The floral homeotic protein APETALA2 recognizes and acts through an AT-rich sequence element

The flower is an evolutionary innovation that contributes to the success of angiosperms. Dicotyledonous flowers are composed of four major types of organs: sepal, petal, stamen and carpel. Four major classes of homeotic genes, A, B, C and E, specify the four floral organ types in a combinatorial manner (reviewed by Krizek and Fletcher, 2005). A-class genes, APETALA1 and APETALA2 (AP1 and AP2), together with the E-class genes, confer sepal identity in the first whorl. Petal identity is determined by the activities of A-, B- and E-class genes in the second whorl. The C-class gene, AGAMOUS (AG), together with B- and E-class genes, specifies stamen identity in the third whorl. Carpel identity is conferred by C- and E-class activities in the fourth whorl. With the exception of AP2, all floral homeotic genes encode MADS domain-containing proteins for which DNA binding and dimerization/multimerization specificities have been extensively characterized (reviewed by Immink et al., 2010). By contrast, the activity of AP2 as a transcription factor in flower development is poorly understood. In addition to promoting sepal and petal identities, AP2 restricts AG expression to the inner two whorls (Bowman et al., 1991; Drews et al., 1991). Previous studies using a GUS reporter system have shown that the 3-kb AG second intron contains sequence elements required for its proper expression, including responsiveness to repression by AP2 (Bomblies et al., 1999; Deyholos and Sieburth, 2000).

AP2 is the founding member of a family of 144 genes that encode at least one AP2 DNA-binding domain in Arabidopsis; the biological functions of this family range from development to stress and defense responses (Jofuku et al., 1994; Weigel, 1995; Okamuro et al., 1997; Riechmann and Meyerowitz, 1998). This DNA-binding domain was thought to be plant specific but computational analysis has identified this DNA binding domain in species such as Tetrahymena (Wuitschick et al., 2004) and Plasmodium (Yuda et al., 2009). Plant AP2 domain-containing genes were categorized into five subfamilies (Sakuma et al., 2002). Members of the AP2-like subfamily generally contain two AP2 DNA-binding domains, AP2R1 and AP2R2 (Kim et al., 2006; Shigyo et al., 2006). In the second to fifth subfamilies, the ERF-like, DREB-like, RAV-like and others, members contain one AP2 DNA-binding domain and are involved in abiotic and biotic stress responses.

The DNA-binding properties of AP2 domain proteins from various subfamilies have been studied. Members of the ERF-like and DREB-like subfamily bind well-documented GC-rich motifs (Ohme-Takagi and Shinshi, 1995; Stockinger et al., 1997; Hao et al., 1998; Liu et al., 1998). The AP2 domain in a RAV-like family member binds a CAACA motif (Kagaya et al., 1999). ANT is the only protein in the AP2-like subfamily for which DNA-binding properties have been studied (Nole-Wilson and Krizek, 2000; Krizek, 2003). Despite the crucial role of AP2 in flower development and the diversity of its targets (Yant et al., 2010), the DNA binding specificity of AP2 has never been characterized.

To better understand the molecular functions of AP2, we sought to determine its binding consensus sequence in vitro and characterize the relevance of the sequence in vivo. Here, we report that AP2R2 specifically binds the TTTGTT or AACAAA motif in vitro. We show that these motifs within the 2nd intron of AG are important for restricting AG expression to the inner two whorls in vivo. In silico analysis of 2nd intron sequences from AG orthologs uncovers strong conservation of this element in the Brassicaceae family. Furthermore, we found that AP2 directly regulates AG in young flowers through these elements. Curiously, the AP2 full-length protein binds DNA with no apparent specificity in vitro, suggesting that other factors influence its DNA binding specificity in vivo. These findings establish a missing link in the mechanisms underlying flower development, shed light on the molecular function of AP2, and set the foundation for further appreciation of the molecular basis for the broad biological functions of AP2.

To express the AP2R1, AP2R2 and AP2R1R2 domains of AP2 in E. coli, the corresponding coding regions from the AP2 cDNA were amplified by PCR (supplementary material Table S1) and cloned into the pET21-A vector using BamHI and EcoRI sites (Novagen). To express the full-length AP2 protein, the entire coding region of AP2 was cloned in-frame to an N-terminal MBP and His tag using BamHI and EcoRI sites in the pmCSG7 XF0510 MBP-LIC vector (a gift from Dr Xiaofeng Cao, Institute of Genetics and Developmental Biology, Beijing, China).

