NUCLEAR FACTOR Y Transcription Factors Have Both Opposing and Additive Roles in ABA-Mediated Seed Germination

In the model organism Arabidopsis thaliana the heterotrimeric transcription factor NUCLEAR FACTOR Y (NF-Y) has been shown to play multiple roles in facilitating plant growth and development. Although NF-Y itself represents a multi-protein transcriptional complex, recent studies have shown important interactions with other transcription factors, especially those in the bZIP family. Here we add to the growing evidence that NF-Y and bZIP form common complexes to affect many processes. We carried out transcriptional profiling on nf-yc mutants and through subsequent analyses found an enrichment of bZIP binding sites in the promoter elements of misregulated genes. Using NF-Y as bait, yeast two hybrid assays yielded interactions with bZIP proteins that are known to control ABA signaling. Accordingly, we find that plants mutant for several NF-Y subunits show characteristic phenotypes associated with the disruption of ABA signaling. While previous reports have shown additive roles for NF-YC family members in photoperiodic flowering, we found that they can have opposing roles in ABA signaling. Collectively, these results demonstrated the importance and complexity of NF-Y in the integration of environmental and hormone signals.


Introduction
Successful acclimation of plants to temporal environmental changes requires the integration of multiple intersecting signals. These environmental signals, such as changing light and water availability, profoundly modify growth and developmental programs. Phytohormones often play a central role in these responses and recent studies identified several integrators of light and hormone signaling pathways (reviewed in [1] and [2]). For example, the basic leucine zipper (bZIP) transcription factor ELONGATED HYPOCOTYL 5 (HY5) has been intensely studied for its roles in light regulated development, but was only recently found to additionally mediate abscisic acid (ABA) signaling in germinating seeds [3]. Thus, pathway integrators can act as hubs for multiple environmental inputs, but how they coordinate these variable inputs to generate unique transcriptional outputs remains unknown [2].
One possibility for coordinating multiple environmental inputs is through interactions with combinatorial transcription factors. Increasing evidence suggests that the combinatorial transcription factor NUCLEAR FACTOR Y (NF-Y) plays important roles in facilitating plant responses to various environmental signals and hormones [4][5][6][7]. In fact, like HY5, NF-Y complexes can regulate both light signaling (blue light) and hormone perception (ABA, [7]). NF-Y transcription factors bind at CCAAT cis-elements and function as trimeric complexes consisting of three distinct protein subunits, NF-YA (NF-Y, subunit A), NF-YB, and NF-YC [8]. These individual subunits are each encoded by small gene families found throughout the plant lineage -e.g., Arabidopsis thaliana (Arabidopsis) has 36 NF-Y encoding genes (10 NF-YA, 10 NF-YB, and 10 NF-YC [9][10][11]). Because the mature DNA-binding NF-Y complex is thought to contain only one of each subunit type, hundreds of unique combinations are theoretically possible. Thus, unique combinations of NF-Y transcription factors may provide a flexible system for fine-tuning the integration of environmental signals to transcriptional outputs [9,12,13].
Although NF-Y itself represents a multi-protein complex that can independently integrate multiple signals, recent studies have shown additional interactions with other transcription factor families, especially those in the bZIP family. Yamamoto et al. [4] demonstrated that central regulators of embryogenesis, LEAFY COTYLEDON 1 (LEC1/NF-YB9) and LEC1-LIKE (L1L/NF-YB6), interact with bZIP67 to regulate the expression of genes with ABA-responsive elements (ABRE) in their promoters. More recently, Liu and Howell [25] proposed a model where a complex of bZIP28, NF-YA4, NF-YB3, and NF-YC2 binds to endoplasmic reticulum stress response elements (ERSE) to regulate genes related to the unfolded protein response (UPR). In both cases, full activation of target promoters required the presence of NF-Y and bZIP components. Therefore, the complexity of NF-Y transcription factors can be further increased through interactions with bZIP proteins.
Like NF-Ys, bZIPs are also found throughout the eukaryotic lineage. In Arabidopsis bZIP proteins represent a large family containing 75 members in ,20 phylogenetically related subgroups [28]. Similar to NF-Y, bZIP proteins are involved in diverse processes ranging from environmental stress tolerance to development. Group A bZIP proteins are particularly important for the integration of environmental stress signals (e.g., drought) with hormone-related responses (especially ABA, [29]). Examples of Group A bZIP proteins include the previously discussed bZIP67, as well as ABSCISIC ACID INSENSITIVE 5 (ABI5) and the ABA RESPONSE ELEMENT BINDING PROTEINS/ABA BINDING FACTORS 1-4 (AREB/ABF1-4). Mutations in ABI5 allow developing seeds to germinate and grow on normally restrictive levels of ABA [30]. Mutant and overexpression lines for ABF1-4 show characteristic morphological and molecular phenotypes typically associated with ABA responses, including altered growth, transpiration rates, and seed germination in response to abiotic stress [31]. Although not in Group A, the light and ABAsignaling integrator HY5 is also a bZIP protein. HY5 directly regulates the expression of ABI5 and hy5 mutants have reduced sensitivity to ABA in a variety of assays, including seed germination and root growth [3].
Previously, we showed that NF-YC3, NF-YC4, and NF-YC9 have overlapping roles in photoperiod-dependent flowering [16]. In the current study we used transcription profile analysis of nf-yc3 nf-yc4 nf-yc9 triple mutants to investigate other pathways potentially affected by these genes. This analysis revealed an over-representation of cis-elements with putative bZIP binding sites. We hypothesized that NF-Y mutants would share phenotypic similarities to various bZIP mutants and tested this idea by assaying seed germination, seedling greening, and root elongation for abnormalities as well as quantitative reverse transcriptase polymerase chain reaction analyses (qPCR) on known bZIP regulated genes. We found that various mutant and overexpression combinations of NF-YB and NF-YC genes significantly altered characteristic morphological and molecular phenotypes associated with ABA signaling. Additionally, we found that NF-Y subunits from the same family can have opposing roles in a given process. Overall, these data suggested that NF-Y and bZIP transcription factors coordinately regulate ABA-related phenotypes.

