Chromosomal domains in Drosophila are marked by the insulator‐binding proteins (IBPs) dCTCF/Beaf32 and cofactors that participate in regulating long‐range interactions. Chromosomal borders are further enriched in specific histone modifications, yet the role of histone modifiers and nucleosome dynamics in this context remains largely unknown. Here, we show that IBP depletion impairs nucleosome dynamics specifically at the promoters and coding sequence of genes flanked by IBP binding sites. Biochemical purification identifies the H3K36 histone methyltransferase NSD/dMes‐4 as a novel IBP cofactor, which specifically co‐regulates the chromatin accessibility of hundreds of genes flanked by dCTCF/Beaf32. NSD/dMes‐4 presets chromatin before the recruitment of transcriptional activators including DREF that triggers Set2/Hypb‐dependent H3K36 trimethylation, nucleosome positioning, and RNA splicing. Our results unveil a model for how IBPs regulate nucleosome dynamics and gene expression through NSD/dMes‐4, which may regulate H3K27me3 spreading. Our data uncover how IBPs dynamically regulate chromatin organization depending on distinct cofactors.
Insulator‐binding proteins (IBPs) separate chromatin domains. The IBP dCTCF/Beaf32 and the H3K36 histone methyltransferase NSD/dMes‐4 co‐regulate nucleosome dynamics, chromatin accessibility, and expression of insulator‐flanked genes.
Insulator proteins interact with H3K36 histone methyltransferase NSD/dMes‐4.
Expression data show that dMes4 acts as a cofactor of IBPs.
H3K36me2/3 regulate transcription‐coupled nucleosome dynamics and RNA splicing.
IBPs/dMes‐4 regulate H3K27me3 deposition for hundreds of genes.
Recent Chromosome Capture Conformation (3C/Hi‐C) data have highlighted that chromosomes are physically partitioned into distinct epigenetically marked chromatin domains bordered by insulators (Ghirlando et al, 2012; Hou et al, 2012; Sexton et al, 2012). The genomic distribution of the thousands of Drosophila insulator protein binding sites tightly correlated with the “physical borders” that restrict long‐range interactions between chromosomal domains (Sexton et al, 2012). This may provide with a distinct mechanism of gene regulation through higher‐order physical organization of chromatin into topologically associating domains (TADs) that is not strictly linked to the demarcation of chromosomes into epigenetically marked domains (Hou et al, 2012; reviewed in Phillips‐Cremins & Corces, 2013).
The function of chromatin insulators in long‐range interactions was initially suspected genetically from “enhancer‐blocking” assays showing that, when interposed, insulators can block the long‐range interactions between distant regulatory elements (Cai & Levine, 1995; reviewed in Maeda & Karch, 2007; Gohl et al, 2011). Enhancer‐blocking activity requires the binding of CCCTC‐binding factor (CTCF) or one of the additional insulator‐binding proteins (IBPs) identified in Drosophila—GAGA Factor (GAF), Boundary Element‐Associated Factor (Beaf32), or suppressor of Hairy‐wing [Su(Hw)]; (reviewed in Vögelmann et al, 2011). Only a fraction of the identified genomic IBP sites were shown to function in standard enhancer‐blocking assays (Negre et al, 2010), maybe as such function may depend on genomic contexts, the presence of nearby insulators (Gohl et al, 2011) or on additional cofactors. Genome‐wide analyses of long‐range contacts highlighted that IBPs may actually favor long‐range interactions between distant sites, thereby contributing to the physical organization of TADs (reviewed in Phillips‐Cremins & Corces, 2013). This requires additional cofactors for proper insulation of chromatin into domains, including CP190 (Bushey et al, 2009; Liang et al, 2014), that interact with all types of IBPs, chromator (Hou et al, 2012; Sexton et al, 2012), or key chromatin regulators like cohesin (reviewed in Dorsett, 2011; Phillips‐Cremins & Corces, 2013).
IBPs were also proposed to function as “insulator barriers” that could block the spreading of condensed regions toward euchromatin (Gaszner & Felsenfeld, 2006; Maeda & Karch, 2007). Whether barrier and enhancer‐blocking activities are conferred by distinct types of factors has been questioned (Vögelmann et al, 2011). Genome‐wide, the binding sites of IBPs including CTCF and Beaf32 are largely enriched at the borders of repressed domains epigenetically marked by histone H3 tri‐methylated on lysine 27 (H3K27me3) (Guelen et al, 2008; Cuddapah et al, 2009; Sexton et al, 2012). Only a fraction of the IBP sites could actually restrict H3K27me3 spreading (Schwartz et al, 2012), again suggesting that additional cofactors may be required. Remarkably, the barrier and enhancer‐blocking activities of the chicken beta‐globin locus were shown to be conferred by different types of nearby elements recognized by either USF/VEZF or CTCF, respectively (Ghirlando et al, 2012; Gowher et al, 2012). Nucleosome occupancy is particularly low at the IBP binding sites including those flanking H3K27me3 domains (Emberly et al, 2008; Bartkuhn et al, 2009; Cuddapah et al, 2009), yet it remains unknown whether specific histone modifiers or chromatin remodelers play a role in this context. Whether chromatin barriers involve a division of labor through distinct IBPs, cofactors or distinct chromatin regulators required for barrier activity, remains open questions.
Here, we report that Beaf32 depletion specifically regulates the expression of hundreds of genes flanked by marked Nucleosome Free Regions (NFRs). Biochemical purification of Beaf32 identifies nuclear SET domain‐containing proteins (NSD)/Drosophila Maternal‐effect sterile gene 4 (dMes‐4), an essential HMT needed for dimethylation of histone H3 on lysine 36 (H3K36me2) and involved in recruiting histone acetyltransferases (HATs) (Bell et al, 2007). Strikingly, dMes‐4 depletion recapitulates gene expression defects upon Beaf32 depletion, supporting its role as a co‐regulator. H3K36me2/3 genome‐wide patterns tightly correlate with NFRs/nucleosome positioning, respectively, in complete agreement with recent data showing the role of H3K36 methylation in regulating histone exchange (Venkatesh et al, 2012). Beaf32/dMes‐4 depletions impair subsequent recruitment of DREF, a key transcriptional activator required for Set2/Hybp‐dependent H3K36me3. As such, H3K36me2 presets chromatin for transcription‐coupled, H3K36me3‐dependent nucleosome positioning that is further implicated in regulating H3K27me3 spreading. Taken altogether, our results suggest a pivotal role of H3K36 HMTs in regulating the expression of genes flanked by IBPs, through chromatin dynamics.