For in vivo analysis of the AG 2nd intron, the region of the AG 2nd intron in the KB31 construct (Bomblies et al., 1999) was amplified and cloned into PCR2.1 (Invitrogen). Site-directed mutagenesis was performed (supplementary material Table S1) to introduce mutations into each of the two AP2-binding sites. The wild-type and mutant KB31 fragments were then cloned into pD991 (Tilly et al., 1998) using BamHI and HindIII sites. The 35S::AP2m3-GR construct was generated as described previously (Yant et al., 2010).

The pET21A-AP2R1, AP2R2 and AP2R1R2, and the MBP-AP2 full-length protein plasmids were transformed into E. coli BL21. Protein expression and purification were carried out as previously described (Smith et al., 2002; Husbands et al., 2007) and purified proteins were quantified against BSA.

Either 200 ng or 500 ng of doubly affinity-purified (with Ni²⁺ beads and T7 antibody) and desalted AP2R2, AP2R1 or AP2R1R2 was subjected to a SAAB assay as previously described (Smith et al., 2002). Briefly, the protein-bead mixture was divided into six tubes. In the first tube, a pool of random, double-stranded oligonucleotides (supplementary material Table S1) was added and incubated for 4 hours with the protein-bead mixture. The DNA bound by the protein-bead complex was eluted and PCR was performed to amplify the bound sequences. An aliquot of the PCR reaction was added to the second tube of the protein-bead mixture to allow protein-DNA binding to occur. This process of affinity binding and amplification was reiterated a total of six times. The PCR product from either cycle 5 and/or 6 was cloned via TA cloning and sequenced. The sequences were analyzed with the motif finding program MEME to identify consensus motifs.

EMSAs were performed as described (Husbands et al., 2007) with some modifications. For EMSAs shown in Figs 1, 2 and in supplementary material Fig. S4B, Figs S5, S7, probes were generated by annealing 100 pmol of sense and antisense oligonucleotides (supplementary material Table S1) and 1-2 pmol of probe was used in each reaction. For gel-shifts shown in Fig. 3B,C and supplementary material Fig. S8, the DNA fragment was amplified by PCR from wild-type or mutant versions of KB31, or from Col genomic DNA, and 0.1-0.2 pmol of labeled probe was used in each reaction. Probes were prepared as previously described (Broitman-Maduro et al., 2005).

Gel shift reactions were conducted at 4°C in 20% glycerol, 20 mM Tris (pH 8.0), 10 mM KCl, 1 mM DTT, 12.5 ng poly dI/C, 6.25 pmol of random, single-stranded oligonucleotides, Herring sperm DNA, BSA and the probe in the amount specified above.

All samples involving AP2R2 were loaded on an 8% gel, whereas as those involving the AP2 full-length protein were loaded on a 6% gel to resolve protein-DNA complexes. Gels were then dried and either exposed to X-ray films or imaged and quantified using the Typhoon PhosphorImager. For quantification analyses in Fig. 2 and supplementary material Fig. S5, the percentage bound was calculated by dividing the shifted amount after incubation with the mutated probes by the shifted amount after incubation with the Δα-probe.

In reactions with cold competitors, 5-40× unlabeled probes were included in the reactions. Furthermore, anti-His and anti-T7 antibodies were added to some reactions at 1-2× the amount of the AP2R2 protein to obtain super-shifts.

Inflorescences were stained for GUS activity and processed for sectioning as previously described (Sieburth and Meyerowitz, 1997). Slides were viewed under a Leica DMR compound microscope and images were taken with a Spot digital camera (Diagnostic Instruments).

The inflorescences of 35S::AP2m3-GR ap2-2 plants were treated once with a solution of 10 μM dexamethasone (DEX)/0.015% Silwet with or without 10 μM cyclohexamide (CHX) (Fisher). Six hours later, the treated inflorescences were dissected to remove stage 8 and older flowers. Total RNA was isolated from the dissected inflorescences and subjected to DNaseI treatment and reverse transcription. RT-PCR was performed on the cDNAs using primers specific for AG and UBQ5 (supplementary material Table S1). Real-time RT-PCR was performed on the same cDNAs using a real-time PCR SYBR Green system (BioRad). Three technical replicates were performed for each real-time RT-PCR. Three biological replicates of DEX induction and real-time RT-PCR were performed. Error bars represent the standard deviation from three technical replicates.

Chromatin immunoprecipitation (ChIP) experiments were performed on two biological replicates following previously described protocols (Gomez-Mena et al., 2005; Mathieu et al., 2009; Yant et al., 2010). The input and ChIP samples were subjected to real-time PCR (BioRad). Three technical replicates were performed. The data were analyzed as previously described (Wierzbicki et al., 2008).

