Cap ‘n’ collar B cooperates with a small Maf subunit to specify pharyngeal development and suppress Deformed homeotic function in the Drosophila head

Segmental diversification of the posterior head is a part of a conserved genetic system for anterior-posterior patterning of a metameric animal body, in which homeotic selector genes play a central role (Lawrence and Morata, 1994; Mann and Affolter, 1998; McGinnis and Krumlauf, 1992). In Drosophila, Hox genes labial (lab), proboscipedia (pb), Deformed (Dfd) and Sex combs reduced (Scr) are required for determining the identity of the head segments (Popadic et al., 1998). Their action is assisted by many other head-specific genetic functions, whose role is to elaborate on Hox signals and to specify the differentiation of head-specific structures (Rogers and Kaufman, 1997). One such function is encoded in the CncB protein isoform. In a previous study, we showed that the cnc locus encodes three transcript and protein isoforms. The cncB transcript is expressed in an embryonic pattern that includes the labral, intercalary and mandibular segments, while cncA and cncC are expressed ubiquitously (McGinnis et al., 1998; Mohler et al., 1991). Mutant embryos with reduced levels of CncB protein have a duplicated set of mouth hooks and cirri in place of mandibular structures (Harding et al., 1995; McGinnis et al., 1998; Mohler et al., 1995). Mouth hooks and cirri derive from the maxillary segment and are specified by the Dfd segmental identity function. Transient ectopic expression of CncB can repress Dfd function in the maxillary segment, inducing a loss of normal maxillary structures, while having no effects on the functions of homeotic proteins in the trunk (McGinnis et al., 1998). The cncB loss-of-function phenotype has therefore been interpreted as a homeotic transformation in which the maxillary-promoting activity of Dfd is derepressed in the mandibular segment.

Lower levels of CncB function also result in other head defects, including the loss of the median tooth and dorsal pharynx (labrum-derived structures), loss of the floor of the pharynx (an intercalary segment-derived structure), and loss of the anterior lateralgraten and the base of the mouth hook. The lateralgraten and mouth hook base both originate in part from the posterior mandibular segment and both require Dfd and CncB function for normal development. Thus, in addition to CncB’s role in suppressing the maxillary-promoting function of Dfd, it is also required to promote the development of labral, intercalary and posterior mandibular structures (McGinnis et al., 1998; Mohler et al., 1995). The mechanisms by which CncB accomplishes these functions are poorly understood.

All three Drosophila Cnc protein isoforms contain a basic-leucine zipper region that is highly homologous to the p45 subunit of mammalian NF-E2 (Andrews et al., 1993a; Mohler et al., 1991). Numerous studies have demonstrated transcriptional activation properties of CNC-class b-Zip proteins from mammals: e.g. p45 NF-E2 (Amrolia et al., 1997; Bean and Ney, 1997; Deveaux et al., 1997; Nagai et al., 1998), TCF11/LCR-F1/Nrf1 (Johnsen et al., 1998), ECH/Nrf2 (Alam et al., 1999; Itoh et al., 1997, 1995; Wild et al., 1999) and Nrf3 (Kobayashi et al., 1999). Such activation has been shown to be dependent on heterodimerization of the CNC family members with small Maf subunits (reviewed in Blank and Andrews, 1997; Motohashi et al., 1997) and on the presence of specific heterodimer binding sites in target enhancers. In addition to NF-E2-like consensus sites, enhancers targeted by CNC family members contain other sites that are required for full activity and bound by cofactors such as GATA-1, Ets-like proteins and Sp1 (see Magness et al., 2000; Deveaux et al., 1997; Stamatoyannopoulos et al., 1995 and references therein). Only those CNC class proteins that contain BTB domains, such as Bach1 and Bach2, have been shown to function primarily as transcriptional repressors (Igarashi et al., 1998; Muto et al., 1998; Oyake et al., 1996).

In this study we have investigated the requirements for CncB function in Drosophila embryos, and find that it acquires sequence-specific DNA binding, and biological function, in combination with a Drosophila homolog of vertebrate small Maf proteins. We find that among the three Cnc isoforms, only CncB is involved in specification of embryonic head skeletal structures. The CncB protein contains a strong transcriptional activation domain and can activate simple response elements in vivo. We provide several lines of evidence suggesting that CncB-mediated repression on at least some Dfd response elements is likely to be indirect. Finally, we have found that CncB is sufficient for the development of ventral pharyngeal structures. Based on these results, we propose a model explaining the diverse functions of CncB in anterior-posterior patterning of the Drosophila embryonic head, and discuss possible links between genetic systems involving CNC family members in Drosophila and other organisms.

Flies were reared on a standard medium at 25°C. To generate new cnc alleles, ve st e ca males were fed ethylmethanesulfonate (EMS), according to Grigliatti (1986). Mutagenized males were crossed en masse to virgin TM3 Sb / TM6B Hu Tb females, and ve st e ca * / TM6B males were then singly crossed to virgin cnc^VL110/ TM6B females. Candidate cnc mutants were identified from these crosses by the absence of empty non-Tb pupal cases. A total of 8600 chromosomes were analyzed, and eight new lethals recovered.

Broad-range denaturing gradient gel electrophoresis (DGGE) was used to identify locations of the new cnc mutations, following the procedure described in Guldberg and Güttler (1994). All three Cnc reading frames were covered by overlapping fragments of 300-600 bp that were PCR-amplified from genomic DNA extracted from heterozygous adults. The fragments were chosen using MacMelt software (BioRad). In each amplification reaction, one of the primers contained a 36-nucleotide 5′ GC-clamp (Myers et al., 1985). PCR products were resolved using the D-GENE apparatus (BioRad) on a 0-80% denaturant 6% polyacrylamide gel run at 60°C according to the manufacturer’s instructions. Candidate fragments containing mutations were sequenced on both strands along with parental lines to identify exact locations of mutations, which are summarized in Table 1.

Heat-shock treatment: embryos were collected on apple juice/agar plates for 2-4 hours at 25°C, aged for 6 hours, then the plates were immersed in a 37°C water bath for 1 hour. After heat shock, embryos were immediately analyzed for mRNA expression or aged over 24 hours for cuticular preparations. Antibody and in situ staining and cuticular preparations were done as previously described (McGinnis et al., 1998).

