs in Arachnida, with a Hox3 ortholog in Centruroides exilicauda, Parasteatoda tepidariorum, and Stegodyphus mimosarium as well as a pb ortholog in Centruroides and Parasteatoda having the downstream motif (Motif 16); we also found a 12-mer matching the intron motif (Motif 17) in one of the pb paralogs in Stegodyphus (Figure 6—figure supplement 9, Supplementary file 1). By contrast, we observed largely no conservation of 40 randomly generated 12mers (also allowing for a 1-bp mismatch) even when examining the Schizophoran regions, for example, with a single motif instance occurring every 772,189 bp on average (Supplementary file 2). Together our analysis indicates that while functional assumptions are limited to the Schizophora, sequences resembling the EO053 motif exhibit patterns of conservation across Arthropod clades within the Hox2/Hox3 genomic region, particularly within clades emerging after the Hox3/zen divergence.

Perhaps the most curious feature of EO053 is its requirement by distinct genes for the reliability (pb) or temporal progression (zen2) of their expression. Given its intronic location, we suggest that this regulatory arrangement would be mediated by a looping interaction between EO053 and each target promoter (Levine et al., 2014; Matharu and Ahituv, 2015). Such interactions are likely permitted by the distinct temporal activation profiles of each target gene, allowing the enhancer to separately engage only a single active promoter at a time (Figure 7). In addition, we have gained insight into the temporal dynamics with which EO053 operates on the zen2 locus, whereby the initial activation of zen2 expression is mediated by the promoter-proximal zen2US segment and then switches to control by EO053.

The regulatory interactions between EO053 and both pb and zen2 are likely influenced by the overall regulatory architecture of the Antp Complex. Recent high-resolution analyses of topologically associated domains (TADs) suggest that pb, zen2, zen, and bcd all reside in a single TAD (Eagen et al., 2017; Stadler et al., 2017), potentially biasing regulatory activities between these loci and separate from Dfd, which is a regulatory island (Stadler et al., 2017). The three-dimensional architecture facilitating interactions between EO053 and its target promoters is likely mediated by Polycomb Response Elements (PREs) that have been mapped within the pb locus (Figure 5A; Kapoun and Kaufman, 1995b; Nègre et al., 2011; Stadler et al., 2017) and exhibit chromosomal pairing experimentally (Kapoun and Kaufman, 1995b). The observed establishment of Polycomb bodies in the nucleus at stage 5 (Cheutin and Cavalli, 2012) also correlates well with the temporal shift in regulation of zen2 from promoter-proximal to distal regulatory regions.

We have shown that the pb-like and zen2-like specificities overlap within the FG region of EO053. While the pb-like expression is largely restricted to this region, the zen2-like expression appears to be much more broadly extended throughout EO053. Moreover, within the FG region, mutation of FG3 or FG7 affects both specificities, suggesting the specificities may have one or more motifs in common in their regulatory logic. Such a motif could be bound by the same transcription factor in both settings or related factors with different temporal and/or spatial profiles. We raise the possibility that the conserved EO053 motif may represent such a site. First, its pattern of conservation and location within the functionally important FG3 region suggests that this sequence itself may be required. Second, we show that mutating the core Hox motif affects both the pb-like and zen2-like expression in the context of EO053 (Figure 6). Similar sequences have been identified as functionally relevant to the expression of Dfd (Chan et al., 1997; Zeng et al., 1994; Bergson and McGinnis, 1990; Regulski et al., 1991; Lou et al., 1995) and pb itself (Rusch and Kaufman, 2000). Both of these examples involve regulation by Dfd, and we suspect the pb-like expression mediated by this motif would likely also involve Dfd. We also notice instances of this motif upstream of the Dfd orthologs themselves in many of the species we have analyzed (Figure 6—figure supplements 2–9), suggesting that Dfd auto-regulation within the arthropods may be ancient. Given the similar binding specificities of Hox proteins (Noyes et al., 2008; Berger et al., 2008), this site could be utilized by one or more of these proteins, even operating as a promiscuous auto-regulatory enhancer. This is consistent with the demonstrated role of EO053 in the temporal dynamics of zen2 expression where its activity is preceded by zen2US activity, and also with our observation that region 2.1 deletion significantly affects the early pb expression in maxillary and labial segments (Figure 5—figure supplement 3E). We expect activation and specificity to also involve key regulators binding non-Hox motifs, which again could participate in one or both specificities.