All 2nd intron sequences from AG orthologs were downloaded from GenBank (supplementary material Table S2). The start and end positions, provided by Hong et al. (Hong et al., 2003) and those specified in the GenBank annotations files, were used to parse the introns from their source sequences and to bring them into their proper sense orientation (supplementary material Table S2). Sequence manipulations and analyses were performed with custom scripts that are based on the Biostrings package of the statistical programming environment R (Morgan et al., 2009; R Development Core Team, 2010). Multiple sequence alignments (MSAs) were computed with the diaglign2-2 software from Morgenstern (Morgenstern, 2004) using the default parameters in the DNA mode. A sliding window analysis was performed to visualize the degree of conservation in the final MSA. For this, the relative conservation of each base was calculated at each position where a value of 1.0 indicates perfect conservation of one base (disregarding gaps) and a value of 0 indicates equal representation of all four bases. For plotting purposes, these conservation values were smoothed by calculating their mean for a sliding window size of 20 nucleotides along all MSA positions. Pattern searches were performed with the matchPattern function of the Biostrings package (Morgan et al., 2009).

To begin uncovering the molecular mechanisms underlying the role of AP2 in development, we sought to identify its binding sequence by performing a SAAB assay with AP2R1, AP2R2 or AP2R1R2 doubly purified based on their N- and C-terminal tags.

When the SAAB assay was performed for the purified AP2R2 (supplementary material Fig. S1A), amplified DNA from the bound fraction could be detected starting from cycle 4 (supplementary material Fig. S1B). After cloning and sequencing of the AP2R2-bound DNA after cycle 6, 20 unique sequences were obtained, 17 of which contained two perfect AT-rich motifs – AACAAA or the complementary TTTGTT (supplementary material Fig. S1C) – whereas the other three clones had either one site (clone 11), a single nucleotide change at each site (clone 12) or a single nucleotide change at one of the sites (clone 7) (supplementary material Fig. S1C). Therefore, all recovered clones contained at least one copy of the consensus sequence with no more than one nucleotide change.

When the same procedure was applied to purified AP2R1 (supplementary material Fig. S2A), no bound DNA was detectable by PCR, indicating that AP2R1 did not bind DNA in vitro (supplementary material Fig. S2B). The lack of DNA recovered from the AP2R1 SAAB assay was not due to loss of the protein during the procedure because the AP2R1 protein was present on the beads throughout the experiment (supplementary material Fig. S2C). The SAAB assay was also performed for purified AP2R1R2 (supplementary material Fig. S3A). DNA bound to AP2R1R2 was detectable from cycles 3 to 6 (supplementary material Fig. S3B); however, sequencing of cloned DNA bound to AP2R1R2 from cycle 6 did not reveal any obvious consensus sequence (supplementary material Fig. S3C). Ten out of 25 unique sequences contained one of the sites bound by AP2R2 (supplementary material Fig. S3C; and data not shown), one had a site with one nucleotide change (clone 3), and some of the other sequences were GC rich.

Next, to confirm the SAAB assay results, we performed EMSAs with the AP2R2 protein. Sense and antisense strands corresponding to one of the sequences obtained from the SAAB assay containing both AACAAA (termed ‘α’ site) and TTTGTT (termed ‘β’ site) were used as a probe (Fig. 1A). DNA binding as revealed by a shift in mobility was observed with as little as 50 ng of AP2R2 (data not shown), and with 100 or 200 ng of AP2R2, which yielded more strong and consistent binding (Fig. 1B, lanes 1 and 2, arrow). To confirm the observed binding, 20-fold cold competitor was added to the binding reaction. Indeed, the intensity of the shift was greatly diminished (Fig. 1B, lane 3). As the experiments were conducted with AP2R2 purified from E. coli, there was a possibility that the observed shift may be due to a contaminating protein instead of AP2R2 (although AP2R2 was the only protein detected by Coomassie staining in the protein fraction; supplementary material Fig. S1A). If the observed binding was specifically caused by AP2R2, the addition of His and T7 antibodies would generate a supershift as AP2R2 had both a T7 and a His tag. We observed that the inclusion of the His and T7 antibodies resulted in super-shifted bands when compared with AP2R2 alone (Fig. 1B, compare lanes 4 and 5 to lane 2, stars), confirming that it was indeed AP2R2 itself that bound the probe. As a control, we also added 400 ng His and T7 antibodies to the reaction in the absence of AP2R2 and we did not observe any of the same band shifts as seen in the presence of AP2R2 (Fig. 1B, lane 6).

