CTCF depletion alters chromatin structure and transcription of myeloid-specific factors

Differentiation is a multistep process tightly regulated and controlled by complex transcription factor networks. Here, we show that the rate of differentiation of common myeloid precursor cells increases after depletion of CTCF, a protein emerging as a potential key factor regulating higher-order chromatin structure. We identified CTCF binding in the vicinity of important transcription factors regulating myeloid differentiation and showed that CTCF depletion impacts on the expression of these genes in concordance with the observed acceleration of the myeloid commitment. Furthermore, we observed a loss of the histone variant H2A.Z within the selected promoter regions and an increase in non-coding RNA transcription upstream of these genes. Both abnormalities suggest a global chromatin structure destabilization and an associated increase of non-productive transcription in response to CTCF depletion but do not drive the CTCF-mediated transcription alterations of the neighbouring genes. Finally, we detected a transient eviction of CTCF at the Egr1 locus in correlation with Egr1 peak of expression in response to lipopolysaccharide (LPS) treatment in macrophages. This eviction is also correlated with the expression of an antisense non-coding RNA transcribing through the CTCF-binding region indicating that non-coding RNA transcription could be the cause and the consequence of CTCF eviction.


Introduction
Macrophages, the resident tissue phagocytes and sentinels of the innate immune response, arise from the differentiation of haematopoietic stem cells (HSCs) through a hierarchical process of cell fate decision and lineage commitment.This process is dictated by the instructive action of complex transcription factor networks, which will progressively restrict the fate of differentiation as well as reducing the proliferation potential of the cells.Commonly, the transition from proliferation to differentiation involves the downregulation of pluripotency transcription factors like c-Myc (Henriksson and Luscher, 1996;Grandori et al., 2000) and activation of lineage-specific genes.The emergence of common myeloid precursors (CMPs) from multipotent progenitors is considered to depend on the simultaneous expression of both transcription factors PU.1 and GATA.1 (Nerlov and Graf, 1998;Seshasayee et al., 1998;Heyworth et al., 2002).Upregulation of PU.1 activity and expression of the CCAAT/enhancer-binding protein a (C/EBPa) are critical for CMPs to commit to the granulocytic and monocytic lineages (Zhang et al., 1997;Rhodes et al., 2005).Downstream of PU.1 and C/EBPa, Egr1 and Egr2 function redundantly to both activate macrophage cell fate and repress neutrophil differentiation programme (Laslo et al., 2006).The DNA-binding factor CTCF has also been shown to control myeloid cell growth and differentiation but the molecular mechanism driving this function is still unclear (Torrano et al., 2005).CTCF was first identified as a transcriptional repressor of the chicken lysozyme (cLys) and c-Myc (Baniahmad et al., 1990;Lobanenkov et al., 1990) and its impact on myeloid differentiation was at least partially dependent on its role in the regulation of the latter (Torrano et al., 2005).CTCF is differentially expressed and posttranslationally modified, depending on the particular differentiation pathway within the myeloid lineage (Delgado et al., 1999).Interestingly, CTCF plays a specific role in some developmental processes but not in others.For example, CTCF depletion in mouse thymus leads to cell growth inhibition, cell cycle arrest, and lack of differentiation of ab T cells with no effect on gd T cell maturation (Heath et al., 2008).CTCF can also regulate differentiation of nonhematologic cells (Soshnikova et al., 2010;Hirayama et al., 2012).
CTCF is a highly conserved 11-zinc finger protein ubiquitously expressed and implicated in multiple regulatory functions, including transcriptional activation and repression, silencing, enhancer blocking, gene insulation, imprinting, X chromosome inactivation, and long-range chromatin interactions (Phillips and Corces, 2009).CTCF occupancy studies revealed up to 55000 CTCF-binding sites in the various cell lines investigated, with most of these sites being cell type non-specific and with binding patterns which do not correlate with transcriptional start sites (Barski et al., 2007;Kim et al., 2007;Wang et al., 2012), arguing that CTCF is not a classical transcription factor.CTCF is the only identified vertebrate protein with insulator activity.Insulators affect gene expression by blocking unsuitable interaction between neighbouring regulatory elements and by preventing the spread of repressive heterochromatin (Wallace and Felsenfeld, 2007).It has been proposed that CTCF drives enhancer-blocking insulation through long-range chromatin interactions, in particular through loop formation between distal elements (Hou et al., 2008;Phillips and Corces, 2009), in association with cohesin (Parelho et al., 2008;Wendt et al., 2008).Increasing evidence suggests that changes in higher-order genome structure and subnuclear chromatin localization are crucial for lineage specification and tissue-specific transcriptional regulation (Misteli, 2007).Because of its involvement in mediating long-range chromatin loops at specific developmentally regulated genomic loci, CTCF has been suggested to control the genome-wide organization of chromatin architecture (Phillips and Corces, 2009).However, given the conservation of CTCF occupancy across different cell types, it is unlikely that CTCF alone could dynamically organize the three-dimensional structure of the genome.CTCF is also believed to impact locally on the expression of neighbouring genes.Studies conducted in Drosophila have shown that new CTCF binding appeared during evolution with the gain or loss of CTCF binding impacting on the expression levels of nearby genes (Ni et al., 2012).Furthermore, at least 20%-25% of CTCF binding is proximal to promoters (Barski et al., 2007;Kim et al., 2007;Jothi et al., 2008), with CTCF interacting with RNA polymerase II (RNAPII) and conferring promoter activity to a promoter-less reporter gene when stably integrated to the genome (Chernukhin et al., 2007).On the other hand, CTCF can provoke RNAPII pausing at promoters (Paredes et al., 2013) and also within coding regions where the presence of CTCF plays a role in alternative splicing and where this presence is regulated by DNA methylation (Shukla et al., 2011) as observed for 41% of CTCF sites not occupied in all cell types (Wang et al., 2012).CTCF association with DNA can also be regulated by transcription of non-coding RNA (Lefevre et al., 2008).CTCF-binding sites are flanked by at least 20 positioned nucleosomes enriched in the nucleosome variant H2A.Z and histone marks associated with active chromatin structure (Fu et al., 2008;Jothi et al., 2008).
