Evidence for multiple ARF-Aux/IAA protein interactions in vivo using ChIP-seq
Two families of proteins, ARFs and Aux/IAAs, play pivotal roles in auxin responses in plants (13, 32, 55-57). ARFs are transcription factors that have a DNA binding domain while the auxin co-receptor Aux/IAAs have no DNA binding domain, but are thought to interact with ARFs and other transcription factors (58). To investigate the direct regulation of the transcriptional response to auxin by these proteins in vivo, we generated new ChIP-seq data (including ARF5, ARF7, ARF10, and IAA19 and IAA12) (49) and analyzed all publicly available genome-wide binding data (ChIP-seq and DAP-seq). ARF7 and IAA19 ChIP-seq datasets were generated using native expression, epitope-tagged lines constructed using the recombineering technique (44). These results were corroborated by ChIP-seq experiments using an inducible expression system (Fig. 1a, Supplementary data1). The numbers of ChIP-seq binding sites for different factors under different conditions ranged from a few hundred up to 7-8 thousand (Fig. 1b, Supplementary data 1).
We assessed the global genome-wide similarity among ARF/IAA binding sites by clustering the top 10% most strongly enriched ChIP-seq peaks across all conditions. This analysis revealed two large clusters of binding profiles (Fig. 1c). ARF10 and gain-of-function versions of IAA19 and IAA12 bound to similar sites genome-wide, as indicated by the similarity of their binding profiles by Pearson correlation. The in vitro ARF5 DAP-seq dataset stood alone in correlation analyses, indicating it was distinct from the in vivo experiments, though it did bind most of the auxin response genes (49). Gene Ontology (GO) analysis using the ChIP-seq targets that were associated with the common peaks shared by the two clusters (Fig. 1c) were enriched for auxin response signaling pathway and transcriptional responses genes, indicating that both ARFs and Aux/IAAs were associated with the auxin response genes in vivo (Fig. 1d). The auxin primary response gene families (Aux/IAAs, SUARs, GH3s) were all found as direct ChIP-seq binding targets of ARFs and Aux/IAAs (Supplementary data 1) (59-62).
ARF family proteins were previously classified as activators (5 members) and repressors (18 members) (32, 34, 40). Most studies have focused on the activator ARFs and the only evidence that repressor ARFs function to repress gene expression is provided by a transient repression of a gene reported in a protoplast-based assay (36). To further examine repressor ARFs function, we performed ChIP-seq studies using putative repressor ARF (ARF10) to identify its target genes. To achieve this, we used an ARF10 overexpression line, generated by expressing a miR160 resistant ARF10 (mARF10) under the control of a dexamethasone (Dex)-inducible promoter (63). This was necessary because ARF10 transcripts are normally rapidly degraded by miR160, preventing the use of native expression constructs. DNA binding profiles of three activator ARFs (ARF5, 6, 7) were compared to that of mARF10 to determine if they target a common set of genes; 230 common genes were identified (Fig. 1e). In all four cases, both the activator and repressor ARFs had higher peak scores for the common binding sites than for their unique sites (Fig. 1f). Gene Ontology analysis indicated that the 230 common targets were enriched in auxin response genes (GO terms; response to auxin, corrected p-value= 1.99E-12, auxin-activated signaling pathway, corrected p-value= 7.05E-12), included in this list were: ARF19, Aux/IAAs (1, 2, 3, 4, 5, 7, 9, 13, 19), GH3 (3.3, 3.5, 3.6), PINs (3, 4, 7) (Supplementary Fig.1a and Supplementary data1). Taken together, these results provide further evidence that the auxin response mechanism involves the binding of target genes by multiple factors by both activator and repressor ARFs, enabling regulation of a common set of genes.