Next, we tested whether both sites were necessary for AP2R2 binding. We mutated the AACAAA site to AGGTGA and the TTTGTT site to TCCACT (Fig. 1A). The EMSA showed that AP2R2 was still able to bind probes that had one intact site (Fig. 1C, lanes 2 and 5, arrow). The shift was lost upon addition of the corresponding cold competitor (Fig. 1C, lanes 3 and 6). Loss of both sites, however, completely abolished binding (Fig. 1C, lane 8). These results demonstrate that AP2R2 binds AACAAA and/or TTTGTT in vitro.

ANT, the only protein characterized in terms of DNA-binding properties in the AP2-like subfamily of AP2 domain-containing proteins, has been shown to bind a loose and long consensus sequence, gCAC(A/G)N(A/T)TcCC(a/g)ANG(c/t) (Nole-Wilson and Krizek, 2000). We wondered how tight the AP2R2 consensus binding sequence might be. As AP2R2 was able to bind probes containing one site, probes containing just one site were used for testing the nucleotide specificity at each position. The nucleotide at each position was converted into one of the other three nucleotides, whereas the other five positions remained unchanged. All possible perturbations were made and EMSAs were performed with AP2R2. Our results indicated that the binding site was extremely tight as any mutation at any position greatly compromised binding by AP2R2 (Fig. 2, supplementary material Fig. S4). Notably, any change at the fourth position resulted in loss of binding (Fig. 2, supplementary material Fig. S4).

Proteins in the ERF-like and DREB-like subfamilies containing a single AP2 DNA-binding domain have been shown to bind specific GC-rich consensus sequences (Ohme-Takagi and Shinshi, 1995; Stockinger et al., 1997; Hao et al., 1998; Liu et al., 1998). Thus, we wanted to test whether AP2R2 could bind the conventional GCC-box (Hao et al., 1998), GCCGCC or other random sequences. We performed EMSAs with a GCC-box, two random probes (supplementary material Fig. S5A) and the Δα probe as a positive control. We observed binding of AP2R2 to the Δα probe (supplementary material Fig. S5B, lane 2) but not to the canonical GCC-box or two other random sequences (supplementary material Fig. S5B, lanes 5, 7 and 9).

Next, to determine whether the full-length AP2 protein could bind the consensus sequence TTTGTT, we cloned the AP2 full-length protein into a vector containing an N-terminal His and MBP tag (MBP-AP2), and purified and desalted it (supplementary material Fig. S6A). Owing to the large size of MBP-AP2, 400 ng was used to perform the EMSA. Using the same probes as in the AP2R2 EMSAs (Fig. 1A), we found that MBP-AP2 was able to bind probes containing one or both consensus sequences (supplementary material Fig. S7A, lanes 2 and 6, arrow). Interestingly, MBP-AP2 was also able to bind the probe with both sites mutated (supplementary material Fig. S7A, lane 9, arrow). To assess the binding specificity of the AP2 full-length protein further, we performed EMSAs with a probe containing the canonical GCC-box and two random probes (supplementary material Fig. S5A). Interestingly, MBP-AP2 could bind all of the probes in vitro (supplementary material Fig. S7B, lanes 2, 5 and 8, arrow). To address whether the MBP tag may be binding the probes, an equal amount of MBP was added to the reactions for the EMSA (supplementary material Fig. S6B). MBP alone did not bind any DNA sequences (supplementary material Fig. S7A, lanes 3, 7 and 10; S7B, lanes 3, 6 and 9).

Considering that AP2R2, but not AP2 full-length protein, specifically bound the consensus sequence, we sought to test whether the AP2R2-binding site had biological relevance. Previous studies have shown that AP2 represses AG expression (Drews et al., 1991), but it is still unknown whether AG is a direct target of AP2, although AP2 has been shown to bind the AG 2nd intron in vivo in our previous ChIP-seq analysis (Yant et al., 2010). Characterization of the AG 2nd intron has identified a 750 bp region that, when fused to the GUS reporter in a construct termed KB31, confers AP2 responsiveness to GUS (Bomblies et al., 1999; Deyholos and Sieburth, 2000). In silico analysis indicated that the KB31 region contained two AP2R2-binding sites at the 3′ end, which we termed A and B (Fig. 3A). Thus, we proceeded to test whether AP2R2 was able to bind this region of the AG second intron in vitro. Using primers encompassing this region (supplementary material Table S1), we generated a 167 bp probe (Fig. 3A) to perform an EMSA. Indeed, a shift was found with AP2R2 and the binding was stronger as we added increasing amounts of the protein (Fig. 3B, lanes 2-7, arrow). Furthermore the binding was lost upon the addition of 20× or 40× cold competitor (Fig. 3B, lanes 8 and 9, respectively) to the binding reaction. To test whether the observed binding required the two sites within the 167 bp sequence, we performed site-directed mutagenesis to mutate both sites (Fig. 3A). With the ΔAB probe, binding was diminished greatly (Fig. 3B), showing that AP2R2 binds the 167 bp region via the two elements in vitro. In addition, and consistent with prior results, gel shifts showed binding of the MBP-AP2 full-length protein to both the wild-type and the ΔAB probe (Fig. 3C, lanes 3-8, 12 and 13, arrow). This binding was abolished upon addition of 40× cold competitor (Fig. 3C, lanes 9 and 14) and MBP alone did not bind either probe (Fig. 3C, lanes 2 and 11).

