Discovery of directional and nondirectional pioneer transcription factors by modeling DNase profile magnitude and shape

Abstract

We describe protein interaction quantitation (PIQ), a computational method for modeling the magnitude and shape of genome-wide DNase I hypersensitivity profiles to identify transcription factor (TF) binding sites. Through the use of machine-learning techniques, PIQ identified binding sites for >700 TFs from one DNase I hypersensitivity analysis followed by sequencing (DNase-seq) experiment with accuracy comparable to that of chromatin immunoprecipitation followed by sequencing (ChIP-seq). We applied PIQ to analyze DNase-seq data from mouse embryonic stem cells differentiating into prepancreatic and intestinal endoderm. We identified 120 and experimentally validated eight 'pioneer' TF families that dynamically open chromatin. Four pioneer TF families only opened chromatin in one direction from their motifs. Furthermore, we identified 'settler' TFs whose genomic binding is principally governed by proximity to open chromatin. Our results support a model of hierarchical TF binding in which directional and nondirectional pioneer activity shapes the chromatin landscape for population by settler TFs.

Figure 1

Accurate detection of dynamic TF binding using DNase-seq and PIQ. Schematic outlining the PIQ algorithm. See text and Supplementary Information for details.

Figure 2

Benchmarking PIQ. (a) Comparison of AUC values (the probability of correctly ranking a bound TF site above an unbound one) comparing PIQ versus ChIP-seq (X-axis) and DGF/CENTIPEDE versus ChIP-seq (Y-axis) for 303 matched ChIP-seq experiments in K562 cells. The Y-axis value plots the higher AUC value between DGF and CENTIPEDE for each experiment. (b) ROC curves (which show the tradeoff between true positives to false positives as the cutoff for defining what is bound is varied) comparing mESC-stage PIQ binding calls for the TFs Ctcf, c-Myc and Esrrb against matched ChIP-seq binding calls. To calculate ROC curves, we ranked all above-threshold genomic motif instances for each TF according to their PWM motif strength (purple), total adjacent DNase hypersensitivity in a 400 bp window (red) or the per-site binding score given by PIQ (black). Plots compare true positives (Y-axis) to false positives (X-axis) at progressively lower ranked sites. Inset plot displays average, minimum and maximum AUC values for six mESC-stage PIQ versus ChIP-seq comparisons.

Figure 3

Systematic identification of pioneer TFs. (a) Flow chart outlining mESC-derived populations used for dynamic DNase-seq analysis. (b) Differences in PIQ-detected binding sites for eight selected TFs with strong microarray expression values in mESC (green), mesendoderm (blue) and PancE (red). For each TF at each stage, PIQ calculates a score representing the overall number and strength of binding sites, plotted in natural log PIQ binding strength units normalized to mesendoderm values. (c) Pioneer index log odds scores for all PIQ motifs. (d) Chromatin opening index log odds scores for all PIQ motifs. (e) Social index scores for all PIQ motifs. Scores of selected pioneer and non-pioneer TFs in A-C are noted. (f) Schematic of modular Tol2 transposon-based pioneer reporter system to test pioneer and non-pioneer motifs for chromatin opening ability. Chromatin openness is read out by the level of RA-induced RAR:RXR DNA binding and consequent GFP transcriptional activation, as measured by flow cytometric fluorescence. (g) Average increase in flow cytometric fluorescence after RA addition for 18 pioneer reporter lines grouped as predicted pioneer (red) and non-pioneer (blue) TFs, normalized to RA-induced GFP of the control reporter line. Error bars indicate SEM, and dotted line represents 99% prediction interval based on control RA-induced GFP, indicating lines with RA-induced GFP out of the predicted control range. Predicted pioneers as a group have significantly higher average RA-induced GFP than predicted non-pioneers. n=4, P<0.01 in t-test.

