LATE MERISTEM IDENTITY2 acts together with LEAFY to activate APETALA1

Flowering plants transition through a number of distinct developmental phases in their lifecycle. During each phase, different organs are generated from a group of cells at the flanks of the shoot apical meristem, which form the primordia (Steeves and Sussex, 1989). Developmental phase transitions have been studied extensively in the plant model system Arabidopsis thaliana (Araki, 2001; Blazquez et al., 2006; Poethig, 2003; Steeves and Sussex, 1989). During the vegetative phase, the primordia give rise to a series of leaves, which form the basal rosette. During the early reproductive phase, the shoot apical meristem grows upward (bolts) and the newly formed primordia develop into secondary inflorescence branches subtended by cauline leaves. Finally, after the meristem identity transition, the first flowers are formed.

The precise timing of flower formation is crucial for reproductive fitness, as plants must ensure that the energy and resources accumulated during the vegetative phase are optimally allocated to the production of offspring (Roux et al., 2006). Plants rely on both environmental and endogenous cues to fine-tune the onset of reproductive development (Araki, 2001; Koornneef et al., 1998; Simpson et al., 1999). These signals modulate the level and activity of flowering-time regulators, which initiate the reproductive phase and induce expression of the meristem identity genes (Amasino, 2010; Baurle and Dean, 2006; Kobayashi and Weigel, 2007; Komeda, 2004; Turck et al., 2008). The meristem identity regulators then trigger formation of the first flower (Blazquez et al., 2006; Liu et al., 2009a; Parcy, 2005).

Two key meristem identity regulators in Arabidopsis are the plant-specific transcription factor LEAFY (LFY) and the MADS box transcription factor APETALA1 (AP1). LFY is considered to be a central meristem identity regulator because lfy null mutants cause a very dramatic delay in the meristem identity transition (Huala and Sussex, 1992; Weigel et al., 1992). Moreover, LFY upregulation in the initiating primordia flanking the shoot apical meristem is one of the first steps in the regulatory cascade that leads to the meristem identity transition (Blazquez et al., 1997; Hempel et al., 1997). LFY executes its meristem identity role in part by activating AP1 expression directly (Parcy et al., 1998; Wagner et al., 1999; William et al., 2004). AP1 upregulation marks commitment to flower formation (Blazquez et al., 1997; Bowman et al., 1993; Hempel et al., 1997; Liu et al., 2007; Mandel and Yanofsky, 1995; Yu et al., 2004). AP1 promotes floral fate by upregulating floral identity pathways and by repressing inflorescence identity pathways (Ferrandiz et al., 2000; Kaufmann et al., 2010; Liljegren et al., 1999; Liu et al., 2007; Yu et al., 2004). Two LFY-independent pathways can also upregulate AP1: one involves the photoperiod flowering-time regulators, FLOWERING LOCUS T (FT) and FD, and the other involves components of the age-sensing flowering-time pathway, the SBP transcription factors (Abe et al., 2005; Wang et al., 2009; Wigge et al., 2005; Yamaguchi et al., 2009). In agreement with this, simultaneous loss-of-function mutations in both LFY and AP1 results in plants that essentially lack flowers (Bowman et al., 1993; Huala and Sussex, 1992; Schultz and Haughn, 1993; Weigel et al., 1992).

Although the meristem identity transition is a key developmental switch, our understanding of the events that lead from LFY upregulation to flower formation is still incomplete. Previously, we used a genomic approach to define direct targets of LFY during the meristem identity transition (William et al., 2004). This approach identified the meristem identity regulators and direct LFY targets CAULIFLOWER (CAL), a close AP1 homolog, and LATE MERISTEM IDENTITY1 (LMI1), a class I HD-Zip transcription factor (Saddic et al., 2006; William et al., 2004). Another direct LFY target identified was AtMYB17 (William et al., 2004). AtMYB17 is a member of the R2R3 class of MYB transcription factors, which have important roles in many processes in plants, including cell fate specification, metabolism, and biotic and abiotic stress responses (Dubos et al., 2010; Kranz et al., 1998; Martin and Paz-Ares, 1997; Stracke et al., 2001). The Arabidopsis homologs of AtMYB17, AtMYB16 (MIXTA) and AtMYB106 (NOECK), have been reported to function in the determination of cell shape in the petal epidermis and in the repression of trichome branching (Baumann et al., 2007; Jakoby et al., 2008). The biological function of AtMYB17 is not understood. Here, we show a role for AtMYB17 in the meristem identity transition upstream of AP1; based on these findings, we renamed this gene LATE MERISTEM IDENTITY2 (LMI2).