To evaluate the importance of the AP2R2 binding sites in the AG 2nd intron in vivo, we used the KB31 GUS reporter, which had been shown to report faithfully the endogenous domains of AG expression and to respond to the regulation by AP2 (Bomblies et al., 1999; Deyholos and Sieburth, 2000). We cloned the 750 bp KB31 region from the AG 2nd intron containing either the wild-type or mutant (ΔAB) sites into a GUS expression vector with a minimal (–60) 35S promoter (Tilly et al., 1998). The constructs were introduced into rdr6-11 to prevent post-transcriptional gene silencing of the transgenes (Dalmay et al., 2000; Mourrain et al., 2000). For the wild-type construct, GUS staining of inflorescences from 99 independent T1 transgenic plants showed that 74 recapitulated the proper AG expression patterns (Fig. 4A,C,E). Twenty-two inflorescences did not show any GUS staining and three did not recapitulate the proper AG expression patterns. For the ΔAB construct, however, 71% of the 27 independent transformants showed expansion of the GUS expression domain to the outer two whorls (Fig. 4B,D,F) in all stages of flower development, with the remainder showing the correct expression patterns. Therefore, the A and B sites are important for the restriction of AG expression to the inner two floral whorls.

The fact that the AP2R2 binding sites in the 2nd intron of AG are important for the restriction of AG expression to the inner two whorls implies that AP2 is a direct regulator of AG. To address whether AP2 acts on AG directly, we used a rat glucocorticoid receptor (GR)-induction system that has been widely used in Arabidopsis as a method to establish direct relationships between a transcription factor and its targets (Sablowski and Meyerowitz, 1998; Wagner et al., 1999; Ito et al., 2004; William et al., 2004). Because AP2 is targeted by miR172 and transgenes containing miRNA target sites are readily silenced in vivo, we fused a miR172-resistant AP2 cDNA (AP2m3) (Chen, 2004) to GR. The 35S::AP2m3-GR construct was transformed into the progeny of ap2-2/+ plants. After obtaining single-locus insertion transformants of the ap2-2/+ genotype, single and continuous treatments of 10 μM DEX were performed to determine the functionality of the transgene. A single DEX treatment of inflorescences was not sufficient to induce the AP2m3 phenotype (data not shown) (Chen, 2004). Continuous treatments (once a day for 1 week), however, led to the induction of the AP2m3 phenotype, thus showing that the transgene was functional (Fig. 5A,B).

To determine whether AP2 directly represses AG expression, we subjected 35S::AP2m3-GR ap2-2 inflorescences to a single treatment of cyclohexamide (CHX) with or without DEX. After 6 hours, inflorescence tissue was micro-dissected to remove stage 8 and older flowers. RT-PCR was performed to measure AG mRNA levels. We found that upon induction of AP2m3-GR, AG mRNA levels decreased in young flowers (Fig. 5C). Real-time RT-PCR of three biological replicates revealed a 50% decrease in AG transcript levels upon AP2m3-GR induction (Fig. 5D). Therefore, AG is likely to be a direct target of AP2.

Next, we sought to determine whether AP2 regulates AG through the two AP2R2-binding sites. If AP2 acts through the two sites, we would expect KB31, but not KB31ΔAB, to be repressed by AP2. KB31 and KB31ΔAB transgenic lines harboring a single transgene locus were identified and crossed into the 35S::AP2m3-GR ap2-2 background. Homozygous KB31 35S::AP2m3-GR ap2-2 or KB31ΔAB 35S::AP2m3-GR ap2-2 inflorescences were treated with DMSO or DEX for 6 hours and GUS expression was determined by real-time RT-PCR. DEX induction caused a decrease in GUS mRNA levels in KB31 but not in KB31ΔAB (Fig. 5E,F). Therefore, AP2 represses AG through the two AP2R2-binding sites in vivo.