RNAs were synthesized from PCR products containing a T7 promoter sequence on each side, according to Kennerdell and Carthew (1998). The presence of double-stranded self-annealed products was verified on an agarose gel. RNAs were complementary to cDNA sequences spanning the following nucleotides: 22-464 for CncA (GenBank accession AF070062), 117-498 for CncB (GenBank accession AF070063), 61-994 for CncC (GenBank accession AF070064), 45-601 for Maf-S (GenBank accession AI238611). All three Cnc fragments contained 5′ UTR sequences, while the Maf-S RNA covered the whole open reading frame plus parts of the 5′ and 3′ UTR. dsRNA was injected into the head region of precellular blastoderm embryos at a concentration of 0.5 μg/μl of the double-stranded product. Injected embryos were then aged for 26-28 hours at 25°C and analyzed for cuticular phenotypes.

Full-length CncA, CncB and Maf-S proteins were cloned into the pSP64ATG or pSP64Flag vectors (a gift from C. Murre). For site selection and EMSA, the proteins were produced using a TNT in vitro transcription/translation system (Promega). Site selection protocol followed the procedure of Neuteboom and Murre (1997), except the first four washes were performed using the salt concentration of 50 mM KCl/50 mM NaCl, and the final two washes used 50 mM KCl/250 mM NaCl. For site selection, Flag-tagged CncA or CncB were cotranslated with the untagged Maf-S. A 62-mer oligonucleotide used in site selection procedure was 5′-GTAAAAGCACGGCCAGTGA-ATTCT(N)₁₄GTCTAGAGTGCTGACTGGGAAAAC, where (N)₁₄ indicates a core of 14 ambiguous positions. Primers A (5′-GTAAAAGCACGGCCAGT) and B (5′-GTTTTCCCAGTCAGCAC) were used for amplification of the 62-mer. After 6 cycles of binding, washes and amplification, the resulting pool of oligonucleotides was digested with EcoRI and XbaI and cloned into pBluescript (Stratagene). 20 clones were randomly chosen and sequenced.

Electrophoretic mobility shift assays were performed as in Neuteboom and Murre (1997). The double-stranded Cnc/Maf binding site oligonucleotide (CBS) was 5′-GGTTCTGCTGAGTCATCGTG (consensus binding site is underlined), and the mutant Cnc/Maf binding site oligonucleotide (mCBS) was 5′-CTTCTTGTACGGACT-CGTGGAATT. To assay heterodimer binding, Cnc proteins and Maf-S were contranslated, though we observed no decrease in binding when separately translated proteins were mixed.

Transgenic lines were established using standard methods (Spradling, 1986), and multiple lines were analyzed for each construct. All fragments made by PCR were verified by sequencing. To generate CncB^Ala, a part of CncB was amplified in a PCR reaction using a primer containing the nucleotide substitutions required to change the IRRRGKNKV sequence in the DNA binding domain of CncB to IRAAGAAKV (Fig. 1E). Full-length CncB^Ala was then regenerated by substituting a wild-type part of the sequence with the mutated fragment. This cDNA was subcloned into pHSBJ-CaSpeR (Jones and McGinnis, 1993) for heat-shock induction in the embryos and into pSP64ATG for in vitro translation. P{hs-cnc.A} (HS-CncA) and P{hs-cnc.B} (HS-CncB) were described in McGinnis et al. (1998). Descriptions of published transgenic lines can also be found in FlyBase (The FlyBase Consortium, 1999). P{UAS-hth.P} (UAS-Hth) was a gift from Y. Henry Sun. A full-length Maf-S open reading frame is present in the clone GH14688 (GenBank accession AI238611), which is available from Research Genetics. P{UAS-Dfd.L} (UAS-Dfd) is described in Li et al. (1999a). GAL4 drivers are described in the following references: P{GawB}69B (69B-GAL4), Brand and Perrimon (1993); P{GAL4-arm.S} (arm-GAL4⁴), Sanson et al. (1996); P{GAL4-Hsp70.PB} (HS-GAL4), Brand et al. (1994).

To make P{CBS Dfd-lacZ.4CE} (4CE), a 20-bp oligonucleotide containing the Cnc/Maf binding site (CBS) was cloned into PstI/EcoRI sites of plasmid pB-E upstream of module E (Zeng et al., 1994) to generate pB-CE. The resulting CE cassette was quadruplicated in tandem to generate 4CE and cloned into the XbaI/XhoI sites of CZIII lacZ reporter vector (Zeng et al., 1994). P{mCBS Dfd-lacZ.4ME} (4ME) was created in the same way, except the mutant Cnc/Maf binding site (mCBS) was used in place of CBS. To make P{CBS-lacZ.4C} (4C), a 40-bp oligonucleotide containing two copies of CBS was cloned in tandem into the EcoRI site of pBluescript, and head-to-tail orientation was selected by sequencing individual clones. The final cassette was then cloned into XbaI/XhoI sites of CZIII.

To make a GAL4 DNA binding domain (DBD)∷CncB-specific domain chimera, PCR-generated CncB-specific domain (amino acids 1-272) was ligated with GAL4 DBD in pBXG1 vector (Quong et al., 1993). This construct was cotransfected with the UAS-luciferase reporter into Bosc23 cells, and enzyme activity was measured as previously described (Li et al., 1999a).

P{UAS-cnc.B} (UAS-CncB) was created by subcloning an EcoRV fragment from a full-length CncB cDNA (McGinnis et al., 1998) into the pUAST vector (Brand and Perrimon, 1993). To make a P{UAS-VP16∷cnc.A} (UAS-VP16-CncA) expression construct, CncA (common) region was ligated with the VP16 fragment from pGEX-VP16 (Li et al., 1999a) and inserted into the pUAST vector, so that VP16 is at the amino-terminal end. To test for coexpression effects of CncB and Dfd in Dfd^w21 mutant background, the following chromosomes were prepared by recombination and combined by crossing appropriate lines: UAS-Dfd Dfd^w21, arm-GAL4⁴Dfd^w21 and UAS-CncB UAS-Dfd Dfd^w21. To test for coexpression effects of Cnc proteins and Hth, HS-CncA; HS-GAL4 or HS-CncB; HS-GAL4 flies were crossed to HS-CncA; UAS-Hth or HS-CncB; UAS-Hth, respectively.