Beaf32 differentially regulates genes harboring marked NFRs flanking high nucleosome positioning
Drosophila IBP sites including Beaf32 and dCTCF or GAF were shown to be highly enriched within marked Nucleosome Free Regions (NFRs) (Emberly et al, 2008; Gurudatta & Corces, 2009; Jiang et al, 2009; Negre et al, 2010; Mukhopadhyay et al, 2011). Similarly, human CTCF sites were shown to overlap with as much as 28% of the mapped DNase I hypersensitive sites (DHSs) linking transcription programs to such chromatin landscape (Natarajan et al, 2012; Thurman et al, 2012). In the case of Drosophila, the binding sites of IBPs often localize near promoters (reviewed in Raab & Kamakaka, 2010; Vögelmann et al, 2011), as illustrated by ChIP‐Seq data of Beaf32 (Fig 1A; < 500 bp from TSS; P‐value < 1e‐300), prompting us to test whether such genomic features reflect a role of IBPs in regulating gene expression through chromatin organization.
In such contexts, Beaf32 may regulate hundreds of genes as shown by genome‐wide expression analyses (Supplementary Fig S1). These data were obtained through Digital Gene Expression‐sequencing (DGE‐Seq) in S2 cells depleted of Beaf32 (Beaf32‐KD ~95% efficiency) as compared to mock‐depleted control cells (“WT control”), and the impact of Beaf32 on gene expression could be verified by RT‐qPCR and microarray analyses (Supplementary Fig S1F and G). Intersecting expression data with ChIP‐Seq data showed that 57% of the differentially expressed (DE) genes harbor a Beaf32 binding site in their promoter (P‐value = 1e‐291). 28.7% of the promoters harboring a Beaf32 site were differentially regulated as compared to 7.9% without such site. The fold changes in expression as measured by DGE‐Seq were limited (WT/Beaf32‐KD ~twofold on average) showing a moderate influence of Beaf32, as expected if it is not a transcription activator (Cuvier et al, 1998). Supporting this interpretation, no effect of Beaf32 depletion was detected using in vitro transcription assays choosing Beaf32 promoters as naked DNA templates (unpublished data).
To investigate whether IBPs such as Beaf32 regulate gene expression through their role in chromatin organization, we performed MNase‐Seq (see Materials and Methods) (Barski et al, 2007; Schones et al, 2008; Gilchrist et al, 2010) and ranked genes according to the influence of Beaf32 on their expression (Fig 1B). The generated heat maps highlighted a good correlation between DE genes upon Beaf32‐KD and the presence of NFRs or of high nucleosome positioning signals in promoters or in gene bodies, respectively (Fig 1B, NFR/“+1”), providing Beaf32 was bound to promoters (Supplementary Fig S2A). In agreement, high nucleosome positioning has been observed for highly active, housekeeping genes such as those regulated by Beaf32/DREF or dCTCF (Emberly et al, 2008; Bushey et al, 2009; Gilchrist et al, 2010).
Beaf32‐KD significantly affected nucleosome positioning in approximately 2,000 genes (Fig 1C, see “+1” for “Beaf32KD”), as evidenced by changes in MNase‐Seq reads along their bodies upon Beaf32‐KD compared to control cells (see Materials and Methods). By contrast, the averaged nucleosome levels slightly increased in the corresponding NFRs (Fig 1C), as illustrated by the tsp39D gene promoter region (Fig 1D; NFR). Such variations upon Beaf32‐KD were not systematically found in every gene flanked by a Beaf32 binding site (differentially expressed or not). The thousands of genes with marked changes in nucleosome positioning were however specifically enriched among genes flanked by a Beaf32 binding site as well as genes differentially regulated upon Beaf32‐KD (Supplementary Fig S2B and C), supporting the functional implication of Beaf32 in regulating nucleosome dynamics.
Nucleosome positioning may function in regulating gene expression by preventing spurious transcription (Carrozza et al, 2005; Gilchrist et al, 2010) or RNA splicing (reviewed in Schwartz & Ast, 2010). In agreement, further RNASeq analysis in Beaf32‐KD highlighted increasing levels of spurious/aberrant intronic RNASeq reads compared to WT cells (Supplementary Fig S2D). This phenomenon was specifically encountered for genes bound by Beaf32 or where significant changes in nucleosome positioning were scored upon Beaf32‐KD (P‐values = 1e‐37 and 1e‐3, respectively). Aberrant transcripts were found in as much as 48% of the differentially regulated genes harboring nucleosome positioning changes (226/469 genes; P‐value = 3e‐21; see below). Our data therefore strengthen the view that the function of Beaf32 in regulating gene expression may be linked to nucleosome dynamics, which very likely involves additional cofactors.
Beaf32 interacts with the histone methyltransferase dMes‐4
Genome‐wide analysis by ranking genes according to the amount of ChIP‐Seq reads of Beaf32 in their promoters confirmed that Beaf32 binding in the top quartile (q1) correlated with marked NFRs in promoters as well as high nucleosome positioning over gene bodies (Fig 2A). We thus sought to identify its cofactors through high‐salt elution of factors biochemically co‐purified with the Beaf32 complex (Fig 2B; see Materials and Methods). The specificity of the assay was confirmed by the presence of the Beaf32 doublet and of CP190, a key cofactor that interacts with all types of Drosophila IBPs including dCTCF (Bushey et al, 2008; Wood et al, 2011; Liang et al, 2014). In addition, this analysis further identified a unique histone modifier, dMes‐4 (Maternal‐effect sterile 4; Fig 2B; see arrow), an essential HMT that dimethylates lysine 36 of histone H3 (H3K36me2) (Bell et al, 2007) involved in epigenetic mechanisms (Pirrotta, 2002). Co‐immunoprecipitation experiments using anti‐Beaf32 antibodies confirmed the specific interaction of Beaf32 and dMes‐4 (Fig 2C) compared to H3 loading control, as confirmed by two‐hybrid assays (Supplementary Fig S3A). Ranking of genes according to the ChIP‐Seq reads of Beaf32 highlighted a genome‐wide correlation between its binding and H3K36me2 levels in promoters (Fig 2D) as measured by ChIP‐Seq (see Materials and Methods). Beaf32‐KD actually led to a reduction of approximately 50% of the chromatin levels of H3K36me2/me3 compared to control cells (Fig 2E), unlike what was found for the soluble pool of H3K36 (Supplementary Fig S3B and C). H3K36me2/3 chromatin levels are readily dependent on dMes‐4 as shown (Fig 2E, left blot) (Bell et al, 2007). ChIP analysis showed that Beaf32‐KD impaired the recruitment of dMes‐4 specifically for promoters bound by Beaf32 compared to control promoters (Fig 2F), in complete agreement with data showing that Beaf32‐KD decreased H3K36me2 levels in such promoters as compared to control cells (Supplementary Fig S3E).
