Catalytic-Independent Functions of PARP-1 Determine Sox2 Pioneer Activity at Intractable Genomic Loci

Abstract

Pioneer transcription factors (TFs) function as genomic first responders, binding to inaccessible regions of chromatin to promote enhancer formation. The mechanism by which pioneer TFs gain access to chromatin remains an important unanswered question. Here we show that PARP-1, a nucleosome-binding protein, cooperates with intrinsic properties of the pioneer TF Sox2 to facilitate its binding to intractable genomic loci in embryonic stem cells. These actions of PARP-1 occur independently of its poly(ADP-ribosyl) transferase activity. PARP-1-dependent Sox2-binding sites reside in euchromatic regions of the genome with relatively high nucleosome occupancy and low co-occupancy by other transcription factors. PARP-1 stabilizes Sox2 binding to nucleosomes at suboptimal sites through cooperative interactions on DNA. Our results define intrinsic and extrinsic features that determine Sox2 pioneer activity. The conditional pioneer activity observed with Sox2 at a subset of binding sites may be a key feature of other pioneer TFs operating at intractable genomic loci.

Keywords: PARP-1; Sox2; chromatin; embryonic stem cells; nucleosome; nucleosome rotational positioning; pioneer transcription factor; transcription.

Figure 1. PARP-1 colocalizes with Sox2 genome-wide

A) Top, Summary of the known roles of PARP-1 protein and PARylation activity in embryonic stem cells. Bottom, Summary of known changes in Sox2 protein and PARylation levels, total cellular PARylation levels during embryonic stem cell differentiation. B) The level of PARP-1-mediated PARylation is very low in undifferentiated mESCs and increases upon differentiation. Western blots of poly(ADP-ribose) (PAR) and PARP-1 showing their relative levels in mESCs during a 9 day time course of differentiation upon LIF removal. C) Distribution of significant peaks of PARP-1 binding from ChIP-seq across genomic features in mESCs. D) Distribution of significant peaks of PARP-1 binding relative to the TSSs of all RefSeq genes. E) PARP-1 colocalizes with Sox2 genome-wide. Correlation matrix of genome-wide enrichment for chromatin- and transcription-related factors from ChIP-seq data in mESCs, organized and ordered using hierarchical clustering. F) Left, Heatmap of high confidence Sox2 peaks in mESCs from ChIP-seq data (n = 16,042; p-value < 10⁻⁷) centered on the Sox2 binding sites (± 2 kb) and ordered top to bottom by signal intensity. Middle and right, Oct4 and PARP-1 ChIP-seq signals associated with the corresponding Sox2 binding sites. G) Genome browser tracks of PARP-1, Sox2 and Oct4 ChIP-seq data around the Nanog gene in WT mESCs. H) PARP-1 binds to Sox2, but not Oct4. Flag-tagged PARP-1 was expressed in HEK293T cells with Sox2 and Oct4, and immunoprecipitated using a Flag antibody. Western blots showing the relative levels of PARP-1, Sox2, and Oct4 in the input and the co-immunoprecipitated (co-IP) material. See also Fig. S1.

Figure 2. PARP-1 is required for maintaining transcriptional program in embryonic stem cells

A) Western blots showing the relative levels of three pluripotency factors (Sox2, Oct4, and Nanog) and PARP-1 in WT and Parp1^−/− mESCs. Actin is used as an internal loading control. B) Effect of Parp1 knockout on gene expression in mESCs as determined by RNA-seq. The heatmap shows the relative expression levels of genes whose expression significantly (FDR < 5%) increased upon Parp1 knockout (“Negatively Regulated” by PARP-1) or decreased upon Parp1 knockout (“Positively Regulated” by PARP-1). The data are log₂(Parp1^−/− RPKM/WT RPKM; RPKM = Reads per kilobase of transcript per million mapped reads). C) Effect of Parp1 knockout on the expression of pluripotency-associated genes in mESCs, as determined by RT-qPCR. The data for Parp1^−/− mESCs are expressed relative to WT ESCs. Each bar represents the mean plus the SEM, n ≥ 3. The differences observed for Nanog and Parp1 are significant (Student’s t test, p-value < 0.05). D) Effect of Parp1 knockout on the expression of differentiation-associated genes in mESCs, as determined by RT-qPCR. The data for Parp1^−/− mESCs are expressed relative to WT ESCs. Each bar represents the mean plus the SEM, n ≥ 3. The differences observed for all of the genes shown are significant (Student’s t test, p-value < 0.05). E) Genome browser tracks of mRNA-seq data (top) and H3K4me3 ChIP-seq data (bottom) around the Nanog, Oct4, and Gata6 genes in WT and Parp1^−/− mESCs. F) The expression of Nanog (a pluripotency-associated gene), as well as Pax6 and Gata6 (differentiation-associated genes), in undifferentiated (‘Day 0”) mESCs is not affected by treatment with the PARP inhibitor PJ34. RT-qPCR was performed using total RNA isolated from mESCs treated with 5 M PJ34 for 24 hrs. The expression levels we standardized to the expression of the Gapdh gene. Each bar represents the mean plus the SEM, n ≥ 3. The small differences are not significant. G) Analysis of mRNA expression for a panel of 19 differentiation-associated genes in mESCs during a 9 day time course of differentiation upon LIF removal. The expression levels of each mRNA are normalized to Gapdh mRNA levels and scaled. The individual genes are listed in Fig. S2D. * p = 0.01, ** p = 0.0002. See also Fig. S2.