To clarify the role of CTCF in myeloid differentiation, we knocked down CTCF in CMPs from murine bone marrow and monitored differentiation in the absence of differentiation-driving stimulus.By analysing CTCF-binding site databases (Barski et al., 2007), we identified binding sites in the vicinity of the transcription start sites (TSSs) of the myeloid-specific genes Egr1, Egr2, C/EBPa, C/ EBPb, and Runx1.Though Runx1, which belongs to the family of Runt-related transcription factors, is essential to the establishment of definitive HSCs but no longer to haematopoiesis (Chen et al., 2009), Runx1-depleted mice demonstrate expansion of the myeloid lineage (Growney et al., 2005).Similarly, ectopic expression of C/EBPb, a transcription factor from the C/EBP family, induces granulocytic differentiation in vitro (Popernack et al., 2001).Changes in the expression of all the selected transcription factors would therefore affect myeloid differentiation.Using the PUER model of myeloid terminal differentiation (Laslo et al., 2006), we looked at changes in gene expression and chromatin organization around CTCF-binding sites in the vicinity of the selected genes.We report that CTCF/cohesin enrichment at these binding sites varies and correlates with local modifications of the chromatin structure but does not allow prediction of associated modification in gene expression.In addition, all the selected genes show a loss of the histone variant H2A.Z at the promoter, as well as an increased intergenic transcription upstream of the TSS.These data suggest that CTCF locally preserves promoter chromatin organization, controls RNAPII pausing, and prevents cryptic transcription.However, this common feature of CTCF-associated promoters does not appear to play a major role in the transcription of the genes, since these genes are variably affected by CTCF knockdown.

CTCF knockdown increases the rate of CMP differentiation
To obtain an insight into the role of CTCF in macrophage differentiation, we purified CMPs from murine bone marrow using a previously published strategy (Tagoh et al., 2002) and transfected these cells by retroviral infection using a retroviral construct expressing an shRNA targeting CTCF.The CMPs (c-kit + /PECAM-1 + /Ly6C 2 ) were purified by magnetic cell separation followed by cell sorting (Figure 1A) or by magnetic cell separation only (Supplementary Figure S1A).After transfection, these CMPs were cultured for 7 more days in 'expansion' medium.PECAM-1 is a marker for morphologically undifferentiated myeloid blast cells, whereas Ly6C recognizes committed macrophage precursors (late CFU-M) and monocytes.In these conditions, we did not detect any difference in cell growth or apoptosis when comparing cells transfected with the shRNA anti-CTCF or the shRNA control targeting the firefly luciferase gene (FF3).However, FACS analysis revealed an 2-fold reduction of the PECAM-1 + /Ly6C 2 compartment in CTCF-depleted cells compared with the control and, reciprocally, a 60% increase in the PECAM-1 2 /Ly6C + compartment (Figure 1A and Supplementary Figure S1A), without any effect on colonyforming capacity (Figure 1B).These results indicate that CTCF depletion favours differentiation over cell proliferation, without changing the myeloid cell fate decision.CTCF has been identified in the promoter regions of several key factors of myeloid differentiation in human (Barski et al., 2007).By comparing human and mouse sequences, we identified CTCF consensus sequences conserved in both species for Egr1, Egr2, C/EBPa, and C/EBPb and in the two promoters regions of Runx1 (Supplementary Table S2 and Figure S1B).Then, we analysed the expression of c-Myc and these lineage-specific transcription factors in the three described fractions after separation by cell sorting (Figure 1C).We measured Runx1c and Runx1b expression, two isoforms transcribed from Runx1 upstream (P1) and downstream (P2) promoters, respectively.No difference in expression was detected in the earliest differentiation stage (PECAM-1 + /Ly6C 2 ), but Egr1 and Runx1c were upregulated and C/EBPa downregulated in the CTCF-depleted double-positive compartment (PECAM-1 + /Ly6C + ) (Figure 1C).The expression levels for these three genes were comparable between CTCF-depleted cells in the double-positive compartment and the control cells (PECAM-1 2 /Ly6C + ).Moreover, in the PECAM-1 2 /Ly6C + compartment Egr2 and Runx1c were upregulated and c-Myc downregulated (Figure 1C).Taken together, these data revealed that CTCF knockdown induced spontaneous differentiation of CMPs towards the myeloid lineages.However, in this system, we were not able to establish whether the detected changes in expression of the selected genes were CTCF or differentiation dependent.

