Characterization of DNA-Binding Specificity of 63 A. thaliana TFs.

We cloned ∼100 TFs as N-terminal Maltose Binding Protein (MBP)-tagged fusions for expression in Escherichia coli and analyzed their DNA-binding specificity by incubation of PBMs (10). We obtained specific DNA-binding sequences for 63 TFs (Fig. 1), representing 2.5–4.5% of the total complement of the TFs in A. thaliana. This dataset includes 25 families or subfamilies of TFs, 15 of them plant-specific (Fig. 1; SI Appendix, Tables S1 and S2; see also SI Appendix, Text S1 and Figs. S1–S27 for a detailed description of the proteins and sequences determined).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

DNA-binding specificity of plant TFs. Position weight matrix (PWM) representation of top-scoring 8-mers for the TFs indicated. Secondary and tertiary motifs were considered when they differed substantially from primary ones among the lists of top-scoring motifs or after reranking the motifs (Methods). TFs are grouped into families or subfamilies according to a previous classification (1).

Overall, for TFs for which there is information our data are consistent with available data on binding specificity of corresponding families, although some remarkable differences were appreciated. The DEHYDRATION RESPONSE ELEMENT BINDING (DREB) family members DREB2c and DEAR3 (DREB and EAR motif protein 3) recognized a GCC-like element (GCCGCC) with similar affinity than the expected DRE (RCCGAC; R: A or G) for this family (ref. 11 and Fig. 1; see below). DNA motif for WUSCHEL (WUS)-related homeobox WOX13 is partially compatible to that described for the related WUS (ref. 12; TTAATSS; S: G or C). GATA12 recognized the palindromic motif AGATCT, nearly identical to the consensus motif described for this family (13) but differing at a critical residue (WGATAR; W: A or T). The class I TCP16 yielded a binding motif slightly different from the other class I TCPs analyzed, TCP15 and TCP23 (Fig. 1), but matching a class II TCP-motif, similarly to that observed (14, 15).

The Auxin Response Factor (ARF) ETTIN recognized the DNA sequence TGTCGG, partially coincident with the canonical AuxRE (ref. 16; TGTCTC) but differing at the 3′-terminal dinucleotide. However, binding affinity of ETTIN to the AuxRE was notably lower than to the motif that we obtained (SI Appendix, Text S1), suggesting that DNA-binding specificity of the ARF family may be broader than initially suspected. The LOB domain protein LBD16 specifically recognized the palindromic sequence TCCGGA, partially differing from the core sequence described for other member of this family (ref. 17; SI Appendix, Text S1). Finally, the short internodes/stylish (SHI/STY) family member STY1 recognized a palindromic DNA sequence containing the core CTAG, differing from that proposed for this TF (18).

More interestingly, we determined DNA-binding specificity for several TFs belonging to classes for which there was little or no previous information. The GARP (GOLDEN2,ARR-B,Psr1) members KAN4 and KAN1 recognized similar sequences (Fig. 1), compatible with the motif proposed for KAN1 (19). GLK1 recognized a DNA element containing the palindromic sequence RGATATCY (Y: C or T), compatible to DNA sequences recognized by other GARP or Myb(Myeloblastosis)-related TFs (Fig. 1). TOE1 and TOE2, together with AP2, SMZ, and SNZ, belong to the same phylogenetic clade and they act redundantly in the repression of flowering (7, 20). Both proteins recognized similar sequences, representing a consensus DNA motif for this group of TFs (Fig. 1). Studies on the molecular mechanisms of the function of YABBY (YAB) members are controversial, because these proteins have been proposed to bind DNA either specifically or nonspecifically (21, 22). We analyzed YAB1 and YAB5 and obtained similar binding specificity to A/T-rich elements (Fig. 1), helping us to define a consensus binding-motif for this family as WATNATW. Similarly, the only protein studied so far from the REM class of B3 superfamily is VRN1, which interacts with DNA in a nonspecific manner (23). We obtained a DNA motif for a member of this family (REM1; ref. 24) containing the core sequence TGTAG, representing a cognate sequence for B3 proteins that differs from the consensus binding motifs described for other groups belonging to the same superfamily (Fig. 1).

Previous work to determine DNA-binding specificities of mouse TFs suggested that even closely related TFs have distinct DNA-binding profiles (3). Thus, although proteins with up to 67% amino acid sequence identity may share similar high-affinity binding sequences, they prefer different low-affinity sites (3). To determine whether our data allow differentiating specific DNA-binding patterns among closely related TFs, we analyzed with more detail DNA-binding specificities of members from two families of TFs, MYB and AP2/EREBP. As expected, we found a strong correlation between amino acid similarity and DNA-binding sequence preference. Structural subfamilies were clearly differentiated, because proteins belonging to the same subfamily showed similar DNA-recognition patterns, reproducing amino acid identities among the members of the same family (SI Appendix, Figs. S28 and S29). However, we could identify subtle differences in DNA-recognition patterns of proteins showing up to 79% amino acid identity, suggesting that, despite their similarity, different TFs have distinct DNA-binding profiles.