We have previously shown that the cnc locus encodes three transcript and protein isoforms: CncA (whose amino acid sequence is included in all Cnc protein isoforms), CncB and CncC (McGinnis et al., 1998). Based on our genetic and molecular analysis, only one isoform, CncB, was implicated in the suppression of Dfd function and formation of the intercalary and labral structures. However, available mutations in the cnc gene were limited to P element insertion and excision lines, disrupting the function of the whole locus (Mohler et al., 1995), and to three EMS alleles identified in a Dfd modifier screen that were not associated with sequence polymorphisms in the open reading frames of the Cnc isoforms (Harding et al., 1995; McGinnis et al., 1998). We therefore screened for additional EMS-induced lethals at the cnc locus to better understand the function of individual isoforms. Eight new lethal mutations were identified that failed to complement cnc^VL110, a deletion removing the common cnc exons. One of these mutations turned out to be a complex chromosomal rearrangement and was not analyzed further. The remaining seven chromosomes were analyzed for embryonic cuticular phenotype as well as for Cnc mRNA and protein levels as transheterozygotes over cnc^VL110 (Table 1). The mutants were also analyzed for the presence of nucleotide changes in the open reading frames and adjacent intronic regions covering all three Cnc isoforms. We were able to identify the nature of mutations in four of these alleles by using PCR amplification, denaturing gradient gel electrophoresis, and sequencing of the candidate mutation-containing fragments (see Materials and Methods; schematic locations of mutations are shown in Fig. 1A).

Four of the new alleles (cnc^C5, cnc^C29, cnc^G15 and cnc^K22) displayed a strong cncB mutant phenotype (Mohler et al., 1995; McGinnis et al., 1998), characterized by duplicated mouth hooks (Fig. 1B, arrow), loss of median tooth, shortened lateralgraten and loss of the posterior pharynx (Fig. 1B, arrowhead; compare to Fig. 1C). The mutations in cnc^C5 and cnc^C29 are predicted to result in truncated forms of all Cnc isoforms (Fig. 1A, Table 1), while no mutations were detected in the open reading frames of the cnc^K22 and cnc^G15 alleles. In all four strong alleles, no protein is detected with an antibody specific for the Cnc common region (Table 1). The levels of CncB mRNA were normal or only slightly reduced in cnc^C5 and cnc^C29, moderately reduced in cnc^K22 and strongly reduced in cnc^G15. This suggests that cnc^G15 has a mutation affecting a regulatory region required for the transcription of all cnc isoforms, and cnc^K22 has a mutation affecting mRNA abundance as well as the accessibility of the mRNA to the translation apparatus.

The remaining three alleles (cnc^C15, cnc^J26 and cnc^K6) are hypomorphs. cnc^C15 and cnc^J26 mutant embryos each show mild defects in lateralgraten and mouth hooks, but have normal median teeth and pharynx (data not shown). The mutation in cnc^C15 is not in the open reading frame, whereas cnc^J26 has a sequence change that affects the donor splice site in the last intron of the common region, which is predicted to change the carboxy-terminal sequence of all three isoforms downstream of the leucine zipper. cnc^J26 may affect translation efficiency or protein stability, as it results in slightly lower protein levels without any effect on mRNA levels (Table 1). cnc^K6 homozygotes have normal embryonic morphology (data not shown). Interestingly, the molecular defect in cnc^K6 is a nonsense mutation in the CncC-specific region, which is predicted to truncate CncC protein, without affecting CncA or CncB. The absence of cuticular defects in the cncC^K6 mutant strongly suggests that the CncC isoform is not required for the development of embryonic head skeleton. Since cncC^K6 is lethal over cnc^VL110, CncC is apparently required for some other essential function at later stages.

Since each Cnc transcript has at least one unique 5′ exon (McGinnis et al., 1998), the method of double-stranded RNA interference (dsRNAi; Fire et al., 1998; Kennerdell and Carthew, 1998) allowed us to test the function of each isoform independently. Double-stranded RNA fragments specific for the 5′ UTR of CncA, CncB or CncC (see Materials and Methods) were injected into the head region of precellular blastoderm embryos, and cuticular preparations were made to assess the extent of head skeleton development. For both CncA- and CncC-specific injections, the first instar larval hatch rate was comparable to control injection of buffer alone (40% of all injected embryos), and in the unhatched embryos, no consistent head defects were detected. The absence of any head phenotype and high survival rate of CncC-injected embryos agrees with the cncC^K6 mutant data. The normal morphology of CncA-RNAi-injected embryos suggests that this isoform is not required for embryonic morphological development, but in the absence of CncA specific mutants, the lack of a dsRNAi-induced phenotype might also mean that dsRNAi was incapable of influencing CncA expression in our experiments.

The hatch rate for CncB-specific injections was significantly lower than in controls (less than 1%). Embryonic cuticles displayed partial or complete duplication of mouth hooks, and approximately 10% had a phenotype indistinguishable from strong mutations in cncB (Fig. 1D, compare to B). We also observed a similar phenotype when CncB dsRNA was injected into the posterior of the embryo. The percentage of severe head transformations in this case was slightly lower than for head-injected embryos. The results from the dsRNA interference experiments, combined with the analysis of the new cnc mutants, provide strong evidence that among the three Cnc isoforms, only CncB plays a role in specifying the development of the embryonic head skeleton, while CncA and CncC perform other functions.

The presence of a basic-leucine zipper region in all three Cnc isoforms, including CncB, suggests that they function as DNA binding proteins. However, several experiments have indicated that p45 NF-E2, a mammalian protein that is closely related to the Drosophila Cnc isoforms, is involved in functional protein-protein interactions that do not require DNA binding (Cheng et al., 1997; Gavva et al., 1997). To test whether CncB activities are dependent upon DNA binding function, we generated a transgenic strain with a heat-shock-inducible form of CncB in which alanines were substituted for conserved residues in the basic domain (construct HS-CncB^Ala; Fig. 1E). Mutations of these residues abolish DNA binding in other b-Zip proteins (Carroll et al., 1997), as well as CncB (see Fig. 4C). Heat-shock-induced ubiquitous expression of wild-type CncB between 6 and 8 hours of development is lethal for 99% of embryos, and most lack mouth hooks and other Dfd-dependent structures in their defective heads (Fig. 1F, arrows) (McGinnis et al., 1998). When HS-CncB^Ala is expressed under the same conditions, about half of the embryos hatch, and the embryos that fail to hatch have normal head morphology (Fig. 1G). As measured by embryonic immunostaining and quantitative western blots, the mutant protein is present at the same level as wild-type protein, and is concentrated in nuclei (data not shown). Therefore, the residues in the b-ZIP region of CncB that provide its DNA binding function are required both for the repression of Dfd-dependent structures as well as for the other cuticular defects resulting from ectopic expression.