To test whether AP2 is associated with the AG 2nd intron in vivo, we performed ChIP assays using anti-AP2 antibodies (Mlotshwa et al., 2006). The antibodies were directed against a C-terminal region of AP2 that is predicted to be absent in the ap2-2 mutant. From a ChIP-seq experiment conducted with these antibodies on whole inflorescences, genome-wide AP2-binding sites were uncovered (Yant et al., 2010). The ChIP-seq effort identified a region in the 5′ end of the KB31 fragment that was bound by AP2, which we named region II (Fig. 6A). This region did not overlap with the region containing our binding sites, AB (Fig. 6A). To specifically test whether AP2 binds the AB region, especially in young flowers, we performed ChIP experiments with dissected inflorescences containing stages 7 and younger flowers and used region II as the positive control. We were able to find enrichment of AP2 within the AB region as well as region II in two biological replicates (Fig. 6; data not shown). As our ChIP assays showed that AP2 bound both sites but the genome-wide study did not show enrichment in our AB site (Yant et al., 2010), we further dissected the II region to test which part of that region may be bound by AP2. We divided the region into four parts (1-4; supplementary material Fig. S8A and Table S1), including a region directly upstream and downstream. We found that AP2R2 bound part 3 of region II (supplementary material Fig. S8B) and upon further inspection, the 5′ end of part 3 contained an AP2R2 consensus sequence with one nucleotide change (TTTGTT to TTTGTG). Consistent with our site mutagenesis analysis (Fig 2; supplementary material Fig. S4), AP2R2 is still able to bind the TTTGTG sequence in a medium manner. Binding was not visibly observed with this probe (p3) until 400 ng of protein was added (supplementary material Fig. S8, lane 5) when compared with 200 ng of the wild-type probe (supplementary material Fig. S8, lane 2). Consistent with all of our previous results, MBP-AP2 was able to bind all four probes (supplementary material Fig. S8C).

The repression of AG expression by AP2 is specific to Arabidopsis as the AP2 orthologs in Antirhinnum, LIPLESS 1 and 2 (LIP1 and LIP2), do not negatively regulate C-class function (Keck et al., 2003). However, it is possible that AP2-mediated restriction of AG expression is conserved in species closer to Arabidopsis. Thus, we sought to determine whether the AP2R2-binding consensus sequence was conserved in Brassicaceae. Using AG 2nd intron sequences from 29 Brassicaceae species (Hong et al., 2003), we performed multiple sequence alignment as well as a sliding window conservation analysis to identify regions in the 2nd intron that are conserved both in sequence and in position. Although the TTTGTT (or AACAA) motif was present multiple times in each of the introns (supplementary material Fig. S10), the A-site (Fig. 3A) was embedded within a short region that was conserved throughout Brassicaceae both in sequence and in position within the introns, as revealed by the sliding window analysis (supplementary material Figs S9, S10). The B-site was not found at invariant positions among the introns (supplementary material Fig. S10). At the A-site, 28 of the 29 Brassicaceae species showed a perfect match to the TTTGTT pattern. The apparent exception appeared to be Thysanocarpus (AY253255) with a single nucleotide change in the motif. When the analyses included the 2nd intron of AG homologs from Antirrhinum majus (AY935269), Lycopersicon esculentum (AY254705) or Cucumis sativus (AY254702 and AY254704) belonging to Veronicaceae, Solanaceae or Cucurbitaceae, respectively, the divergence in these sequences was too high to compute reliable multiple sequence alignments of the introns, thus precluding any conclusions on the conservation of this motif outside of Brassicaceae.

Genes encoding one or more AP2 DNA-binding domains are categorized under five subfamilies: DREB-like, ERF-like, RAV-like, AP2-like and others (Sakuma et al., 2002). The AP2-like subfamily, which can be further divided into two lineages, ANT and euAP2, is the only subfamily that contains two AP2 domains. Single AP2 domain containing proteins of the other subfamilies bind to highly specific, mostly GC-rich sequence motifs (Ohme-Takagi and Shinshi, 1995; Stockinger et al., 1997; Hao et al., 1998; Liu et al., 1998). Only the target sequence of a single member of the AP2-like subfamily, ANT, has been reported. ANT binds a long and loose consensus sequence that is also GC rich (Nole-Wilson and Krizek, 2000). In contrast to conventional GC-rich target sequences of these characterized AP2-domain proteins, AP2R2 is highly specific for the AT-rich consensus sequences TTTGTT or AACAAA. The AP2 domain in RAV1 also binds a non-GC rich sequence CAACA (Kagaya et al., 1999). Two AP2 domain proteins from Plasmodium were found to bind the consensus sequences TGCATGCA and GTGCAC, which are different from the target sequences of all plant AP2 domain proteins characterized to date (De Silva et al., 2008). Collectively, these studies show that AP2 domains have wide ranging target specificities. Consistent with this, the AP2 domain of AtERF1 and the AP2 domains of ANT appear to use largely non-conserved amino acids for DNA binding (Allen et al., 1998; Krizek, 2003) (supplementary material Fig. S11). The fact that AP2R1 does not appreciably bind any DNA sequences in vitro raises the possibility that some AP2 domains function in processes other than DNA binding.