T-DNA insertion lines were obtained from the SALK collection (Alonso and Stepanova, 2003) and twice backcrossed to Columbia (wild type). lfy and ap1 alleles used were described previously (Saddic et al., 2006; Yamaguchi et al., 2009). lfy-2 and lfy-10 carry the same lesion (Schultz and Haughn, 1993; Weigel et al., 1992) and were used interchangeably. For all genotyping primers, see Table S2 in the supplementary material.

All plant growth was in inductive photoperiod. Seeds were stratified for seven days at 4°C and either grown in white fluorescent lights at 22°C in soil in long-day conditions (16 hours light, 8 hours dark; 110 μmol/m²s) for experiments involving phenotyping and inflorescences, or on plates (0.5× MS media) in long-day conditions for three days followed by growth in continuous light (90 μmol/m²s) for seedling experiments.

For genomic rescue, the LMI2 locus including 2150 bp upstream of the translational start site was PCR amplified, sequenced and Gateway cloned into pGWB1 (Nakagawa et al., 2007). The resulting construct was transformed into lmi2-2 lfy-10 plants. A representative pLMI2:LMI2 lmi2-2 lfy-10 transgenic line was characterized further. For all cloning primers see Table S3 in the supplementary material.

Developmental age was determined based on number of days of growth and adjusted by developmental stage (emergence and size of true leaves) (Saddic et al., 2006). RNA was extracted from entire seedlings except for the study of LMI2 mis-expression in lmi2-1 mutants. RNA purification, reverse transcription and qRT-PCR were described previously (Yamaguchi et al., 2009). All real-time RT-PCR experiments were normalized over the ubiquitously expressed EIF4A gene (AT3G13920). The mean and s.e.m. were calculated for each biological replicate using three technical replicates. One representative experiment is shown. See Table S4 in the supplementary material for qRT-PCR primers used.

Upstream and downstream intergenic regions (2150 bp upstream of the translation start site and 2699 bp downstream of the translation termination site) were PCR amplified, sequenced and cloned into pBI101 (Clontech, Mountain View, CA, USA). Wild-type plants (Col) were transformed and a representative transgenic line was characterized. To investigate the role of LFY on LMI2:GUS expression, LMI2:GUS was crossed to lfy-9, 35S:LFY-GR in Ler (Wagner et al., 1999), and Ler (wild type). GUS assays were performed as described by Saddic et al. and Yamaguchi et al. (Saddic et al., 2006; Yamaguchi et al., 2005) using seven-day-old seedlings or 1-2 cm bolted primary inflorescences. For transient induction assays, seven-day-old F1 seedlings (LMI2:GUS × Ler or LMI2:GUS × 35S:LFY-GR) were incubated overnight with 10 μM dexamethasone at room temperature as previously described (Wagner et al., 1999) prior to GUS staining. Whole-mount samples and histological sections were visualized using an Olympus SZX12 dissecting or an Olympus BX51 compound microscope.

The LMI2:GUS reporter showed ectopic expression in the L1 layer of stems, petioles and leaves not detected by LMI2 in situ hybridization analyses. This might be due to missing cis regulatory elements located in LMI2 introns (Liu et al., 2007; Oh et al., 2009; Sieburth and Meyerowitz, 1997).

For the LMI2 antisense and sense probes, the genic region downstream of the MYB DNA binding domain was used. The AP1 in situ probe contained the genic region downstream of the MADS box. The constructs were PCR amplified, cloned into pGEM T-easy (LMI2) and pGEM-T (AP1; Promega, Madison, WI, USA), and sequenced. Sense and antisense LMI2 probes were digested with SalI and transcribed with the T7 polymerase, whereas the antisense AP1 probe was transcribed using the T7 polymerase following digestion with EcoRI. The Riboprobe Combination System (Promega) and DIG RNA labeling mix (Roche, Branchburg, NJ, USA) were used for probe synthesis. In situ hybridization was performed as described by Long and Barton (Long and Barton, 1998).