dMes‐4‐driven methylation is required for histone acetylation by favoring the recruitment of histone acetyltransferases (HATs) (Bell et al, 2007; Venkatesh et al, 2012). This may account for the observed enrichment of Beaf32 binding sites among genomic sites harboring high acetylated histone levels as shown for H4K16ac (Fig 2G). ChIP of H4K16ac in Beaf32‐KD showed a significant reduction of H4K16ac levels compared to control cells (Fig 2H; P‐value = 1e‐7), strengthening its implication in chromatin accessibility. In agreement, intersection analyses suggested that genes differentially expressed upon Beaf32‐KD were also highly enriched among those harboring high acetylation levels (Supplementary Fig S3D). Altogether, our data supported a functional interplay between Beaf32 and dMes‐4 as a key co‐regulator of chromatin dynamics, as further addressed below.
dMes‐4 is a key co‐regulator of Beaf32 in gene expression
The functional interplay of dMes‐4 and Beaf32 was addressed by genome‐wide expression analyses after efficient depletion of either factor (Fig 3A, “dMes‐4‐KD” and “Beaf32KD”, respectively; Supplementary Fig S3F–H). Strikingly, a large overlap was found between genes whose expression was impaired by Beaf32 and differentially expressed (DE) genes upon dMes‐4‐KD (Fig 3B), strongly supporting a key role of dMes‐4 as a co‐regulator of Beaf32. Accordingly, DE genes upon dMes‐4‐KD were largely enriched in the same gene ontologies as found upon Beaf32‐KD, including the cell cycle and/or cell death (Supplementary Fig S4A). Such regulations likely implicate DREF as a transcriptional activator enriched within Beaf32 sites (Supplementary Fig S4B; see below) (Emberly et al, 2008) or dCTCF that shares many binding sites with Beaf32 and that regulates similar cellular functions (Bushey et al, 2009; Gurudatta et al, 2013).
Given the overlapping binding sites between Beaf32 and dCTCF, we then tested whether dMes‐4 might specifically influence genes that are flanked by dCTCF sites. The percentage of DE genes uniquely bound by dCTCF and under the influence of dMes‐4 was lower compared to those bound by Beaf32 (Fig 3C; 36.5 and 49.9%, respectively). The influence of dMes‐4 on genes flanked by dCTCF binding sites was however specific (207 genes uniquely bound by dCTCF; P‐value = 1e‐8) as confirmed by the enrichment in dCTCF sites in the promoters of genes uniquely regulated by dMes‐4 but not Beaf32 (122 genes; 50.8%; P‐value = 1e‐70). Such specificity was confirmed by inspecting our data at various threshold settings using receiver operating characteristic (ROC) curve analysis (Supplementary Fig S4E). We therefore conclude that dMes‐4 is an important co‐regulator of Beaf32 and dCTCF, which may in part account for high expression levels of genes flanked by IBP sites (Bushey et al, 2009; Negre et al, 2010).
H3K36me2/me3 patterns correlate with NFRs and nucleosome positioning
dMes‐4‐mediated H3K36me2 appears to be a key mechanism for the recruitment of various HATs that would favor nucleosome remodeling/eviction over NFRs (Bell et al, 2007; Venkatesh et al, 2012). We thus sought to test its impact on nucleosome positioning. Heat maps generated by ranking genes according to the ChIP‐Seq counts of H3K36me2 (see Materials and Methods) showed a good correlation between H3K36me2 levels and the presence of NFRs in promoter regions (Fig 4A), as confirmed by inspection of averaged nucleosome profiles (Fig 4C). H3K36me2 levels further correlated with nucleosome positioning (Fig 4A and C), yet a more significant correlation was detected by ranking genes with H3K36me3 levels (Fig 4, compare panel A and B, C and D). These results are highly consistent with recent findings showing that by preventing interaction of histone H3 with chaperones, H3K36me3 prevents histone exchange along gene bodies (Venkatesh et al, 2012), which may in turn drive nucleosome positioning as shown (Fig 4B and D).
Tri‐methylation of H3K36 occurs over the bodies of genes, and it requires the HMT Hypb/Set2 family (Bell et al, 2007). Beaf32 binds to promoters and it may not be directly responsible for nucleosome positioning over gene bodies. dMes‐4 is however pre‐required for subsequent tri‐methylation of H3K36 (Bell et al, 2007) that further involves Hypb/Set2‐mediated H3K36me3 upon transcription elongation (Joshi & Struhl, 2005; Govind et al, 2010) (see below). As such, Beaf32/dMes‐4 may preset chromatin for subsequent H3K36me3‐driven nucleosome positioning. Supporting this view, DE genes upon dMes4‐KD were enriched among genes harboring high nucleosome positioning in wild‐type cells (Supplementary Fig S5A), as well as genes with defects in nucleosome positioning upon Beaf32‐KD (Supplementary Fig S5B). These DE genes were mostly enriched in high H3K36me2 levels even in the absence of Beaf32 (Supplementary Fig S5C), supporting a strong linkage between this mark and the impact of dMes‐4 on gene expression. Higher H3K36me3 levels were found in DE genes upon dMes‐4‐KD, providing Beaf32 was bound. In agreement, DE genes by dMes‐4‐KD harbored high levels of nucleosome positioning when Beaf32 was bound to their promoters, as shown in wild‐type cells (Supplementary Fig S5A and D). Taken altogether, these data suggested that Beaf32/dMes‐4 are involved in presetting chromatin for subsequent H3K36me3 deposition upon transcriptional activation, as addressed below.