Misregulated Genes in nf-yc Mutants are Enriched for bZIP Binding Sites
We previously used qPCR and microarray analyses to demonstrate that FLOWERING LOCUS T (FT) was downregulated in the 10 day-old seedlings of long day grown nf-yc3 nf-yc4 nf-yc9 triple mutants (hereafter nf-yc triple) [16]. To generate new hypotheses and extend the nf-yc triple analysis here, we further examined this published microarray data by searching for overrepresented promoter motifs within the 59 upstream sequences of the misregulated gene set (both up and down regulated genes, p,0.05 without false discovery rate correction; see Materials and Methods). This set consisted of 83 nuclear-encoded genes that were at least 1.5 fold misregulated in the nf-yc triple. Initially, we used the online software package Athena to search this gene set for enriched promoter motifs (http://www.bioinformatics2.wsu.edu/ cgi-bin/Athena/cgi/home.pl, [32]).
Using the Athena analysis suite, three statistically overrepresented promoter motifs were discovered (Table I). All three were bZIP-binding sites with the core-conserved sequence ACGT. The most highly significant motif was the 3.3-fold over-represented CACGTG motif, also called the G-box [33]. The G-box is found in the promoters of many light responsive genes and is a known binding site for HY5 [34][35][36]. The other two significantly enriched sequences were related to the ABRE and also have the ACGTG portion of the G-box. ABREs are also bound by bZIP proteins, specifically the Group A bZIP transcription factors AREB/ABF1-4 [28,37,38]. These data are consistent with a previous report that L1L (NF-YB6) [24,39] physically interacts with the Group A bZIP protein bZIP67 to regulate ABRE containing promoters [4]. Also of note in Table I is the enrichment of the unfolded protein response (UPR) element, as identified in animal systems [40]. Although just missing statistical significance in our dataset, Liu and Howell recently showed that NF-Ys can physically interact with bZIP28 and affect the expression of genes with the UPR promoter element [25].
In addition to Athena, we used MEME (Multiple Em for Motif Elicitation, http://meme.nbcr.net) to search for shared motifs in the misregulated gene set [41,42]. MEME uses a Hidden-Markov model and is not constrained by searching for known motifs (i.e., it can find both known and novel motifs). Three low E-value motifs were discovered and each was tested for resemblance to previously identified transcription factor binding sites using the TOMTOM motif comparison tool [43]. While two of the motifs showed no clear homology to known binding sites, one showed significant resemblance to known bZIP binding sites, including those bound by ABF, HY5 and EmBP-1 ( Figure 1A). Consistent with our previous results, bZIP binding sites were often located between 2100 to 2200 bp of the transcription start site within the promoters of the misregulated gene set ( Figure 1B). These data, as well as several recent reports [4,25], suggest that NF-Y may have a generalizable relationship with bZIP proteins in the transcriptional regulation of numerous genes.

Predicted NF-YC Interaction Network Include bZIP Proteins
Bioinformatic analysis of promoter elements enriched on the nfyc triple microarray revealed a possible interaction with bZIP domain containing proteins. To further investigate the predicted protein-protein interaction network of NF-YC3, NF-YC4, and NF-YC9, we used the web-based software GeneMANIA that is designed to identify functional associations [44,45]. For visualization, individual protein-protein interaction networks from Gene-MANIA were integrated into a single network map using the open source software Cytoscape (Figure 2, [46]). In addition to the predicted interaction network provided by GeneMANIA, we included previously published NF-Y by CCT (Constans, Constans-Like, TOC1) protein interactions. We also included bZIP and CCT protein interactions discovered in a yeast two hybrid (Y2H) screen using NF-YC9 as bait and the directed Y2H data shown below ( [16] and RWK and BFH unpublished). As expected, all 13 Arabidopsis NF-YB proteins, as well as four of 10 NF-YA proteins, were predicted to interact with the NF-YC proteins. In addition to the previously described interactions with NF-YA, NF-YB and CCT proteins [5,6,16], there were two other large classes of proteins represented on the network map -histone 2A and bZIP containing proteins. These data further support the bZIP binding site enrichment found in the nf-yc triple microarray gene set, as well as previously reported interactions between NF-Y and bZIP proteins.