The pLMI2:LMI2 rescue construct excluding the translation termination codon was Gateway cloned into pGWB13 (Nakagawa et al., 2007). pLMI2:LMI2-HA was transformed into lmi2-2 plants followed by testing for rescue. For ChIP, 300 mg tissue from eleven-day-old seedlings of a representative line were used with 3 μg/sample or 4 μg/sample of anti-HA antibody [sc-805 (Santa Cruz, Santa Cruz, CA, USA) or 12CA5 (Roche), respectively] using published procedures (Kwon et al., 2005; William et al., 2004). LMI2 occupancy on genomic DNA was calculated by computing the enrichment over the respective input and normalized over lmi2-2. The mean and s.e.m. were calculated using at least three technical replicates; one representative biological replicate is shown. For ChIP-qPCR primers see Table S5 in the supplementary material.

The LFY coding region was amplified and inserted between the EcoRI and NotI sites into pGEX-5X-1 (GE Healthcare, Piscataway, NJ, USA). The fusion protein was expressed in Escherichia coli (AD494). After induction with 0.1 mM IPTG at 37°C for one hour, cells were harvested by centrifugation and resuspended in ice-cold PBS containing 1 mM EDTA, 1 mM PMSF, 1 mg/ml lysozyme and 1% Triton X-100. Following a 20 minute incubation at room temperature, the cell lysate was cleared by centrifugation. Protein extracts were incubated with Sepharose 4B slurry (GE Healthcare) at 4°C for one hour. The beads were washed five times with PBS containing 1 mM EDTA and 1 mM PMSF. The protein-bound beads were used directly for pull-down assays. In vitro transcription and translation of LFY, LMI2 and NC^a (1-464 amino acid fragment of the chromatin remodeling ATPase SYD) (Wagner and Meyerowitz, 2002) and the pull-down assay were performed as previously described (Sang et al., 2005).

LMI2N consisted of the N-terminal protein coding region of LMI2, including the MYB domain and the subgroup 9 motif, whereas LMI2C contained the remainder protein coding region of LMI2. The LMI2 fragments were amplified and inserted between the SalI and NotI sites of pDBLeu (Invitrogen, Carlsbad, CA, USA). The coding region of LFY was amplified and Gateway cloned into pDEST22 (Invitrogen).

pDBLeu-LMI2N or LMI2C bait constructs were co-transformed into yeast (PJ69-4A) with either pDEST22-LFY or pDEST alone. After transformation, cells were plated on –Trp –Leu/SD media. Double transformants were grown in –Trp –Leu/SD liquid media overnight, adjusted for equal cell density, serially diluted (10^–1-10^–4) and spotted on –Trp –Leu –His/SD plates.

LMI2N and LMI2C fragments were inserted into pENTR3C (Invitrogen) and Gateway cloned into pCL113 (pBATL). The coding region of LFY was cloned into pCL112 (pBATL) to create the nYFP. p35S::2xmCherry was cloned into pEarley102 (Earley et al., 2006). The control protein (NC^b: TDY1-NLS in pCL113) was previously described (Ma et al., 2009). Constructs were transformed into onion epidermal cells using the PDS-1000/He Biolistic Particle Delivery System (BioRad, Hercules, CA, USA) as described by Ma et al. (Ma et al., 2009). Protein interactions were observed using an Olympus MVX10 fluorescent microscope.

To elucidate the role of LMI2 in the meristem identity transition, we analyzed three T-DNA insertion alleles (Alonso and Stepanova, 2003): lmi2-1, lmi2-2 and lmi2-3 (Fig. 1A). In lmi2-1, the T-DNA insertion was located in the promoter region (116 bp from the transcription start site), whereas the insertions in lmi2-2 and lmi2-3 were located in the conserved MYB DNA binding domain (Fig. 1A). All three T-DNA insertions caused deletions in the LMI2 locus ranging in size from 4 to 41 bp (Fig. 1A).

Both lmi2-2 and lmi2-3 expressed RNA upstream of the T-DNA insertion, suggesting that they are not RNA-null alleles (Fig. 1B). However, we did not detect LMI2 expression in either the lmi2-2 or the lmi2-3 mutant using primers flanking the T-DNA insertions (Fig. 1B). Hence, these insertions probably give rise to a truncated non-functional LMI2 protein lacking part of the conserved DNA binding domain. The lmi2-1 mutant, however, expressed elevated levels of LMI2 RNA (Fig. 1B). As the T-DNA insertion in lmi2-1 is located in the promoter region, it is likely that this insertion generates a full length LMI2 transcript. Nonetheless, our combined data (see below) suggests that lmi2-1 is a loss-of-function allele. Because lmi2-2 and lmi2-3 have similar T-DNA insertion sites, we chose to focus on the lmi2-1 and lmi2-2 alleles.