Figure 3. PARP-1 is required for the binding of Sox2 to a subset of its genomic sites in mESCs

A) Genome browser tracks of Sox2 ChIP-seq data around the Nanog and Pou5f1 (encoding Oct4) genes in WT and Parp1^−/− mESCs. B) Left and middle, Heatmaps of Sox2 ChIP-seq signals in WT and Parp1^−/− mESCs centered on the Sox2 binding sites (± 2 kb) and ordered top to bottom by signal intensity. Right, PARP-1 ChIP-seq signals in WT mESCs associated with the corresponding Sox2 binding sites. PARP-1-dependent Sox2 binding sites are defined as those sites whose ChIP-seq signals are significantly decreased upon PARP-1 knockout (p-value < 0.01; n = 1,606,). PARP-1-independent Sox2 binding sites show no change upon PARP-1 knockout (n = 4,556). C) Normalized ChIP-seq read counts for Sox2 (left) and Oct4 (right) at (1) PARP-1-dependent (Dep.) sites where Sox2 and Oct4 binding overlap, (2) PARP-1-independent (Ind.) sites where Sox2 and Oct4 binding overlap, and (3) sites with Sox2 or Oct4 only. Asterisks indicate significant differences (Student’s t-test, p-value < 2.2 × 10⁻¹⁶). D) Fraction of genes associated with different categories of Sox2 binding sites. Dep. = PARP-1-dependent Sox2 binding sites; Ind. = PARP-1 independent Sox2 binding sites; All = All Sox2 binding sites. Yellow, Genes whose expression significantly increases upon Parp1 knockout (“Negatively Regulated” by PARP-1). Blue, Genes whose expression significantly decreases upon Parp1 knockout (“Positively Regulated” by PARP-1). E) GSEA analysis showing the relationship between PARP-1-regulated Sox2 binding sites (n = 1,606) and gene expression changes upon Parp1 knockout in mESCs. Top, The expression of genes associated with PARP-1-regulated Sox2 binding sites is significantly decreased in Parp1^−/− mESCs compared to WT mESCs (p-value < 0.001) based on RNA-seq. Bottom, A randomly selected and equally sized set of genes (n = 1,606) shown as a control (p-value = 0.36). See also Fig. S3.