CTCF and cohesin occupancy at promoters of key transcription factors vary during myeloid differentiation
To investigate the impact of CTCF depletion on the expression of the CTCF-associated genes in myeloid differentiation, we use the PUER system (Laslo et al., 2006).In the absence of 4-hydroxy-tamoxifen (OHT), PUER cells retain their myeloid progenitor morphology and, upon OHT addition and activation of the PU.1 protein, undergo rapid growth arrest, and differentiate into adherent macrophages.In these cells, upon 48 h of OHT addition, Egr1 and Egr2 were upregulated by 4-and 24-fold, respectively, whereas c-Myc, C/EBPa, and C/EBPb were downregulated and Runx1b and c were not affected (Supplementary Figure S2A).This OHT-dependent differentiation into monocytes was linked to variations in CTCF/cohesin (Rad21) binding at the selected binding sites (Supplementary Figure S2B and C).All the selected binding sites were occupied by CTCF in vivo with the weakest site (Runx1P1) showing 3-fold enrichment in CTCF occupancy over a Myc 220 kb control region.Interestingly, all the loci underwent specific changes in CTCF/cohesin (Rad21) binding but these changes were not correlated with the observed changes in gene expression.For example, Egr1 and Egr2, but also C/EBPa and c-Myc, were associated with a similar increase in CTCF and Rad21 recruitment (Supplementary Figure S2).
Next, we knocked down CTCF in these cells by transfection of the retroviral constructs detailed in the 'CMPs' experiments using electroporation.Forty-eight hours after transfection, the efficiency of CTCF depletion was assessed by western blot (Figure 2H).We did not detect any changes in cell morphology after CTCF depletion, indicating that the cells retained their progenitor morphology in absence of both PU.1 and CTCF and were normally differentiating upon OHT addition.Consequently, potential changes in gene expression observed between control and CTCF-depleted cells were not an indirect consequence of CTCF-induced differentiation.In these transfected cells cultured with or without OHT for 48 h, the expression of the CTCF-associated genes was analysed (Figure 2A-G).C-Myc transcription was downregulated upon OHT addition, but was not affected by CTCF depletion in agreement with the results obtained in CMPs (Figure 2G).In addition, CTCF knockdown variably affected the expression of the CTCFassociated genes.C/EBPa and C/EBPb were upregulated in both non-OHT-and OHT-treated cells (Figure 2C and D) and Egr2 was upregulated in non-OHT-treated cells and downregulated together with Runx1c upon OHT addition (Figure 2B and F).Finally, Egr1 was upregulated in differentiated myeloid cells but not in the non-OHT-treated fraction (Figure 2A).Taken together, these data were highlighting an apparent random impact of CTCF depletion on CTCF-associated gene expression.In addition, overall, the activation of Egr2, C/EBPa, and C/EBPb, observed in CTCF-depleted PUER, was coherent with an accelerated differentiation rate of CTCF-less myeloid progenitors compared with wild type.