It is useful to compare and contrast the DNA-binding specificities of ANT and AP2 as representatives of the two lineages within the AP2-like subfamily. In vitro selection of DNA sequences bound by ANT-AP2R1R2 led to the identification of a long consensus sequence (Nole-Wilson and Krizek, 2000; Krizek, 2003). In our study, we found that AP2R1R2 bound DNA in vitro, but no consensus sequence could be identified. We note that both ANT-AP2R1R2 and AP2R1R2 have poor DNA-binding specificities (as exemplified by the loose ANT consensus sequence and the lack of obvious consensus motifs from AP2R1R2-bound sequences). We also note that ANT-AP2R1R2 and AP2R1R2 have differences in their binding preferences. Although ANT-AP2R1R2 binds GC-rich sequences, AP2R1R2 probably prefers the TTTGTT or AACAAA motif as this motif was in 10 out of 25 clones from the SAAB assay. Moreover, not all clones from the SAAB assay were GC rich. Consistently as well, this could be explained by the addition of 10 amino acids in ANTR1 versus AP2R1 and a single amino acid addition in ANTR2 versus AP2R2 (supplementary material Fig. S11).

AP2 full-length protein was able to bind all probes that were tested in vitro (Fig. 3; supplementary material Figs S7, S8). This led us to question the specificity of AP2 DNA binding in vivo, especially in relation to its biological functions. AP2 has diverse biological functions such as seed development (Jofuku et al., 2005; Ohto et al., 2009), shoot apical meristem maintenance (Würschum et al., 2006), control of floral timing (Yant et al., 2010), preventing replum overgrowth during fruit development (Ripoll et al., 2011), establishment of floral meristem identity (Schultz and Haughn, 1993; Shannon and Meeks-Wagner, 1993), floral organ specification (Bowman et al., 1989; Kunst et al., 1989) and the regulation of homeotic gene expression (Drews et al., 1991). Perhaps the lack of strong inherent DNA-binding specificities underlies the diverse biological roles of AP2. Whole-genome ChIP-seq experiments identified more than 2000 sites that are bound by AP2 in vivo (Yant et al., 2010), highlighting the potential of AP2 in influencing the expression of a large number of genes. However, computational analyses failed to uncover a consensus sequence that is enriched in regions bound by AP2 in vivo (data not shown). In addition, owing to its AT-rich nature, we were not able to state that the AP2R2-binding site was statistically significant among the targets. However, it is interesting to note that in the sum sequence space of the 2275 bound sites there are 445 instances of AACAAA and 473 of TTTGTT (Yant et al., 2010). The lack of ability to find an AP2 consensus sequence could be reflective of its diverse roles in development. In addition, the discrepancy in the two sites (II and AB) that we found in the two ChIP experiments (Yant et al., 2010) (this study) could be due to tissue differences.

Despite the large number of in vivo binding sites, AP2 is still selective in its DNA binding in vivo, in contrast to its largely non-specific DNA binding in vitro. One potential mechanism underlying the in vivo specificity is that it might be conferred by other DNA-binding proteins that interact with AP2. In this scenario, the largely non-specific DNA binding by AP2 enhances the binding of other transcription factors at specific sites. The promiscuous binding of MBP-AP2 to all DNA sequences in vitro lends itself to this hypothesis as it could be feasible that AP2 full-length itself, as a regulator of diverse functions, would have specific binding abilities, depending on its protein-binding partners, that may modulate its activity in vivo. Another potential mechanism is that other factors interact with AP2R1 to allow AP2R2 to specifically interact with DNA. We prefer this explanation as the AP2R2-binding sites in the AG 2nd intron are indeed important for the function of AP2 in vivo. Moreover, the AP2R2 consensus sequence was recovered in 10 out of 25 clones in the AP2-R1R2 SAAB assay, suggesting that there is some inherent affinity of the AP2 domains for the TTTGTT or AACAAA consensus sequence. Both scenarios may occur in vivo, in which case the AP2R2 consensus sequence would only be present at some of the in vivo AP2-binding sites.