We assessed the timing of the meristem identity transition in lmi2 mutants compared with wild type by counting the number of secondary inflorescences and cauline leaves formed prior to the formation of the first flower (Saddic et al., 2006; Yamaguchi et al., 2009). Flowering time was measured by counting the number of rosette leaves (see Table S1 in the supplementary material) (Yamaguchi et al., 2009). lmi2-2 displayed a statistically significant increase in the number of cauline leaves and secondary inflorescences formed compared with wild type in five independent experiments (Table 1; Fig. 1C), suggesting that LMI2 plays a non-redundant role in the meristem identity transition. lmi2-1 exhibited a more subtle delay in the meristem identity transition that differed significantly from wild type in only some of the experiments performed (Table 1; Fig. 1C).

All three lmi2 alleles significantly enhanced the meristem identity phenotype of the weak lfy-10 mutant in at least six independent experiments (Table 1; Fig. 1D,E). lmi2-2 lfy-10 double mutants showed the strongest meristem identity delay, essentially phenocopying the lfy-1 null mutant (Fig. 1E). In addition, in the lfy-10 background, lmi2-2, and to a lesser extent lmi2-1, caused a delay in the meristem identity transition in heterozygotes (see Fig. S1 in the supplementary material). Hence, LMI2 is a dosage-sensitive gene, at least under conditions when LFY activity is impaired. LFY itself is also dosage dependent (Blazquez et al., 1997; Okamuro et al., 1996), highlighting the sensitivity of this pathway to the level of both regulators. Finally, lmi2-2 and lmi2-1 displayed a subtle delay in flowering time (see Table S1 in the supplementary material) both as single mutants and in the lfy-10 genetic background.

We next tested whether the mutations in LMI2 caused the delay in the meristem identity transition by performing phenotypic rescue. Transformation of lmi2-2 lfy-10 with a genomic copy of LMI2 (pLMI2:LMI2) restored LMI2 expression to a level similar to that observed in lfy-10 (Fig. 1F). In addition, pLMI2:LMI2 fully rescued the enhanced meristem identity defects of lmi2-2 lfy-10 relative to lfy-10 (Fig. 1G).

To test whether LMI2 has additional LFY-independent roles during the meristem identity transition, we crossed the lmi2-2 allele to the lfy-1 null mutant. lmi2-2 significantly enhanced the meristem identity transition defect of lfy-1 (Table 1) indicating that LMI2 acts both downstream of and in parallel to LFY in this pathway. This is similar to AP1, which also acts downstream of and in parallel to LFY (Bowman et al., 1993).

We first examined the expression of LMI2 during the meristem identity transition using a bacterial β-glucuronidase (GUS) transcriptional reporter. In nine-day-old wild-type seedlings, LMI2:GUS was expressed in the center of the rosette close to the shoot apex (Fig. 2A) in a pattern roughly similar to that of pLFY:GUS (Fig. 2E). In the inflorescence, the LMI2:GUS reporter was expressed in the meristem proper and in young floral primordia, as well as in the carpels of older flowers (Fig. 2B-D; see Fig. S2A in the supplementary material). By contrast, as previously reported (Blazquez et al., 1997), pLFY:GUS expression was absent from the meristem proper but was observed in young floral primordia as well as in older flower primordia (Fig. 2F,G; see Fig. S2B in the supplementary material). In addition, both LMI2:GUS and pLFY:GUS were strongly expressed in secondary inflorescences (Fig. 2D,H). Thus, LMI2:GUS and pLFY:GUS have overlapping, but not identical, expression patterns during reproductive development.

LMI2:GUS expression was reduced in the shoot apex of intermediate lfy-9 mutants compared with wild-type seedlings (see Fig. S2C,D in the supplementary material). Conversely, steroid treatment of an inducible version of LFY, LFY-GR (William et al., 2004), resulted in elevated LMI2:GUS expression in seedlings; this was not observed in steroid treated wild-type seedlings expressing LMI2:GUS (see Fig. S2E,F in the supplementary material). Therefore, LFY acts on LMI2 cis regulatory elements present in this reporter construct, consistent with in vivo LFY binding to this locus (Winter et al., 2011).

We next examined endogenous LMI2 expression by in situ hybridization. LMI2 was expressed throughout the shoot apical meristem of primary inflorescences, with the highest expression observed in the young flower primordia (Fig. 2I). LMI2 expression was reduced, but not absent, in the young flower primordia of lfy-1 null mutant apices (Fig. 2J). No signal was observed using a sense probe (Fig. 2K). The residual LMI2 expression in lfy-1 is consistent with our genetic data that revealed an LFY-independent role for LMI2 in addition to its function downstream of LFY.

