The Border Between the Ultrabithorax and abdominal-A Regulatory Domains in the Drosophila Bithorax Complex

Abstract

The bithorax complex in Drosophila melanogaster includes three homeobox-containing genes—Ultrabithorax (Ubx), abdominal-A (abd-A), and Abdominal-B (Abd-B)—which are required for the proper differentiation of the posterior 10 segments of the body. Each of these genes has multiple distinct regulatory regions; there is one for each segmental unit of the body plan where the genes are expressed. One additional protein- coding gene in the bithorax complex, Glut3, a sugar-transporter homolog, can be deleted without phenotype. We focus here on the upstream regulatory region for Ubx, the bithoraxoid (bxd) domain, and its border with the adjacent infraabdominal-2 (iab-2) domain, which controls abdA. These two domains can be defined by the phenotypes of rearrangement breakpoints, and by the expression patterns of enhancer traps. In D. virilis, the homeotic cluster is split between Ubx and abd-A, and so the border can also be located by a sequence comparison between species. When the border region is deleted in melanogaster, the flies show a dominant phenotype called Front-ultraabdominal (Fub); the first abdominal segment is transformed into a copy of the second abdominal segment. Thus, the border blocks the spread of activation from the bxd domain into the iab-2 domain.

THE ~300-kb expanse of the bithorax complex (BX-C) includes very little protein- coding information; the BX-C includes three transcription units for homeobox transcription factors, plus one open reading frame encoding a glucose transporter homolog (Martin et al. 1995). Many mutations in the complex have been recovered because of their segmental transformation phenotypes; most fall outside the known transcription units (Maeda and Karch 2006). However, such mutations fail to complement with mutations in one of the three homeobox genes, and so they were suspected to alter the regulation of these three major products. Indeed, many of these mutations alter the patterns of expression of the three homeobox genes (Beachy et al. 1985; Celniker et al. 1990; Karch et al. 1990; Sánchez-Herrero 1991).

The regulatory mutations revealed a striking order to the complex; they are aligned on the genetic map in the order of the most anterior segments [or, more precisely, the parasegments (PS)] that they affect (Lewis 1978). The lesions of recessive (loss-of-function) mutants that affect a particular parasegment roughly define a segmental domain; there appears to be one such domain for each parasegment. The domains are named after the mutant names in the following order (proximal to distal): bithorax (bx, affecting the posterior thorax or PS5), bithoraxoid (bxd, affecting the first abdominal segment or PS6), and infraabdominal-2 through infraabdominal-9 (iab-2 through iab-9, affecting the second through the ninth abdominal segments, or PS7–PS14) (Lewis 1978). It has been suggested that each of these segmental domains reflects a large region of the chromosome that is either activated or repressed as a unit, depending on the position of a cell along the body axis (Lewis 1981; Peifer et al. 1987). Support for this view has come from the study of insertions into the BX-C of mobile elements (Bender and Hudson 2000), and of other probes for DNA accessibility (Fitzgerald and Bender 2001).

Unfortunately, the extents of the segmental domains are not precisely mapped by the mutant analysis for several reasons. Many of the available mutations are associated with rearrangement breakpoints, which can affect domains beyond the one that they interrupt, either by position effects or by polar effects. As an example of a polar effect, a break in the iab-7 (PS12) domain separates the iab-6 (PS11) and iab (PS10) domains from the Abdominal-B (Abd-B) transcription unit, which all three domains regulate. Thus, iab-7 breaks cause transformations in the fifth, sixth, and seventh abdominal segments. Four regulatory domains (bx, iab-2, iab-8, and iab-9) lie partly or wholly within the transcription units that they regulate, and are thus more difficult to define, especially with rearrangement breakpoints. In addition, transformations among the second though fifth abdominal segments are difficult to discern, due to the similar morphologies of these segments.

“Boundaries” between segmental regulatory domains have been proposed based on the phenotypes of several small deletions. The Fab-7 deletions were proposed to define the boundary between the iab-6 and iab-7 domains (regulating PS11 and PS12, respectively) (Gyurkovics et al. 1990). Likewise, the Mcp deletions were thought to remove a boundary between the iab-4 and iab-5 domains (affecting PS9 and PS10) (Karch et al. 1994), Fab-8 deletions are associated with the iab-7/iab-8 boundary (affecting PS12/PS13) (Barges et al. 2000), and Fab-6 deletions correspond to the iab-5/iab-6 boundary (affecting PS10/PS11) (Iampietro et al. 2010). In each case, these deletions cause a dominant gain-of-function phenotype, showing a transformation of the more anterior parasegment toward the character of the more posterior one. This has been interpreted as spreading of the activation from one segmental domain into the normally repressed adjacent domain, when the boundary between them has been deleted. However, the limits of the iab-4 through iab-8 domains are not well defined for the reasons described above, and so it is not clear that these deletions fall at the borders between domains—if, indeed, there are discrete borders. The assumption that these deletions remove boundaries has been inferred from their dominant (gain-of-function) phenotypes, but there is no reason to predict what the phenotype of a boundary deletion should be. It is not clear whether there must be a discrete barrier between adjacent active and repressed domains, or whether the spread of marks for activation and repression are short range and graded (Hathaway et al. 2012).

The definition of boundaries is complicated by the presence of Polycomb response elements (PREs) at, or very near, the Mcp, Fab-6, Fab-7, and Fab-8 boundaries. Deletions of the best-studied PRE, in the middle of the bxd domain (far from any proposed boundary), cause a dominant phenotype similar to those of the boundary deletions, a one-parasegment posterior transformation (Sipos et al. 2007). Although the bxd PRE deletion phenotype is weaker than those of the boundary deletions, it is possible that the latter transformations result from the loss of PREs. Efforts have been made to distinguish the boundary from the nearby PREs (Mihaly et al. 1997; Gruzdeva et al. 2005). Mihaly et al. (1997) concluded that deletion of the Fab-7 boundary, but not the associated PRE, resulted in both gain-of-function and loss-of-function phenotypes, due to spreading of both activation and repression. Deletions that define the Fab-6 boundary also show a mixture of gain-of-function and loss-of-function phenotypes (Iampietro et al. 2010). However, these analyses rest on assumed phenotypes of a minimal boundary deletion or on assumed properties of PREs and boundaries on transgenes. In particular, the test for boundaries as insulators (i.e., enhancer blockers) on transgenes is not reliable. The Fab-7 boundary, for example, lies between the iab-6 domain and the Abd-B transcription unit that it regulates; this boundary clearly does not block those enhancer/promoter interactions. It has been proposed that “promoter targeting sequences” or “promoter tethering elements” exist in the BX-C to bypass Fab-7 and other boundaries (Chen Q et al. 2005; Akbari et al. 2008). However, Hogga et al. (2001) converted prototypic insulators (binding sites for the suppressor of Hairy-wing protein, or the scs element adjacent to the HSP70 locus) into the endogenous BX-C, and showed that they are not bypassed and cannot substitute for the Fab-7 boundary.