Beaf32/dMes‐4 preset chromatin for subsequent regulation of RNA splicing
We next sought to test whether interactions between Beaf32 and dMes‐4 are required for subsequent deposition of H3K36me3 and/or for nucleosome positioning. Ranking of genes according to Beaf32 ChIP‐Seq reads in promoters showed a tight correlation with H3K36me3 levels in the corresponding gene bodies (Supplementary Fig S6A). ChIP analysis showed no significant variations in H3K36me3 levels upon Beaf32‐KD compared to control cells, for promoters bound by Beaf32 or not (Supplementary Fig S6A and B). Genome‐wide, little variations of H3K36me3 levels may be found in promoter regions (Supplementary Fig S6C, see arrow), however, which may reflect an equilibrium between dMes‐4/Hybp HMTs and demethylases that are recruited to promoter regions (Lin et al, 2012) (see Discussion). By contrast, H3K36 methylation decorates gene bodies (Supplementary Fig S6C) (Bell et al, 2007; Kolasinska‐Zwierz et al, 2009) where H3K36me3 levels decreased upon Beaf32‐KD providing the gene was flanked by a Beaf32 site (Fig 5A; P‐value = 1e‐4), which would affect nucleosome positioning. Strongly supporting this hypothesis, the nucleosome positioning defects scored upon Beaf32‐KD tightly correlated with H3K36me3 levels (Fig 5B, red bars) and to a lesser extent with H3K36me2 (Fig 5B, grey bars). Altogether, these data therefore strongly supported that Beaf32 binding to promoters is required for subsequent transcriptional elongation, thereby mediating H3K36me3‐coupled nucleosome positioning (see below).
H3K36 methylation may serve to drive nucleosome positioning, thereby preventing spurious transcription and the presence of aberrant RNAs as shown (Supplementary Fig S2D) (Carrozza et al, 2005). Alternatively, such aberrant transcripts (scored by counting RNASeq counts outside exons) may be related to the influence of H3K36me3‐coupled transcriptional elongation on RNA splicing (Kolasinska‐Zwierz et al, 2009; Schwartz et al, 2009; Luco et al, 2011; Shukla et al, 2011). H3K36 marks may therefore play a pivotal role coupling nucleosome positioning and RNA splicing (reviewed in Schwartz & Ast, 2010). Genome‐wide analysis of RNASeq reads using DiffSplice (Hu et al, 2013) showed a significant accumulation of reads from unspliced RNAs upon Beaf32‐KD compared to control cells (Fig 5C), which was specific for genes bound by IBPs (P‐value = 1e‐19), as confirmed by RT‐qPCR analyses using oligos that span exon–intron junctions (Supplementary Fig S7A). They were tightly correlated with H3K36me3 levels over gene bodies, varying from < 3 to approximately 50% (Supplementary Fig S7B, red bars) and to a lesser extent with H3K36me2 levels (Supplementary Fig S7B, grey bars). In complete agreement, a high proportion of the splicing defects observed in Beaf32‐KD were also found upon dMes4‐KD compared to WT cells (> 57%; P‐value = 1e‐300; Fig 5C). They were specifically encountered for genes flanked by a Beaf32 together or not with a dCTCF binding site (Fig 5D; “RI”, retained introns; 391 and 242 genes, respectively), strengthening the overall implication of IBPs in RNA splicing through dMes4‐mediated H3K36 methylation. Specific alternative splicing defects were also scored upon Beaf32‐KD giving rise to 229 alternative transcripts with skipped exons (Fig 5D, “ES”), similar to what was reported for murine CTCF in alternative splicing (Shukla et al, 2011). The alternative splicing defects detected upon Beaf32‐KD were specific of genes harboring both Beaf32 and dCTCF binding sites (Beaf+/− dCTCF; P‐value = 1e‐6 and 1, respectively), which may be related to the role of murine CTCF in alternative splicing (Shukla et al, 2011). Such defects were also encountered upon dMes4‐KD in the presence of dCTCF sites (52 genes; P‐value = 1e‐6) in complete agreement with our results implicating H3K36 methylation in IBP‐mediated regulation of gene expression. Altogether, our data therefore show a role of insulator proteins in RNA splicing involving their interaction with the H3K36 HMTs.
Beaf32/dMes‐4 are required for the recruitment of the transcriptional activator DREF
Our data show that the observed defects upon Beaf32‐KD involve H3K36me3 deposition. This may likely involve the HMT dHypb/Set2 whose activity has been associated with transcription activation/elongation (Joshi & Struhl, 2005; Govind et al, 2010). The binding sites of Beaf32 largely overlap with that of the transcriptional activator DREF (Emberly et al, 2008; Gurudatta et al, 2013) and we thus sought to characterize the interplay between Beaf32/dMes‐4 or DREF and H3K36 methylation.
ChIP analysis showed that Beaf32‐KD specifically impaired the recruitment of DREF to Beaf32 bound promoters as compared to control cells (Fig 6A), which was not found for control promoters. Of interest, dMes4‐KD impaired the recruitment of DREF to Beaf32 bound promoters as compared to control promoters (Fig 6B), supporting its role in opening chromatin. The reduction in DREF recruitment was more significant upon Beaf32‐KD as compared to dMes‐4‐KD, which may involve direct interactions of DREF with Beaf32 as suggested by co‐immunoprecipitation experiments (Supplementary Fig S8F). In contrast to Beaf32‐KD, DREF‐KD did not affect the recruitment of dMes‐4 (Supplementary Fig S8E). Curiously, its depletion was accompanied by a significant increase in H3K36me2 levels for promoters bound by Beaf32 (Fig 6C), similarly to the depletion of HypB (Supplementary Fig S6D). These results raise the possibility that DREF favors the Hypb‐mediated H3K36me2 to H3K36me3 transition, upon transcriptional activation. Strongly supporting this view, DREF‐KD led to a significant decrease over gene bodies of H3K36me3—but not H3K36me2—levels, as compared to control cells (Fig 6D and E). Moreover, ranking of genes according to their relative enrichment in H3K36me3–over H3K36me2 marks showed a tight correlation with the elongation rate of RNA polymerase II (Fig 6F; see Materials and Methods), showing that H3K36me3‐mediated nucleosome positioning reflects transcriptional elongation.
Taken altogether, our result therefore shows that IBP/dMes‐4 play an important role by coupling the recruitment of transcriptional activators, such as DREF, to the presetting of chromatin through H3K36me2 that is required for subsequent H3K36me3‐mediated RNA splicing (see Discussion).