CTCF knockdown alters chromatin structure and increases short transcription at the Egr1 locus
To look in more detail at the consequences of CTCF depletion on gene expression and chromatin structure, we focused on the Egr1 locus.Egr1 was consistently upregulated in CMPs and in PUER.Furthermore, Egr1 was not upregulated in the absence of PU.1, allowing for the determination of CTCF-dependent chromatin alterations specifically associated with changes in gene expression.CTCF depletion was associated with loss of CTCF and Rad21 from their binding sites in both progenitor and monocyte cells (Figure 3A and Supplementary Figure S3A).This loss was correlated with H2A.Z depletion in the promoter region of Egr1 (Figure 3D).This H2A.Z depletion was not a consequence of a global reduction in H2A.Z protein level as confirmed by western blot (Supplementary Figure S4A).Interestingly, upon OHT addition and increased Egr1 transcription, cohesin (Rad21) recruitment to the CTCF-binding site in 5 ′ of the TSS was consistently enhanced (Figure 3A and Supplementary Figure S2C).This change in Rad21 recruitment coincided with an augmentation of both total histone H3 positioned upstream of the CTCF-binding site and H3K9ac detected in the promoter proximal region downstream of CTCF (Figure 3B and C).Such observation would be consistent with a role for the CTCF/cohesin complex as a boundary element between the 'more compacted' upstream region and the 'more opened' downstream region, H3K9ac being a histone mark associated with active chromatin.Moreover, in absence of the CTCF/cohesin complex, the amount of total histone H3 did not vary in the distal promoter region upon OHT addition.This absence of 'compaction' of the region upstream of CTCF did not correlate with any spreading of H3K9ac from the promoter, but with a slight increase in acetylation at the CTCF-binding site and 24.8 kb upstream of the TSS, where a cisregulatory element has been previously identified (Weigelt et al., 2009) (Figure 3B and C).Furthermore, H3K9me1 did not show any CTCF-dependent change (Supplementary Figure S3B).Additional nucleosome mappings performed by quantitative PCR of Micrococcal Nuclease-treated (MNase) chromatin showed two positioned nucleosomes between the CTCF-binding site and the nucleosome-free region upstream of the TSS (Figure 3E and F).In the CTCF-depleted cells, the more upstream of the two nucleosomes was slightly moving over the CTCF site in PUER cells independently of the OHT treatment (Figure 3E and F).In conclusion, we observed local changes in chromatin structure around the Egr1 promoter dependent on CTCF depletion.However, the accumulation of total histone H3 upstream of CTCF observed upon OHT addition and lost in CTCF-depleted cells was the only difference specific of monocytes.In addition, this increase in total histone H3 upstream of the promoter region correlated with the differentiation and upregulation of Egr1, whereas this absence of chromatin reorganization in CTCF-depleted cells led to a further activation of the gene suggesting that compaction of this region of the locus was not conditioning transcriptional activation.Therefore, the described CTCF-associated alterations of the chromatin structure were not sufficient to mediate changes in Egr1 expression.
Because CTCF has been shown to increase RNAPII pausing index, we looked at the recruitment of the different isoforms of RNAPII at the Egr1 locus in PUER cells upon addition of OHT for 48 h in CTCF-depleted or control cells (Figure 4 and Supplementary Figure S3C and D).As expected, the unphosphorylated form of RNAPII C-terminal domain (CTD) decreased at the region immediately upstream of the TSS when CTCF was depleted in correlation with a slight increase in Egr1 mRNA level and primary transcript (Figures 2A, 4B and F).Similarly, the elongating form of RNAPII phosphorylated at serine 2 (RNAPII S2p) was slightly increased in the 3 ′ end of the gene (Figure 4D).However, the major changes in RNAPII distribution within the Egr1 locus were seen upstream of the Egr1 TSS, where the unphosphorylated RNAPII was the main polymerase form enriched after CTCF depletion (Figure 4A and B and Supplementary Figure S3C).This upstream accumulation of RNAPII upon CTCF depletion was correlated with an increase in intergenic transcription (Figure 4E and Supplementary Figure S4B).This transcription was driven in the antisense direction compared with Egr1, as determined by preparing cDNA using biotinylated primers specific to the sense and antisense orientations, the 'sense' cDNA preparation not giving any product (Figure 4E).We also measured intergenic transcription by PCR using several primer pairs designed downstream, upstream, and at the CTCF site.The result indicated that short transcripts terminating before the CTCF-binding site were induced by CTCF depletion  (Supplementary Figure S4B).Therefore, this 'aberrant' antisense transcript did not exist as a short transcript in the presence of CTCF or as a longer one upon CTCF depletion, but was controlled by CTCF at the initiation level.In addition, primary transcription was also activated by CTCF depletion in PUER cells, whereas Egr1 mRNA level was not (Figures 2A and 4F), arguing for an accumulation of short and/or unprocessed transcripts firing from the Egr1 promoter in sense orientation.Changes in chromatin structure and intergenic transcription also characterize other CTCF-associated genes For the other CTCF-associated genes, the loss of CTCF was also associated with the loss of cohesin (Rad21) in the promoter region when the CTCF site is localized upstream of (Egr2) or within (C/EBPa, C/EBPb, and c-Myc) the proximal promoter (Figure 5A, Supplementary Figure S1B).H2A.Z depletion at the promoter of the selected genes was a general characteristic of CTCF knockdown (Figure 5D).In addition, local changes in nucleosome organization were detected for c-Myc, for which downregulation upon differentiation was associated with an intensification of nucleosome occupancy within the promoter, and the loss of CTCF and H2A.Z was correlated with this nucleosome positioning in undifferentiated PUER (Figure 5B).Similarly, CTCF depletion correlated with nucleosome reorganization at the C/EBPa and C/EBPb promoters, but independently of the differentiation stage (Figure 5B).These changes could be local and comparable to the changes in chromatin organization observed at the Egr1 locus (Figure 3E and F).No significant change in histone H3K9 acetylation was seen for these genes (Figure 5C).Furthermore, as described for Egr1, these loci expressed intergenic transcripts, expression of which was amplified by CTCF/H2A.Z depletion (Figure 5F).At the same time, RNAPII occupancy at the promoter decreased (C/EBPa) or was maintained at the same level indicating a change in pausing index and turnover rather than an increased accumulation of RNAPII (Figure 5E).In PUER, RNAPII occupied the promoter of every selected gene prior to and after OHT addition (Supplementary Figure S5), justifying the existence of this detected transcription independently of the differentiation stage.