The transition zone, or “border” between the bxd domain and the adjacent iab-2 domain, presents a unique opportunity because it can be precisely mapped by multiple criteria. (We prefer the term “border” to “boundary” or “insulator” because “border” does not imply a function.) Mutations with lesions proximal to the bxd/iab-2 border fail to complement with Ultrabithorax (Ubx), and primarily affect the first abdominal segment (PS6) (Bender et al. 1985). Mutations distal to the border fail to complement with abdominal-A (abd-A) and affect the second abdominal segment (PS7) (Karch et al. 1985). Enhancer traps proximal to the border drive reporter genes starting in PS6; distal traps mark PS7 (Bender and Hudson 2000). Finally, in Drosophila virilis, Ubx and its regulatory regions are clustered with the genes of the Antennapedia complex, leaving abd-A and Abd-B as a separate cluster (Von Allmen et al. 1996). Mapping the homologies from the edges of the two D. virilis clusters should help to define the limits of the adjacent bxd and iab-2 regulatory domains. If the border region, so defined, lacks a PRE, then a deletion of the border might show whether it is required to separate the regulatory signals of the two domains.

In this article, we consolidate studies carried out over many years, which together map the functions and extent of the bxd regulatory domain, and which define the position and function of the bxd/iab-2 border.

Materials and Methods

Sequence coordinates

The D. melanogaster sequence coordinates follow the SEQ89E numbering of Martin et al. (1995) (GenBank U31961). Base #1 of SEQ89E corresponds to base #12,809,162 in Release 5.37 of the D. melanogaster genome; SEQ89E numbering proceeds from distal (Abd-B) to proximal (Ubx); the assembled genome proceeds proximally to distally. The genome sequence includes a 6134-bp insertion of the Diver retroposon in the bxd domain with a 4-bp target duplication of bases 220,924–220,927 in SEQ89E. The target chromosome used for gene conversion experiments includes 278 bp from the 3′ end of the jockey transposable element, with a duplication of the 10-bp target site of bases 184,640–184,649 in SEQ89E.

Derivation of synthetic deficiencies

Hm:

This breakpoint is derived from the complex rearrangement T(2;3) Hm (new order 100-89E/32-29/89E-88E/32-60; 61-88E/29-21). Both fusion fragments of the translocation have been cloned, and the rearrangement interrupts the BX-C DNA map at ~240,000 (SEQ89E coordinates) (W. Bender, unpublished results). The rearrangement has a dominant phenotype of transformation of the wing blade to the capitellum of the haltere (Lewis 1982), but there are no dominant effects that we have noticed in the embryonic pattern of UBX protein or on the embryonic cuticle. E. B. Lewis generated an insertional translocation into the Y chromosome of the left part of the complex through the Hm breakpoint [Dp(Y;2;3)Hm]; this insertion conferred the dominant phenotype, and the altered Y chromosome was male sterile. This Y insertion was used to generate embryos lacking BX-C DNA to the right of the Hm breakpoint. Females [Dp(3;1)68/FM6/Dp(Y;2;3) Hm] were crossed to males [T(1;3)P115/TM1]. Female offspring of the genotype Dp(3;1)P115/FM6/Dp(Y;2;3)Hm; Df(3R)P115/+ were mated to males with Df(3R)P9/Sb Dp(3R)P5. One-eighth of the embryos from this cross had the desired genotype, Dp(Y;2;3)Hm; Df(3R)P9/Df(3R)P115. The Dp(Y;2;3)Hm stock has been lost since these experiments were done.

bxd¹⁰⁰:

This breakpoint is from an insertional transposition of the proximal part of the BX-C (89B5-6 to 89E2 into 66C); the resulting duplication and deficiency stocks are available separately. The breakpoint has been located by in situ hybridization and genomic blots to the interval 227,500–230,500 (Bender et al. 1983). Males with Dp(3:1)68;Dp(3)bxd¹⁰⁰ Df(3R)P115/TM1 were crossed to Df(3R)P9/Sb Dp(3)P5 females. One-eighth of the resulting embryos were the desired genotype, Dp(3)bxd¹⁰⁰ Df(3R)P115/Df(3R)P9.

bxd¹¹¹:

This breakpoint is from an insertional translocation of the distal part of the BX-C into the X chromosome (89E3-4 to 90B2 into 4D). The breakpoint has been assigned to the interval 213,000–216,000 by genomic blots and in situ hybridizations (Bender et al. 1983). T(3;1)bxd¹¹¹/TM1 males were crossed to Df(3R)P9/Sb Dp(3)P5 females. One-eighth of the offspring were the desired genotype, Df(3R)bxd¹¹¹/Df(3R)P9.

bxd¹⁰⁶⁸:

This is a translocation to 2R heterochromatin; the break in BX-C DNA is between 201,000 and 203,500, as determined by genomic blotting (Bender et al. 1985). To create a synthetic deficiency for BX-C DNA to the right of this breakpoint, this rearrangement was combined with T(2;3)Abd-B¹⁰⁶⁵ (called iab-7¹⁰⁶⁵ in Karch et al. 1985), which is also a translocation to 2R heterochromatin breaking in the BX-C at ~42,000 kb. The Abd-B¹⁰⁶⁵ breakpoint is ~12 kb distal to the Abd-B homeobox, and the synthetic deficiency is completely lacking both abd-A and Abd-B function, as judged by the apparently uniform expression of Ubx protein in the first through eighth abdominal segments. Males with T(2;3)bxd¹⁰⁶⁸/TM1 were crossed to females with T(2;3)Abd-B¹⁰⁶⁵/Sb Dp(3)P5. Male offspring with T(2;3)bxd¹⁰⁶⁸/T(2;3)Abd-B¹⁰⁶⁵ were crossed to Df(3R)P9/Sb Dp(3)P5 females. One-eighth of the embryos had the desired genotype of bxd¹⁰⁶⁸-left Abd-B¹⁰⁶⁵-right/Df(3R)P9.