Beaf32 and dMes‐4 regulate H3K27me3 spreading
The tight correlation among the various genomic features including IBP binding, H3K36me2/3, and the presence of NFRs was statistically relevant as shown by clustering analyses (Fig 7A), in stark contrast with the clear anti‐correlation of these features with H3K27me3 levels (Supplementary Fig S9A). IBP binding sites are enriched at the borders of H3K27me3 domains (Sexton et al, 2012), prompting us to test the function of Beaf32/dMes‐4 in regulating H3K27me3 spreading. Our MNase‐Seq analysis followed by ChIP‐Seq of H3K27me3 showed a significant increase in the averaged levels of H3K27me3 upon Beaf32‐KD as compared to control mock‐depleted cells, for a limited subset of 990 genes (Fig 7B). Of the promoters bound by Beaf32, 11.9% harbored higher H3K27me3 levels upon Beaf32‐KD as compared to only 4.1% without a Beaf32 site (Fig 7C; P‐value = 1e‐12 and 1, respectively). Beaf32 specifically protected genes from H3K27me3 spreading as 49.5% (480/990) of the genes with higher H3K27me3 levels upon Beaf‐KD corresponded to genes bound by Beaf32. Furthermore, a significant proportion of the genes exposed to H3K27me3 spreading intersected with the genes with increasing nucleosome levels within NFRs (326/990 genes; P‐value = 1e‐6; Fig 1) highlighting a good correlation between Beaf32/dMes‐4‐driven chromatin dynamics and such phenotype. Genes exposed to H3K27me3 spreading were further enriched among the differentially expressed genes (~44%), as compared to genes with Beaf32 binding yet without no variation in H3K27me3 levels upon Beaf32‐KD (~26%; Fig 7D; see also Supplementary Fig S10A‐C). Similarly, genes exposed to H3K27me3 spreading upon Beaf32‐KD were more enriched in differentially regulated genes upon dMes4‐KD as compared to genes with no variations in H3K27me3 (52 versus 33%; Fig 7D).
Our data thus suggested a role of Beaf32 and dMes‐4 in controlling the deposition of H3K27me3, in agreement with recent data involving NSD/MES4 in the regulation of H3K27me3 (see Discussion) (Yuan et al, 2011; Gaydos et al, 2012). Accordingly, ChIP analysis of H3K27me3 upon depletion of dMes‐4 or of Beaf32 showed a reproducible and significant increase in H3K27me3 levels as compared to control cells (Fig 7E). The effect of dMes4‐KD on H3K27me3 levels was specific of promoters harboring a Beaf32 binding site as compared to control promoters (P = 1e‐4 and 1, respectively), accounting for the good correlation between the impact of dMes4‐KD on H3K27me3 levels, compared to that of Beaf32‐KD (Supplementary Fig S10D). Taken altogether, our results suggest that the influence of Beaf32/dMes‐4 on H3K27me3 deposition may reflect their activity in chromatin dynamics, yet it may not condition their more general influence on gene expression (see Discussion).
The identification of dMes‐4 as a novel cofactor interacting with IBPs shed new light into how they may impact gene expression through chromatin dynamics. Such regulations involve cycles of NSD/dMes‐4‐mediated H3K36me2 followed by Set2/Hypb‐mediated H3K36me3 (Venkatesh et al, 2012) together with the recruitment and/or activation of various HATs (Bell et al, 2007; Venkatesh et al, 2012) and HDACs (Barski et al, 2007; Schones et al, 2008) including Rpd3 (Joshi & Struhl, 2005; Govind et al, 2010). In turn, such cycles may provide with a highly dynamic regulation of chromatin locally, involving the eviction or fixation of histones/nucleosomes, respectively (Venkatesh et al, 2012), which appears to play a pivotal role in nucleosome positioning. These genomic features are statistically relevant (Fig 7A; Supplementary Fig S9A), summarizing strong links between H3K36 di‐ or tri‐methylation, NFRs, and nucleosome positioning, respectively, depending on the presence of insulator protein sites and of DREF (see model, Fig 8).
NSD/dMes‐4 may account for the enrichment of “active” histone marks, NFRs, or DHSs at IBP sites including CTCF (Natarajan et al, 2012; Thurman et al, 2012). H3K36 methylated marks and IBPs are enriched at the borders of globular H3K27me3 domains (Gurudatta & Corces, 2009; Negre et al, 2010; Schwartz et al, 2012; Sexton et al, 2012). Our data further suggest that the role of IBPs in restricting H3K27me3 spreading involves the recruitment of dMes‐4/NSD. H3K36 methylation actually antagonizes PRC2‐dependent H3K27me3 as shown in HeLa cells (Yuan et al, 2011). In C. elegans, MES‐4 interferes with the spreading of this repressive mark as shown (Gaydos et al, 2012), which may account for the global exclusion of H3K27me3 and H3K36me2/3 marks. Of interest, these two histone marks define bivalent nucleosomes that participate in controlling developmentally regulated genes through PRC2 recruitment (Cai et al, 2013). By interacting with H3K36 HMTs, IBPs like Beaf32 may thus regulate H3K27 deposition through a dynamic interplay with transcription‐coupled chromatin dynamics involving H3K36 methylation.
dMes‐4/Hypb‐driven H3K36 methylation participates in the activation of housekeeping genes flanked by insulator sites that, unlike genes developmentally regulated “paused genes” (Gilchrist et al, 2010), harbor high nucleosome positioning over their bodies. dCTCF/Beaf32 sites flank promoters particularly enriched in genes that regulate specific cell functions including the cell cycle (Emberly et al, 2008; Bushey et al, 2009; Gurudatta et al, 2013). By regulating chromatin accessibility, IBP and dMes‐4 may participate in the formation of NFRs, thereby favoring the recruitment of transcriptional activators including DREF. Beaf32/dMes‐4‐mediated H3K36me2 is pre‐required for subsequent Hybp/Set2‐driven tri‐methylation of H3K36 upon transcription elongation (Bell et al, 2007; Kolasinska‐Zwierz et al, 2009). This directionality is evidenced by increasing levels of H3K36me2 upon depletion of—DREF or more directly of—dHypb (Bell et al, 2007). Such mechanisms involving the evolutionary conserved factors, DREF, or dMes‐4/NSD also identified as Wolf–Hirschhorn syndrome proteins (e.g. WHSC1), are likely to be essential for the regulation of genes involved in cell cycle and/or proliferation.
The role of IBPs/dMes‐4 appears not limited to the regulation of nucleosome occupancy over NFRs as H3K36 methylation functions both in chromatin organization by triggering nucleosome positioning and in RNA splicing (Kolasinska‐Zwierz et al, 2009; reviewed in Schwartz & Ast, 2010; Luco et al, 2011). H3K36me3 regulates RNA splicing by favoring the recruitment of splicing factors concomitantly with transcriptional elongation and vice versa, and splicing activity influences nucleosome organization (Keren‐Shaul et al, 2013) raising the potent interplay between RNA splicing and nucleosome positioning through H3K36 marks (reviewed in Schwartz & Ast, 2010). The role of IBPs and/or dMes4/H3K36me2 may be to couple the recruitment of transcriptional activators such as DREF with the concomitant presetting of chromatin for Hypb‐dependent H3K36me3 deposition, which requires prior H3K36me2 (Bell et al, 2007). Splicing defects may thus be exacerbated when transcriptional elongation is uncoupled from H3K36me3 (and by extension H3K36me2) deposition. Supporting this idea, DREF is required for H3K36me3—but not for H3K36me2—deposition along gene bodies, and its depletion has a lower impact on RNA splicing as compared to Beaf32 (Supplementary Fig S9B). Additional IBPs including CTCF and the insulator barrier protein Vezf1 were shown to affect alternative RNA splicing in other organisms including mouse (Shukla et al, 2011; Gowher et al, 2012), involving their ability to pause RNAPII by binding directly over exon–intron junctions. H3K36 marks are enriched within such genomic locations (Kolasinska‐Zwierz et al, 2009) raising the possibility that CTCF‐mediated regulation of alternative splicing similarly involves its interaction with dMes4/NSD‐driven H3K36 methylation.