NF-Y can Physically Interact with ABA Related bZIP Proteins
Relevant to the promoter analyses, NF-YC interaction network, and hypothesis that bZIP/NF-Y interactions are common, we isolated bZIP proteins in Y2H screens using both NF-YB2 and NF-YC9 as bait. When NF-YB2 was used as the bait, we isolated and sequenced 42 positive clones. Five of the positive clones were either bZIP1 (2 independent clones) or ABF3 (3 independent clones). Further, a screen using NF-YC9 as bait yielded a single clone of ABF2 out of 200 sequenced positives. Although a single clone of ABF2 appears low, it should be noted that 78% of the sequenced positive clones from the NF-YC9 screen were either NF-YB or CCT proteins (155 out of 200).
To further validate and extend the bZIP/NF-Y interactions from the Y2H library screens, we independently cloned full-length versions of five bZIP genes with known roles in ABA responses: ABF1, ABF2, ABF3, ABF4 and HY5. Each of these was tested for interaction with a panel of NF-YB and NF-YC proteins in directed Y2H assays. For the NF-YC proteins, we chose the members of the nf-yc triple used in the above microarray analysis (NF-YC3, NF-YC4, and NF-YC9). These three proteins are known to genetically interact in an overlapping manner to control FT expression and presumably they are interchangeable in the NF-Y complexes controlling FT transcription [16]. Likewise, we tested NF-YB2 and NF-YB3 against the panel of bZIPs. NF-YB2 and NF-YB3 are known in vivo interactors with NF-YC3 and NF-YC4 and also have overlapping functions in the control of FT expression [14,16]. Finally, NF-YB1 was tested because of its known roles in drought resistance [18] -i.e., we hypothesized a possible connection to the known roles of the ABF/AREBs in various ABA-related responses, including drought resistance.
NF-YC4 and NF-YC9 interacted with all five of the tested bZIP proteins ( Figure 3A). NF-YC3 also likely interacted with the four ABFs, but this result was difficult to accurately interpret due to previously reported autoactivation ( [16], compare to empty vector interaction in Figure 3A). Previous reports indicate that NF-YC subunits also mediate bZIP interactions in fungal and mammalian systems [47,48], suggesting broad evolutionary conservation of these interactions in the eukaryotic lineage.
Although the NF-YB2 by ABF3 interaction was reproducible in the directed Y2H tests ( Figure 3B), no other NF-YB by bZIP interaction was detected. This result was unanticipated because NF-YB2 and NF-YB3 share nearly 70% amino acid identity over their entire proteins. Further, their central conserved domains, the 95AA histone fold motif (HFM), are 93% identical and 100% Table 1. Over-represented cis motifs in promoters of genes misregulated in nf-y triple mutants.

Motif Name
Sequence  similar ( Figure S1). This led to the hypothesis that regions outside of the conserved domain in NF-YB2 are responsible for its unique ability to interact with ABF3.
To test this hypothesis we dissected the NF-YB2 protein into three different regions: the N-terminal region up to the conserved domain ( Figure 4A, B2N, AA 1-25), the conserved HFM (B2HFM, AA 26-121) and the C-terminal region (B2C, AA 122-190). Constructs containing each of the regions were tested in Y2H for interaction with ABF3. None of the partial NF-YB2 clones showed interaction with ABF3 ( Figure 4A), although the conserved histone fold motif of NF-YB2 still retained the ability to interact with NF-YC proteins (data not shown). This led to the hypothesis that one of the ends of NF-YB2, in conjunction with the HFM, could drive the interaction with ABF3. Clones were generated from the start of NF-YB2 through the HFM (B2N+B2HFM, AA 1-121) and starting from the first AA of the HFM through the end of the protein (B2HFM+B2C, . Like the partial constructs above, neither of these constructs interacted with ABF3 ( Figure 4A). Thus, if the N or C terminal ends were required for NF-YB2 specificity for ABF3 interactions, these appeared to still require the context of the full-length protein.
To explore this latter idea, we created chimeric proteins between NF-YB2 and NF-YB10. We chose NF-YB10 because it has highly divergent terminal ends outside of the HFM and also showed no interaction with ABF3 ( Figures 4B, S1). The constructs contained either the N-terminal region of NF-YB2 or NF-YB10 through the first half of the HFM fused to the C-terminal region of the other NF-YB ( Figure 4B). Interestingly the NF-YB2 Nterminal/NF-YB10 C-terminal chimeric protein was able to interact with ABF3, although more weakly than the native full length NF-YB2. This is in contrast to the complementary construct Figure 2. NF-YC3, NF-YC4 and NF-YC9 protein-protein interaction network. Both demonstrated and GeneMANIA predicted protein-protein interaction data for NF-YC3, NF-YC4, and NF-YC9 were visualized using Cytoscape [46]. Predicted physical interactions are depicted as dashed lines, while demonstrated interactions ( [6,16] and this work) are depicted as solid lines. Input nodes NF-YC3, NF-YC4 and NF-YC9 are shown as squares. Circle nodes are those predicted data from GeneMANIA [44,45]. Octagonal nodes represent demonstrated physical interactions (e.g., Y2H, some shown in this report -see below). Related protein nodes are color coded as follows: red-bZIP; blue -CCT; green -H2A; orange/tan -NF-YB; yellow -NF-YA; gray -unclassified interacting proteins. A Microsoft Excel file is available to recreate/manipulate this data (File S1). Common names were used where available -File S1 contains all AGI numbers and references for sources of data. doi:10.1371/journal.pone.0059481.g002 (NF-YB10 N-terminal/NF-YB2 C-terminal), which did not interact ( Figure 4B). Further, neither constructs containing the NF-YB10 N-terminal (B10N, AA 1-22) or C-terminal (B10C, AA 123-228) regions alone interacted with ABF3. Collectively, these data suggest that the N-terminal domain of NF-YB2 is necessary but not sufficient for the specific interaction with ABF3. A previous report in Arabidopsis also suggested that bZIP28 interactions with NF-Ys were subunit specific (NF-YB3) and additionally required the initial formation of an NF-YB3/NF-YC2 dimer [25]. Thus, additional interactions may be masked in our two way directed tests.