LPS treatment induces transient CTCF eviction from the Egr1 locus in macrophages
We have previously shown that LPS induces CTCF/cohesin eviction from a binding site 2.4 kb upstream of cLys TSS (Lefevre et al., 2008).Because LPS transiently induces Egr1 in macrophages, we decided to determine whether its transcriptional activation was also correlated with CTCF eviction in activated macrophages.In the macrophage derived cell line Raw264.7,Egr1 expression was slightly higher than in PUER after 48 h OHT treatment (Supplementary Figure S6A).Nucleosome mapping revealed that the two nucleosomes trapped between the proximal promoter and CTCF-binding site were more strictly positioned than in PUER cells in correlation with a strongest binding of CTCF to its binding site in RAW264.7 cell line (Supplementary Figure S6B and C).This data suggested a dynamic competition between CTCF and the adjacent nucleosome for the positioning over the DNA fragment containing the CTCF consensus sequence.Moreover, as expected, Egr1 mRNA level transiently increased to reach a peak at 30 min after LPS addition corresponding to 100-fold of the basal expression level (Figure 6A).Egr1 antisense transcript followed a similar kinetic, peaking 30 min post-LPS addition and being 3-fold more expressed than the observed basal level (Figure 6B).ChIP experiments looking at RNAPII S2p, CTCF, and cohesin (Rad21) recruitment showed that increased Egr1 antisense transcription was correlated with an augmentation of RNAPII S2p occupancy in the region bound by CTCF and with concomitant losses of CTCF and cohesin (Rad21) from their binding site (Figure 6C-E).We observed a 2-fold increase in RNAPII S2p recruitment for 35% loss in CTCF binding (Figure 6C and D) compared with no change in RNAPII S2p for almost 100% of CTCF depletion seen in PUER upon OHT addition.This eviction was transient and seen after 30 min of LPS addition only.This eviction is not seen for sites occupied by CTCF within NOS2, Egr2, SOCS3, PU.1, and LTB loci (Figure 6D and E).In addition, as described for cLys (Lefevre et al., 2008), CTCF/cohesin (Rad21) eviction from the Egr1 locus was accompanied by the transient sliding of the adjacent nucleosome over the CTCF-binding site (Figure 6F).Looking at RNAPII S2p enrichment within the coding or 5 ′ regions of Egr1, we detected 20-fold more polymerase within the coding region compared with the regulatory region (Figure 7A).Interestingly, Egr1 transcriptional activation induced by LPS was accompanied by a transient loss of the histone variant H2A.Z in the proximal promoter (Figure 7C).This decrease in H2A.Z-containing nucleosomes was less marked than that after CTCF depletion in PUER and associated with amplification of H3K9 acetylation without any observable change in total histone H3 (Figure 7B and D).Treatment of cells with 5,6-Dichloro-1b-D-ribofuranosylbenzimidazole (DRB), an inhibitor of transcription elongation, together with LPS treatment, prevented CTCF eviction in agreement with a role for the antisense transcript in CTCF eviction (Supplementary Figure S6D).Taken together, these data revealed that LPS-mediated Egr1 and cLys transcriptional activations follow the same mechanism of non-coding RNA-associated CTCF eviction.For cLys, gene expression and CTCF depletion are maintained for at least 24 h post-LPS stimulation, whereas this mechanism was transient for Egr1.

Discussion
We observed an accelerated spontaneous differentiation of CTCF-depleted CMPs towards the myeloid lineage, but could not detect any changes in c-Myc expression in contrast with previous studies suggesting that CTCF was repressing transcription of this transcription factor (Gombert et al., 2003;Lutz et al., 2003;Torrano et al., 2005).In human embryonic stem cells (hESCs), CTCF depletion, similarly, accelerated loss of pluripotency upon BM4-induced differentiation (Balakrishnan et al., 2012).In contrast with our system, in which spontaneous commitment towards the myeloid lineage occurred in the absence of IL-1, IL-3, and M-CSF, hESCs needed stimulation to escape the pluripotent stage.In consequence, CTCF can be seen, in both studies, as a brake controlling the rate of differentiation without preventing it.Such function would be consistent with a reported downregulation of CTCF during the differentiation of dendritic cells and myeloid leukaemic cell lines (Delgado et al., 1999;Koesters et al., 2007).In opposition to this 'myeloid differentiation repressor' function, CTCF depletion has previously been shown to induce cell proliferation of myeloid leukaemia cells (Torrano et al., 2005).These apparently contradicting observations might be explained by the widespread DNA  methylation characteristic of cancer cells, which prevents CTCF binding to DNA (Wang et al., 2012).In this context, CTCF depletion may have dramatically different outcomes in primary and immortalized cells.Overall, though initial studies mostly conducted with cell lines suggested that CTCF was a tumour suppressor, this work and others' illustrate that the CTCF-induced alterations of cell proliferation and differentiation are context-dependent and potentially mediated by the lineage-specific genes for which CTCF is located within the promoter region.