Remarkably, our analyses also revealed that an unexpectedly high number of TFs recognize secondary elements with similar or slightly lower affinities to their primary ones, which may explain the difficulty to derive consensus binding-sites from ChIP-seq data. Thirty-three of 63 proteins bound to DNA elements partial or completely differing from their primary motifs (Fig. 1). In most cases, secondary motifs represented sequence variants of their corresponding primary elements. In the case of MYBs, two DNA-binding motifs have been described (MBSI and MBSII; ref. 25). Interestingly, only MYB52 recognized both MBSI and MBSII, whereas the other MYBs tested only recognized variants of the MBSII (MBSII and MBSIIG; ref. 25). In the case of B type-Arabidopsis response regulator (ARRs), secondary motifs found corresponded to DNA elements described for other members of this family (26). Similarly, ERFs(ethylene response factors), SPL1(squamosa promoter binding protein like), bZIPs(basic leucine zipper), ANAC55(Arabidopsis NAM, ATAF and CUC), and REM1(reproductive meristem) recognized secondary DNA elements, some of them previously proposed (27⇓⇓–30), partially differing from their corresponding primary ones. AT-hook containing TFs (AHLs) yielded several A/T-rich motifs with similar affinities, consistent with their role as modifiers of the architecture of DNA through binding to A/T-rich stretches in the minor groove of DNA (30). Both heat shock factors tested (HSFB2A and HSFC1) recognized identical motifs, representing inverted repeats of the trinucleotide GAA. Actually, both DNA elements may be considered as overlapping halves of a longer motif containing three GAA inverted repeats (TTCNNGAANNTTC), as described for several eukaryotic HSFs (31).

Several AP2/EREBP TFs belonging to the DREB subfamily were recognized with high-affinity GCC and DRE motifs (Fig. 1). These results indicate that DNA binding of DREBs may be more complex than initially suspected, where some DREBs posses a broader range of DNA-binding by recognizing DRE- and GCC-related elements, similarly to that observed for TINY (32). Dof5.7 recognized a canonical binding motif containing the core sequence AAAG (33), but evaluation of secondary motifs yielded a different cognate one. To date, no other Dof protein has been shown to recognize this element, and it may represent an alternative DNA target sequence. The fact that approximately half of the TFs recognized secondary motifs, in some cases completely unrelated to the primary element, suggests that recognition of cis-regulatory elements by TFs may be much more complex that initially anticipated.

In line with our results, previous studies with mouse TFs showed that approximately half of them also recognized secondary elements, which were annotated into four different classes (3). In our case, we find more realistic a different classification in which the secondary elements fall into one of the following categories: (i) Secondary elements involving variations or extensions on the primary motif, including most (22) of the TFs analyzed. (ii) Secondary motifs reflecting overlapping halves of longer motifs (HSFB2A and HSFC1). (iii) Secondary motifs corresponding to inverted repeats of a core sequence (ETTIN and STY1). (iv) Multiple model, as in ref. 3 (GLK1). (v) “Truly” secondary elements found for TOE1, TOE2, WOX13, Dof5.7, and AHLs. It is worth noting that this is an arbitrary classification, and even secondary motifs derived from variations of the primary ones (class i) may retain biological relevance on their own with important implications in the recognition patterns of the TFs, as described above for DREB proteins.

Binding Motifs Are Associated with DNase I Hypersensitive Regions.

Binding of TFs to regulatory DNA regions triggers displacement of nucleosomes and chromatin remodeling, resulting in DNase I hypersensitivity (34⇓–36). The ENCODE Project includes mapping of DNase I hypersensitivity sites (DHS) from more than 160 human cell types and tissues, providing an almost saturated mapping of regulatory cis-elements (34⇓–36). In A. thaliana, although the information is scarce, mapping of DHS from leaf and floral tissues has been useful for the identification of cis-regulatory elements (37). We used these datasets for mapping the TF-binding motifs obtained in our PBM assays. Overall, we observed a higher density of TF-binding motifs in DHS fragments from both tissues than their corresponding negative controls (Fig. 2A). In addition, we observed particular tissue specificities for some motifs. For instance, density of the HD-ZIP protein ICU4-binding motif was higher in the leaf dataset than in the flower one (Fig. 2B), consistent with its role in leaf patterning (38). By contrast, binding sites for the similar ATHB51 were more abundant in DHS derived from floral tissues, indicative of the role of this TF in flower development (39). Taken together, these results indicate that binding sequences identified in PBM assays correlate well with genomic regions exposed to DNase I and, thus, with cis-regulatory elements. Moreover, this analysis highlights that combination of TF-binding motifs identified in vitro and genome-wide characterization of DHS in multiple tissues and/or experimental conditions may be an alternative for mapping the TF-binding sites in vivo.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Binding motifs are associated with DNase I hypersensitive regions. (A) Density plots of TF-binding sites in DNase I hypersensitive sites (DHS) from leaf (Left) and flower (Right) described in ref. 40. Density of TF-binding sites (red diamonds and line) is measured as the number of sites at each position per number of fragments with that length, along DHS fragments (1 kb showed, centered at the middle nucleotide of each DHS). In blue is represented the average density for 100 randomized PWMs. (B) Density plots of binding sites recognized by HD-ZIP proteins ICU4 (Upper) and ATHB51 (Lower).