Uab⁵:

This rearrangement is a translocation to the tip of the X chromosome (1F); the break is associated with recessive lethal mutations and a dominant male sterile mutation, presumably due to the X chromosome breakpoint. Both fusion fragments have been cloned, and the breakpoint maps to 186,000 in the BX-C DNA (Barbara Weiffenbach and W. Bender, unpublished results). The dominant Uab phenotype is associated with the X/distal 3R fusion chromosome. Synthetic deficiencies for the right half of the BX-C were generated with T(Y;3)B116, which breaks in 90E, distal to the BX-C. Females with T(1;3)Uab⁵/FM6/Y were crossed to T(Y;3)B116 males. Nondisjunction in the mothers leads to female offspring with T(1;3)Uab⁵/T(Y;3)B116/FM6; these were crossed to Df(3R)P9/Sb Dp(3)P5 males. One in 12 of the resulting zygotes that were not grossly aneuploid were of the desired genotype, T(1;3)Uab⁵-left T(Y;3)B116-right/Df(3R)P9.

P10:

This is an insertional translocation of the proximal part of the BX-C into 2L (89C1-2 to 89E1-2 into 29A-C). It breaks the BX-C DNA at 174,000, within the abd-A transcription unit (Karch et al. 1985). Dp(3;2)P10 homozygous males were crossed to T(3;1)P115/TM1 females. Male offspring with Dp(3;1)P115; Dp(3;2)P10; Df(3R)P115 were crossed to Df(3R)P9/Sb Dp(3)P5 females. One-eighth of the zygotes were of the desired genotype, Dp(3;2)P10/+; Df(3R)P9/Df(3R)P115.

D. virilis clones

Phage clones overlapping the D. virilis abd-A gene were obtained from François Karch (Von Allmen et al. 1996). D. virilis clones homologous to the distal bxd region of D. melanogaster were isolated by Barbara Weiffenbach, using a phage library constructed by Ron Blackman (Charon 30 vector). Subclones from the virilis phage clones were sequenced from both ends using primer walking. Sequences were assembled and homologies were mapped using MacVector software; they are listed in GenBank under accession nos. JX877552 (for virilis bxd) and JX877553 (for virilis iab-2). These sequences are collinear with the comparable regions of the more recent D. virilis genomic scaffold (Clark et al. 2007), with a 1–2% mismatch. The genomic scaffold sequence was the more reliable since our sequence was usually from one strand, often with only single coverage. The genomic sequence was used for the Glut3 protein prediction of Figure 3.

Figure 3.

Sugar transporter alignment. The predicted amino acid sequence for the Glut3 sugar transporter homologs from D. virilis and D. melanogaster are aligned with each other and with glucose transporters from Arabidopsis thaliana (STP1) (Sauer et al. 1990) and human erythrocytes (GLUT1) (Mueckler et al. 1985). Below these sequences are shown the residues most conserved in a comparison of many sugar transporters from diverse species (Baldwin 1993). Some positions in the conserved line indicate groups of amino acids: +, positively charged, R or K; −, negatively charged, D or E; °, hydroxyl-bearing, S or T; Ø, aromatic, F, W, or Y (after Baldwin 1993). The shading indicates agreement with one of the conserved amino acids; many other sequence homologies are not marked. Note that most of the conserved positions that are shaded in the melanogaster sequence are also shaded in virilis and vice versa.

Recombination between P elements

The “homing pigeon” P element was used to recover enhancer traps in the BX-C (Bender and Hudson 2000). It contains two FRT sites flanking the rosy⁺ transformation marker. When two such P insertions with the same chromosomal orientation are in trans, recombination between them can be generated by a heat-inducible flipase gene (Kopp et al. 1997). The resulting crossovers can generate either a duplication or a deletion for the chromosomal region between the insertion sites, and the hybrid P element can have no, one, or two copies of the rosy⁺ marker (Bender and Hudson 2000). Flanking markers (Sb and Fab-7) were used to recognize recombinants, and to indicate the direction of the crossover (i.e., duplication or deletion).

Deletion of the bxd/iab-2 border

The donor plasmid diagrammed in Figure 5 included three genomic fragments, cloned by PCR from cn; ry flies (the background strain for the HC184B P-element insertion) into the Bluescript II KS+ vector (Strategene-Agilent). The fragments covered sequence coordinates 189,407–186,680 (red), 182,352–181,405 (blue), and 186,670–183,675 (orange). A synthetic linker between the blue and orange fragments included the following sites: HindIII/I-SceI/NheI/NotI/BamH1. A cassette containing the Gal4-VP16 fusion gene flanked by I-SceI sites was recovered as a 2.4-kb XbaI fragment from the plasmid psce-G4VP16 3L, a gift from László Sipos (Sipos et al. 2007). This XbaI fragment was cloned into the NheI site of the synthetic linker.

Figure 5.

Generation of bxd/iab-2 border deletion. The HC184B P element was mobilized by P recombinase in embryos injected with a plasmid with homology to genomic sequences on both sides of the P-element insertion site (red and orange segments). Between these two segments, the conversion donor plasmid contained a short DNA segment homologous to the distal edge of the desired deletion (blue segment), followed by a marker for successful integration, the GAL4-VP16 transcriptional activator (pink triangle). The GAL4-VP16 marker was flanked by restriction sites for the rare-cutting restriction enzyme I-SceI. The injected animals were crossed to flies homozygous for a UAS-GFP reporter, and the progeny were screened for fluorescence. Animals with the conversion chromosome were crossed to a heat-inducible source of the I-SceI restriction enzyme, and progeny carrying the conversion chromosome were screened for loss of GAL4-VP16-driven GFP. The structure of the final deletion chromosome was verified by PCR. The pictured fly is heterozygous for the 4.3-kb Fub deletion; the arrow marks the first abdominal tergite, transformed to the character of the second abdominal tergite.