Subsequent to the meristem identity transition, LMI2 was expressed in stage 2 to stage 4 flowers (Fig. 2L; data not shown) (Smyth et al., 1990) and in the developing stamens and carpels of older flowers from stage 6 to stage 8 (Fig. 2M,N). Eventually, in stage 8 flowers, LMI2 expression decreased in the developing stamens but persisted in the carpels (Fig. 4N).

lmi2-1 displayed elevated LMI2 expression in seedlings based on semi-quantitative RT-PCR (Fig. 1B), yet behaved as a loss-of-function allele (Table 1; Fig. 1C; see Fig. S1 in the supplementary material). Moreover, the defect in lmi2-1 lfy-10 was rescued by pLMI2:LMI2 (Fig. 3A). In contrast to the wild type, LMI2 expression was undetectable in lmi2-1 shoot apices and young flower primordia (Fig. 3B,C), similar to the sense control (Fig. 3D). Thus, in lmi2-1 mutants, LMI2 is absent from the initiating floral primordia, where it is required for the meristem identity transition. This suggests that the increased LMI2 levels observed by RT-PCR could be due to ectopic LMI2 expression. Indeed, whereas LMI2 expression was very low in the roots and leaves of nine-day-old wild-type seedlings, it was strongly expressed in these tissues in lmi2-1 (a 40-fold and 400-fold increase, respectively; Fig. 3E). Based on our combined findings, we conclude that the T-DNA insertion in lmi2-1 apparently disrupts the LMI2 promoter, causing loss of LMI2 expression in the shoot apical meristem and in the young flower primordia. At the same time, the insertion causes ectopic and elevated LMI2 expression, perhaps from a promoter located in the T-DNA insertion.

To place LMI2 in the meristem identity pathway, we examined the expression of the direct LFY targets AP1, CAL, LMI1 to LMI5, and that of another meristem identity regulator, FRUITFULL (FUL) (Ferrandiz et al., 2000; Wagner et al., 1999; William et al., 2004) in lfy-10 single mutants compared with lmi2-2 lfy-10 double mutants during the meristem identity transition (Fig. 4A, see Fig. S3A in the supplementary material). We conducted a time-course experiment spanning time points prior to, during and immediately subsequent to the meristem identity transition for all genotypes tested (Fig. 4A,B-E) (William et al., 2004; Yamaguchi et al., 2009). Although we did not observe a reduction in the expression of LMI1, LMI3 or LMI5, we observed a subtle reduction in the expression of LMI4 and a pronounced (approximately fourfold) reduction in the expression of AP1 in lmi2-2 lfy-10 compared with lfy-10 at day 13 (Fig. 4A; see Fig. S3A in the supplementary material). Indeed, AP1 expression was induced more slowly in the double mutant compared with lfy-10 (Fig. 4A). By contrast, induction of CAL and FUL expression was very similar in lfy-10 and lmi2-2 lfy-10 plants, suggesting that the observed defect in AP1 upregulation is specific. AP1 expression was also reduced in lmi2-2/+ lfy-10 plants relative to lfy-10 mutants (see Fig. S3B in the supplementary material), consistent with the observed dosage sensitivity of LMI2, as well as in lmi2-2 single mutant seedlings compared with wild type (Fig. 4G). Our combined data suggest that LMI2 acts upstream of AP1.

lfy null mutants cause a delay, but not a loss in AP1 expression; AP1 is expressed in the flowers that eventually form in these mutants (Ruiz-Garcia et al., 1997; Wagner et al., 1999). Likewise, based on qRT-PCR, AP1 is upregulated in lmi2-2 lfy-10, reaching expression levels similar to those observed in lfy-10 at day 15 (Fig. 4A), when flower patterning is initiated (see Fig. S3C in the supplementary material).

We next examined AP1 upregulation in wild-type, lfy-10 and lmi2-2 lfy-10 seedlings using in situ hybridization. By day 13, all three genotypes had initiated the first flowers. AP1 expression was much reduced in stage 1 or 2 flower primordia in thirteen-day-old lmi2-2 lfy-10 and the lfy-10 mutants relative to wild type (Fig. 4B-E; data not shown). In addition, AP1 expression levels were slightly more reduced in developing flower primordia of lmi2-2 lfy-10 compared with lfy-10 (Fig. 4C-E), and in the double mutants especially in the shoot meristem proximal region of stage 2 flower primordia (Fig. 4D,E).