NF-YC have Opposing Roles in ABA Responses
Our bioinformatic analyses and Y2H screens suggest that NF-Y/bZIP interactions may be commonplace. These findings led us to hypothesize that, like many bZIPs, NF-Y are also involved in ABA responses. Warpeha et al. [7] previously demonstrated that several Arabidopsis nf-y mutants have delayed germination in the presence of ABA. We extended these analyses by first testing the germination of nf-y triple mutants on non-ABA containing media, where no significant differences to control were observed ( Figure 5A). Conversely, on media supplemented with ABA, germination inhibition was significantly reduced in nf-yc triple mutants compared to Col-0 control plants ( Figure 5B-C,J). This result was similar to previous observations of both abf1 or hy5 mutants [49,50].
Although NF-YC3, NF-YC4 and NF-YC9 have overlapping functionality in photoperiod-dependent flowering, it was unclear whether the same would be true for these ABA related phenotypes. In fact, it was previously shown that nf-yc4 single mutants were hypersensitive to ABA [7] -the opposite of what we observed for the nf-yc triple mutant which includes the same nf-yc4-1 loss of function allele. Therefore, we tested the germination of all possible single and double mutant combinations of nf-yc3, nf-yc4, and nf-yc9 (see Materials and Methods for allele designations).
The response of nf-yc3 nf-yc9 double mutants on ABA media was non-distinguishable from the nf-yc triple, indicating that NF-YC4 does not normally play a role in repressing ABA-mediated germination ( Figure 5D-F,J). As was previously reported [7], nf-yc4 single mutants were hypersensitive to ABA in our germination assays ( Figure 4H,J). Like nf-yc4, the nf-yc3 nf-yc4 and nf-yc4 nf-yc9 double mutants were also slightly hypersensitive to ABA and germinated later than controls ( Figure 5E,J). The opposite is true in nf-yc triple mutants where the early germination phenotype of the nf-yc3 nf-yc9 double mutant is observed ( Figure 5B). To extend these ABA analyses beyond germination, we additionally measured cotyledon greening and root elongation.
For cotyledon greening assays, seeds were sown on media containing 1 mM ABA and the percentage of plants with open green cotyledons was scored after 10 days of growth in continuous light [51]. As predicted from the germination data, nf-yc3 nf-yc9 double and nf-yc triple mutants performed much better than controls -after 10 days nearly 100% of these plants had open green cotyledons compared to less than 20% for parental Col-0 ( Figure 5K-L). Poor greening for nf-yc4 single and nf-yc3 nf-yc4 double mutants also correlated with the germination phenotypes. However, nf-yc9 single mutants, which showed no significant differences in germination, had a significantly greater number of seedlings that greened ( Figure 5K). Additionally the nf-yc4 nf-yc9 double mutants greened similarly to nf-yc9 single mutants. This is in contrast to the ABA germination phenotype exhibited by the nf-yc4 nf-yc9 double mutant where the nf-yc4 phenotype was observed. Unlike cotyledon greening, NF-YC mutants showed no clear root elongation differences on media supplemented with ABA ( Figure  S2). These observations indicate that NF-YCs not only have opposing roles in ABA responses, but also have separable roles in germination and post-germination growth in the above ground plant.

NF-YB2 and NF-YB3 Overexpression Significantly Delays Germination
Because of the known physical interactions between NF-YB2/ NF-YB3 and NF-YC3/NF-YC4/NF-YC9 [16], as well as our demonstration that NF-YB2 can physically interact with ABF3, we additionally tested these NF-YBs in ABA germination assays and root elongation assays. As with nf-yc3, nf-yc4, and nf-yc9 mutants, we measured no root growth defects in response to ABA for the nf-yb2 and nf-yb3 mutants ( Figure S2). Further, we found that neither nf-yb2 or nf-yb3 single knockdown mutants, nor nf-yb2 nf-yb3 double mutants showed any significant alteration in germination on either regular media or ABA-supplemented media ( Figure 6A-B; note that the same nf-yb alleles have clear late flowering phenotypes [12,14,15]. We additionally used the 35S promoter from the cauliflower mosaic virus [52] to overexpress both NF-YB genes in transgenic plants [Note that the same experiment with the NF-YC genes was not possible due to the consistent inability to obtain stable overexpressing transgenic lines]. Stable, single insertion, third generation NF-YB overexpressing plant lines were chosen. When assayed on non-ABA growth media, p35S::NF-YB2 and p35S::NF-YB3 seeds reached 100% germination approximately 24 hours later than control plants ( Figure 6C). In the presence of 1 mM ABA, p35S::NF-YB2 and p35S::NF-YB3 seeds were much later germinating and never reached 100% germination (typically peaking at ,50% total germination by 120 HAI, Figure 6D). These data demonstrate that NF-YB2 and NF-YB3 overexpression can significantly alter seed germination responses, but knockdown data suggests this may not be their normal biological role. These results are similar to what is observed with abf mutants, where single loss of function alleles did not show strong phenotypes, but ABF overexpression could inhibit germination on ABA media [51,53,54]. Thus, higher order mutants combining nf-yb2, nf-yb3, and additional nf-yb knockdown alleles might uncover a biological role in seed germination for NF-YB2 and NF-YB3.
To help sort through these possibilities, we examined the tissuespecific expression of NF-YB2 and NF-YB3 in imbibed seeds using previously described transgenic plant lines expressing the reporter gene b-glucuronidase (GUS) driven by native NF-Y promoters (pNF-Y::GUS lines, [9]). If these NF-Y were likely to be normally involved in seed germination, expression would be expected in either the developing embryos or endosperm. As controls, we also examined the pNF-Y::GUS lines for NF-YC3, NF-YC4, and NF-YC9 where our loss of function data already clearly demonstrated a biological role in germination. As expected from the nf-yc mutant data above, all three NF-YC genes were consistently expressed in seeds, including both the embryos and endosperm layer ( Figure 7C-F, I-L). Although there were variations in specific expression patterns, all three NF-YC had expression throughout the embryonic root with the strongest GUS staining in the meristematic regions. Additionally, staining was typically stronger in the vascular regions of the cotyledons than the remainder of the seed leaf (this contrast was strongest for NF-YC3, Figure 7J). However, most pNF-YB2::GUS seeds showed no staining in the embryo or endosperm ( Figure 7B, H) and pNF-YB3::GUS seeds showed only weak and inconsistent staining in the embryo and endosperm ( Figure 7C, I). The observed expression patterns for the promoter fusion constructs are consistent with publically available seed microarray data ( Figure S3). The addition of ABA did not significantly change the expression patterns for any of the constructs or mRNA patterns from public microarray studies (data not shown). These data support the hypothesis that NF-YB2 and NF-YB3 are not normally the major NF-YB components of the germination-influencing NF-Y complexes. Nevertheless, both can clearly and reproducibly alter germination when ectopically expressed in seeds.