Formation of maxillary structures such as mouth hooks depends on the activation of downstream target gene transcription by Dfd protein (Zhu and Kuziora, 1996). This activation has been shown to depend on parallel input from Extradenticle (Exd) and other cofactors (Li et al., 1999b; Pinsonneault et al., 1997; Zeng et al., 1994). A MEIS-like homeodomain protein Homothorax (Hth) is required for the normal expression levels and the nuclear localization of Exd, and has also been recently shown to participate in trimeric complexes with Hox proteins and Exd (Ryoo et al., 1999). Since Exd and Hth act in parallel to potentiate the transcriptional activation function of Dfd, and CncB appears to do the opposite, we asked whether increasing the levels of Hth and therefore of nuclear Exd would affect the ability of CncB to repress Dfd-dependent structures. When CncB and Hth are both ectopically expressed throughout the embryo, a reduction in severity of all aspects of the CncB ectopic phenotype is observed. Although overexpression of CncB and Hth still results in embryonic lethality, most head skeletal elements are close to wild type, and mouth hooks and other Dfd-dependent structures are seen in all embryos (Fig. 1H, compare to F). Control embryos that coexpress CncA (which has no effect on head development) and Hth under the same conditions show mild or no head defects (Fig. 1I). We conclude from this experiment that increasing the concentration and nuclear localization of Dfd activating cofactors such as Exd and Hth can partially rescue the repressive effects of CncB.

CncB can suppress Dfd function in a way that is independent of its effects on Dfd transcription Previous results showed that CncB acts as a selective suppressor of Dfd function, and can also act to repress Dfd transcription (McGinnis et al., 1998). It was still unclear, however, whether CncB-mediated repression of Dfd-dependent maxillary structures actually requires silencing Dfd transcription. To answer this question, we ectopically expressed CncB and Dfd proteins in Dfd^w21 homozygous mutant embryos, which lack the endogenous Dfd protein (Zeng et al., 1994). As shown in Fig. 2A, Dfd^w21 mutants lack mouth hooks, cirri and other Dfd-dependent structures (compare to Fig. 1C). When Dfd protein is provided in the Dfd^w21 mutant embryos from a UAS-Dfd construct under the control of the arm-GAL4⁴ driver, maxillary mouth hooks and cirri are restored to near-normal morphology, and some maxillary structures are produced in other segments (Fig. 2C). The rescue of the Dfd^w21 mutant phenotype by Dfd protein under the control of the arm driver is suppressed when CncB protein is coexpressed with Dfd using the same driver (Fig. 2D) – mouth hooks, cirri and other Dfd-dependent structures are missing or severely underdeveloped. This is similar to the phenotype obtained when CncB is driven by arm-GAL4⁴ in otherwise wild-type embryos (Fig. 2B). It is possible that CncB might interfere with Dfd expression from the GAL4-induced heterologous promoter by interfering with the activation of the UAS in the UAS-Dfd construct. We tested this by staining the ectopic CncB, ectopic Dfd embryos, and found that Dfd protein is produced at similar levels and localized in nuclei, as seen in controls in which no ectopic CncB is present (data not shown).

Thus, even when Dfd protein is persistently expressed from a heterologous promoter, its maxillary-promoting function is inhibited if CncB protein is present in the same cells. The normal repression of Dfd transcription observed in anterior mandibular cells in wild-type embryos may be a secondary effect resulting from CncB dependent inhibition of Dfd protein function on the Dfd autoactivation circuit (Kuziora and McGinnis, 1988).

Mammalian homologs of Cnc (p45 NF-E2 and Nrf proteins) were shown to bind DNA as obligate heterodimers with small Maf proteins (MafK/p18, MafF and MafG) (Andrews et al., 1993b; Igarashi et al., 1994; Johnsen et al., 1996). An apparent Drosophila ortholog of the mammalian small Mafs was identified in a yeast two-hybrid screen for genes encoding peptides capable of interacting with the common b-Zip domain of Cnc proteins (J. M., unpublished results). BLAST searches with a corresponding cDNA sequence detect only one small maf gene in the near complete Drosophila genome (Adams et al., 2000) (FlyBase Genome Annotation Database (GadFly) identifier CG9954, http://flybase.bio.indiana.edu/annot/). We call the gene Drosophila maf-S (S for Small), and the protein Maf-S. In the basic region that contacts DNA, Maf-S shares a domain of perfect identity with mammalian small Maf proteins (Fig. 3A), and partial identity in the leucine zipper and other protein regions. In situ hybridizations show that maf-S RNA is maternally deposited in the Drosophila egg, and the transcripts are present in all or nearly all embryonic cells throughout development, making the Maf-S protein available for potential interactions with all Cnc isoforms (Fig. 3B). In addition, there is an enrichment in maf-S transcripts in certain areas of the embryo, against the ubiquitous expression background. At stage 8, slightly higher levels are detected around cephalic furrow and in the mesoderm. At stage 11 and beyond, there is an enrichment in maf-S transcripts in the anterior and posterior midgut, in the nervous system, as well as in the involuting intercalary segment that will give rise to ventral pharynx (Fig. 3C, arrow).

We have used dsRNA interference in an attempt to test the phenotypic effects of loss of maf-S function. After injection of maf-S dsRNA into the head region of precellular blastoderm embryos, only 1.5% of embryos hatched as first instar larvae. The cuticles of injected embryos showed head defects that are remarkably similar to those found in cncB mutants (Fig. 3D): duplications of mouth hooks, shortened lateralgraten, missing or deformed median tooth and truncated ventral pharynx (compare Figs 3D and 1B). Immunostaining of maf-S dsRNA-injected embryos with Cnc-specific antibody revealed that the levels of CncB protein are not significantly different than wild type (data not shown). Results of the dsRNA interference experiment suggest that Maf-S is an obligate functional partner of CncB. Without the small Maf subunit, CncB is unable to promote the development of intercalary, mandibular and labral head structures and to repress the function of Dfd.