DNA-Binding Sites Have Biological Relevance.

As a validation of DNA motifs, we searched in databases for transcriptomic assays of mutant or overexpressing genotypes directly involving the TFs under study. We selected the gene sets deregulated in each of these experimental conditions and searched at their promoters the DNA elements identified for the TF under study. We observed strong overrepresentation of DNA motifs determined in vitro with sequences in the promoters of deregulated genes in mutant or overexpressing genotypes. In particular, we observed overrepresentation of at least one DNA motif for all of the proteins analyzed by these means (14 in total) and 16 of 20 DNA elements (80%). In some cases (e.g., MYB46 and TGA2), overrepresentation was observed for both primary and secondary motifs obtained in vitro, suggesting their role as cis-regulatory sequences (SI Appendix, Fig. S30A).

We extended this analysis to transcriptomic data indirectly related with TFs tested, either involving structurally similar TFs (SI Appendix, Fig. S30B), supposed to recognize similar DNA elements, or growth conditions in which the TFs are involved (SI Appendix, Fig. S30C). In the case of structurally similar TFs, we observed overrepresentation of 30 of 40 (75%) corresponding to 25 different proteins (SI Appendix, Fig. S30B). It is worth noting that this correlation does not necessarily involve functional redundancy, but rather similarity in high-affinity binding motifs of structurally related TFs. Similarly, 22 of 23 DNA elements (95.6%) corresponding to 11 TFs were enriched in the promoters of genes differentially expressed in the experimental conditions in which the TFs are involved (SI Appendix, Fig. S30C).

A further confirmation of the relevance of TF-binding motifs in vivo can be obtained by analyzing enrichment within bound genomic fragments identified by ChIP-seq. In addition to the correlation between in vitro and in vivo data that we observed for PIF5-binding motifs (8), we found a fourfold enrichment of PIF3-binding motifs within ChIP genomic fragments compared with a negative control (SI Appendix, Fig. S31), evidencing the good correlation between in vitro and in vivo data.

Validity of Coregulation Data and Binding Motifs to Define Putative Targets.

We further evaluated the biological relevance of DNA motifs obtained in vitro in the context of the TF-dependent gene regulation. We hypothesized that, in the case of transcriptional activators, TF-coding genes should have similar expression patterns than their corresponding targets. By contrast, in the case of repressors, TF-coding genes and targets should display opposite expression patterns. We obtained the lists of genes positively and negatively coregulated with the genes encoding the TFs, and searched at their promoters the DNA elements identified for the TF under study. We also included data relating previously characterized proteins (8, 10, 41). Remarkably, we obtained significant overrepresentation of at least one DNA element for all of the TFs analyzed (69 in total; Fig. 3). When all of the DNA elements were considered, 99 of 122 (81%) were significantly overrepresented (Fig. 3). When considering globally the sets of genes coregulated and deregulated in mutant or overexpressing genotypes, 101 of 122 (82.8%) of the elements had a biological role in the context of gene regulation. These numbers support the extraordinary accuracy of the DNA targets identified in our PBM experiments and underscores the surprisingly high correlation between DNA-binding sequences determined in vitro with cis-regulatory elements in vivo. This correlation has been already pointed out in other biological systems (42, 43), and our work shows that it is also the case in plants.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Evaluation of biological relevance of DNA-motifs from coregulation information. Lists of coregulated genes with the TF-coding genes were scanned for the presence of DNA-motifs obtained in vitro at their promoter regions (1 kb). Frequencies (in %) of genes positively (blue bars) and negatively (red bars) coregulated genes containing the indicated DNA elements are shown. Proportions of genes represented in the ATH1 microarray containing the corresponding elements, and thus representing a random distribution, are represented as green bars. Asterisks indicate the degree of statistical significance in the differences of the proportions indicated as follows: *P < 0.005; **P < 0.05.