Figure 4. Genomic determinants of PARP-1-dependent Sox2 binding

A) Identification of a minimal set of genomic features that accurately predict the PARP-1 dependence of Sox2 binding sites using the machine learning algorithm, LASSO (Tibshirani, 1996). The features included nucleosome occupancy, transcription factor co-occupancy, DNaseseq signal strength, Sox motif sequence, rotational orientation of the Sox motif in the nucleosome, and predicted structural features of the DNA at Sox2 binding sites (e.g., nucleosome rotational positioning, minor groove width, roll, x-displacement, slide). Prediction accuracy was evaluated by 10-fold cross validation. The combination of features that produced the best prediction accuracy with the smallest mean squared error (MSE; Y-axis) was selected (red arrow). X-axis, Log10 values of the penalty score lambda. Top, Number of features corresponding to the respective lambda values. B) List of the optimized feature combination from the LASSO algorithm producing the best prediction accuracy. The complete list of 86 features tested in this analysis is listed in the Supplemental Materials. C) Genome browser tracks of Sox2 ChIP-seq data and MNase-seq data around the Nanog gene in WT and Parp1^−/− mESCs. The green shading highlights the relationship between Sox2 binding and nucleosome occupancy. D) Average MNase-seq signals surrounding PARP-1-dependent (red) and PARP-1-independent (blue) Sox2 binding sites in mES cells. The data are centered on the Sox2 binding sites determined by ChIP-seq (± 2 kb). E) Average DNase-seq signals surrounding PARP-1-dependent (red) and PARP-1-independent (blue) Sox2 binding sites in mES cells. The data are centered on the Sox2 binding sites determined by ChIP-seq (± 2 kb). F) Heatmaps showing the binding of other transcription factors at Sox2 binding sites. Top, Results for PARP-1-independent Sox2 binding sites. Bottom, Results for PARP-1-dependent Sox2 binding sites. Red, significant binding. Black, no significant binding. G) Fraction of PARP-1-independent (Ind.) and PARP-1-dependent (Dep.) Sox2 binding sites associated with the specified number transcription factors (TFs; 0, 1, 2, 3, >3) based on ChIP-seq in mESCs. Asterisks indicate significant differences (Fisher’s exact test, p-value < 0.0001). H) Summary of genomic features for PARP-1-independent and PARP-1-dependent Sox2 binding sites based on Fig. 4 and Fig. S4. 76% of PARP-1-dependent Sox2 binding sites have the features shown in the bottom row, while 68% of PARP-1-independent Sox2 binding sites have the features shown in the top row. See also Fig. S4.

Figure 5. PARP-1 stabilizes Sox2 binding to nucleosome in vitro

A) In vitro nucleosome binding assays show PARP-1-dependent binding of Sox2 to nucleosomes, but not naked DNA. Ten nM of biotin-labeled 601 NPE DNA (207 bp) with or without a Sox motif (Sox+ or Sox−, respectively), or the same DNA assembled into a mononucleosome (Nuc), was immobilized on streptavidin beads and incubated ± recombinant Sox2 (10 nM) and PARP-1 (10 nM), as indicated. After washing, the bound proteins were analyzed by Western blotting. B) In vitro DNase I footprinting assays show PARP-1-dependent binding of Sox2 to nucleosomes, but not naked DNA. Left, Footprinting assay with naked 3x 601 NPE DNA containing a Sox motif. Right, Footprinting assay with a trinucleosome containing a Sox motif located at the dyad axis of the middle nucleosome. Addition of 10 nM WT PARP-1, but not a DNA binding domain mutant PARP-1 (DBDmut), enhances Sox2 (25 nM) binding, as indicated. PARP-1 binding to the nucleosome is also evident. C) Quantification of the DNase I footprinting assays shown in panel C. The bands marked with a red asterisk in panel B were quantified using a phosphorimager and ImageJ software. Each bar represents the mean plus the SEM, n = 3. Asterisks indicate significant differences versus the control (−Sox2) (Student’s t test, ** p < 0.01 and * p < 0.05). D) PARP-1 stabilizes Sox2 binding to nucleosomes with or without linker DNA. In vitro nucleosome binding assay, as in panel A, using 601 NPE mononucleosomes with (Nuc 207; 207 bp) or without (Nuc 147; 147 bp) linker DNA, or the corresponding naked DNA (“DNA”). See also Fig. S5 and Fig. S6.

Figure 6. The BRCT domain of PARP-1 is required for stabilizing Sox2 binding to nucleosomes