Only a sub‐fraction of the binding sites of Beaf32 or dCTCF appears to influence H3K27me3 levels (Schwartz et al, 2012) (this work). The large overlap between Beaf32 and dCTCF binding sites may in part account for such limited impact, if these IBPs play a redundant function. Alternatively, the influence of IBPs on H3K27me3 may depend on the recruitment of additional cofactors of IBPs such as CP190, cohesin or chromator and their contribution to the physical organization of chromosomes into topological domains (Wood et al, 2011; Dixon et al, 2012; Hou et al, 2012; Sanyal et al, 2012; Phillips‐Cremins & Corces, 2013). By regulating the chromatin accessibility, the interaction of IBPs with HMTs such as dMes‐4 may act as a platform for the recruitment of additional IBP cofactors (see Fig 8) as illustrated for cohesin loading (reviewed in Dorsett, 2011; Fasulo et al, 2012). A division of labor was suggested in the context of the beta‐globin insulator, where USF1/2‐VEZF and CTCF distinctly contribute to local chromatin accessibility involving histone acetylation and higher‐order chromatin organization (Ghirlando et al, 2012). The division of labor is thus not limited to various sets of DNA–protein interactions, defining which IBP (e.g. dCTCF or Beaf32) interacts with chromosomal borders, but also extends to interactions with multiple cofactors that influence each other. How IBPs participate in restraining H3K27me3 spreading over domain borders may further reflect such a dependence on additional IBP cofactors participating in transcription‐coupled chromatin dynamics, such as dMes‐4.
dMes‐4 is an histone modifier implicated in epigenetic mechanisms of germ line cells involving an interplay with additional HMTs of the Maternal‐effect sterile (MES) or set family involved in X‐chromosome regulations (Pirrotta, 2002). Such HMTs may dictate specific genomic landscapes contributing to regulate genes, including those associated with dosage compensation (Kharchenko et al, 2011; Regnard et al, 2011; Schwartz et al, 2012). The recruitment of dMes‐4 by IBPs may thus provide with key information regarding genomic features specific to cell types (male or female germ cell lines) or to chromatin domains. Following dMes‐4/HATs and dSET2/HDACs, the resulting chromatin is deacetylated, yet resetting H3K36 methylation additionally requires the JmjC‐domain‐containing factor dKdM4A that also demethylates H3K9 (Fig 8) (Lin et al, 2012), which may be controlled by IBPs including Beaf32 (Emberly et al, 2008). Of interest, dKdm4A is brought to the borders of repressed domains through its interaction with HP1a, thereby regulating repressive marks (Lin et al, 2012). The regulation of genes at chromosomal borders thus likely involves both a highly dynamic organization of chromatin locally, through multiple interactions of IBPs with NSD/dMes‐4 HMTs and additional players including HP1 with histone demethylases, and higher‐order chromatin organization through a distinct set of IBP cofactors.
Materials and Methods
RNAi, genome‐wide expression data (RT‐qPCR, microarray, DGE‐Seq, and RNASeq)
Drosophila Schneider SL2 cells were treated with specific interfering RNAs to knock down Beaf32, dMes‐4, DREF, or control RNA as a mock control essentially as previously described (Emberly et al, 2008). For RNAi‐mediated depletion, T7‐driven synthesis (Fermentas TranscriptAid™ T7 High Yield Transcription Kit) of double‐stranded RNAs (dsRNA) specific for beaf32, dmes‐4, dref, or control dsRNA (against luciferase; for mock‐depleted control cells) was checked for potential off‐target effects using NCBI primer designing tool and dsCheck (http://dsCheck.RNAi.jp/; Emberly et al, 2008) was used. 400/400/900 μg of dsRNA were added to 30 millions cells in 10 ml media without FBS for beaf32/dref/dmes‐4 depletion, respectively. Cells were incubated for 2 h at 25°C, and 20 ml of media with FBS were added. After incubation for 5 days, the cells were harvested, followed by RNA extraction (Qiagen RNeasy; deep‐seq and RT‐qPCR) or formaldehyde cross‐linking (ChIP). Samples were analyzed as replicates by RNA sequencing (DGE/RNA‐seq; HiSeq2000; Illumina; INRA platform (GeT‐PlaGe; (http://genomique.genotoul.fr/intranet/)) or BGI) using Tophat or the Burrows Wheeler Alignment tool (BWA) software (default parameters) on genome annotations of release 5.41 of Flybase for parsing, HTseq for counting the reads, and DEGseq package to identify differentially expressed genes (P‐value < 0.01) from independent replicates. Data were independently verified through analyses by RT‐qPCR and microarray analysis for validation (Supplementary Fig S1) using Fisher's exact test by intersecting the groups of differentially expressed genes. Parsing was performed using the TopHat or Burrows Wheeler Alignment tool (BWA) software (default parameters) on genome annotations of release 5.41 of Flybase for parsing, HTseq for counting the reads, and DEGseq package to identify differentially expressed genes (adjusted P‐value < 0.01) from four independent replicates for RNASeq. Data were independently verified through analyses by RT‐qPCR and microarray analysis for validation (Supplementary Fig S1F and G) using Fisher's exact test by intersecting the groups of differentially expressed genes. Gene expression was measured by real‐time PCR analysis using cDNAs prepared from S2 control, Beaf32 KD, and dMes‐4‐depleted or control cells, for the following (control or Beaf32 bound promoters) genes: control: CG9988, CG13766, CG9520, CG17715, sara, gapdh2; Beaf32 bound: B‐tub56D, vha100‐2, crc, ter‐94, CG5284, tsp39D, spin, ror, kel, CG4210. Data were analyzed using the absolute quantification component of pyQPCR. Quality of mRNA levels was controlled with Experion (Biorad) and quantified in parallel with at least five different concentrations of cDNA and genomic DNA for standard curves (see example in Supplementary Fig S10E) for all oligos used in this study using Biorad iQ SYBR Green Supermix in an Eppendorf Realplex. qPCR or RT‐qPCR data were analyzed using the relative quantification component of pyQPCR software developed by M. Hennion and T. Gastine accessible at http://pyqpcr.sourceforge.net/ verifying that the quality of the data obtained with Eppendorf Realplex by confidence interval (95%) or Applied Biosystems Viia7. Further analyses of post‐transcriptional defects (Supplementary Figs S2 and S9) were performed using DiffSplice to quantify significant splicing defects (Hu et al, 2013) or by measuring RNASeq reads in introns normalized to the reads in exons. Data were verified by RT‐qPCR using Applied biosystem Viia7 with oligos that span exon–intron junctions, and immature RNA levels were normalized to mRNA levels measured with exon‐specific oligos for the same list of genes used to measure expression except CG9988 (due to low expression) and the addition of the Ucp4A gene. For intersection analysis between genome‐wide expression analyses of Beaf32 or dCTCF and/or dMes‐4, receiver operating characteristics (ROC) analysis was performed using R as a convenient classifier tool (Fawcett, 2006) to visualize true positives (e.g. dCTCF or Beaf32 binding) independently of the threshold of differential expression. In the context of genes flanked by IBPs, Fisher's exact was used to measure the relative enrichment in differentially regulated genes as a function of which binding sites were present with respect to the total number of genes of the same category (binding site or other genomic features).