The maf-S gene, which maps to chromosomal segment 57A, is uncovered by Df(2R)exu². Homozygotes for this deletion, which lack the zygotic dose of maf-S and other nearby genes, exhibit mild head defects that include disruptions of the pattern of ventral pharyngeal transverse ribs (T-ribs) (Fig. 3E, arrowhead; compare to Fig. 8B). The weakness of this phenotype relative to the RNAi result may be due to the maternal contribution from maf-S, which is unaffected in the zygotic deletion mutants. Further genetic characterization of maf-S mutant embryonic defects is impossible with Df(2R)exu², since it also uncovers exuperantia (exu), and maternal mutations in exu are associated with severe head defects (Schupbach and Wieschaus, 1986).

Despite a high degree of primary sequence similarity between the DNA binding domains of CncB and NF-E2, the optimal recognition site for the Cnc isoforms in association with Drosophila Maf might be different. We sought to find CncA and CncB optimal binding sites by selecting specific sequences from a pool of degenerate oligonucleotides by Cnc/Maf-S heterodimers (see Materials and Methods). Flag-tagged full-length CncA or CncB proteins were cotranslated in vitro with Maf-S and bound to an oligonucleotide containing a core of 14 ambiguous positions. Four rounds of selection, washes, elution and PCR were performed under mild conditions (100 mM total salt concentration), after which the selected pools of oligonucleotides were tested in a gel-shift assay. CncA/Maf-S- and CncB/Maf-S-selected pools of oligonucleotides displayed indistinguishable specificity and strength of binding when bound by either CncA or CncB with Maf-S (data not shown). The CncA/Maf-S-selected pool of oligos was subjected to two more rounds of selection with stringent washes (300 mM total salt) and then cloned. 20 randomly chosen clones were sequenced. The resulting consensus binding site is quite homogeneous, with seven perfectly conserved nucleotides (shown in bold in Fig. 4A). All selected sites started from the same position (T, marked with an arrow in Fig. 4A), suggesting that this and nearby nucleotides influenced the selection process.

The sequence selected in our experiment by Cnc/Maf-S heterodimers (TGCTGAGTCAT) is identical to the preferred site selected by TCF11/MafG (Johnsen et al., 1998), and is very similar to the consensus binding site derived for p45 NF-E2/p18 MafK from gel shift competition experiments (Andrews et al., 1993a) (Fig. 4B). The TGCTGAG part of the site corresponds to the Maf-bound sequences in the sites for mammalian CNC family proteins, and also to the T-MARE (TRE[phorbol-12-O-tetradecanoate-13-acetate (TPA)-responsive element]-type Maf recognition element) consensus half-site for large Maf homodimers (Motohashi et al., 1997).

The GTCAT half-site is contacted by Cnc homologs, and also is a perfect match to the binding site for the C. elegans Skn-1 protein that binds DNA as a monomer (Blackwell et al., 1994). We tested the consensus sequence (CBS, for Cnc/Maf binding site) in an electrophoretic mobility-shift assay (EMSA) with in vitro translated CncA, CncB and Maf-S proteins, along with a mutated site, mCBS (Fig. 4B). As shown in Fig. 4C, monomers of CncA or CncB do not appreciably bind to the Cnc/Maf consensus. Interestingly, in our assays, Maf-S does not bind as a homodimer either to this sequence or to the T-MARE consensus (Fig. 4C and data not shown). This property of the Drosophila small Maf protein is different from mammalian small Mafs, which form stable homodimer complexes on T-MARE-like sites (reviewed in Blank and Andrews, 1997; Motohashi et al., 1997). When CncA or CncB are cotranslated with Maf-S, abundant and stable complexes form on Cnc/Maf binding sites (Fig. 4C), suggesting that DNA binding is achieved only by Cnc/Maf heterodimers and that such binding is highly cooperative. CncA or CncB and Maf-S can be easily coimmunoprecipitated from separate in vitro translation reactions (data not shown), suggesting that stable heterodimers form in solution without DNA. Mutating DNA contact residues in the CncB protein abolishes protein/DNA binding (B^Ala+Maf, Fig. 4C), as does changing the nucleotides at critical positions in the Cnc/Maf binding site sequence (Fig. 4D, mCBS lanes). These data demonstrate that Maf-S and Cnc proteins can form specific heterodimer complexes on NF-E2-like sites, and that the presence of both protein subunits is required for complex formation.

The requirement of both CncB and Maf-S for specific DNA binding agrees well with the similarity of the loss of function phenotypes generated with dsRNA interference for either gene alone. Taken together, our data suggest that the biological function of CncB is achieved via its association with specific sites in the genome, in heterodimer complexes with the small Maf subunit. With the knowledge of requirements for specific DNA recognition by CncB, we set out to investigate the mode by which CncB achieves Dfd suppression and promotes the development of labral and intercalary structures.

We have previously shown that overexpression of CncB can partially repress Dfd response elements (McGinnis et al., 1998). However, none of the known Dfd response elements have close matches to the CncB/Maf-S consensus binding site (CBS, Fig. 4B). We have therefore positioned this site near a well-characterized Dfd response element, module E (Zeng et al., 1994), to test whether Dfd-dependent expression of module E would be influenced when ectopic CncB is supplied. Four copies of such a hybrid element (CE) were inserted in tandem into a lacZ reporter vector with a minimal promoter (construct 4CE, Fig. 5A). As a control, a mutated binding site (mCBS, Fig. 4B) was fused to module E with the same spacing and orientation (construct 4ME, Fig. 5D). Element CE was readily bound by CncB/Maf-S heterodimers in an EMSA, whereas no association was observed between element ME and CncB/Maf-S (data not shown).