The donor plasmid was injected into embryos derived from a cross of males homozygous for the HC184B P-element insertion with females of the genotype ry, Ubx109, Δ2-3 (99B ry+)/MKRS. G₀ survivors were crossed to UAS-GFP homozygotes, and resulting larvae were screened for expression of GFP, marking the convertants. Adults carrying the conversion chromosome had unfolded wings and a reduced tergite on the first abdominal segment; these phenotypes were likely due to the toxicity of GAL4-VP16. Conversion heterozygotes were crossed to [hs-FLP][ hs-ISceI] Sco/CyO; ry Fab-7, and offspring were heat-shocked as young larvae for 1 hr at 37°. Resulting adult males were crossed to UAS-GFP; pbx Fab-7/MKRS females, and non-Fab7, Sb progeny were screened for loss of GFP expression (indicating I-SceI cutting and removal of GAL4-VP16). Selected adults, potentially containing border deletions, were screened by PCR for the expected fusion of the deletion endpoints, and PCR products were sequenced to confirm the expected junction.

Results

Mapping the bxd domain by mutant lesions

The PS6 regulatory domain was first defined by the mapping of bxd and postbithoraxoid (pbx) mutant lesions (Bender et al. 1983, 1985; Karch et al. 1990). These mutations transform structures of PS6 to those of PS5 [posterior haltere transforms into posterior wing, first abdominal tergite is removed, and extra legs appear on the first abdominal segment (Lewis 1963)]. These mutant alleles included primarily chromosomal rearrangements, although several alleles were associated with insertions of the gypsy mobile element, and the pbx alleles were X-ray-induced deletions. Figure 1A illustrates some of these lesions, including previously unreported rearrangement breakpoints associated with very weak bxd phenotypes. They are spread across a 50-kb region upstream of the Ubx transcription unit. Several of these bxd breaks were induced on a homozygous-viable inversion, In(3R)1000, which breaks in 81F and 90C, and puts the BX-C near the centromere. The breaks were recovered in a screen for the disruption of transvection between the mutagenized chromosome and a copy of In(3R)1000 with the Cbx¹ and Ubx¹ mutations (Celniker et al. 1990). For the weakest bxd breaks (the group from bxd²⁶⁶ through bxd⁵⁵¹), hemizygous embryos show weak thoracic-like ventral pits on the anterior abdominal segments, and hemizygous adults show only a slight reduction of the first abdominal tergite. The rightmost break in the bxd series is associated with the Uab⁵ rearrangement, which is a translocation between the BX-C and section 1F on the X chromosome. The dominant Uab phenotype [transformation of the first abdominal (A1) tergite to the second abdominal (A2)] is likely due to misexpression of abd-A. This transformation obscures any weak bxd phenotype in the adult, but when Uab⁵ is tested over a deficiency, it also shows ventral pits in the abdominal segments. By that criterion, Uab⁵ is the most distal break associated with a bxd phenotype. Figure 1A also illustrates a variety of DNA elements that have been uncovered in the bxd domain, including enhancers that drive PS6 expression in early embryos (Pirrotta et al. 1995), a noncoding RNA that is transcribed in PS6 (Lipshitz et al. 1987), a prominent PRE (Sipos et al. 2007), and the coding region for the glucose transporter homolog (Martin et al. 1995).

Figure 1.

The extent of the bxd regulatory domain. (A) The long horizontal line indicates the DNA map with the upper coordinates according to SEQ89E (Martin et al. 1995) and lower coordinates from D. melanogaster Genome Release 5.37. Splicing patterns are shown for the abd-A transcription unit and the major embryonic bxd noncoding RNA, and the start site of the Ubx transcription is shown at the left. Colored triangles above the DNA map show insertion positions of enhancer traps that have anterior expression limits in PS6 (green) or PS7 (blue). Four mapped embryonic enhancers are indicated by the pink boxes on the DNA line, and the red box shows the site of the bxd PRE. Mutant lesions are shown below the line; rearrangement breakpoints are shown with vertical arrows (dashed bars show mapping uncertainties), the pbx¹ and pbx² deletions are indicated by horizontal dashed lines, and the iab-2^Kuhn gypsy insertion is shown by the lower triangle. The Glut3-coding region is shown in orange. (B) The series of horizontal black lines diagram the BX-C DNA remaining in synthetic deficiencies that end at the indicated breakpoints. The pictures below show pelts of stage 14 embryos stained for UBX protein, which has an anterior limit in PS5. The UBX staining in PS6–PS13 increases in pattern and intensity with additional DNA sequences from the bxd region. The maximal expression is reached in the Uab⁵ embryo; additional sequences up to the P10 breakpoint do not add to the pattern.

Three iab-2 mutations are shown on the right in Figure 1A. The iab-2^S3 rearrangement breakpoint lies just a few kilobases distal to the Uab⁵ break (Karch et al. 1985); homozygotes show a weak transformation of the A2 tergite toward A1, but the A1 segment is normal in embryos and adults (Bender et al. 1985). The other iab-2 alleles include the iab-2^Kuhn gypsy insertion and the iab-2⁶⁷¹ breakpoint (Karch et al. 1990). Breaks or deletions farther distal impinge on the abd-A transcription unit, so that it is not easy to determine how the segmental regulation of abd-A might be disturbed. This analysis would put the transition between the bxd (PS6) and iab-2 (PS7) regulatory regions within a few kilobases between Uab⁵ and iab-2^S3. This analysis is tentative, however, since any individual break may involve some position effect that affects BX-C sequences distant from the breakpoint.

The definition of the bxd regulatory region is based not only on mutant transformations of the adult cuticle, but also on changes in the expression pattern of UBX protein in the embryo. UBX protein is strongly expressed in most nuclei in PS6 of the embryo, and the series of bxd rearrangement breaks make it possible to map cis-regulatory sequences responsible for that PS6 pattern. UBX antigen patterns for various bxd rearrangements have been reported (Beachy et al. 1985; White and Wilcox 1985; Irvine et al. 1991; Camprodón and Castelli-Gair 1994), but the results were confusing, with some breaks giving apparent increases in UBX expression in PS5. We also saw increased PS5 expression in the case of bxd¹¹³, but not for other bxd breaks examined (not shown). It was more revealing to examine UBX expression in the absence of the abd-A and Abd-B genes, in part because potential cross-regulatory interactions are removed, and in part because the PS6 pattern is repeated in PS7–PS12. It was possible to create synthetic deficiency genotypes for six rearrangements, each of which has one copy of the Ubx transcription unit, with various extents of the bxd region intact, as shown by the bars in Figure 1B. Each genotype lacks BX-C DNA distal to the break (see Materials and Methods).