Chromatin immunoprecipitations (ChIP‐Seq)
ChIP‐Seq data of H3K36me2, H3K36me3, H3K27me3, and RNA polymerase II (“Pol II”) were performed using exponentially growing S2 cells essentially as previously described (Emberly et al, 2008; Schones et al, 2008). Chromatin pellets were analyzed using Qubit fluorometer and Agilent (2100) before library construction and SE‐ or PE‐sequencing at BGI. Data were analyzed using a rMAT package by comparison with read counts obtained by precipitating chromatin with IgG controls (not shown). For pair‐end sequencing, data were further processed by mapping positive and negative reads separately (average distance ~82.5 bp) followed by Gaussian smoothing to generate genome‐wide profiles. The average fragment size and variance was calculated using the spatial correlation function between the + and − reads (25 bp reads from either the 5′ (+) or 3′ (−) end of each fragment) by making a histogram of the distances between all + and − reads across the genome. Each read's contribution was obtained by shifting (~41 bp) +/− by +/− half the average fragment size providing a Gaussian density with a fixed standard deviation (50 bp) for genome‐wide profiles (the total score at a given location is the sum of all the read densities). For ranking of genes according to the log ratio of H3K36me3 over H3K36me2 levels, normalized ChIP‐Seq reads were counted over gene bodies (+500 to end of gene). RNAPII elongation was estimated by the ratio of ChIP‐Seq reads in the same region normalized to the amount of reads in promoters, as previously done (Rahl et al, 2010). For motif search, the MEME program (http://meme.sdsc.edu/meme4_6_1/intro.html) was used.
ChIP of dMes‐4, DREF, H4K16ac, H3K36me2/me3, and H3K27me3 were analyzed in depleted (Beaf32, dMes‐4, DREF, or mock‐deplete control/WT) cells, using equivalent amounts of chromatin prepared from Beaf32‐KD or WT cells used for immunoprecipitations in triplicates. DNA was analyzed by real‐time qPCR using Applied Biosystems Viia7 and a two‐sided Wilcoxon paired test. Statistical analysis was performed using a two‐sided Wilcoxon paired test between two conditions for each of the two groups (control or Beaf32 bound promoters): control: CG9988, CG13766, CG9520, CG17715, sara, gapdh2; Beaf32 bound: B‐tub56D, tina‐1, vha100‐2, crc, chico, CG5284, tsp39D, spin, ror, CG4210. Data were analyzed using the absolute quantification component of pyQPCR. For ChIP with anti‐H3K27me3 antibodies in dMes4‐KD, Beaf32‐KD versus control cells, a total of 25 genes were analyzed including the same list of genes or additional genes selected based on variations in H3K27me3 levels as measured by ChIP‐Seq in Beaf32‐KD compared to control cells (CG3812, CG9422, CG11851, CG3408, sav, CG11975, CG2264, CG14683, Prx). To score such variations in H3K27me3 levels, ChIP‐seq reads were counted over a +/− 1 kbp region surrounding every Drosophila TSS genome‐wide. A z‐score was then defined for each gene/TSS as the number of ChIP‐Seq reads in the window in Beaf32‐KD minus WT control normalized to the square root of the mean of reads, providing a list of 990 genes increasing H3K27me3 levels as shown (z > 2; Fig 7). The significance of such variations and further correlation with dMes4‐KD was further tested through three independent ChIP analyses by qPCR measurements of the variations in Beaf32‐KD/WT compared to variations in dMes4‐KD/WT, providing a determination coefficient (R2) of 0.57 and 0.08 for Beaf32 bound and control (unbound) promoters, respectively (Supplementary Fig S10D).
Nucleosome positioning by MNase‐Seq
Genome‐wide nucleosome positioning (MNase‐seq) was measured from exponentially growing control (WT control) or Beaf32‐depleted (“Beaf32KD”) cells after purifying nucleosomes followed by pair‐end sequencing essentially as previously described (Schones et al, 2008). Nucleosome profiles upon Beaf32 depletion were generated after normalization to the total number of reads in WT using Gaussian smoothing of the raw read counts at each position in the genome for both plus and minus reads. A shift of 80 bp and a standard deviation of 20 bp were applied to generate smoothened nucleosome profiles. ChIP‐Seq of H3K27me3 was performed on chromatin digested with MNase (Schones et al, 2008).
Antibodies, biochemical purification of IBP cofactors
Affinity‐purified antibodies specific of dMes‐4 or DREF were generated essentially as described (Cuvier & Hirano, 2003) using recombinant DREF protein injected in mouse or using the middle region of dMes‐4 injected in rabbit as previously (Bell et al, 2007). Anti‐Beaf32 antibodies were generated in the laboratory as described (Liang et al, 2014). Antibody specificity was confirmed by Western blotting of nuclear extracts and by competition with peptides (Supplementary Fig S1) and by RNAi‐mediated depletion. Additional antibodies were purchased: anti‐H3 (Abcam ab1791), H3K36me3 (Abcam, ab9050), H3K27me3 (Upstate #07‐449), H4K16ac (Active motif), anti‐RNAPII (Covance), H3K36me2 (Upstate #07‐369) anti‐actin (Sigma).