Why does neither pb nor zen2 mRNA expression reflect the complete regulatory capacity of EO053? A common feature of many enhancers is their ability to interact reliably with a heterologous promoter to drive reporter gene expression in a manner that not only largely recapitulates specificity encoded by the enhancer, but also represents a subset of the expression pattern of the endogenous gene. EO053, however, produces a pattern that would be considered an ectopic specificity relative to the pattern of either target gene (i.e. if EO053 were only a pb enhancer the blastoderm spatiotemporal activity is ectopic and the mouthpart activity does not recapitulate zen2 expression). Promoter selectivity/interpretation is a likely model to explain the different transcriptional outputs of the separate promoters that utilize EO053. The collection of core promoter elements at an individual promoter can bias promoter-enhancer compatibility (Butler and Kadonaga, 2001), and the pb promoter itself is required for expression driven by certain enhancers (Kapoun and Kaufman, 1995a). Replacing the promiscuous synthetic core promoter (Pfeiffer et al., 2008) in the pBPGUw vector used here (TATA box present) with the minimal core promoter from either pb or bcd (TATA box absent) or replacing the initiator sequence with that of zen2 had no effect on reporter gene expression (Supplementary file 3, Figure 7—figure supplement 1), which may suggest that promoter interpretation may require additional promoter-proximal elements to mediate the appropriate spatial/temporal output. It is additionally possible that an EO053-proximal element may facilitate appropriate promoter targeting and output (Chen et al., 2005; Zhou and Levine, 1999), or another separate region (i.e. dominant repressor) may be influencing output (Perry et al., 2011).

The stable colinearity of vertebrate Hox genes has been attributed to sharing of regulatory elements (Gould et al., 1997; Gérard et al., 1996; Sharpe et al., 1998). Arthropods may not share this paradigm across the complete complex, as suggested by the occurrence of rearrangements (Pace et al., 2016; Faddeeva-Vakhrusheva et al., 2017; Wu et al., 2017), inversions (Negre and Ruiz, 2007), gene loss (Chipman et al., 2014; Grbić et al., 2011), regulatory independence (Gellon and McGinnis, 1998; Shippy et al., 2008), and local chromatin organization (Eagen et al., 2017; Stadler et al., 2017). We provide evidence here for the possibility of limited shared regulation within an Arthropod Hox complex, between a true Hox (pb) and a neighboring Hox-derived gene (zen2) both sharing EO053. While we do not know how deeply this regulatory relationship extends, the organization of Hox complexes across the phylum exhibits features consistent with shared regulation. The strong synteny of pb and zen and reduced gene loss (a subset of ant genomes being the exception to date) suggests shared regulation may have existed from the point of Hox3/zen divergence, if not earlier. The most likely scenario based upon the available data would be a pb-Hox3 shared duplicate enhancer that drives a pattern of expression common to both genes. Acquisition of a novel expression specificity by the shared enhancer would then be buffered by the additional independent regulatory sequences of each gene. Alternatively, the change in spatial/temporal expression of the shared enhancer might also impart selective pressure alleviated by differential interpretation of the enhancer by each gene, and/or functional divergence. Regardless of the mechanism, EO053 serves as an unusual example of an enhancer maintaining a promiscuous relationship with two distinct gene promoters even in the context of disparately evolving expression specificities.

The EO053 region was identified by enriched CBP binding in embryonic rather than later stages [i.e., ‘Embryo Only’ (EO) (Nègre et al., 2011)]. It was amplified from genomic DNA using the primers EO053-f (CCCGGAGCGGCACAATTAGTCTTG) and EO053-r (CGGTAATGCTGAATGAACCTTTCAA). ΔFG, FG, and TTAAm mutations were generated using overlap extension PCR (Ho et al., 1989). The DvEO053 primers were DvEO053-f (TGCCCTGGTTCTTTGGCTAACACG) and DvEO053-r (TTTCTTGTACATAATCGTTCTTGG). PCR products amplified from genomic DNA were cloned into pBPGUw (Pfeiffer et al., 2008) through an LR Clonase II (ThermoFisher—Waltham, MA) Gateway reaction from a pCR8/TOPO/GW intermediate. All variants were inserted upstream of the DSCP promoter in the same orientation with respect to EO053. zen2US was amplified using zen2US-f (TTATATACCCCAGAAGCCCTTCGTGACG) and zen2US-r (TGATGTGATGACACCAATTTATCTGAGC) and cloned as above. See also Supplementary file 4 for complete sequences.

Preparation of guide RNA and Cas9 mRNA was done as described previously (Hwang et al., 2013). The sequences for guide RNAs directed against EO053 (GGAGTCGGTCGGACACAGAG) or region 2.1 (GAGAAAGATTTTCTCCCCTC and GCTGTGCCTCATTTAATGCA) were cloned into DR274 (Addgene—Cambridge, MA; deposited by J Keith Joung) cut with BsaI. For pb^M2:20, sequence-verified clones were linearized with DraI and 1 μg transcribed using the MAXIscript T7 kit (ThermoFisher). Cas9 mRNA was transcribed using the mMESSAGE mMachine T7 kit, using 1 μg MLM3613 (Addgene; deposited by J Keith Joung) linearized with PmeI. RNAs were precipitated according to the manufacturer’s instructions and resuspended in injection buffer. The final injection cocktail for injecting w¹¹¹⁸ embryos contained 900 ng/μL Cas9 mRNA and 100 ng/μL EO053 guide RNA. For pb^11A, pb^11C, pb^11D, and pb^11E, pb^M2:20 flies were crossed to Cas9-expressing flies and a cocktail containing both guide RNAs and the homology-directed repair template were injected by BestGene, Inc The region 2.1 HDR template was constructed by separately cloning each arm (~2 kb each, see Supplementary file 4) into pGEM-T (Promega). The left arm was flanked by BamHI and SalI sites and the right arm contained tandem BglII and XhoI sites on the 5’ end. The left arm was cut with BamHI and SalI and subcloned into the plasmid containing the right arm, digested with BglII and XhoI. Injected flies were crossed to w;;TM2/TM6C individually, and then F1 males from each injected fly were crossed individually to w;;TM2/TM6C. After viable larvae were detected, the F1 males were removed from the vials and pooled into groups of four for gDNA extraction and PCR screening for deletion. F2 vials from positive pools were screened and sequenced to uncover the pb^M2:20 1255-bp deletion and the precise region 2.1 deletion. Only a single injected fly harbored the 2.1 deletion, and we were ecstatic to obtain progeny with deletions on both the EO053 deletion chromosome and the wildtype homologous chromosome to provide us with both the single and double mutants.