Additional information extracted from this analysis refers to the molecular activity of the TFs studied. Binding motifs for 21 TFs were enriched in the promoters of negatively coregulated genes (Fig. 3), and some of them confirmed in transcriptomic profiles (SI Appendix, Fig. S30), suggesting that they act as transcriptional repressors. Of these 21 TFs, 16 were previously proposed to act as repressors, whereas At5g28300, ATHB51, AHL12, LBD16, and Dof5.7 were not (SI Appendix, Table S3). We analyzed with more detail the transcriptional properties of these proteins in kinetic assays of luciferase (LUC) activity transiently expressed in Nicotiana benthamiana leaves. We cloned the promoter region of one putative target for each TF, selected among negatively coregulated genes containing their corresponding cognate motif. Reporter constructs were coinfiltrated with their corresponding effectors expressing the TF under the control of the constitutive 35S promoter. The results show that four of the five TFs (At5g28300, AHL12, Dof5.7, and LBD16) suppress the expression of their corresponding target promoters, confirming that these four TFs behave as repressors (SI Appendix, Fig. S32). In the case of ATHB51, the lack of activity as activator or repressor suggests that we might have failed in the selection of the correct target promoter (SI Appendix, Fig. S32). These results confirm the predictive potential of analyzing the sets of coregulated genes for inferring the transcriptional activity of TFs, given that, of 21 predicted, 20 can actually act as repressors.

ChIP experiments have demonstrated that some TFs may recognize DNA sequences at downstream regions of target genes (7⇓–9). We then evaluated the presence of DNA motifs in vitro in downstream regions of coregulated genes. Overall, sequence enrichment was lower than in upstream genomic regions, and observed significant enrichment for 52 of 122 motifs (42.6%), corresponding to 41 of 69 TFs (59.4%) (SI Appendix, Fig. S33). When considering the genes coregulated and deregulated in mutant or overexpressing genotypes, we obtained 56 of 122 motifs (45.9%) corresponding to 45 of 69 TFs (65.2%) significantly enriched at downstream regions (SI Appendix, Figs. S33 and S34). This enrichment seems to be more conspicuous in the case of transcriptional repressors. Of 21 putative repressors recognizing 36 sites inferred from the analysis of the promoter regions, 25 motifs (69.4%) corresponding to 18 TFs (85.7%) showed significant enrichment at downstream regions of putative targets (SI Appendix, Fig. S35). This finding suggests that spatial restrictions for repressors are less strict than for activators, where binding to promoters is preferred.

We next mapped DNA-binding motifs along upstream and downstream regions (1 kb) of coregulated genes. We observed a higher frequency of DNA-binding sites corresponding to transcriptional activators in the promoters of genes positively coregulated near the transcription start site in relation to negatively coregulated or the overall distribution of binding sites (Fig. 4). An analogous distribution was obtained for binding sites corresponding to transcriptional repressors, but relating in this case to negatively coregulated genes (Fig. 4). Consistent with our previous analysis, a higher frequency of binding sites was also observed at downstream regions of coregulated genes, although at a lower degree than in promoters (Fig. 4). However, we observed some differences between transcriptional activators and repressors. Whereas binding sites corresponding to activators were enriched at downstream regions 500 bp and further the translation stop codon, binding sites for repressors were particularly enriched near the stop codon, most likely laying at 3′-UTRs (Fig. 4). These results indicate that TF-binding sites corresponding to repressors may be affecting the transcription of target genes, rather than acting as cis-regulatory elements. Mapping genome locations associated to transcriptional corepressor complexes will help to unravel the molecular mechanisms underlying transcriptional repression.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Mapping of DNA-motifs along upstream and downstream regions. Frequency plots of the average distribution of binding sites at upstream and downstream regions (1 kb) of the genes represented in the ATH1 microarray (green lines), positively coregulated (blues lines), and negatively coregulated (red lines). “0” represents the transcriptional start site in upstream plots, and to stop codon in downstream ones. Plots on the top correspond to binding sites for activators and bottom plots to transcriptional repressors.

Data presented above indicate that the genes coregulated with TF-coding genes are enriched in their corresponding targets, identified as subsets of genes containing cognate motifs at their promoters. To check whether these subsets are functionally related with the TF under study, we analyzed the lists of Gene Ontology (GO) terms significantly enriched in the list of coregulated genes (CRG) and in the subsets of coregulated genes with DNA element (CRGE) for each TF (SI Appendix, Fig. S36). We observed a higher enrichment in GO terms related with the function of the TF in CRGE, indicating that these subsets containing cognate DNA motifs at their promoters may represent putative target genes of the TFs.

In addition to help identifying the gene targets of a TF, our work also provides a way to explore how a particular set of coexpressed genes is regulated by scanning their promoters with PBM matrices and identifying the TFs responsible for their regulation among those coexpressed that recognize these sequences. Thus, in summary, the description of TF-binding motifs in vitro and evaluation of their potential as cis-elements represent a quick strategy, complementary to ChIP-based methodologies, to define regulatory networks and to help unravel the transcriptional regulatory code.