ABA and Germination Marker Genes have Altered Expression in NF-Y Mutants
To further test NF-Y roles in ABA signaling and germination, we used qPCR to examine transcript levels of several genes that are well-known to be responsive to bZIPs -ABI3, ABI5, AIA, AIL, RAB18, and RD29B [3,31,53]. For these assays we examined transcripts from imbibed seeds after 24 h of light exposure (see Materials and Methods). Based on our phenotypic data, we chose to compare the nf-yc3 nf-yc9 double mutants to the nf-yc4 single mutant. Additionally, we examined gene expression in the ABAhypersensitive p35S::NF-YB2 and p35S::NF-YB3 lines. Consistent with ABA-insensitivity phenotypes, the germination inhibitors ABI3 and ABI5 were downregulated ,5-10 fold in nf-yc3 nf-yc9 mutants ( Figure 8A-B). The ABA-hypersensitive lines nf-yc4, p35S::NF-YB2, and p35S::NF-YB3 all had less dramatic, but increased levels of ABI3 expression. ABI5 expression only increased in the NF-YB overexpressing lines. Collectively, these data are also consistent with previous research showing that ABI3 expression is positively regulated by LEC1 (NF-YB9, [55,56]). For the ABA-regulated genes AIA, AIL, RAB18, and RD29B, we consistently measured downregulated expression in the nf-yc3 nf-yc9 mutants, ranging from 22.5 fold (AIA) to 214.3 fold (AIL, Figure 8C-F). Once again, the two p35S::NF-YB lines consistently had the opposite response. For several genes the differences were quite dramatic. For example, in the p35S::NF-YB2 transgenic seeds, AIL was upregulated 22.5 fold ( Figure 8D) and RAB18 was up 15.5 fold ( Figure 8E). None of these ABA-regulated genes were strongly misregulated in nf-yc4 mutants, although the general trend was unexpectedly slightly down for each gene other than ABI3. The lack of strong differences in nf-yc4 may be due to the relatively weak ABA sensitivity phenotypes reported here and elsewhere [7]. Future experiments examining higher order mutants, especially a double mutant between NF-YC4 and its apparent paralog NF-YC1 [9], may improve the resolution of this analysis.

