Cistrome and Epicistrome Features Shape the Regulatory DNA Landscape

Abstract

The cistrome is the complete set of transcription factor (TF) binding sites (cis-elements) in an organism, while an epicistrome incorporates tissue-specific DNA chemical modifications and TF-specific chemical sensitivities into these binding profiles. Robust methods to construct comprehensive cistrome and epicistrome maps are critical for elucidating complex transcriptional networks that underlie growth, behavior, and disease. Here, we describe DNA affinity purification sequencing (DAP-seq), a high-throughput TF binding site discovery method that interrogates genomic DNA with in-vitro-expressed TFs. Using DAP-seq, we defined the Arabidopsis cistrome by resolving motifs and peaks for 529 TFs. Because genomic DNA used in DAP-seq retains 5-methylcytosines, we determined that >75% (248/327) of Arabidopsis TFs surveyed were methylation sensitive, a property that strongly impacts the epicistrome landscape. DAP-seq datasets also yielded insight into the biology and binding site architecture of numerous TFs, demonstrating the value of DAP-seq for cost-effective cistromic and epicistromic annotation in any organism.

Figure 1. Genome-wide TFBS Discovery by DAP-Seq

(A) Preparation of DAP- and ampDAP-seq libraries. (B) Expression and capture of affinity-tagged TFs. (C) gDNA is bound to immobilized TFs, eluted, and sequenced. (D) ABI5 DAP- and ChIP-seq peaks at a known regulatory element in the ABI5 promoter. (E) Motifs derived from DAP- and ChIP-seq match a published ABI5 motif (Weirauch et al., 2014). See also Table S2.

Figure 2. A Genome-wide Atlas of Arabidopsis TFBS Motifs and Binding Locations

(A) Web portal of TF binding motifs from 529 DAP-seq and 343 ampDAP-seq experiments. (B) Sample screen shot of genome browser with DAP-seq peaks for selected TFs. (C) Overlap between TFs from the DAP-seq, CIS-BP PBM (Weirauch et al., 2014), and PBM datasets (Franco-Zorrilla et al., 2014). (D) Number of informative bases (information content ≥ 0.8 bits) in DAP-seq and PBM motifs. (E) Number of TFBS predicted by peaks (DAP-seq) or motifs (DAP-seq, CIS-BP, and PBM [Franco-Zorrilla et al., 2014]). See also Figure S1 and Tables S1 and S2.

Figure 3. The Global Diversity of Arabidopsis TF Motifs

(A) bZIP family motifs from DAP-seq clustered by motif similarity. (B) 57 TF motifs with GC-rich clusters in blue and AT-rich clusters in red. (C) Multidimensional scaling plot of the full set of 529 TFs highlighting the 57 representative motifs. See also Figure S2 and Table S1.

Figure 4. Critical Biological Processes Are Enriched in DAP-Seq Target Genes

Target genes predicted for 44 diverse TFs (subset of the 57 representatives) are enriched for functional terms associated with basic cellular properties. See also Figure S3 and Table S1.

Figure 5. Concordance of In Vitro and In Vivo Binding Sites for Multiple TF Families

(A) Percent overlap of ChIP-seq peaks with DAP-seq peaks (blue), which increase for peaks associated with higher motif scores (red). (B) Empirical cumulative distribution of motif scores shows shared ChIP- and DAP-seq peaks (DAP-ChIP) contain higher scoring motifs than do ChIP-only peaks, in which motif scores are similar to the motifs not bound in either assay. (C) Percent DAP-seq peaks in DHS that overlap with ChIP-seq peaks. (D) Using DHS data from multiple sources ([a] Sullivan et al., 2014; [b] Zhang et al., 2012) increases coverage of DAP-seq peaks. (E) Precision (y axis) and recall (x axis) curve shows DAP-seq read depth (signal) predicts in vivo ABI5 binding sites better than mapping DAP-seq and PBM-derived motifs to genome, for all motifs (left) and motifs in DHS (right). (F) By area under the precision-recall (PR) curve as in (E), all ChIP-seq datasets are most accurately predicted by DAP-seq read depth. PBM_D, motif directly determined by PBM. PBM_I, motif inferred by PBM based on DNA binding domain similarity. See also Figure S4.

Figure 6. The ARF Family Preferentially Binds to Phased Motif Clusters that Are Enriched in Target Gene Promoters

(A) Three possible orientations of an ARF motif repeat. (B) ARF homodimers could be stabilized at a DR by an interaction of the III/IV domain (top) (Nanao et al., 2014) and at an ER by the dimerization domain (bottom) (Boer et al., 2014). (C) Relative frequencies of DAP-seq peaks at DR, ER, and IR pairs for Arabidopsis (AtARF5 and AtARF2) and maize (ARF5/ZmARF29) proteins interrogating Arabidopsis (At-gDNA) or maize (Zm-gDNA) DAP-seq libraries. (D) A cluster of 13 phased TGTC sites (red ticks) in the promoter of the ARF5 target IAA5. Black ticks are non-phased TGTC sites. (E) DAP-seq signal at the TSS (x axis) of ARF5 direct target genes, non-auxin-responsive background genes, and all genes. See also Figure S5.

Figure 7. Motif Methylation Impacts Binding For 76% of TFs Surveyed

(A) Inset: mC-all regions contain dense methylation in all cytosine contexts. Left: binding fold change (FC) at motifs containing relative to motifs neighboring (within 200 bp) an mC-all site. Right: relative ampDAP-seq binding at the same motifs. Gray boxes indicate TFs with too few (<25) methylated motifs to score or a failed experiment. (B) Inset: an isolated mCG-only site. Left: binding FC at motifs containing relative to motifs neighboring (within 200 bp) an mCG-only site. Right: relative ampDAP-seq binding at the same motifs. (C) TF methylation sensitivity is correlated with cytosine content of the motif, defined as the informative content (IC) of cytosines, divided by total IC of the motif. (D) Cytosine content (left) and informative CG, CHG, and CHH content for each motif (right) of TFs in (A and B). (E) Genome browser showing DAP-, ampDAP-, and ChIP-seq peaks at methylated and unmethylated ABI5 motifs. (F) Waterfall plot of log2 relative binding at methylated motifs for 349 TFs in DAP-seq and 219 TFs in ampDAP sets. In total, 248 of 327 TFs (76%) that had sufficient motif instances for quantification were found to be methylation-sensitive. See also Figure S6 and Table S3.