Genome-wide analysis of CTCF binding revealed that CTCF was frequently found within the promoter of transcription factors important in cell fate decision (Barski et al., 2007;Kim et al., 2007), suggesting that CTCF could repress differentiation by preventing the expression of these lineage-specific transcription factors.This is particularly true for the myeloid lineages, for which CTCF is found within the first 2 kb of the promoter regions of some of the main transcription factors driving differentiation (Egr1, Egr2, C/EBPa, C/EBPb, and Runx1).Transcriptional alteration of these genes would impact on myeloid differentiation as previously described (Zhang et al., 1997;Popernack et al., 2001;Growney et al., 2005;Rhodes et al., 2005;Laslo et al., 2006).Analysing the impact of CTCF knockdown on gene expression in CMPderived cells was difficult since gene expression alterations can induce differentiation and reciprocally.In contrast, PUER cell differentiation was PU.1-dependent and CTCF depletion did not overcome PU.1-dependent differentiation blockage.In CTCF knockdown cells, Egr2, C/EBPa, and C/EBPb were upregulated.In addition, Egr1 was upregulated upon PU.1 activation.Egr1 is a downstream target of PU.1 (Laslo et al., 2006) and, in PUER cells, a PU.1-dependent cis-regulatory element has been identified 24.8 kb upstream of Egr1 TSS (Weigelt et al., 2009).These data, therefore, suggest a CTCF-dependent overexpression of both Egr1/2 and C/EBP in myeloid precursor cells, which is consistent with the acceleration of myeloid differentiation revealed in CMPs.
CTCF-binding to DNA is believed to impact on chromatin structure (Fu et al., 2008;Lefevre et al., 2008).In consequence, local changes in chromatin organization upon CTCF depletion may impact on gene expression.When looking at chromatin structure in the vicinity of the CTCF-depleted binding sites, the most striking observations were the loss of H2A.Z at the promoter of all the selected genes and an increase in intergenic transcription upstream of the TSS.To date, the systematic association between CTCF and H2A.Z at promoters has not been shown.However, the amount of CTCF binding identified in genome-wide analyses is variable depending on the studies, and weak CTCF-binding sites might have been ignored, the threshold between true and false positives being commonly difficult to establish.In consequence, CTCF might be more broadly associated with promoter regions than anticipated and/or H2A.Z might be maintained in CTCF-less promoter regions by CTCF-independent mechanisms.In this respect, several promoter types have been described exhibiting different initiation patterns ranging from dispersed regions to focused start sites, CTCF and H2A.Z being associated with the latter and with a higher presence of RNAPII (Rach et al., 2011).Accordingly, at promoters lacking CTCF protein, RNAPII might not be stalled and could be specifically recruited upon induction.In addition, our results suggest that CTCF stabilizes the stalled polymerase attached to the promoter (Figure 7E).In this process, the order of events is unclear.H2A.Z is commonly associated with CTCF (Guastafierro et al., 2008;Jothi et al., 2008) and CTCF depletion could therefore mechanically lead to H2A.Z depletion.For example, the CTCF/cohesin complex could prevent INO80, the complex controlling either the eviction of the H2A.Z/H2B dimers or the loss of H2A.Z-containing nucleosomes (Papamichos-Chronakis et al., 2011), accessing chromatin, and H2A.Z could prevent cryptic transcription at promoters.On the other hand, INO80 may be targeted to the coding regions of many genes transcribed by RNAPII through interactions with the transcription elongation complex (Klopf et al., 2009;Venters and Pugh, 2009) and H2A.Z removal may therefore be a consequence of cryptic transcription.The second hypothesis is reinforced by the fact that CTCF has been shown to bind directly to RNAPII and to increase the pausing index of the polymerase with an efficiency directly correlated with the distance between the CTCF site and the proximal promoter (Chernukhin et al., 2007;Paredes et al., 2013) (Figure 7E).Furthermore, though H2A.Z is preferentially located at the promoters of active genes in higher eukaryotes, it tends to be reduced from promoters after transcription activation (Papamichos-Chronakis et al., 2011) and loss of H2A.Z has a relatively minor effect on gene expression profiles, at least in yeast, typically affecting only the transcriptional kinetics of a subset of inducible genes (Meneghini et al., 2003).Similarly, the impact of H2A.Z and intergenic transcription on the observed changes in gene expression upon CTCF depletion could be determined by the balance between non-productive and productive transcription associated with the reduction of RNAPII pausing index.However, as illustrated by Egr1, identical H2A.Z depletion and intergenic expression were observed in PUER with or without OHT, but the outcome of CTCF depletion was different.This suggests a limited role of both H2A.Z depletion and cryptic transcription in the variation detected in mRNA level.