Discussion
Through bioinformatics and mutant analyses we add to the growing body of evidence that NF-Y and bZIP transcription factors cooperatively regulate similar subsets of genes and, thereby, some of the same plant processes. It also appears likely that NF-Y and bZIP form higher order regulatory complexes capable of integrating inputs from many signaling pathways. Examination of the cis-regulatory regions of mis-expressed genes in nf-yc triple mutants uncovered an over-representation of G-box and ABRElike bZIP binding motifs. ABRE and G-box elements are common in the promoters of genes responsive to abiotic stress and light [57,58]. Studies of ABRE containing promoters demonstrated that single ABRE sequences are not sufficient to induce transcriptional activation and a coupling element is required for induction of ABA-responsive genes [59]. Our data suggests that the NF-Ybound CCAAT box might play the role of the ABRE coupling element for some bZIP-responsive promoters.
In Arabidopsis, Liu and Howell show that a G-box containing promoter (ERSE-I) required intact CCAAT and CACGTG (G-box) elements for full activation in response to ER stress [25]. Further support that the CCAAT box may act as a coupling element to the G-box comes from outside the plant kingdom. Motifs similar to the ABRE were over-represented in NF-Y bound mammalian promoters [60]. Further, in response to ER stress, mammalian NF-Y must bind to the CCAAT box in ERSE containing promoters before the bZIP protein ATF6 can bind an adjacent G-box-like element [48]. In contrast, Yamamoto et al. demonstrated that NF-YB, NF-YC, and bZIP proteins collectively activate the ABRE-containing promoter of CRUCIFERIN C (CRC), but claim that this activity does not require the CCAAT box sequence [4]. Further, the addition of NF-YA subunits to their protoplast assays inhibits CRC expression. These data suggest a non-canonical use of the NF-YB/ NF-YC dimer where NF-YA subunits are not involved in the final transcriptional complex. In rice, OsMADS18 was shown to interact with an NF-YB/NF-YC dimer, without the NF-YA, suggesting some NF-Y may have evolved to form atypical complexes in plants [27]. Atypical NF-Y complexes (i.e., lacking NF-YA) would be unlikely to bind CCAAT boxes because the NF-YA subunit is thought to make all the direct physical contacts with the CCAAT nucleotide sequence [61,62]. Nevertheless, while it is possible that some NF-Y might not bind the actual CCAAT box, they were still necessary for bZIP67 activation of CRC expression.
NF-Y and bZIP proteins not only bind the same promoters, but also can physically interact [4,25]. Here we show that NF-YC subunits interact with different bZIP proteins fairly indiscriminately in directed Y2H assays. This result is consistent with data from animals and fungi where the NF-YCs appear to make the primary contacts with bZIP proteins [47,48]. While NF-YC proteins indiscriminately bound bZIP proteins in our Y2H assays, the NF-YB proteins were more selective. This suggests that NF-YBs may play a role in discriminating which bZIP binds to an NF-YB/C dimer or the full complex. In addition we demonstrate that regions outside of the highly conserved HFM are likely driving the preference of NF-YB for bZIP partners. This observation is of interest because it is consistent with protoplast assays where only the closely related NF-YB6 and NF-YB9 could interact with bZIP67 to activate the CRC promoter, as well as the finding that bZIP28 only interacts in complexes containing NF-YB3 [4,25]. Together, these data raise an intriguing hypothesis for the large expansion and maintenance of NF-YB subunits in plants as compared to animal and fungal systems. It is possible that NF-YB have evolved specific interactions with bZIP proteins whereas in other systems the interactions are more general.
In addition to promoter analyses and Y2H assays, we show that NF-Y mutant plants have morphological phenotypes related to ABA signaling and that several well-known marker genes are misregulated in the various mutant backgrounds. Initial germination studies with nf-yc triple mutants showed that they are less sensitive to ABA treatments. This was surprising because the nf-yc4 single mutant, one of the mutants comprising the nf-yc triple, was previously reported as being more sensitive to ABA in germination assays [7]. We then tested all possible double mutant combinations as well as the single mutants of NF-YC3, NF-YC4, and NF-YC9. Out of this analysis an epistatic relationship between NF-YC3/NF-YC9 and NF-YC4 emerged. nf-yc3/nf-yc9 double mutants showed identical ABA germination phenotypes to nf-yc triple mutant plants. This suggests that NF-YC3 and NF-YC9 are interchangeable as negative regulators of germination, and NF-YC4 does not have overlapping functionality in this process. The fact that NF-YC3 and NF-YC9 work as functional equivalents is not surprising as their amino acid sequences are 100% identical throughout the conserved DNA binding and NF-YA/NF-YB interaction domains [16]. NF-YC4 is more divergent in these domains and in agreement with previous reports [7] we demonstrated that nf-yc4 single mutants are late germinating on ABA-containing media. Epistasis analysis between the late germinating nf-yc4 mutants and early germinating nf-yc3/nf-yc9 mutants revealed that NF-YC3 and NF-YC9 collectively suppress the role of NF-YC4 as a positive regulator of seed germination.
The epistasis of NF-YC3/NF-YC9 to NF-YC4 suggests a simple linear genetic pathway where NF-YC4 is upstream and possibly directly regulates NF-YC3 and NF-YC9. Another possibility is that NF-YC3, NF-YC4 and NF-YC9 are acting competitively at single hubs to regulate the ABA germination response. Similar functional antagonism between closely related family members has been seen with the seed specific bZIPs, ABI5 and EEL (Enhanced EM Level, bZIP12, [63]). Mutant analysis revealed that EEL is a negative regulator of embryogenesis-abundant genes and is dependent on the presence of the positive regulator ABI5. Interestingly, instead of a simple linear pathway where EEL regulates ABI5, it was shown that ABI5 and EEL could interact and possibly compete for the same ABRE promoter elements in embryogenesis-abundant genes. It will be interesting to determine if positive and negative regulatory NF-Y complexes can compete for the same elements. In addition it stands to reason that bZIP proteins could also be involved in these complexes.
In an attempt to determine the composition of biologically active NF-Y complexes related to ABA signaling, we additionally tested nf-yb2 and nf-yb3 mutants for their roles in germination. Both of their encoded proteins interact in vivo with NF-YC3 and NF-YC4 to regulate flowering, but these assays only utilized seedlings [16]. nf-yb2 and nf-yb3 loss of function mutants showed no clear difference to controls in ABA germination assays, but gain of function p35S::NF-YB2 and p35S::NF-YB3 lines showed strong repression of germination in the presence of ABA. While these gain of function phenotypes suggest a role for NF-YB2 and NF-YB3, their lack of native expression in the endosperm/embryo and the absence of loss of function phenotypes suggest other NF-YB act natively in the germination complex. It is likely that NF-YCs partner with LEC1 and L1L (NF-YB6 and NF-YB9, respectively) as these NF-YB proteins are well-known, embryo-specific regulators [7]. Similar to nf-yc3 nf-yc9 double mutants, lec1 mutant plants are less sensitive to ABA in germination assays [23]. In addition, NFYC3, NF-YC4 and NF-YC9 can all interact with LEC1 and L1L in directed Y2H assays [64,65] and we consistently isolate LEC1 and L1L in library screens using NF-YC9 as bait (BFH, CLS, RWK unpublished data). Although nf-yc3 nf-yc9 and lec1 mutants share reduced ABA sensitivity in germination assays, nf-yc3 nf-yc9 mutants do not have the lec1 leafy cotyledon or desiccation intolerant phenotypes. This suggests that there may be additional, as yet unidentified NF-YC(s) involved in embryogenesis.
Although NF-Y can interact with bZIP proteins and can regulate the expression of a similar subset of targets, it remains to be determined how and if these complexes form in planta. With the recent discovery of the long sought ABA receptor, it has become clear that Group A bZIP transcription factors are directly activated through phosphorylation in the presence of ABA [29,66]. In contrast to these bZIP proteins, relatively little is known about NF-Y complexes. How NF-Y activity is regulated and what cis-regulatory elements they are capable of binding remains to be determined for the plant lineage.
In this study we begin to uncover the complex and sometimes antagonistic roles that NF-Y play in ABA signaling. In addition, we add to the mounting evidence that plant, animal, and fungal NF-Y interact with bZIPs to form multi-protein, transcriptionregulating hubs to affect gene expression. Future studies describing the regulation and formation of these complexes will further our understanding of how plants integrate multiple signals to fine-tune growth and development.
All germination and root growth assays were performed using seeds that were collected from plants grown concurrently in the same growth chamber under standard long day conditions (