Chimeric constructs 4CE and 4ME are expressed in posterior maxillary cells in an anchor-shaped pattern that is characteristic of module E (Fig. 5B,E). Such expression has been previously shown to depend on direct binding and activation of this element by Dfd and other cofactors (Li et al., 1999b; Zeng et al., 1994). In addition, construct 4CE, but not 4ME, is expressed in the intercalary and labral cells that contain high levels of endogenous CncB protein (this aspect of 4CE expression is discussed in the next section). We tested whether ectopic expression of CncB from a heat-shock construct could influence the maxillary expression of 4CE and 4ME. After a 1-hour heat shock, construct 4CE is ectopically activated in many cells in stage-13 embryos, especially in the dorsal-lateral epidermal region (Fig. 5C). In the maxillary segment, expression is also stronger and occurs now in more cells than in the non-heat-shocked embryos. 4ME, on the other hand, does not provide any ectopic expression with HS-CncB (Fig. 5F). In fact, construct 4ME is repressed by overexpression of CncB, in agreement with previous data on the normal 4xE element (McGinnis et al., 1998). Furthermore, HS-CncB^Ala (see Fig. 1E) is incapable of activating ectopic expression of 4CE or repressing maxillary expression of 4ME (A. V. and W. M., unpublished observation). These results suggest that CncB may be a transcriptional activator, with activation occurring through interaction of CncB/Maf-S with heterodimer binding sites in the target element. These data also indicate that when a simple Cnc/Maf binding site is placed near a functional Dfd element (module E), CncB protein does not repress Dfd-mediated activation of the element, but instead enhances it. In addition, ectopic CncB is able to activate expression of the 4CE element in the trunk cells that do not have any Dfd protein.

To test for CncB activation potential in the absence of any influence from adjacent sequences, we tested reporter expression driven by a multimerized Cnc/Maf binding site (construct 4C, Fig. 5G). No reporter expression from the 4C element is detected in embryonic labral, mandibular, or intercalary segments – regions where CncB is normally expressed. This suggests that a minimal sequence like 4C does not contain the information required to provide an activation response to CncB when the protein is provided at wild-type levels. Ectopic expression of CncB using heat shock leads to extensive activation of the 4C element in the trunk epidermal cells immediately after heat treatment (Fig. 5I), indicating that CncB possesses intrinsic activation properties that can be revealed on this element when CncB is provided at higher levels. However, even ectopic CncB protein is still incapable of activating the 4C element in the head region, indicating that a coactivator for CncB is localized in the trunk, and/or a corepressor for CncB is localized in the head. Assuming that a corepressor exists, its function can obviously be overcome on 4CE elements in the head region, as shown above. Ectopic CncA protein does not activate the 4CE, 4ME or 4C elements in embryos (data not shown).

From stage 12 onward, element 4C is activated in epidermal cells fated to become dorsal midline, cells that do not contain CncB protein (Fig. 5H). This 4C activity may be due to other factors such as Jun or Fos proteins, which can recognize the AP-1 core (TGANTCA) contained in the Cnc/Maf binding site. A control construct with mutated core nucleotides gave no detectable staining (data not shown).

We propose that the simple 4CE and 4C elements can be directly activated by CncB/Maf-S heterodimers in certain cells, but not by CncA/Maf-S. The difference between these two Cnc isoforms lies in the CncB-specific amino terminal domain, which is absent from the CncA isoform. We next asked whether this CncB-specific domain is sufficient to confer activation properties on a heterologous DNA binding domain. To that end, the 272 amino acid CncB-specific region was fused to the carboxy terminus of the GAL4 DNA binding domain (DBD). This chimeric protein was tested in mouse tissue culture cells (Bosc23) for its ability to activate a reporter element containing UAS sequences upstream of a basal promoter. We found that the GAL4 DBD∷CncB-specific fusion protein strongly activates reporter expression in a dose-dependent matter (Fig. 6). Taken together, the response element expression in Drosophila embryos and in mouse cultured cells strongly suggests that the CncB/Maf-S complex can act as a sequence-specific transcriptional activator on simple regulatory elements.

As mentioned in the previous section, the hybrid 4CE element is strongly expressed in the intercalary segment and in the clypeolabrum, as well as the posterior maxillary segment. Ventral intercalary cells internalize during head involution (Fig. 7A, arrow) to form the floor of the pharynx (Fig. 7C, arrow). At stage 15 and beyond, the 4CE transgene is expressed in the ventral pharyngeal cells that produce the T-ribs (Fig. 7D, arrowhead). Clypeolabral expression corresponds to the future position of the median tooth (Fig. 7A, arrowhead). Both the labral and intercalary expression domains of 4CE correspond to cells that contain abundant levels of CncB protein (Fig. 7B). To test whether CncB was required for the 4CE driven reporter expression, we tested 4CE activity in the cnc^VL110 null mutant background. In embryos homozygous for this deficiency, both hypopharyngeal and clypeolabral domains of 4CE expression are absent (Fig. 7E), suggesting that CncB is required for activation of the element. Construct 4ME, which contains mutations in the Cnc/Maf binding site, provides no expression in the pharynx or median tooth primordia in wild-type embryos (Fig. 7F). It is still possible that the loss of pharynx-specific expression of 4CE in cnc^VL110 embryos is due to cell loss in the mutants. At present, we do not have a CncB-independent marker to confirm that pharynx precursor cells are still present at this embryonic stage in cncB mutants.

The absence of 4CE expression in the pharynx precursors in cnc^VL110 embryos correlates well with the absence of ventral pharyngeal structures in strong cncB mutants. When CncB is ubiquitously expressed using the GAL4-UAS system (69B driver), the abdominal denticle pattern along the ventral midline of the embryo (Fig. 8A) is replaced with a band of incurved folds, which upon close examination resemble the T-ribs of the ventral pharynx (Fig. 8C,D, compare to B). An even stronger phenotype was observed when ectopic CncB expression is driven by arm-GAL4⁴ (see Fig. 2B). No additional transformations are seen when CncB is coexpressed with Maf-S (data not shown), suggesting that localization of transformation to the ventral region is not due to the limiting levels of Maf-S. Interestingly, 4CE is strongly expressed in epidermal ventral midline cells in 69B-GAL4/UAS-CncB embryos (Fig. 7H), whereas in the wild-type background such expression is absent (Fig. 7G). These 4CE-positive midline trunk cells elongate in transverse directions, similar to cells that form the T-ribs. This result, along with the formation of ectopic T-rib-like cuticle upon misexpression of CncB, suggests that persistent presence of this protein is not only necessary but sufficient for specification of the ventral pharynx.

An important question is whether the transcriptional activiation function of the CncB-specific domain is sufficient to provide the pharynx-promoting function of this isoform. We substituted the CncB-specific domain with the similarly sized VP16 activation domain (Triezenberg et al., 1988). This VP16-CncA fusion protein was tested in an ectopic expression experiment. Remarkably, when UAS-VP16-CncA is ectopically activated with the 69B-GAL4 driver, embryonic ventral cuticle is transformed towards pharyngeal identity, with a somewhat weaker expressivity than the one observed with wild-type CncB (Fig. 8E,F). In addition, UAS-VP16-CncA embryos have reduced maxillary structures and have general head defects that resemble those induced by CncB (data not shown). These data suggest that the CncB-specific domain encodes a general activation function, which can be partially mimicked by a heterologous activation domain. This activation function is required for the pharynx selector role of CncB as well as its function in suppressing Dfd-dependent maxillary development.