Figure 4

Asymmetrical chromatin opening by directional pioneers. (a) Per-base chromatin opening index log odds scores, which represent expected local increase in hypersensitivity induced by TF binding at all above-threshold genomic motifs for Creb1, Klf7, NFYA, and Zfp161. X-axis for each plot is +/−200 bp from motif center. (b) Experimental validation of directional pioneers. Average increase in flow cytometric fluorescence after RA addition for pioneer reporter lines for the stated motifs. For each TF noted, the left plot (labeled as RC for reverse complement) shows reporter results when the motif orientation is such that the RAR site is on the left of the motif with respect to the plot in a, and the right plot (labeled as Fw for forward) shows results when the RAR site is on the right of the motif with respect to a. All plots are normalized to control line RA-induced GFP as in Figure 3f, error bars indicate SEM, and a 99% prediction interval is shown as in Figure 3f.

Figure 5

Binding of settler TFs is governed by underlying chromatin state. (a) Comparing motif dependence (X-axis) versus chromatin opening index (Y-axis) for all 733 motifs in mouse lineage. Positions of select TFs are denoted and a linear trendline is displayed which shows imperfect but statistically significant positive correlation. (b) Comparing chromatin dependence (X-axis) versus chromatin opening index (Y-axis) for all 733 motifs in mouse lineage. Classes of pioneer TFs (blue), settler TFs (red), and migrant TFs (green) as defined by their chromatin opening and dependence properties are shaded, and select members of each class are listed. (c) Comparing K562 DNase-seq chromatin openness score (X-axis) vs. binned K562 ChIP-seq binding probability at strong motifs (Y-axis) for Elf1 (ETS family, pioneer), c-Myc (settler), and the average of all ChIP-seq experiments. (d) Contour plots showing log odds binding probability (contour) for bins of strong motifs at varying chromatin openness scores (X-axis) and PWM scores (Y-axis) for the K562 ChIP-seq TF clusters displaying chromatin-dependence only (left) or combinatorial motif-dependence and chromatin-dependence (right). For chromatin-dependent TFs, binding probability is predominantly dependent on chromatin openness score, whereas binding probability scores of combinatorially-dependent TFs increase as both chromatin openness and PWM score are increased. (e) Change in number of true positive PIQ calls per TF motif at a 10% false discovery rate as a result of incorporating motif-dependence and chromatin-dependence as prior information for all K562 ChIP-seq motif comparisons. Prior information improves PIQ accuracy for most TFs.

Figure 6

Pioneer TFs control chromatin state and settler TF binding. (a–b) Per-base average DNase hypersensitivity (HS) (a) and number of PIQ binding sites (b) within 4 kb of Klf7 and NFYA motifs for sites conserved (dotted lines) or lost (solid lines) between mesendoderm and endoderm stages. Both DNase HS and adjacent TF binding are diminished when Klf7 and NFYA binding are lost between successive stages. (c) Schematic of pioneer dominant negative (DN) competition experiments in which Doxycycline (Dox) induces DN pioneer TF expression (DBD), which should block pioneer-induced chromatin opening and prevent settler binding to opened chromatin. (d) Mean DNase hypersensitivity at several strong binding sites for NFYA (left) and Nrf1 (right) in wildtype (wt) (green) or DN NFYA/DN Nrf1 (red) mES, normalized to background DNase activity at non-hypersensitive sites. Asterisk indicates statistically significant difference between average DNase HS between wt and DN (n=4, P<0.01) using t-test. (e) Mean ChIP enrichment for four c-Myc sites downstream (in direction of predicted pioneer activity) of NFYA (left column), upstream (in direction of predicted non-pioneer activity) of NFYA (middle column) or adjacent to Nrf1 (right column) in wt (blue) or DN NFYA/DN Nrf1 (red) mES, normalized to positive and negative control genomic c-Myc sites. N=3, P<0.01 using t-test. (f) Model of TF binding hierarchy. Pioneers open chromatin, some directionally, and open chromatin is populated by settler TFs and by certain combinations of migrant TFs.