Embryo pelts from the six genotypes are shown in Figure 1B, all stained for UBX antigen. All six show UBX expression in PS5 that looks identical to that of wild-type embryos. Embryos with the complex truncated at the Hm break, 3 kb upstream of the Ubx RNA start site, show UBX staining in PS6–PS11 that is nearly identical to that of PS5. There are some additional stained nuclei in PS6 in the ventral-lateral region, including muscle nuclei, as judged with Nomarski optics. When the complex is truncated at the bxd¹⁰⁰ break, 12 kb upstream of the Ubx start, the embryos look very similar to the Hm embryo, except that there is more staining in the epidermal cells in the lateral region around the tracheal pit. With additional sequences to the bxd¹¹¹ breakpoint, 27 kb from the Ubx start, there are many additional stained nuclei in the ventral nerve chord and the lateral epidermis. Additional sequences to the bxd¹⁰⁶⁸ breakpoint, 35 kb from the Ubx start, give additional staining in nuclei of the ventral nerve chord. Embryos with DNA up to the Uab⁵ breakpoints (50 kb from the Ubx start) show more intense staining in the nuclei of the ventral nerve chord; the pattern and intensity resembles the staining in PS6 of the wild type. Additional DNA up to the P10 breakpoint (within the ABD-A coding region), does not notably alter the UBX expression pattern. The P10 pattern was previously reported by Struhl and White (1985).

The analysis of UBX expression patterns shows that the PS6 regulatory region extends beyond the bxd¹⁰⁶⁸ break, but not necessarily beyond the Uab⁵ breakpoint. Both the antigen patterns and the adult phenotypes suggest that the cis-regulatory region controlling UBX expression in PS6 extends at least 35 kb upstream of the Ubx promoter, and that it contains a succession of elements, each controlling a part of the UBX pattern. There is no single site [such as the major PRE (Chan et al. 1994; Chiang et al. 1995; Sipos et al. 2007)] necessary or sufficient for PS6 expression and maintenance.

Mapping by Enhancer traps

The regulatory domains of the bithorax complex can also be defined by the expression patterns of enhancer trap insertions. Many P-element insertions carrying marker genes such as LacZ, Gal4, or GFP have been recovered in the bithorax complex; in nearly every case, the marker gene is repressed in head segments. More specifically, the anterior-most parasegment showing marker expression corresponds to the anterior-most parasegment affected by mutations in the neighborhood of the enhancer trap insertion (Galloni et al. 1993; McCall et al. 1994; Casares et al. 1997; Barges et al. 2000; Bender and Hudson 2000; Fitzgerald and Bender 2001; Herranz and Morata 2001; Estrada et al. 2002). Enhancer traps are particularly informative in regions where the mutant lesions give only subtle or ambiguous segmental transformations, such as the iab-3 and iab-4 domains. In the distal extent of the bxd domain and in the proximal part of the iab-2 domain, the phenotypes of mutant lesions are subtle, as discussed above. There are several enhancer traps in this region, all of which were “trimmed down,” so that the P element contained only the lacZ reporter (Bender and Hudson 2000). These minimal enhancer traps divide clearly into two groups, with anterior expression limits in PS6 or PS7 (Figure 1). This assay delimits the bxd/iab-2 border to a 12-kb interval; the most distal PS6 enhancer trap (HC184B) is at the same position as the most distal rearrangement breakpoint with a bxd phenotype (Uab⁵).

D. virilis sequence comparison

The extent of the bxd regulatory region might also be inferred from a DNA sequence comparison with the comparable region in D. virilis. The D. virilis homeotic genes lie in two clusters, as do those of D. melanogaster, but the split in virilis lies between Ubx and abd-A, instead of between Antp and Ubx (Von Allmen et al. 1996) (Figure 2A). Assuming that the virilis Ubx and abd-A genes have intact PS6 and PS7 regulatory regions, respectively, then the limits of those regions should be revealed by the position of the split in the ancestral sequence. Recombinant bacteriophage were recovered from D. virilis genomic libraries using D. melanogaster probes from the distal bxd region or from the proximal end of the abd-A transcription unit. These hybridized to the 24E or 26D regions of the virilis chromosome 2, respectively (Von Allmen et al. 1996). The initial D. virilis phage clones were used to recover additional overlapping phage clones in both locations. Primer walking was used to sequence 16,660 bp from the 24E region and 19,267 bp from the 26D region.

Figure 2.

Homology matrix comparison of D. virilis and D. melanogaster sequences. (A) Diagram of the breakup of the homeotic gene complex in Drosophila evolution (adapted from Lewis et al. 2003). In melanogaster, the ancestral complex was split between Antennapedia and Ubx; in virilis, the split occurred between Ubx and abd-A. The cytological positions of the resulting clusters are shown for chromosome 3R (melanogaster) and chromosome 2 (virilis). The left/right orientation of the two virilis clusters is not yet established. (B) The horizontal line in the middle is a map of the melanogaster sequence. The Glut3 open reading frame is shown below the map on the left, and the final two exons of the abd-A gene are shown above the map on the right. The grid above the map shows a homology comparison with the ~16.7 kb of virilis sequence from 24E on chromosome 2. The grid below the sequence line shows the comparison with ~19.3 kb of virilis sequence from 26D. The vertical axes are virilis sequence coordinates, and the horizontal axes are melanogaster sequence coordinates, aligned with the map. The sequences of both virilis DNA strands are included in the comparison, so that sequence inversions appear as diagonals from bottom left to top right. Note the gap in homology, indicated by the parentheses near the middle of the melanogaster map. The matrices were constructed by the MacVector program, with a window size of 30, a minimum percentage score of 42, and a hash value of 5.

The virilis sequences could be easily aligned with the corresponding sequences from melanogaster (Martin et al. 1995). Figure 2B diagrams the alignment in a dot-matrix format. Throughout most of the sequence, there were blocks of 20–100 bp with high homology (~80%), spaced by regions of low homology, often A/T rich, and often different in length between the two species. The homology blocks are uniformly collinear, except for four segments of the 24E virilis sequence that appear to be inverted relative to the melanogaster sequence (Figure 2B). The virilis 24E homology to the melanogaster bxd region abruptly ends at ~185,200 on the melanogaster sequence, just beyond the position of the Uab⁵ breakpoint. Coming from the other direction, the virilis 26D homology to the melanogaster iab-2 region stops abruptly at ~183,400, close to the iab-2^S3 breakpoint. The intervening ~1.8 kb of melanogaster sequence has no clear homology anywhere in the virilis genomic scaffold.