Aux/IAA- and ARFs-bound genomic loci are highly correlated
The Aux/IAA family, one of three primary auxin response gene families (SAURs, Aux/IAAs, and GH3s) (57, 58, 64-66), has no proven direct DNA binding ability (33). Aux/IAA proteins share a dimerization domain (domain III/IV, also called PB1 domain) with ARFs (23). Thus, Aux/IAA proteins are assumed to be associated with chromatin through their partner ARFs. However, there is currently no genome-wide experimental evidence that supports this hypothesis. To determine if both Aux/IAA proteins and the ARFs bind to the same genomic loci, ChIP-seq experiments were performed on a yellow fluorescent protein (YPet)-tagged IAA19 native expression line (generated by recombineering, IAA19:YPet) using a modified cross-linking protocol that employed both long-arm and standard cross-linkers (Fig. 2). These crosslinkers enabled the recovery of DNA associated with IAA19, which is not directly bound to DNA. The binding sites of IAA19:YPet were enriched in the promoter region around transcriptional start sites (TSSs) suggesting Aux/IAAs play a role in transcriptional responses in vivo (Fig. 2a). Specifically, approximately 70% of the binding sites of IAA19:YPet were associated with promoter regions (Supplementary Fig 2a). IAA19:YPet and ARF6 binding sites were significantly associated compared to random genomic regions (with 10,000 shuffles, p=0.0001. Fig. 2 c-d overlaps between IAA19:YPet and other ARFs see Supplementary Fig. 2b-c). These results suggest that the peaks of IAA19:Ypet ChIP-seq are likely to be genomic loci where one (or more) ARFs also bind (Supplementary data 1).
Gene Ontology analyses on the gene targets of IAA19:Ypet ChIP-seq peaks revealed they were enriched with auxin response genes (Fig. 2e-f). The top GO terms were SAUR proteins (p-value=1.2E-25), Aux/IAA proteins (p-value=4.3E-10), Aux/IAA-ARF dimerization (p-value=2.9E-10), and auxin signaling pathway (p-value=2.9E-10). To further confirm these findings, ChIP-seq was carried out using inducible Green Fluorescent Protein (GFP) tagged gain-of-function IAA19 (msg2-1) lines and IAA12 (bdl). IAA19 (msg2-1) and IAA12 (bdl) have a point mutation (proline to serine (PtoS)) in the degron motif (Five conserved amino acids (GWPPV)) of their domain II that prevents them from being degraded by the proteasome (67, 68). Both had DNA binding motifs that were like that of ARFs (Fig. 4a, Supplementary data2). Their gene targets were enriched auxin response genes like IAA19:Ypet except for stronger peaks and more binding sites due to high protein stability compared to IAA19:Ypet (Supplementary data1). The ChIP-seq results of IAA19 (msg2-1) and IAA12 (bdl) provided further evidence for the conclusion that Aux/IAA is bound to genomic loci, (Fig. 1a, Supplementary data1). Altogether, results provide strong evidence that auxin responses are initiated at the genomic loci where both Aux/IAAs and ARFs co-localize. Moreover, these data support a model that degradation of Aux/IAA initiates on the chromatin, which might allow rapid transcriptional responses to the hormone.
ARF10 and IAA19 function together to regulate ABA sensitivity
While nine AUX/IAA proteins were identified in the top 20% of IAA19:YPet gene targets, there was only one ARF protein (ARF10) found, suggesting possible autoregulation of this specific ARF (ARF10) and its corresponding IAA (IAA19). Thus, we also carried out ChIP-seq using an inducible mutant of ARF10. Interestingly, a strong similarity between the motifs underlying IAA19-bound peaks and ARF10-bound peaks was observed (Supplementary Data2). In fact, IAA19(msg2-1) shared many more targets with mARF10 than it did with other ARFs (such as ARF5/MP) (Supplementary data1).
Compared to uninduced mARF10 (Fig. 3a, LS), inducible expression of mARF10 in plants produces a dwarf phenotype (Fig. 3a, LS + 10 µM Dex). Expression of IAA19 (msg2-1) (stabilized IAA19) completely rescued the phenotype of mARF10 plants (Fig. 3a, IAA19(msg2-1) mARF10 LS + 10 µM Dex), suggesting a direct interaction of ARF10 and IAA19 in vivo when these two proteins are ectopically expressed. Importantly, the measured expression of GFP::mARF10 excluded the possibility that the introduction of the second transgenic IAA19(msg2-1) caused gene silencing of GFP::mARF10 (Supplementary Figure 4). Moreover, gene expression studies using RT-PCR indicated that both IAA19 (msg2-1) and ARF10 were expressed in IAA19 (msg2-1) mARF10 double transgenic plants (Fig. 3a).