To test whether LMI2 can regulate AP1 expression directly, we scanned the AP1 locus for the presence of plant MYB binding sites using AthaMap (http://www.athamap.de/) (Steffens et al., 2004). Eight predicted MYB binding sites were found in the 5′ upstream region; two in the introns and one in the first exon of AP1 (Fig. 4F). We next examined whether LMI2 binds to AP1 regulatory regions in vivo by anti-HA chromatin immunoprecipitation (ChIP) followed by qPCR using plants expressing a HA-tagged genomic version of LMI2 driven from its own promoter (pLMI2:LMI2-HA). The LMI2-HA fusion protein is biologically active, as pLMI2:LMI2-HA lmi2-2 rescued the reduced AP1 expression observed in lmi2-2 mutants (Fig. 4G). LMI2-HA was recruited to the AP1 promoter and bound to region six of AP1, which is very close to the known or predicted binding sites of other regulators of AP1, including LFY (Fig. 4F,H) (Abe et al., 2005; Parcy et al., 1998; Wang et al., 2009; Wigge et al., 2005; William et al., 2004; Winter et al., 2011; Xu et al., 2010; Yamaguchi et al., 2009). By contrast, we did not see enrichment of LMI2-HA relative to control lmi2-2 plants in the remaining regions of the AP1 locus, suggesting that the binding of LMI2 at region six is specific (Fig. 4H). Taken together, our data suggest that LMI2 directly activates AP1 expression during the meristem identity transition.

To test whether LMI2 acts solely to induce AP1 or whether it regulates other factors during the meristem identity transition, we crossed lmi2-2 to the strong ap1-10 mutant and examined the timing of the meristem identity transition. We did not observe an increase in the number of secondary inflorescences in lmi2-2 ap1-10 compared with ap1-10. There was, however, a significant increase in the number of cauline leaves produced in lmi2-2 ap1-10 compared with ap1-10 (Table 2). AP1 does not play a significant role in cauline leaf suppression during the floral transition (Bowman et al., 1993; Schultz and Haughn, 1993). Thus, like LFY (Liljegren et al., 1999), LMI2 functions through an AP1-independent pathway to suppress cauline leaf formation. We conclude that LMI2 acts both upstream of and in parallel to AP1 during the meristem identity transition.

LMI2 binds very close to the known LFY binding site in the AP1 locus (one putative LMI2 binding site in region six is 6 bp downstream of the LFY binding site; data not shown) (Winter et al., 2011). Hence, LMI2 and LFY might interact physically. Indeed, based on pull-down assays, LMI2 interacted with GST-LFY (Fig. 5A). Full length LFY protein homodimerized, as previously proposed (Hames et al., 2008), serving as a positive control. A negative control protein (see Materials and methods for details) did not interact with GST-LFY, confirming the specificity of the observed interactions (Fig. 5A).

Based on yeast two-hybrid assays, the N-terminal half of LMI2 (LMI2N) showed a weak, but reproducible interaction with LFY (Fig. 5B). The C-terminal domain of LMI2 (LMI2C) also interacted with LFY in yeast (data not shown). This interaction was more difficult to observe because, as previously reported (Zhang et al., 2009), this domain of LMI2 displays transcriptional activation activity. Finally, bimolecular fluorescence complementation (BiFC) was used to test for an in vivo interaction between LMI2 and LFY. Both the LMI2N and, to a lesser extent, LMI2C interacted with LFY (Fig. 5C). Again, LFY interacted with itself. By contrast, a negative control protein did not interact with LFY, suggesting the observed interactions were specific. The combined data suggest that LFY and LMI2 can form heterodimers.

During the floral transition, LFY and AP1 act in a positive feedback regulatory loop (Ferrandiz et al., 2000; Kaufmann et al., 2010; Liljegren et al., 1999). In light of this, we examined whether LMI2 can also feedback to regulate LFY. Indeed, LFY levels were reduced in lmi2-2 lfy-10 compared with lfy-10 mutants throughout the meristem identity transition (Fig. 6A). Furthermore, LFY levels were reduced in eleven-day-old lmi2-2 seedlings compared with wild type (Col) (Fig. 6B). To determine whether the reduction in LFY in lmi2 mutants was an indirect consequence of reduced AP1 expression in these mutants or whether LMI2 directly regulated LFY levels, we used ChIP to examine LMI2 binding to LFY regulatory regions. We tested binding of LMI2 to three predicted MYB binding sites in the 5′ upstream regulatory region: two sites in exon one and two sites in the second intron of LFY (Fig. 6C). We did not see binding of LMI2-HA to the promoter or intron regions of LFY, but we did observe a subtle enrichment at region four in exon one (Fig. 6D). Although one other LFY regulator has previously been shown to bind this region (Yamaguchi et al., 2009), further experiments are needed to determine whether the feedback from LMI2 to LFY is direct.

