Coregulation of transcription factor binding and nucleosome occupancy through DNA features of mammalian enhancers

Abstract

Transcription factors (TFs) preferentially bind sites contained in regions of computationally predicted high nucleosomal occupancy, suggesting that nucleosomes are gatekeepers of TF binding sites. However, because of their complexity mammalian genomes contain millions of randomly occurring, unbound TF consensus binding sites. We hypothesized that the information controlling nucleosome assembly may coincide with the information that enables TFs to bind cis-regulatory elements while ignoring randomly occurring sites. Hence, nucleosomes would selectively mask genomic sites that can be contacted by TFs and thus be potentially functional. The hematopoietic pioneer TF Pu.1 maintained nucleosome depletion at macrophage-specific enhancers that displayed a broad range of nucleosome occupancy in other cell types and in reconstituted chromatin. We identified a minimal set of DNA sequence and shape features that accurately predicted both Pu.1 binding and nucleosome occupancy genome-wide. These data reveal a basic organizational principle of mammalian cis-regulatory elements whereby TF recruitment and nucleosome deposition are controlled by overlapping DNA sequence features.

Fig. 1. Regular arrays of nucleosomes centered at Pu.1-bound enhancers in macrophages

A) Cumulative distribution of midpoints of nucleosomal sequencing fragments centered on the summit of TSS-distal Pu.1 sites in macrophages. The number of fragments in each 10-bp bin was normalized by the total number of fragments in the area. The same information is shown in B) as heatmap (first from the left), ordered from top to bottom based on decreasing occupancy of the NDR and divided in deciles. Heatmaps of Pu.1 and Pol_II are also shown on the right of the MNase data. The counts exceeding the 95th percentile of the overall distribution were set to its value. Considering MNase data, these counts were then normalized in the range 0-1 separately for each region. The same procedure was applied to ChIP-seq data except that the 0-1 normalization was applied to the entire dataset. Sequence logos on the right show the Pu.1 binding motifs identified de novo in individual deciles and their E-values. C) ChIP-seq scores (MACS) of the Pu.1 peaks in in the different deciles. D) Cumulative distributions of the midpoints of the nucleosomal fragments at Pu.1 bound enhancers (divided in deciles according to panel B). E) A representative snapshot showing two NDRs of the 1^st and the 10^th decile. See also Fig. S1.

Fig. 2. Sequence features discriminate among enhancers with different nucleosome occupancy and positioning

A) Cumulative distribution of AAAA tetranucleotides (top panel), AA dinucleotides (middle panel) and G/C containing dinucleotides (bottom panel) are shown relative to the summit of TSS-distal Pu.1 peaks (the strong enrichment of CC/GG dinucleotides at the anchor point is enhanced by the central invariant nucleotides of the Pu.1 site, 5′-AGAGGAAGTG-3′). G+C content (B) and distribution of AA dinucleotides (C) in deciles at Pu.1-bound distal (left) and TSS-proximal (right) sites. See also Fig. S2. D) Statistical over-representation of binding sites for TF families at Pu.1 bound distal sites divided in deciles (according to Fig. 1B).

Fig. 3. Pu.1-bound, nucleosome-depleted macrophage enhancers are covered by nucleosomes in unrelated cell types and in vitro

(A) Cumulative distributions of the midpoints of the nucleosomal fragments centered on distal Pu.1 sites in macrophages and in unrelated cells that do not express Pu.1 (ESCs, NPCs and MEFs). The number of midpoints in each 10-bp bin was scaled according to the total number of regions and sequencing depth. In (B) data were split in deciles (only the 1^st, 5^th and 10^th deciles are shown). C) Midpoint distributions from in vitro assembled nucleosomes. Data for distal and TSS-proximal sites are shown. See also Fig. S3. D) MNase-seq data from in vitro nucleosomes are shown divided in deciles.

Fig. 4. Effects of Pu.1 depletion on nucleosome occupancy

A) Acute depletion of Pu.1 in terminally differentiated macrophages using a retrovirus-encoded Tet-regulated shRNA. Data from two biological replicates are shown. Vinculin was used as loading control. B) Nucleosomal occupancy in Pu.1-depleted macrophages. Pu.1 peaks were divided in quartiles based on the degree of signal reduction in Pu.1-depleted vs. control cells. The 4^th quartile corresponds to Pu.1 peaks with the higher reduction in binding in Pu.1-depleted cells. Quartile-specific distributions of nucleosome fragments midpoints centered on the summit of Pu.1 peaks are shown. Midpoints found within + 80 bp from the Pu.1 summit are also summarized in the box plots on the right. For each quartile, the statistical significance of the difference is expressed by the p-value of a Wilcoxon signed-rank test. See also Fig. S4.

Fig. 5. In vitro analysis of Pu.1 binding to nucleosomal DNA

A) Pu.1 ChIP-seq on in vitro assembled nucleosomes. Macrophage nuclear lysates (used as a source of Pu.1) were incubated with in vitro assembled chromatin. Prior to incubation with nucleosomes, nuclear lysates were reacted twice either with control rabbit IgG or anti-Pu.1 antibody coupled to paramagnetic beads. Immunodepletion with anti-Pu.1 antibodies resulted in an almost complete loss of Pu.1 from lysates. Vinculin: loading control. B) Venn diagram showing the overlap between in vitro and in vivo Pu.1 binding. C) Heatmap of in vitro Pu.1 binding, showing the relative density of nucleosome midpoints. In vitro ChIP signals were sorted according to nucleosome occupancy in macrophages (Fig. 1B). See also Fig. S5. D) A representative ChIP-seq snapshot is shown. E) Pu.1 sites and flanks (40nt) from the 6^th or 4^th decile were transferred to sequences of higher (10^th) or lower (2^nd) deciles and used for nucleosome assembly and ChIP-qPCR. Mock transfected extracts and extracts from HEK-293 cells transfected with a Pu.1 expression vector were used as indicated.

Fig. 6. Pu.1 binding site usage correlates with nucleosome occupancy

A) Venn diagram showing the overlap between Pu.1 peaks identified in ChIP-seq experiments from multiple cell types and computationally identified genomic Pu.1 sites. B) Cumulative distributions of nucleosome midpoints in macrophages, ESCs and NPCs at Pu.1-bound high affinity consensus sites (red), Pu.1-bound non-canonical sites (orange) and computationally identified consensus sites that are not bound in vivo (grey). C) Schematic representation of the relationship between Pu.1 binding and nucleosome occupancy. D) Schematic of the SVM approach used to predict in vivo binding competence of Pu.1 sites and to identify their distinctive DNA sequence and shape features. E) Computationally measured affinity of Pu.1 towards bound and unbound genomic sites. F) Bar plots showing the prediction accuracies of the most predictive features selected by the SVM, divided in categories (blue) or all in combination (red). See also Fig. S6.

Fig. 7. Nucleosome positioning prediction using DNA sequence and shape features associated with engaged TF binding sites

A) Schematic of the SVR approach used to predict nucleosome occupancy from the DNA sequence and shape features predictive for Pu.1 binding. B) Smoothed scatterplots of the predicted values against the observed log2-transformed values of nucleosome occupancy in ESCs over Pu.1 sites (using the set of features selected for one of the training-test randomizations that served as input for the SVM). The scatterplot on the right shows the results on the test dataset using only theoretical nucleosome occupancy. The one on the left shows the results using all the features selected except for it. C) Box plots showing the distribution of the R² for the sets of features from the ten training-test randomizations of the SVM.