It has been reported previously that transgenic expression of ARF10 alters ABA sensitivity, suggesting ARF10 contributes to the regulation of ABA signaling (63). We examined our ChIP-seq data to understand how this might occur. Compared to other ARF datasets, the gene targets of mARF10 binding sites overlapped with that of IAA19 (msg2-1) (Supplementary Data1), including many genes that participate in ABA pathways from a previous study (Fig. 3b) (69). Close inspection of the mARF10 and IAA19(msg2-1) ChIP-seq datasets, revealed that ABA receptors, kinases, ligases, and other ABA metabolic categories were bound by both proteins (Fig. 3c) (69). Taken together, these data reveal a new ARF10-IAA19 module that functions in the regulation of plant sensitivity to ABA through auxin-ABA cross-regulation.
To test if ARF10-IAA19 functions in conjunction with the ABA pathway in planta, mutants and transgenic lines with altered Aux/IAA and ARF function including triple mutants of iaa19/iaa5/iaa6, inducible pDEX:GFP:IAA19(msg2-1), and pDEX:GFP:mARF10 lines were examined in an ABA response assay (Fig. 3d, Supplementary Figure 5). These studies showed that inducible mARF10 plants were more sensitive to the plant hormone ABA than wild-type plants, consistent with results from a previous study (63). Unlike the loss-of-function of the clade of ARF10, 16, 17 by p35S:miR160 that targets ARF10, 16, 17 (63), the iaa5/6/19 triple mutant was sensitive to the ABA treatment. On the other hand, the gain-of-function IAA19 (msg2-1) plants were resistant to ABA treatment, which was very similar to the resistance level of p35S:Mir160a to ABA treatment observed in a previous study (63) (Fig. 3d).
A motif spacing model for ARF binding to targeted loci
From inspection of both in vitro and in vivo datasets, the overall binding motif of ARFs can be summarized as K(T/G)GTCBB(B(T/G/C)). Specifically, although all ARF motifs have a GTC core sequence, the sequences flanking the core are degenerate (Fig. 4a, Supplementary Data 2). The first base was less degenerate, being T for most ARFs but could be either T or G in the ARF5 motif. On the other hand, the fifth and sixth bases were found to be B (T or G or C) but never A in all available datasets published and in this study. Thus, the appearance of A at either fifth or sixth or both positions was used to decide the p-value cutoff in the motif scanning within ChIP-seq peaks (p-value< 0.0014 was used).
Besides the classical ARF motifs shown, another interesting and important observation was that other TF binding motifs, such as TCP, bZIP, and PIF, were identified from the motif analyses of ARF and IAA binding sites (Supplementary data 2). For example, the TCP motif TGGTCC was identified among multiple datasets (p-value ranging from 1.1E-10 to 3.3E-70). For this reason, a ChIP-seq dataset for TCP 3 (AT1G53230) was generated to compare our motif analysis side-by-side. A motif TGGTCC (p-value=3.9E-580) was identified for TCP3 and validated the motif analysis (Supplementary data 2).
With the structure of the ARF DNA binding domain and in vitro binding data available for ARFs (48-50), motif spacing models have been proposed to explain the possible cooperative binding of ARFs. These include the clipper model and the helix turn spacing model (48-50). When we examined the spacing models using our in vivo data, we used motif orientations (direct repeats (DR), everted repeats(ER), and inverted repeats (IR)) similar to that previously reported but the representative KGTCBB motif in our analyses was longer than TGTC that was previously used (Fig. 4b) (47-49). Thus, the spacing number (n) is defined as the number of bases between two adjacent motifs and no motif bases were included (Fig. 4b). If two motifs overlap, n can be -1, -2 in DR (Fig.4b), in ER. However, IR cannot have negative spacing (n>=0).
The in vivo data revealed a few important and novel discoveries (Fig. 4c). First. The direct repeats, everted repeats, and inverted repeats (DR, ER, IR) did not strictly follow the rule of appearance every helix turns (10.5 bps) as reported using in vitro binding assay, suggesting intrinsic differences between in vitro and in vivo experiments since our analysis could repeat the result using in vitro dataset (Fig.4c ARF5DAP) (49). Second, the most striking feature was the negative DR spacing (n<=0) which was not previously reported. Third, peaks with 2 motifs had higher peak scores than peaks with just 1 motif (Wilcoxon rank-sum test, adjusted p-value < 0.01), while peaks with 3 or 4 motifs had a higher peak score than peaks with just 2 motifs (Wilcoxon rank-sum test, adjusted p-value < 0.01) (Supplementary Figure 3). Thus, the number of motifs within the promoter of targets was positively correlated to the binding of ARFs to their targets.