Nuclear extracts, Western blotting, chromatin purification, purification of IBP cofactors
Chromatin was prepared from purified nuclei by mild detergent extraction following purification of chromatin‐associated proteins through a 30% sucrose cushion essentially as previously described (Cuvier et al, 2008). Nuclear extracts were prepared by lysing purified nuclei in NEB‐360 mM KCl (NEB: 10 mM Hepes KOH pH 7.6; 3 mM MgCl2; 0.1 mM EDTA/K; 0.1% Trasylol; 0.2 mM PMSF; 10% glycerol) for 30 min at 4°C and spinning them at 158,000 g for 1 h. nuclear extracts (10 μg/well) were separated on 4–16% SDS–PAGE and passively transferred to nitrocellulose membranes as previously described (Cuvier et al, 2002). Filter strips were then incubated with the indicated antibody ± 0.4 mg/ml of antigen peptide (e.g. Supplementary Fig S1), followed by horseradish peroxidase‐conjugated donkey anti‐rabbit (1:10,000) (GE Healthcare). Signals were developed using an ECL Plus detection kit (GE Healthcare) and FUJIFILM Luminescent Image Analyzer LAS‐4000 for images acquisition. Signal quantification was performed through MultiGauge software following the manufacturer instructions. Identification of dMes‐4 was allowed by high‐salt elution of cofactors associated with affinity‐purified Beaf32 complex using affinity‐purified anti‐Beaf32 antibodies as compared to IgG control essentially as previously described (Cuvier & Hirano, 2003). Columns were then extensively washed before bound proteins were eluted by adding one bed volume (0.5 ml) of 0.6 M NaCl (“high‐salt” extraction) followed by analysis on SDS–PAGE (Fig 2B) and mass‐spectrometric analyses.
Yeast two‐hybrid assays
For yeast two‐hybrid, bait constructs and target constructs including beaf32, dmes‐4, or cp190 were cloned into pDest22 and pDest32 to express fusion proteins and interactions were evaluated for various combinations as a function of yeast ura3 reporter expression essentially according to the manufacturer's instructions (Invitrogen) by growing the yeast strain MaV203 in replicated cells on nonselective (+Ura) and selective (−Ura) media followed by incubation at 30°C for 2 days on SD media.
Statistical analyses by principal component analysis (PCA) and clustering ascendant hierarchical (CAH)
Statistical analyses by PCA were performed using the package FactoMiner from R (Josse et al, 2011) by taking the total read counts as measured from our ChIP‐Seq data of Beaf32, H3K36me2, H3K36me3, from additional ChIP‐Seq available at modENCODE (H4K16ac and H3K36me1) (Kharchenko et al, 2011; Alekseyenko et al, 2012), or from our MNase‐Seq data in Beaf32‐KD or control cells. ChIP‐Seq or MNase‐Seq data were analyzed by comparison with read counts obtained by precipitating chromatin with IgG controls by counting reads within the indicated windows corresponding to gene bodies and/or promoters defined by Flybase (release 5.41). Reads were counted within windows corresponding to gene bodies (“H3K36me2/3”, gene body from +500 to the end of each gene) or from promoter regions (from −300 to 0 with respect to TSS). Nucleosome positioning was estimated by adding the maximum peak intensity (in read counts) obtained from three separated widows (+100, +150), (+290, +320), (470, 500) and corresponding to +1, +2 and +3 nucleosome windows. Nucleosome Free Regions (NFRs) were calculated by normalizing read counts in the promoter window (−150 to 0 from each TSS) normalized to the average read counts in neighboring regions (−1,000 to −500). The generated averaged values were then provided for each ChIP‐Seq or MNase‐Seq for each individual gene allowing to perform clustering ascendant hierarchical (CAH) for all centered and normalized data using the package FactoMiner from R. CAH measures Ward distances reflecting the minimal variance among all data sets provided (Josse et al, 2011), which was performed separately for Beaf32 bound or control genes. Analysis of the influence of dMes‐4 with respect to Beaf32 or dCTCF peaks was based upon intersection analysis between RNA‐Seq data and ChIP‐Seq using Fisher's exact test or pairwise Wilcoxon test.
All genomic sequencing data including ChIP‐Seq of H3K36me2 /H3K36me3/Pol II, MNase‐Seq of bulk or H3K27me3 nucleosomes in WT or Beaf32‐KD cells (“nuc‐WT/ nuc‐Beaf32‐KD”), and RNASeq data in WT, Beaf32‐KD or dMes‐4‐KD cells are available at NCBI (GEO accession number GSE57168) and are accessible through our genome browser links: http://insulators_chromosome-dynamics.biotoul.fr/IBPs
PL, MH, SC, and OC designed experiments, performed research, and interpreted data. MNase‐ and ChIP‐Seq experiments were performed by MH, SC, and PL under the supervision of OC and KZ. AG conducted all bioinformatic and statistical analyzes with OC and with the help of EE. GM performed RNA splicing analyses. Protein purifications were done and analyzed by SU and OC. Confirmations of interactions were done by PL and JL. PL, MH, DS, PM, and SQ performed the RNAi experiments and expression studies by qPCR, DGE, or RNASeq. OC designed experiments and directed the study, with the help of KZ and EE. PL, MG, JL, AG, and OC prepared the figures. OC wrote the manuscript.
Conflict of interest
The authors declare that they have no conflict of interest.
Supplementary Figure S1
Supplementary Figure S2
Supplementary Figure S3
Supplementary Figure S4
Supplementary Figure S5
Supplementary Figure S6
Supplementary Figure S7
Supplementary Figure S8
Supplementary Figure S9
Supplementary Figure S10
Supplementary Table S1
We thank Artem Barski and Kairong Cui for performing the Hi‐Seq sequencing of MNase data, S. Gadat and G. Fichant for suggestion regarding statistical analyses, L. Lacroix and additional OC's laboratory members for critical reading of the manuscript, C. Carles for help with the web server, P. Martin for technical help. E.E.'s laboratory was supported by NSERC and the Canadian Institute For Advanced Research (CIFAR) and by the University of Toulouse (while in O.C.'s laboratory). K.Z.'s laboratory was supported by the Division of Intramural Research Program of the National Heart, Lung and Blood Institute, NIH. P.L. and M.H. were supported by a fellowship from La Ligue Nationale Contre le Cancer (LNCC). O.C.'s laboratory was supported by grants from the Region Midi‐Pyrénées, the Cancer Research funding of the ARC, the ATIP‐AVENIR program of the CNRS and Inserm joint program and the ANR “INSULa”.
- © 2014 The Authors