Therefore, CTCF-dependent transcriptional alterations of the identified lineage-specific transcription factors happened through different mechanisms.Additional local and gene-specific CTCF-dependent chromatin modifications could, for example, play a role in transcription.At cLys, CTCF prevents full occupancy of the enhancer immediately upstream by blocking a nucleosome over a C/EBP-binding site (Lefevre et al., 2008).For Egr1, in PUER, CTCF eviction was associated with nucleosome repositioning over the CTCF-binding site and, at the promoters of C/EBPa and C/EBPb, with changes in total histone H3 in the region bound by CTCF.However, how these local nucleosome reorganizations would impact on the expression of the downstream genes is unclear.Moreover, CTCF is, to date, essentially described as an insulator interfering with inappropriate communication between neighbouring regulatory elements and/or independent chromatin domains (Phillips and Corces, 2009).In this respect, we were expecting to see a correlation between changes in gene expression and the CTCF/cohesin ratio at CTCF-binding sites.Though the binding pattern of CTCF does not appear to predict ES cell-specific gene expression (Chen et al., 2008) and though CTCF occupancy is constitutive and remains unchanged during B cell differentiation at the mouse immunoglobulin heavy-chain (Igh) locus, cohesin is progressively recruited to CTCF-bound sites in a cell type-specific manner that parallels conformational changes in locus topology (Degner  , 2009).However, at the selected genes, neither CTCF nor cohesin (Rad21) occupancy followed a pattern, which would have allowed prediction of the expression level of the downstream genes.This absence of correlation between expression and CTCF/ cohesin occupancy could be explained by the fact that CTCF function is modulated by neighbouring DNA-binding factors (Weth and Renkawitz, 2011) or by post-translational modifications (PTMs) altering CTCF function without change in DNA binding (Klenova et al., 2001;Yu et al., 2004;Torrano et al., 2006;MacPherson et al., 2009).However, to date none of these PTMs have formally linked CTCF to any specific function.In addition, a recent study has shown that CTCF-mediated chromatin loops were marginally regulated during pre-pro-B cell into pro-B cell differentiation, supporting a role for CTCF in the organization of intra-chromosomal domains, either transcriptionally inert or permissive (Dolnik et al., 2012).Traditional regulatory functions of CTCF, including transcriptional activation or repression, would be secondary effects of its ubiquitous and essential role as organizer of chromatin architecture (Phillips and Corces, 2009).Consequently, CTCF depletion was accelerating differentiation, but the 'normal' differentiation process may not depend on CTCF and CTCF/cohesin might therefore not play any role in the regulation of the selected lineage-specific genes.This hypothesis is challenged by the experiments performed with LPS in macrophages, in which both Egr1 and cLys gene activations are accompanied by the dynamic eviction of CTCF.The presence of a PU.1-dependent cis-regulatory element upstream of the CTCF-bound region at the Egr1 locus suggests that CTCF/cohesin may block promoter-enhancer communication in the absence of LPS stimulus and that this active enhancer would be the main driver of Egr1 activation in CTCF-depleted cells.
Interestingly, in activated macrophages, the picture at the Egr1 locus after 30 min of LPS treatment was very similar to CTCF-depleted PUER cells.CTCF and H2A.Z were transiently down, whereas the expression of Egr1 mRNA and antisense transcript increased.However, DRB treatment blocked CTCF eviction, suggesting that antisense transcription mediated CTCF eviction after LPS treatment and not the other way around as observed in CTCF-depleted cells (Figure 7E).The state of the polymerase driving this antisense transcription in either CTCF-depleted or LPS-treated cells could explain how CTCF eviction could be the cause or the consequence of non-coding RNA transcription.At the LPS-inducible promoter driving cLys-associated non-coding RNA transcription, RNAPII recruitment and transcription initiation were simultaneous and the presence of CTCF had no impact on the transcription of this non-coding RNA (Lefevre et al., 2008).Similarly, LPS-dependent transcriptional activation of the Egr1associated antisense RNA was concomitant with enrichment in elongating polymerase within the transcribed region.In contrast, the same transcribing region was mainly enriched for the unphosphorylated form of RNAPII after CTCF knockdown, CTCF interacting more robustly with this form compared with the phosphorylated one (Chernukhin et al., 2007).Taken together, these observations suggested that CTCF controls the stabilization of the RNAPII at the promoter prior to transcription initiation but does not repress transcription initiation per se.In absence of stimulus, CTCF contributes to the stability of the stalled polymerase, which will drive transcription of non-productive RNA in CTCF-depleted cells.Upon LPS addition, phosphorylation of the polymerase will reduce its interaction with CTCF and will allow productive transcription, a low percentage of the polymerase transcribing backward and inducing CTCF eviction (Figure 7E).