Germination and Root Growth Assays
Matched seed sets were harvested and allowed to after ripen for between 2 to 4 months before use in assays. Seeds were surface sterilized and plated onto Gamborg's B5 media (SIGMA, St. Louis, MO, Cat#G5893) or B5 media containing the appropriate amount of (+/2) ABA (SIGMA, Cat#A4906). For germination and cotyledon greening assays, plates were cold stratified for 3 days and placed in continuous light at 90 mmol m 22 s 21 at 22uC in a Conviron model ATC13 chamber. Seeds were scored every 12 hours post incubation for visible radical emergence as a proxy for seed germination. Greening was assayed by counting plants with open green cotyledons on day 10 [51]. Plants for root growth assays were sown to B5 plates and incubated under long day conditions for four days before transfer to plates supplemented with ABA. Plates were oriented vertically for an additional seven days. All plates were photographed and primary root length was measured with image J [67]. All germination, greening, and root growth assays were repeated a minimum of three times with at least two independent sets of matched seeds with consistently similar results (see Figures S5-S7 for additional replicates). All statistics were performed in either INSTAT or Prism (GraphPad Software -La Jolla, CA).

Promoter Analysis
Over representation analyses on nuclear encoded genes from the nf-yc triple were performed on genes misregulated #21.5 and $1.5 fold (p,.05) compared to Col-0. Details of the microarray experiment and public access to MIAME compliant data were previously reported [16]. Because the purpose of the current analysis was hypothesis generation (which were later tested), we relaxed the stringency of the microarray analysis here by removing the Benjamini-Hochberg false positive correction. For MEME denovo motif discovery, 2500 bp of upstream sequence of the misregulated genes was obtained using the bulk data retrieval tool at The Arabidopsis Information Resource (TAIR, www. arabidopsis.org). Upstream DNA sequences were fed to the MEME program and analyzed using the following parameters: motif width between 5 and 12 bp, any number of repetitions of motifs, and search for up to 6 motifs. All other options were left as default (http://meme.nbcr.net/meme4_5_0/intro.html, [42]). Motifs discovered through MEME analysis were then compared against known transcription factor binding sites from Jasper, Transfac and Uniprobe using the TOMTOM motif comparison tool [43]. Positions of CACGTG motifs relative to the transcriptional start were adapted from the MEME analysis. The same misregulated gene list was input into the Athena analysis suite using a 500 max bp upstream cutoff and, otherwise, default settings (http://www.bioinformatics2.wsu.edu/cgi-bin/Athena/ cgi/home.pl, [32]).

Protein-Protein Interaction Network
Individual protein-protein interaction networks were built for NF-YC3, NF-YC4 and NF-YC9 using GeneMANIA (http:// www.genemania.org, [44,45]). Selection criteria to develop the network map in GENEMANIA were predicted interactions and physical interactions with a 50-gene output. Default settings in GeneMANIA were used for network weighting. The individual network maps built in GeneMANIA for NF-YC3, NF-YC4 and NF-YC9 were downloaded as text files and combined to build a protein interactome in Cytoscape 2.8.0 (http://www.cytoscape. org, [46]). Data from Y2H library screens and directed Y2H assays done in the Holt lab and published interactions of NF-YC3, NF-YC4 and NF-YC9 [16] were manually added to the protein interactome in Cytoscape. See File S1 for a fully annotated list of genes (with AGI numbers and references) used to build the protein interactome.

RNA Isolation and qPCR
Total RNA was isolated from 20 mg of matched, stratified seeds using the E.Z.N.A. Plant RNA Kit per the manufacturer's instructions for difficult samples (Omega Biotek, Inc., Norcross, GA, CAT#R6827-01). Prior to RNA extraction, seeds were sown on beds of Whatman paper saturated with liquid Gamborg's B-5 media, cold stratified for 2 days, and exposed to 24 hours of continuous light at 22uC before harvesting. To completely remove genomic DNA, samples were DNAse treated on E.Z.N.A. RNA isolation columns (Omega Biotek, CAT# E1091). Quality and quantity of RNA samples were confirmed by spectrophotometry (Thermo Scientific, Waltham, Massachusetts, NanoDrop TM 1000). First-strand cDNA synthesis was performed using the Superscript III First-Strand Synthesis System (Invitrogen, Carlsbad, California, Cat#18080-051) with supplied oligo dT primers. qPCR was performed as previously described [14], except we used an Applied Biosystems Prism 7500 analyzer (Life Technologies, Carlsbad, California) and the Fermentas Maxima SYBR Green qPCR Master Mix (Fermentas, Glen Burnie, Maryland, Cat#K0222). For each genotype, we analyzed three independent, biological replicates in two separate experiments with similar results. All samples were normalized to the constitutively expressed gene At2g32170 as previously described [68]. Sample comparisons were performed by the 2 (2DDC T ) method [69], and errors (standard deviation) were computed as previously described [70]. qPCR primers for ABI3, ABI5, AIA, AIL, RAB18, and RD29B (AT3G24650, AT2G36270, AT1G64110, AT3G17520, AT5G66400 and AT5G52300, respectively) were designed using Primer3 in Genious Pro 5.1.4 (www.genious.com, [71]). Primer sequences for qPCR and cloning are in File S2.