Previous genetic and molecular data have established that CncB has several functions in the developing Drosophila embryo: it is required for the formation of labral and intercalary structures such as median tooth and pharynx, and it is required to inhibit Dfd function in the mandibular segment (McGinnis et al., 1998; Mohler et al., 1995). Based on results presented in this study, we propose a summary model (Fig. 9) in which these functions are mediated in part by the transcriptional activator properties of CncB, and are achieved through specific DNA binding with Maf-S to an 11-bp CncB/Maf-S heterodimer binding site. In the intercalary and anterior mandibular segments of the embryo, heterodimerization of CncB with Maf can activate some downstream target genes, and the heterodimers are required to develop ventral pharyngeal structures (T-ribs and ventral arms). Furthermore, we find that CncB, in the context of the ventral epidermis of other head and trunk segments, is sufficient to promote pharynx identity, and thus cncB functions as a pharynx selector gene.

The function of CncB in suppressing the maxillary-promoting function of Dfd is carried out in the mandibular cells. Early in development CncB and Dfd are coexpressed throughout the mandibular segment, but by stage 13 Dfd expression retracts and becomes localized to a row of cells in the posterior of the segment (yellow region in Fig. 9). In these cells, CncB and Dfd are both required to specify lateralgraten and the base of the mouth hook (Mohler et al., 1995). Although Dfd function is required for the normal structures that derive from the posterior mandibular segment, the maxillary-promoting function of Dfd is inhibited in these cells (McGinnis et al., 1998). In the anterior mandibular segment, CncB represses both the function and expression of Dfd. Since Dfd protein activates its own transcription, inhibition of Dfd function inevitably results in silencing the endogenous transcription of the Dfd gene. Persistent expression of Dfd in the posterior mandibular cells is likely to be mediated by regulatory regions that are different from the autoactivation enhancers and are insensitive to CncB-mediated repression.

Based on the present data, we suggest that CncB-mediated repression of Dfd function is at least in part indirect (Fig. 9), for several reasons. First, there are no close matches to the consensus CncB/Maf-S binding site in known Dfd response elements, despite the fact that these elements can be repressed by ectopic expression of CncB, e.g. modules C, E, and F from the 2.7 kb Dfd epidermal autoregulatory enhancer (McGinnis et al., 1998) (see also Fig. 5F). Second, the CncB-specific amino terminal region, crucial for the suppression of Dfd function, contains a strong transcriptional activation domain. Third, a CncA fusion protein with a simple heterologous transcriptional activation domain from VP16 can also partially repress Dfd-dependent maxillary structures upon overexpression. Fourth, chimeric Dfd response elements containing Cnc/Maf binding sites adjacent to Dfd binding sites are not repressed by overexpression of CncB, as might be expected if the action of CncB on Dfd were direct. Instead, as shown in Fig. 5C, the endogenous Dfd-dependent maxillary expression of 4CE is enhanced in HS-CncB embryos when compared to non-heat-shocked controls. Suppression of Dfd function by CncB may therefore be achieved by transcriptional activation of an intermediate gene encoding a repressor or corepressor that would interfere with the function of Dfd. Another possibility is that CncB might compete for a common cofactor or coactivator (such as Exd, Hth or CBP), or it might activate a specific Dfd modifying enzyme such as a kinase or phosphatase. We note however, that it is still possible that the suppressive effect that CncB protein exerts on Dfd function may involve CncB mediated transcriptional repression on as yet uncharacterized response elements. We have observed a binding interaction between Dfd and CncB proteins in GST pull-down assays, but the biological significance of this result is as yet unclear (X. L., unpublished observation).

Repression of Dfd function by CncB is analogous to the phenomenon of phenotypic suppression (also known as posterior prevalence; González-Reyes et al., 1990), which denotes the ability of more posterior Hox proteins to suppress the function of more anterior Hox proteins in the same cells. ‘Anterior prevalence’ of CncB over Dfd provides additional morphological diversity in the mandibular and maxillary segments. The indirect way by which CncB acts on Dfd maxillary function has a precedent in Hox phenotypic suppression. The inhibitory effects of posterior Hox proteins on Antennapedia (Antp) function can apparently be exerted via an indirect pathway involving phosphorylation of Antp protein by casein kinase II (Jaffe et al., 1997). Dfd protein is also phosphorylated in vitro by CKII (A. V., unpublished observation), but whether this phosphorylation is relevant to the suppression by CncB is unknown.

Drosophila apparently has only one small maf gene. Ubiquitous expression of maf-S RNA suggests that Maf-S protein can potentially form heterodimers with all Cnc isoforms. The CncB/Maf-S heterodimer binds a consensus sequence (TGCTGAGTCAT) that is very similar to the binding site for NF-E2 and is the same as the sequence selected in vitro by TCF11(LCR-F1/Nfr1)/MafG heterodimers (Johnsen et al., 1998). Vertebrates possess at least three small Maf homologs: p18/MafK, MafF and MafG (Blank and Andrews, 1997; Motohashi et al., 1997), which are ubiquitously expressed and heterodimerize with vertebrate CNC-type proteins on DNA regulatory sites (Blank et al., 1997; Igarashi et al., 1994; Itoh et al., 1995; Kobayashi et al., 1999). Single mutations in the mafF or the mafK genes do not lead to obvious phenotypic defects (Kotkow and Orkin, 1996; Onodera et al., 1999). mafG mutant mice are also viable and fertile, but have defects in megakaryopoiesis and neurogenesis (Shavit et al., 1998). Recently, mafK/mafG double mutant mice were characterized and found to have erythroid defects, as well as more severe defects in neurogenesis than those observed in mafG mutants alone (Onodera et al., 2000). Hematopoietic defects in mammalian small maf mutants resemble the phenotypes observed in mice mutant for NF-E2 or Nrf1 proteins, which suggests that in mammals small Mafs and CNC family proteins can cooperatively activate a set of common targets (Chan et al., 1998; Farmer et al., 1997; Shivdasani and Orkin, 1995). Here, we have shown by dsRNA interference that Drosophila embryos with lower levels of Maf-S have a phenotype that is very similar to strong cncB mutations. This, in combination with the heterodimer binding, suggests that Maf-S protein functions in parallel to CncB as a required cofactor.