Sugar transporter sequence

The distal bxd domain includes the coding region of the sugar transporter homolog, Glut3, which was first identified by examination of the BX-C sequence (Martin et al. 1995). Glut3 is the only “foreign” gene in the bithorax complex, but the Antennapedia complex includes many nonhomeobox genes (including Amalgam, cuticle proteins, and transfer RNAs). It was not clear if the melanogaster GLUT3 represents a pseudogene without function, and so we looked for evidence of protein sequence conservation. The D. virilis homology to the D. melanogaster Glut3 is nearly exactly delimited by an ~1500-bp inversion (Figure 2B). The dot-matrix diagram of Figure 2B also illustrates that the transporter region shows less DNA homology than adjacent, noncoding regions. The virilis region contains an open reading frame of 478 amino acids, with clear homology to the melanogaster transporter homolog over most of that length. Figure 3 presents an alignment of the virilis and melanogaster predicted amino acid sequences, along with plant and human sugar transporter proteins for comparison. The overall homology between the two fly peptides is only 36% (compared to ~92% for UBX), but the presence of a full-length transporter homolog (without stop codons) argues for functional conservation. Certain amino acids are tightly conserved among sugar transporter genes from bacteria, fungi, plants, and animals (Baldwin 1993). These amino acids are also well conserved in fly evolution; there are 45 positions where the melanogaster protein matches the most conserved amino acids, and the virilis sequence matches 37 of them (82% homology) (Figure 3). Thus, the fly sequences appear to be evolutionarily selected for function as a transporter, but it is not possible to predict what molecule they might transport (Marger and Saier 1993).

Sugar transporter function

Our collection of P-element insertions into the bithorax complex permitted us to delete Glut3 fairly precisely. Since the P elements include FRT sites for the yeast FLP recombinase (Golic and Lindquist 1989), it is simple to create duplications or deletions between any pair of such insertions, as long as their FRT sites are in the same orientation (Bender and Hudson 2000). Two such insertions, HC154A and HC148A, flank the sugar transporter homolog, <1 kb from either end of the open reading frame (Figure 4A). FLP-induced recombination produced a 3.2-kb deletion, with one copy of the P element remaining at the site of the recombination. Since the starting P elements each have two internal FRT sites, the final P element can be any of three sizes, depending on the FRT sites involved in the recombination event (Bender and Hudson 2000). Figure 4B illustrates the smallest final P element, which contains only a LacZ reporter fused to the P promoter, plus a single FRT site. Such deletion derivatives are homozygous viable, but they have a strong bxd phenotype, which is most obvious in the absence of the first abdominal tergite (Figure 4C). This phenotype is unexpected because rearrangement breaks to the left of the transporter homolog, including bxd²⁶⁶, bxd⁵¹⁷, and bxd^L69A, all have much weaker phenotypes, with little or no reduction in the A1 tergite. This implied that the transporter homolog has a function in specification of the A1 segment, which is independent of Ubx, or that the phenotype is due to some action at a distance from the remaining P element.

Figure 4.

Deletion of the Glut3 sugar transporter homolog. (A) Diagram of the two chromosomes bearing P elements flanking the Glut3 coding region. The P-element insertions, indicated by the large triangles, are drawn to scale, showing their internal maps. The segment marked “homing” is the 7.5-kb fragment from the BX-C used to target P insertions into the BX-C (Bender and Hudson 2000). The two FRT sites in each P element are sites for FLP, the yeast site-specific recombinase. (B) Deletion recombination product. Recombination between the left FRT site of the HC154B P element and the right FRT site of the HC148B element produces a deletion of ~3 kb of BX-C DNA, and leaves a small P element at the site of the deletion. (C) Dorsal view of an adult fly homozygous for the deletion diagrammed in B. Note the severely reduced tergite on the first abdominal segment (labeled “A1”). (D) Elimination of the remaining P element. A deletion chromosome with a P element remaining was crossed to a source of P transposase. From the offspring of the dysgenic fly, chromosomes were recovered with the P sequences excised. (E) Dorsal view of a fly homozygous for the deletion diagrammed in D. Note that the first abdominal tergite (labeled “A1”) appears normal in size.

The remaining P element could be removed by P transposase (Figure 4D); this was accomplished by crossing the original deletion strain to the 99BΔ2-3 source (Robertson et al. 1988). We actually used a deletion derivative carrying a full-length copy of the starting P element, including the rosy⁺ transformation marker, so that P excisions could be recognized by the loss of rosy. Nineteen independent rosy⁻ derivatives were analyzed; two appeared by Southern blots to be clean deletions. For both lines, the site of the deletion was recovered in a polymerase chain reaction, and the products were sequenced. One had 31 bp remaining at the site of the P element, derived from the P-element terminal repeats; the other had 36 bp of P sequence. Other workers have described similar fragments of P remaining at the sites of P excisions (Staveley et al. 1995; Beall and Rio 1997). Figure 4E illustrates an adult homozygous for one of these deletion lines; the A1 tergite is restored to the full wild-type size. The small P element in the initial derivative (Figure 4B) must have been the cause of the bxd phenotype.

Both lines of the final deletion derivative (Figure 4, D and E) were healthy and fertile as homozygotes, with no apparent segmental transformation. Glut3 is expressed predominantly in the adult male testis (Chintapalli et al. 2007), but 10 of 10 tested single males lacking Glut3 were fertile. Thus, the Glut3 transporter homolog has no obvious function under our culture conditions. We have no rationale for the presence of Glut3 in the BX-C, but its location in the bxd domain should not inhibit expression in the testis. The testis is derived from the rudimentary ninth abdominal segment (Chen EH et al. 2005), where we expect that nearly the entire BX-C is released from Polycomb repression.

Border deletion

The most distal mutant lesion still within the bxd domain is the Uab⁵ rearrangement breakpoint, and the most proximal lesion in the iab-2 domain is the iab-2^Kuhn gypsy element insertion (Figure 1). We sought to generate a deletion that would extend between these two markers, which would unequivocally remove the border between these two domains.