A) Schematics of the PARP-1 deletion mutants used in the nucleosome binding assays. B) Coomassie blue staining of purified bacterially-expressed wild-type and deletion mutant PARP-1 proteins used in the nucleosome binding assays (lanes 1-5). Purified wild-type PARP-1 protein expressed in Sf9 insect cells is shown for comparison (lane 1*). C) The PARP-1 BRCT domain is required for PARP-1-dependent binding of Sox2 to nucleosomes. In vitro nucleosome binding assays using various PARP-1 deletion mutants, with biotin-labeled 601 NPE DNA (207 bp) as in Fig. 5A. D) Quantification of the nucleosome binding assays in panel C. The bands were quantified by densitometry. Each bar represents the mean plus the SEM, n = 3. Asterisks indicate significant differences versus the control (−PARP-1) (Student’s t test, ** p ≤ 0.01 and * p < 0.02). E) The competitive BRCT domain inhibitor (±)-gossypol blocks PARP-1-dependent Sox2 binding to nucleosomes. In vitro nucleosome binding assays ± 60 M of (±)-gossypol, as in panel C, but using the Position 32 nucleosome shown in Fig. 7B. F) Quantification of the nucleosome binding assays in panel E. The bands were quantified by densitometry. Each bar represents the mean plus the SEM, n = 3. Asterisks indicate significant differences versus the control (−gossypol) (Student’s t test, ** p ≤ 0.01). G) PARP-1 interacts with Sox2 protein in vitro. Sox2 was incubated with immobilized PARP-1. Bound material was analyzed by Western blotting for PARP-1 and Sox2, as indicated. H) The PARP-1 BRCT domain mediates interactions between PARP-1 and Sox2. Sox2 was incubated with immobilized wild-type or deletion mutant PARP-1 proteins, as described in panel A. Bound material was analyzed by Western blotting for PARP-1 and Sox2, as indicated. I) Quantification of the binding assays in panel H. The Sox2 bands were quantified by densitometry and the signals were normalized to the molarity of the PARP-1 protein immobilized on the beads. Each bar represents the mean plus the SEM, n = 3. Asterisks indicate significant differences versus the control (WT PARP-1) (Student’s t-test, * p = 0.018; ** p = 0.005).

Figure 7. Cooperative interactions with PARP-1 facilitate Sox2 binding to suboptimal sites in nucleosomes

A) Orientation of the minor groove base relative to the histone octamer for each base pair in the Sox motif in the 601 NPE nucleosome, as determined from high resolution X-ray crystal structures (PBD: 3LZ0, 3LZ1, 3MVD). Heatmaps showing the orientation of each position in the Sox motif (orange shading, minor groove facing away from the histone octamer; gray shading, minor groove facing toward the histone octamer). Letters indicate the nucleotides at each position of the Sox motif (N = A, C, G, or T; W = A or T; R = A or G) and numbers indicate the position of the Sox motif on the nucleosome, as shown in panel B. B) Position of the “NAC” in the Sox motif mapped on the 601 NPE nucleosome for a series of different constructs where the Sox motif has been moved along the length of the DNA. Red, PARP-1-dependent Sox2 binding site. Blue: PARP-1-independent Sox2 binding site. Only half of the nucleosome is shown. The numbering is based on the distance of the “C” in the first half of the Sox motif to the dyad axis of the nucleosome (the −64 position is not shown). C) The rotational orientation of the Sox motif determines the PARP-1-dependence of Sox2 binding to nucleosomes. In vitro nucleosome binding assays, with biotin-labeled 601 NPE DNA (207 bp) as in Fig 5A, using a set of nucleosome templates with a Sox motif placed in different translational positions across the nucleosome, as shown in panel B. D) X-ray crystal structure of Sox2 HMG domain binding to DNA (1GT0) (Remenyi et al., 2003) with the locations of three key arginine residues (Arg2, Arg15, Arg76) indicated (side chains shown in purple). E) Arginine residues in the HMG domain mediate PARP-1-dependent binding of Sox2 to nucleosomes. In vitro nucleosome binding assays, as in panel C, using a set of Sox2 Arg → Ala mutant proteins (R2A, R15A, R76A). The Sox2 mutants were assayed with two different nucleosome constructs: “position 32” (PARP-1-dependent) and “position 27” (PARP-1-independent). F) NRP scores associated with PARP-1-dependent and PARP-1-independent Sox2 binding sites in mESCs. See Fig. S7H for an outline of the methods. The x-axis represents the position in nucleotides of the Sox motif associated with the Sox2 binding site, as determined by ChIP-seq, with the “C” in the first half of the Sox motif set to 0. The y-axis represents average NRP score associated with each nucleotide position (dark lines), with the SEM range indicated by lighter shading. Vertical grey shading highlights regions exhibiting a significant difference in NRP score between PARP-1-dependent and PARP-1-independent Sox2 binding sites. Higher NRP scores indicate a greater tendency for the sequence to face away from the histone octamer (see Fig. S7B). G) Model for Sox2 binding to nucleosomes at PARP-1-dependent sites. PARP-1 interacts with nucleosomal DNA through its DNA binding domain (DBD) and Sox2 through its BRCT motif. PARP-1 stabilizes Sox2 interaction with nucleosomal DNA when the Sox motif sequence is positioned unfavorably along the nucleosome for Sox2 to bind. See also Fig. S7.