We have found an important difference between the binding of the vertebrate Maf proteins and Drosophila Maf-S. Mammalian small Maf proteins were shown to homodimerize on NF-E2-like and T-MARE sites and to function as transcriptional repressors (Igarashi et al., 1994; Johnsen et al., 1998; Nagai et al., 1998; Wild et al., 1999). In contrast, we have observed no detectable binding of Maf-S to Cnc/Maf binding sites (Fig. 4C) or to the T-MARE binding site (data not shown). The leucine zipper region in the Maf-S protein is quite divergent from that of its mammalian counterparts (Fig. 3A). This may prevent homodimerization and therefore the binding of Maf-S to DNA.

In addition to maf-S, a gene for a large Maf protein is present in the Drosophila genome (GadFly identifier CG10034) (Adams et al., 2000) that encodes a Drosophila homolog of mammalian Kreisler/MafB and NRL (reviewed in Blank and Andrews, 1997; Motohashi et al., 1997). We suggest that this gene should be called maf-L (for Maf-Large). Binding tests of the Drosophila Maf-L protein show that it does not heterodimerize with CncB on the CncB/Maf-S binding site, but is capable of binding this site as a homodimer (data not shown). While the function of the large Maf in Drosophila is unknown, its ability to bind the CncB/Maf-S sites presumably adds to the complexity of regulatory interactions on these sites.

In the precursors of ventral pharynx and labrum, endogenous CncB is required to activate transcription from the CncB/Maf-S binding sites in element 4CE, and ectopic expression of CncB using a heat-shock construct results in significant ectopic activation of 4CE, as well as another simple reporter element, 4C. This activation is likely to be direct, since it requires the DNA binding function of CncB and the presence of Cnc/Maf binding sites in the target elements, and is observed immediately after heat-shock induction. Moreover, we have demonstrated that in tissue culture cells the CncB-specific domain acts as a strong transactivation interface. These results support the view that the CncB/Maf-S heterodimer can act as a transcriptional activator in vivo, at least on simple sites in certain cells.

The complex, natural enhancers through which mammalian CNC family members act as activators include and require binding sites for a variety of cofactors, such as GATA-1, Ets-like factors, and Sp1, in addition to conserved NF-E2-like sites (see Magness et al., 2000; Deveaux et al., 1997; Stamatoyannopoulos et al., 1995 and references therein). Similarly, naturally evolved direct target elements for CncB/Maf-S heterodimers are likely to be quite complex. Hints of this are seen in the spatial responses of 4C and 4CE elements – neither element is activated in mandibular cells where CncB and Maf normally overlap, even when CncB is overexpressed. The 4C element is not activated in any head cells even under conditions where CncB levels are high. So CncB in the mandibular segment appears to have its activation function ablated when compared to cells in other body regions. It is possible that CncB is converted into a repressor on many targets in mandibular cells, although the present study provides no direct evidence for or against this idea.

CncB is crucial for the development of the pharynx in Drosophila. In the nematode C. elegans, the Cnc homolog Skn-1 has been shown to be a factor required for the establishment of pharyngeal and midgut fates (Blackwell et al., 1994; Bowerman et al., 1992). GATA-like proteins have been shown to be important for the function of NF-E2 response elements, and it is interesting to note that two GATA-related factors, Elt-2 and End-1, are required for midgut development in the nematode (reviewed in Labouesse and Mango, 1999). This evidence is consistent with the idea that nematodes and arthropods (and perhaps other ecdysozoans) conserve a similar genetic system involving Cnc-like and GATA-like factors to specify anterior midgut and pharyngeal fates. Other Drosophila proteins, such as Knot/Collier and Crocodile, are specifically expressed in, and required for, pharynx and anterior midgut development (Crozatier et al., 1999; Häcker et al., 1995; Merrill et al., 1989). knot is upstream of cnc in the pathway leading to formation of the pharynx (Seecoomar et al., 2000), while the regulatory relationship between cnc and crocodile is not known.

Cross-talk between the functions of CncB and Hox proteins such as Dfd may also be present in Tribolium, since Tribolium Cnc and Dfd homologs are expressed in patterns similar to those observed in Drosophila (Rogers and Kaufman, 1997). As yet there is no evidence suggesting the involvement of mammalian CNC homologs in Hox pathways. It has been shown that p45 NF-E2 and its Nrf paralogs map near the four Hox clusters in mammals, which suggests that a single ancestral CNC gene existed near the ancestral Hox cluster and was duplicated twice along with the clusters (Kobayashi et al., 1999). The lineage including Drosophila has apparently evolved a different means of diversifying the functions of Cnc proteins by evolving three isoforms generated by alternative promoters and splicing.

Numerous experiments have established a major function for mammalian CNC homologs in controlling the expression of genes involved in hematopoiesis. Recently, it has been shown that CNC homologs are also activated in response to xenobiotics, and play a role in activation of several phase II detoxifying enzymes (Alam et al., 1999; Itoh et al., 1997; Wild et al., 1999). It is possible that other Cnc isoforms, CncA or CncC, or even CncB, may be involved in these processes. CncC is ubiquitously expressed and is essential for life, since cnc^K6 mutants stop in development at some point after embryogenesis. It will be of interest to test whether this and other cnc alleles have any impairment in hematopoiesis or detoxification processes.

Target genes that are directly activated by CncB/Maf-S are still unknown, but their identification is crucial for advancing our understanding of the diverse functions of CncB. With the availability of a complete Drosophila genome sequence (Adams et al., 2000) that allows analysis of the structure of natural regulatory elements, and with the use of new techniques for gene expression profiling such as microarrays, it should be possible to identify candidate targets for CncB activation in the near future.

The authors would like to thank Genevieve Buser and Sarah Gobuty for their help with the generation and analysis of the new mutant alleles in the cnc locus, Alex Paredez for participation in making Cnc regulatory constructs, Y. Henry Sun for UAS-Hth flies, Kees Murre for vectors and EMSA expertise, and members of the McGinnis laboratory for comments on the manuscript. A. V. is an HHMI predoctoral fellow. This work was supported by a research grant NICHD 28315 to W. M.