We show here that the direct LFY target and MYB transcription factor LMI2 is required for correct timing of the meristem identity transition in Arabidopsis. LMI2 was identified by two independent genomic approaches as a direct LFY-regulated and LFY-bound target during meristem identity transition (William et al., 2004; Winter et al., 2011). Notably, unlike two other known meristem identity regulator mutants (cal and lmi1) (Bowman et al., 1993; Saddic et al., 2006), lmi2 single mutants displayed a statistically significant delay in the meristem identity transition, suggesting a central role for this transcription factor in the timing of flower formation. Thus far, only one other direct LFY target has a non-redundant role in this vital developmental transition: AP1 (Bowman et al., 1993; Weigel et al., 1992).

The observed LMI2 expression pattern suggests that LMI2 might have a broad role in reproductive development. Like many flowering-time regulators (Abe et al., 2005; Hempel et al., 1997; Lee et al., 2000; Wigge et al., 2005), LMI2 was expressed in the shoot apex, and LMI2 controls the timing of bolting. In addition, both LFY and LMI2 were expressed in older flower primordia. Unlike lfy mutants (Huala and Sussex, 1992; Weigel et al., 1992), lmi2 mutants did not display noticeable floral homeotic defects nor did they enhance the floral homeotic defects of weak lfy mutants (data not shown), suggesting that LMI2 might have a different role in flower development.

AP1 upregulation signals commitment to flower formation and, therefore, must be tightly controlled for proper timing of the meristem identity transition (Bowman et al., 1993; Kaufmann et al., 2010; Mandel et al., 1992; Wellmer and Riechmann, 2010). Here, we provide evidence that LMI2 directly upregulates AP1 expression during the meristem identity transition. The effect of LMI2 on AP1 expression is specific and is not due to a general delay in phase transitions, because accumulation of other meristem identity regulators, such as FUL or CAL (Bowman et al., 1993; Ferrandiz et al., 2000), are not altered in lmi2-2 lfy-10 mutants. LMI2 induction precedes that of AP1 and both are expressed in stage 1 floral primordia, where AP1 directs flower development (this study) (Liljegren et al., 1999; Mandel et al., 1992; Schmid et al., 2005). LMI2 binds to a region of the AP1 locus also occupied by many other transcription factors in vivo, including LFY (Wang et al., 2009; William et al., 2004; Xu et al., 2010; Yamaguchi et al., 2009), thus defining a critical AP1 cis regulatory module (Jeziorska et al., 2009; Wilczynski and Furlong, 2010).

The LMI2 and LFY binding sites on the AP1 promoter are very close to each other and, based on three independent assays, the LMI2 and LFY proteins interact physically. MYB proteins are known to interact with other transcription factors to regulate gene expression (Li et al., 2009; Shin et al., 2007; Zimmermann et al., 2004). LFY also interacts with cofactors, including at least one other downstream target, to regulate gene expression (Chae et al., 2008; Lenhard et al., 2001; Liu et al., 2009b; Lohmann et al., 2001; Winter et al., 2011). For example, LFY directly upregulates the floral homeotic regulator SEPALLATA3 (SEP3) and, in turn, these two factors interact physically to activate the class B and C floral homeotic genes (Liu et al., 2009b; Winter et al., 2011).

Based on the recent finding that LFY acts as both a direct transcriptional activator and repressor (Parcy et al., 2002; William et al., 2004; Winter et al., 2011), it seems likely that cofactors modulate the effect of LFY on gene expression. Consistent with this idea, LFY alone is unable to activate gene expression from the AP1 promoter in yeast: it can only act as a transcriptional activator in this system when fused to a strong activation domain (Parcy et al., 1998; Winter, 2011). It is likely that LFY also needs a co-activator for AP1 induction in vivo. LMI2 is a good candidate for this LFY co-activator: it has strong transactivation activity based on yeast assays, it is induced by LFY prior to AP1 upregulation and can form heterodimers with LFY (this study) (Blazquez et al., 1997; Hempel et al., 1997; Schmid et al., 2005; Schmid et al., 2003; Zhang et al., 2009). Moreover, the temporal delay in the formation of the first flower is very similar in ap1 and lmi2 single mutants (this study) (Xu et al., 2010); thus, LMI2 might be sufficient for LFY-dependent activation of AP1 expression. However, we cannot rule out that other LFY co-factors contribute to this process.