We initially attempted to generate deletions by imprecise excisions of the enhancer trap P elements. These efforts yielded a variety of mutant lines showing dominant gain-of-function “Ultraabdominal” phenotypes (first abdominal tergite transformed to second or third abdominal tergite), but most of these were associated with rearrangements of the P elements (Bender and Fitzgerald 2002). One deletion of 1890 bp was recovered (shown as “Δ1.9 kb” in Figure 5 and Figure 7), which did not span the entire border region. Flies homozygous for this deletion had no apparent segmental transformations.

Figure 7.

Summary map of the bxd/iab-2 border region. The thick black line gives DNA coordinates according to SEQ89E (above the line) or D. melanogaster Genome Release 5.37 (below). The small triangle at ~184.6 shows the site of a 278-bp insertion of the Jockey element present in the background strain for the enhancer trap insertions and their derivatives. Above the DNA line are shown the mutant lesions closest to the border, the regions of homology to the split complex in D. virilis, and the closest enhancer traps. Green coloration indicates inclusion in the bxd domain (by phenotype, homology, or expression pattern); blue indicates the iab-2 domain. All three criteria confine the bxd/iab-2 border to a 2-kb interval that includes a prominent binding site for the CTCF factor. The Δ1.9-kb deletion, shown in black, retains the CTCF-binding site and had no phenotype. The Fub deletion, shown in red, removes the CTCF site and causes a dominant PS6 to PS7 transformation. The yellow double-headed arrow marks the 7.5-kb SalI fragment that confers P-element homing to the bithorax complex.

P-element-mediated gene conversion can be used to introduce small deletions, but it seemed unlikely that a conversion interval would be large enough to include the ~4-kb border region. Xie and Golic (2004) developed a DNA cut-and-repair strategy that can be used in conjunction with homologous recombination to generate large and precise deletions. We used a similar strategy, combined with P-element gene conversion, as diagrammed in Figure 5. The procedure yielded a 4328-bp deletion (182,353–186,679 in SEQ89E numbering), spanning the distance between the most distal bxd lesion and the most proximal iab-2 lesion. Flies heterozygous for this deletion, called Front-ultraabdominal (Fub), showed a dramatic transformation of the first abdominal segment to the character of the second. The A1 tergite shows black pigmentation and large bristles, like those of A2 (Figure 5), and the A1 sternite, normally lacking bristles, has bristles like those of A2. Homozygotes usually die as pharate adults, with an apparently complete transformation of A1 to A2, and often missing one or both halteres and (rarely) one or both third legs. Parasegment 6 includes the posterior compartments of the halteres and metathoracic legs, and so PS6 to PS7 transformations should also affect these appendages, causing their failures to emerge. Homozygotes do not have apparent bxd or pbx loss-of-function phenotypes (anterior transformations in PS6, such as posterior haltere-to-wing transformation, loss of the first abdominal tergite, or appearance of extra legs from the first abdominal segment). When embryos from heterozygous adults were stained for ABD-A protein, there were three staining patterns (Figure 6). Most of the embryos showed ectopic ABD-A in PS6, some with half the intensity of the PS7 level, and some (the presumed Fub homozygotes) with PS6 and PS7 equal in pattern and intensity. The appearance of ABD-A in PS6 is not obviously delayed relative to that in PS7; ABD-A is equally intense in both parasegments in homozygous embryos at stage 10 (~5 hr old).

Figure 6.

ABD-A expression in embryos with the bxd/iab-2 border deletion. Embryos were collected from adults heterozygous for the Fub deletion and stained for ABD-A. Stage 15 embryos (~12 hr old) are illustrated; three classes were apparent. The left embryo shows the wild-type ABD-A pattern. The middle embryo, presumed to be a heterozygote, shows ectopic ABD-A expression in PS6, at ~1/2 the intensity of PS7. The embryo on the right is the presumed deletion homozygote; it shows ABD-A misexpression in PS6 equivalent to the normal expression in PS7. Brackets show the extent of PS6 in the epidermis (]) and in the CNS ([).

Discussion

Dissection of the bxd domain

The deletion series shown in Figure 1 looks strictly cumulative, in that additional DNA upstream of the Ubx promoter gives additional expression of UBX protein. But the details are surprising in several respects. The DNA segments ending at the Hm and the bxd¹⁰⁰ breakpoints show UBX expression in additional cell types in PS6-PS12 beyond those seen in PS5. Yet these DNA segments lack all of the mapped embryonic enhancers in the bxd region that drive PS6 expression in transgene assays [designated as bxd, S1, S2, and pbx in Figure 1 (Poux et al. 1996)]. They also lack the prominent PRE, but the UBX expression pattern does not spread to PS5 or more anterior parasegments in late embryos. It is likely that there are additional PS6-specific embryonic enhancers, not yet mapped, and perhaps cryptic PREs. It is also possible that enhancers and/or PREs within the Ubx transcription unit contribute to this pattern. The DNA segments extending up to the bxd¹¹¹ breakpoint, ~28 kb from the Ubx start, also lack the promoter for the major noncoding RNA, which spans much of the bxd domain (Lipshitz et al. 1987). It has been proposed that transcription of noncoding RNAs across a domain (or across a PRE) is required for that region to be functionally active (Schmitt et al. 2005); such a function cannot be assigned to the major embryonic bxd noncoding RNA.

In the embryos of Figure 1B, there are distinct posterior limits to the UBX expression driven by portions of the bxd domain. The embryonic enhancers included in the bxd¹⁰⁰ DNA segment drive expression in PS6–PS12, while the more distal enhancers work in PS6–PS13. Thus, the lack of UBX expression in PS13 and PS14 is not solely a function of repression by ABD-B.

Fub border

The border between the bxd and iab-2 regulatory domains is positioned by (1) enhancer trap patterns, (2) mutant lesions, and (3) homology to D. virilis. Figure 7 presents the limits on the position (or extent) of the border from each of these criteria. The tightest limit is provided by the edges of homology to D. virilis, and this 1.8-kb interval is contained within the limits of the two phenotypic assays. The deletion of this border interval gives a homeotic protein expression pattern (spread through the anterior adjacent segment) and an adult phenotype (dominant, one-segment posterior transformation) that are both analogous to those of Mcp, Fab-6, Fab-7, and Fab-8. Thus, the original assumption that these latter four deletion mutations remove barriers to the spread of activation seems validated.