DNA Manipulations
All target DNA fragments were generated by PCR using Pfu Ultra II (Agilent Technologies, Santa Clara CA, Cat#600670-51) and cloned into the Gateway TM entry vector pENTR/D-TOPO (Invitrogen, Carlsbad, California, Cat#45-0218). The full length coding regions of NF-YB2, NF-YB3, NF-YB10 NF-YC3, NF-YC4, NF-YC9, ABF1, ABF2, ABF3, ABF4, and HY5 (AT5G47640,  AT4G14540,  AT3G53340,  AT1G54830,  AT5G63470,  AT1G08970, AT1G49720, AT1G45249, AT4G34000, AT3G19290, and AT5G11260, respectively) were generated from Col-0 cDNA populations by standard methods. Partial clones of NF-YB2 and NF-YB10, as well as chimeric constructs between NF-YB2 and NF-YB10, were generated by PCR using Pfu Ultra II and cloned into the Gateway TM entry vector pENTR/D-TOPO. In partial clones that do not contain a native start codon an ATG was added in front of the region of interest. All constructs were sequenced and found to be identical to the expected sequences found at The Arabidopsis Information Resource database [72].

Gus Staining and Microscopy
All pNF-Y::GUS fusions used in seed coat expression assays were previously described [9]. Seed coats were dissected and GUS staining was performed as previously described [74,75]. Seeds coats and embryos were visualized using a Zeiss AxioImager Z1 m with Apotome (Zeiss -Oberkochen, Germany), using the DIC/BF filter and recorded using onboard Axiocam MRm and MRm5 camera. Images are compressions of a 3D Z-stack into a 2D image using the deconvolution and extended focus feature in the Axiovision software (version 4.8.1). Figure S1 Protein alignment of NF-YB2, NF-YB3 and NF-YB10. Full-length amino acid sequences for NF-YB2, NF-YB3 and NF-YB10 were aligned and visualized using ClustalW within the software package Geneious Pro5.6 (www.geneious. com). The junction site used to create chimeric constructs between NF-YB2 and NF-YB10 is annotated. (TIF) Figure S2 NF-YB and NF-YC mutants show no significant differences in root growth on ABA. Mutant lines were germinated and grown on B5 media for four days and then transferred to B5 media, B5 media +10 mM ABA, or B5 media +50 mM ABA and grown vertically for 7 days. Bars represent the mean primary root length (n $12 plants from 2 separate experiments). Error bars are 95% confidence intervals. No statistical significance between any samples on the same growth media was measured using ANOVA (p.0.05). (TIF) Figure S3 Public microarray data visualized by eFP browser show NF-YC3 and NF-YC9, but not NF-YB2 and NF-YB3 are expressed in seeds and during early germination. Absolute levels for NF-YC3, NF-YC4, NF-YC9, NF-YB2, and NF-YB3 were queried in the eFP browser with a signal threshold of 250 [78,79]. Note that the lack of NF-YC4 signal on public microarrays is likely due to problems with the probe (which is predicted to detect more than one gene) and not our GUS fusion. For example, according to public microarrays, NF-YC4 is not expressed in leaf tissues, although we have previously published mRNA and protein data (using a native antibody) showing this is incorrect and there is a clear genetic requirement for NF-YC4 in the leaf-initiated process of photoperiod-dependent flowering [16]. (TIF) Figure S4 Rescue of the nf-yc triple mutant early germination phenotype on ABA media. Seeds were plated on media supplemented with 0.5 mM ABA as previously described (approximately 30 seeds/replicate, three replicates per line). Percent germination is shown at 84 hrs. These rescue lines were previously described and used to show rescue of the late flowering phenotypes of nf-yc triple mutants [16]. Asterisks represent significant differences in Fisher's Exact Test comparing nf-yc triple mutants to all other lines (p,0.05). (TIF) Figure S5 NF-YC mutants show opposing phenotypes in response to ABA. Seeds were plated on media supplemented with 1 mM ABA as previously described and germination was scored every 12 hours. Each replicate of at least 30 seeds was independently graphed along with wild type controls for A) nf-yc3, B) nf-yc4, C) nf-yc9, D) nf-yc3/c4, E) nf-yc3/c9, F) nf-yc4/c9, and G) nf-yc triple. (TIF) Figure S6 NF-YC mutants show opposing phenotypes in response to ABA. Seeds were plated on media supplemented with 2 mM ABA as previously described and germination scored every 12 hours. Each replicate of at least 30 seeds was independently graphed along with wild type controls for A) nf-yc3, B) nf-yc4, C) nf-yc9, D) nf-yc3/c4, E) nf-yc3/c9, F) nf-yc4/c9, and G) nf-yc triple. File S1 List of genes that interact with NF-YC3, NF-YC4 and NF-YC9. AGI numbers are included and common names are given. Additionally the type of interaction (published physical interaction/predicted interaction) and the reference/source of information are given. The.xlsx file can be directly uploaded to Cytoscape (http://www.cytoscape.org/) to build a interactive protein interactome.