The GAL4 digoxygenin probe was prepared as previously described (Nègre et al., 2011; Pfeiffer et al., 2008). The large exons from pb and zen2 were amplified from genomic DNA and cloned into pGEM-T (Promega—Madison, WI). Linearized plasmids served as template for in vitro transcription of digoxygenin-labeled RNA probes as described (Kosman et al., 2004) using T7 RNA polymerase (Promega or Roche) and DIG-UTP or biotin-UTP RNA labeling mixes (Roche). Embryo in situ hybridizations using GAL4, LacZ, pb, or zen2 digoxygenin probes and HLHmβ biotin probes were performed as previously described (Kosman et al., 2004; Nègre et al., 2011; Reeves and Posakony, 2005). GAL4 in situ hybridizations with related mutant constructs were performed in parallel batches and representative images presented.

Following in situ hybridization and mounting of w¹¹¹⁸ and pb^M2:20 embryos in parallel, images were collected of lateral views of stage 10–12 embryos. Filenames of experimental and control in situ hybridization images were gathered, randomly shuffled, and renamed for double-blind scoring image analysis using ImageJ (Fiji). The area of visible stain in maxillary and labial segments was selected for each image and values for area, mean intensity, min intensity, and max intensity were recorded. After data collection, values were reassigned to the corresponding genotype and statistically analyzed.

Following in situ hybridization and mounting of w¹¹¹⁸ and pb^M2:20 embryos in parallel, slides were manually screened for dorsal visibility at stage 4 or stage 5 (staging based upon cellularization under DIC optics). The in situ hybridization signal at these stages was scored as ‘dorsal-weak,’ ‘dorsal-strong,’ ‘poles+dorsal,’ ‘poles-strong,’ ‘poles-weak,’ or ‘no expression.’ In Figure 5, ‘dorsal-weak’ and ‘dorsal-strong’ represent the ‘dorsal’ category, and ‘poles+dorsal,’ ‘poles-strong,’ and ‘poles-weak’ represent the ‘polar’ category, to distinguish between EO053-like and zen2US-like expression.

RNA was prepared using Trizol (Ambion) from embryos collected for 2 hr and aged 4 hr at 25 °C (4–6 hr embryos). Sample embryos were examined with a compound microscope to verify desired age (approximately stage 11). First-strand cDNA was synthesized with a SuperScript II kit (Invitrogen). Quantitative RT-PCR was performed on an iQ5 cycler (BioRad) using iQ SYBR Green Supermix (BioRad).

We queried genomic scaffolds of sequenced arthropods by BLAST using Drosophila melanogaster amino acid sequence for Pb, Zen2, Zen, or Dfd, as well as orthologous sequences from other annotated species. Scaffolds or accession numbers were obtained from http://metazoa.ensembl.org/, http://hymenopteragenome.org/, http://www.vectorbase.org/, http://i5k.nal.usda.gov/, http://www.ncbi.nlm.nih.gov/genbank/, http://genome.wustl.edu/, http://www.collembolomics.nl/collembolomics/. Gene structures were inferred from either annotations or database gene predictions. For some species, only the homeodomain sequence for each gene was determined and located. In some cases (e.g., Dendroctonus ponderosae), the pb and zen homeodomains are encoded on separate scaffolds but the zen scaffold includes near one end a YPWM-encoding ORF that by BLAST is most similar to pb, suggesting that the scaffolds are likely adjacent.

Each sequence was opened in GenePalette (Rebeiz and Posakony, 2004; Smith et al., 2017), and the location of each gene identified by either GenBank or manual annotation. Sequences were searched for instances of ATCATTAATCAT (allowing for 1-bp mismatch), ‘ATTAAT’ (ATCATTAAT or ATTAATCAT), and ‘YPWM’ (TAYCCNTGGATG). Instances found at similar relative locations in related species were analyzed for similarity in both core and flanking sequence to suggest orthology within clades. All figure gene diagrams were generated by creating a postscript export from GenePalette and compiling/editing in Adobe Illustrator.