The Fub border interval does not include a PRE, in that it lacks binding sites for known components of the Polycomb Group repression machinery. Specifically, genome-wide chromatin immunoprecipitation profiles from several cultured cell lines show reduced levels of POLYCOMB at this border, relative to adjacent regions, and little or no binding for other proteins of PRC1 (POLYHOMEOTIC, POSTERIOR SEX COMBS), of PRC2 (ENHANCER OF ZESTE), or of PhoRC (PLEIOHOMEOTIC, SFMBT) (Schwartz et al. 2006; Schuettengruber et al. 2009; modENCODE Consortium 2010; Enderle et al. 2011). Likewise, chromatin immunoprecipitation from 4- to 12-hr-old embryos shows reduced POLYCOMB levels at this border, and little or no binding of POLYHOMEOTIC or PLEIOHOMEOTIC (Schuettengruber et al. 2009). The Mcp, Fab-6, Fab-7, and Fab-8 borders are associated with prominent peaks of POLYHOMEOTIC and PLEIOHOMEOTIC (Schuettengruber et al. 2009). The position of the iab-3/iab-4 border can be guessed from the positions of enhancer traps marking PS8 (HCJ200) (Bender and Hudson 2000) and PS9 (HF608B) (Fitzgerald and Bender 2001); this interval (125,800–127,370) also corresponds to peaks of POLYHOMEOTIC and PLEIOHOMEOTIC. Although the coincidence between borders and Polycomb Group binding sites is striking, the Fub border indicates that PRE function is not essential for border function. There are no apparent loss-of-function phenotypes caused by the Fub deletions, as there were with the Fab-6 deletions and the Fab-7 deletions retaining the PRE (Mihaly et al. 1997; Iampietro et al. 2010). Perhaps the spread of repression from a PRE is short range, and there is no PRE in the iab-2 domain sufficiently close to the border for repression to spread across in the Fub deletion mutants.

The Fub border region includes a prominent binding site for the CCCTC-binding factor (CTCF) (Figure 7), as assayed in early embryos and in cultured cells by chromatin immunoprecipitation (Holohan et al. 2007; Nègre et al. 2010). CTCF is a zinc-finger DNA-binding protein proposed to be involved with intrachromosomal looping and with locus boundaries (Herold et al. 2012). The Mcp, Fab-6, and Fab-8 borders are also sites for CTCF binding, and there are additional sites that may mark the iab-2/iab-3 and iab-3/iab-4 borders (Negre et al. 2010). The Fab-7 border has only a very weak association with CTCF and lacks consensus CTCF-binding sites (Holohan et al. 2007). There are also CTCF sites within the bxd region and within the Ubx transcription unit that do not correspond to suspected border regions. The correspondence between borders and CTCF sites is striking, although evidence for CTCF function at borders is limited. Mohan et al. (2007) examined zygotic null mutants for CTCF and reported partial posterior transformations of the fourth abdominal (A4) and fifth abdominal (A5) segments in pharate adults. The A4 to A5 transformation appeared less severe than that seen in Mcp/+ animals, and there were no apparent Fab-7 (A6 to A7) or Fub (A1 to A2) transformations. It should be informative to remove maternal as well as zygotic CTCF, and to mutate the CTCF sites in the BX-C to learn if CTCF is essential for blocking the spread of activation at borders.

Homing

The Fub border lies within a 7.5-kb SalI fragment (Figure 7) shown to confer homing of P elements to the BX-C (Bender and Hudson 2000). In that study, LacZ transgenes positioned proximal or distal to this “homing fragment” showed expression with anterior limits in PS6 and PS7, respectively, while a LacZ reporter placed between two copies of the homing fragment showed no segmental limit to LacZ expression. It was argued that this fragment included a boundary to the domain of Polycomb-mediated repression (Bender and Hudson 2000). The best-characterized sequence with analogous homing properties has been mapped at the edge of the even skipped (eve) locus. Fujioka et al. (2009) have defined an ~600-bp fragment called “Homie” that directs P-element insertions into or near the eve locus; it also includes a prominent binding site for CTCF (Nègre et al. 2010). P-element homing has also been well documented for small fragments near the promoters of engrailed (Kassis et al. 1992; Cheng et al. 2012) and linotte/derailed (Taillebourg and Dura 1999). In both of these cases, the minimal homing fragment coincides with a CTCF-binding site in embryos or cultured cells (Nègre et al. 2010; modENCODE ID 2638 and 2639). The association of CTCF with P-element homing seems clear, but there must be distinctive properties to each of these CTCF regions since each homing fragment targets its own locus.

Possible border functions

Mapping the borders is merely a first step in the larger investigation of how they function. In the cells of PS6, the Fub border lies between an active bxd domain and a silenced iab-2 domain. When the border is deleted, activation spreads through the iab-2 domain. The deeper question, then, is how activation spreads. An appealing model is that noncoding RNAs might extend across the bxd domain, leaving activating chromatin marks as they go, until they are blocked or terminated at the border. However, no such transcripts approaching the Fub border from the proximal side have been detected in embryos (B. Pease and W. Bender, unpublished results). Moreover, borders between the iab-2 and iab-8 domains (including Mcp, Fab-6, Fab-7, and Fab-8) are not barriers to transcripts in the distal-to-proximal direction, since the iab-8 noncoding RNA proceeds through them all (Gummalla et al. 2012). However, all these domains are in the active mode in PS13, where the iab-8 RNA is made. Spreading activation need not involve transcription; perhaps nucleosome modification or remodeling could be propagated from the early enhancers that respond to gap and pair rule genes. The PS6 misexpression of ABD-A in very early Fub embryos suggests that the spread of activation is fast, although it is not clear how much of the iab-2 domain must be activated before Abd-A transcription begins.

The border’s blockage to spreading activation could be entirely passive; each domain could be activated or repressed, regardless of the state of adjacent domains. By this model, the structure of the border need not change from one parasegment to another. Alternatively, the borders could be active attachment points for segregating DNA domains into repressive structures or nuclear compartments. The Fub border might be attached in PS1–PS6, but released in PS7–PS13. This model would require that a border is modified according to its segmental position. The positional sensor is not likely to be intrinsic to the border, since swapping experiments—i.e., switching the Fab-7 and Fab-8 borders—suggest that borders are largely interchangeable (Iampietro et al. 2008). To distinguish between the passive and the active models, it would be instructive to assay the proteins bound at a border, and to see if the composition changes from one parasegment to another.