AP2 has long been known to be essential in establishing the inner two whorl-specific pattern of AG expression (Drews et al., 1991). In ap2 loss-of-function mutants, AG expression expands into the outer two whorls. Using the GUS reporter system, elements responsive to AP2 regulation have been mapped to at least two regions in the AG 2nd intron (Bomblies et al., 1999; Deyholos and Sieburth, 2000). However, it was not known whether AP2 regulates AG expression directly. We found that a 750 bp AP2-responsive region contains two AP2R2 consensus sequences. Site-directed mutagenesis experiments indicated that the two sites were important for the negative regulation of AG by AP2. In addition, we found that this negative regulation was direct through an inducible system (AP2m3-GR) as well as ChIP experiments. Therefore, AP2 is a direct, negative regulator of AG.

Our data also suggest that AP2 represses AG most effectively during early stages of flower development. Initially, when AP2m3-GR whole inflorescences (composed of both young and old flowers) were used in the induction experiments, no obvious changes in AG mRNA levels were seen. However, upon micro-dissection of the inflorescences after induction to retain only flowers of stages 7 and younger, we observed a 50% decrease in AG mRNA levels upon AP2 induction. It is feasible that AP2 only negatively regulates AG during early stages of flower development as it has been shown that a myriad of other genes, such as CURLY LEAF (CLF), LEUNIG, SEUSS and RABBIT EARS also negatively regulate AG (Goodrich et al., 1997; Liu et al., 1998; Franks et al., 2002; Krizek et al., 2006). It is possible that in the outer two whorls, AP2 establishes the initial repression of AG, and other mechanisms, such as CLF-mediated histone modifications, are responsible for the maintenance of the repressed state throughout flower development.

In addition, it has been shown that AP2 acts through different regions of the AG 2nd intron at different developmental timepoints (Bomblies et al., 1999; Deyholos and Sieburth, 2000). Thus, it is feasible that region II and the AB site may both be important at different time-points. In the Bomblies study, KB31 is sufficient to confer AG expression and contain elements by which AP2 negatively regulates AG, and this construct has both region II and the AB site. Constructs that do not have region II or the AB site did not provide much information because, although they did not respond to the loss of AP2, there was no basal AG expression (KB28 encompasses p3 and the A site only). In this study, we show that the AB site is functional, as loss of AB resulted in expansion of GUS expression (Fig. 4). Interestingly, AP2R2 was able to bind to both region II and the AB site, albeit at a higher affinity with the AB site.

The euAP2 lineage predates the divergence of gymnosperms and angiosperms, but the biological functions of AP2 and its orthologs differ amongst flowering plants (reviewed in Litt, 2007). In Arabidopsis, AP2 specifies perianth identities and restricts C-function to the inner two whorls. However, characterized AP2 orthologs from Antirrhinum and petunia do not appear to share the role of AP2 in flower development (Maes et al., 2001; Keck et al., 2003). For example, LIP1 and LIP2, AP2 orthologs in Antirrhinum, promote sepal identities but do not control petal identity or restrict the expression of PLENA (C-class gene) (Keck et al., 2003). In fact, mutations with ectopic C function in the outer whorls in Antirrhinum and petunia map to a microRNA, miR169 (Cartolano et al., 2007). Interestingly, the petunia ortholog of LIP/AP2, PhAP2A, was able to rescue the ap2-1 mutant when expressed in Arabidopsis (Maes et al., 2001). The ability of the petunia AP2 protein to regulate AG in the Arabidopsis context suggests that the DNA-binding properties of the petunia AP2 are similar to those of Arabidopsis AP2 and implies that divergence in C-class regulatory sequences or in AP2-interacting proteins may be responsible for the divergence in the ability of AP2 to regulate C-class genes.

In this study, we show that AP2R2 recognizes an AT-rich motif in vitro and that two such motifs within the AG 2nd intron mediate the regulation of AG by AP2 in vivo. Given the AT richness of introns, this sequence motif is present multiple times in the introns of AG and AG orthologs from other species. The positions of the motifs relative to other transcription factor binding sites may influence the ability of AP2 to act upon them. We show that the A site recognized by AP2R2 in the AG 2nd intron is conserved both in sequence and in position in Brassicaceae. This implies that AP2-mediated regulation of C-class gene expression is conserved in this family. Although the motif is present in the 2nd introns of AG orthologs from non-Brassicaceae species, the overall large divergence in 2nd intron sequence precluded confident alignments to determine whether the positions of the motifs are conserved.

We are grateful to Drs Aman Husbands, Harley Smith and Gina Maduro for advice on SAAB assays and EMSAs. We thank Xiaofeng Cao from the Institute of Genetics and Developmental Biology, Chinese Academy of Sciences for the pmCSG7 XF0510 MBP-LIC vector and Patricia Springer for sharing plasmids and equipment. We thank Lijuan Ji, Shengben Li, Rae Eden Yumul and Yuanyuan Zhao for comments on the manuscript.