The observed interactions between LFY, LMI2 and AP1 represent a coherent feed-forward loop (Fig. 7) (Alon, 2007), a regulatory circuit with crucial roles in control of developmental processes in many organisms (Alon, 2007; Mangan et al., 2003; Shen-Orr et al., 2002). The type of coherent feed-forward loop observed here serves as a persistence detector for inductive signal(s) and as a temporal delay element (Alon, 2007). Thus, transient inductive cues that cause a temporary increase in LFY, but not in LMI2, will delay LFY-dependent upregulation of AP1.

This finding is consistent with prior observations. For example, LFY upregulation is directed by environmental cues, such as changes in day length or ambient temperature (Amasino, 2010; Kobayashi and Weigel, 2007; Liu et al., 2009a); these stimuli are inherently noisy inputs, yet the transition to flower formation is abrupt in Arabidopsis, without formation of intermediate structures (Parcy, 2005). In addition, as discussed above, AP1 induction is delayed with respect to that of LFY and LMI2, and is reduced in both single mutants. Finally, as predicted by the feed-forward loop model, LMI2 was a haplo-insufficient, rate-limiting factor for AP1 induction downstream of LFY, at least under conditions when LFY activity was compromised.

In addition to the feed-forward loop uncovered here, LFY directs at least two additional coherent feed-forward loops, one of which is also linked to the meristem identity transition and involves the direct LFY targets LMI1 and CAL (Fig. 7) (Kaufmann et al., 2009; Liu et al., 2009b; Saddic et al., 2006; William et al., 2004; Winter et al., 2011). Among these feed-forward loops involving LFY, the LFY-LMI2-AP1 feed-forward loop stands out as it alone comprises three regulators that have non-redundant roles in the process they regulate; hence, it might represent a crucial regulatory module in the meristem identity transition.

In Arabidopsis, the meristem identity transition is not only precise (it occurs after formation of a defined number of secondary inflorescences subtended by cauline leaves), but also robust (no reversion from flower to inflorescence fate is observed) (Amasino, 2010; Blazquez et al., 2006; Liu et al., 2009a; Tooke et al., 2005). As outlined above, the LFY-LMI2-AP1 feed-forward loop is likely to contribute to the precision of this developmental transition; its robustness, however, might be due to positive feedback (Alon, 2007). Indeed, a positive direct feedback from AP1 to LFY has recently been described (Kaufmann et al., 2010; Liljegren et al., 1999). We show here that LMI2 also positively regulates LFY: LFY expression was reduced in lmi2-2 single and double mutants. This reduction of LFY expression could be an indirect effect, triggered by the reduced AP1 expression levels observed in lmi2-2 mutants. However, the positive feedback might, in part, be direct as LMI2 was weakly recruited to the LFY locus. The observed enhancement of the ap1 mutant meristem identity defect by lmi2 is consistent with this hypothesis. It is likely that the AP1 and possible LMI2 feedback loops keep the LFY-LMI2-AP1 feed-forward loop active after full AP1 upregulation has been achieved. Indeed, AP1 directly downregulates upstream activators of itself and of LFY (Kaufmann et al., 2010), providing further support for the idea that the combined feed-forward and feedback loop is self-maintained.

It will be of interest to examine these regulatory interactions in other flowering plant species. In light of this question, we note that LMI2 separated from its closest homologs, the MIXTA/MYB16 and MYB106 genes, before the split of the monocots from the eudicots ∼100 million years ago (Baumann et al., 2007). This raises the possibility that the function of LMI2 in reproductive development evolved early in the flowering plant lineage and might be conserved in other angiosperm species.

We are grateful to David M. Braun for the BiFC negative control protein and to Jennifer Pfluger for the yeast two-hybrid and BiFC LFY constructs. We thank Kim Gallagher, John Wagner, Matthew Willmann and Cara M. Winter for critical comments and suggestions on this manuscript. This research was funded by NSF grant IOS 0849298 to D.W. and a NIH Training Grant 5-T32-HD007516 (Developmental Biology) as well as a University of Pennsylvania, School of Arts and Sciences, Dissertation Research Fellowship to J.J.P. Deposited in PMC for release after 12 months.