This CTCF-dependent non-coding RNA transcription did not significantly interfere with productive transcription in the absence of CTCF.Nonetheless, with up to 55000 binding sites, most of them located in intergenic regions, CTCF might have a global impact on non-coding RNA transcription by repressing cryptic promoters.In addition, because CTCF is associated with active chromatin marks (Fu et al., 2008) and is known to confer promoter activity to promoter-less DNA (Chernukhin et al., 2007), the presence of CTCF might also induce RNAPII recruitment to DNA outside of promoter regions.If so, CTCF could repress transcription in regions where its presence prevents silent chromatin formation.A study of the global impact of CTCF depletion on non-coding RNA transcription to evaluate the function of this protein on cryptic transcription would merit further scrutiny.

Figure 2
Figure 2 CTCF knockdown alters expression of the downstream genes in PUER cells.(A-G) mRNA level for the indicated genes in PUER cells treated with or without OHT for 48 h and transfected with shRNAs targeting CTCF or FF3 (as indicated on the horizontal axis).Results are expressed relative to GAPDH expression.Error bars represent SD from three independent biological replicates.*P , 0.05, **P , 0.01, ***P , 0.001 (unpaired t-test).Data are representative of three independent experiments.(H) Quantification of total CTCF and b-actin protein levels by western blot with total cell extract, after PUER transfection with shRNA targeting CTCF or FF3.

Figure 3 Figure 4
Figure 3 Loss of H2A.Z histone variant is induced by CTCF depletion at the Egr1 promoter.(A-D) ChIP performed in PUER cells treated or not with OHT for 48 h and transfected with shRNAs targeting CTCF or FF3, with anti-Rad21 (cohesin; A), anti-histone H3 (B), anti-H3K9ac (C), and antihistone H2A.Z (D) antibodies.Horizontal axis indicates primers used for the real-time PCR.Data are normalized versus input and then versus the control region Myc 220 kb (B and D) or TNF 20.2 kb (A and C).Error bars represent SD from three independent qPCR replicates.Data are representative of three independent experiments.(E and F) Nucleosome mapping at the Egr1 locus performed with MNase-treated chromatin from PUER cells treated without (E) or with (F) OHT for 48 h and transfected with shRNAs targeting CTCF or FF3.Horizontal axis indicates primers used for the real-time PCR depending on the distance from the TSS in kb.Data are normalized versus input and then versus the control region Egr1 22.7 kb.Secondary y-axis (right) represents paired t-test probabilities of four independent experiments formatted as a percentage and shown as black dots.Only P , 0.05 are shown in the figures.

Figure 5
Figure 5 Loss of H2A.Z histone variant and intergenic transcription characterizes CTCF-associated genes.(A -D) ChIP performed in PUER cells treated or not with OHT for 48 h and transfected with shRNAs targeting CTCF or FF3, with anti-Rad21 (cohesin; A), anti-histone H3 (B), anti-H3K9ac (C), and anti-histone H2A.Z (D) antibodies.Horizontal axis indicates primers used for the real-time PCR.Data are normalized versus input and then versus the control region Myc 220 kb (B and D) or TNF 20.2 kb (A and C).Error bars represent SD from three independent qPCR replicates.Data are representative of three independent experiments.(E) ChIP performed in PUER cells treated with OHT for 48 h and transfected with shRNAs targeting CTCF or FF3, with anti-RNAPII CTD and anti-RNAPII S2p antibodies.Data are normalized versus input and then versus the control region Myc 220 kb.Error bars represent SD from three independent qPCR replicates.Data are representative of two different experiments.(F) Intergenic transcripts expression analysis using primers indicated on the horizontal axis.Results are expressed relative to GAPDH expression.Error bars represent SD from three independent triplicates.Data are representative of three independent experiments.

Figure 6
Figure 6 CTCF is transiently evicted from its binding site at the Egr1 locus after LPS activation.(A and B) Time course of Egr1 mRNA levels (A) or antisense transcript (B) in the region occupied by CTCF in RAW264.7 cells in response to LPS treatment.Results are expressed relative to GAPDH expression.Error bars represent SD from three independent experiments.(C -E) ChIP performed with RAW264.7 cells treated with LPS for the indicated time points in minutes, with anti-RNAPII S2p (C), anti-CTCF (D), and anti-Rad21 (cohesin; E) antibodies.Horizontal axis indicates primers used for the real-time PCR.Data are normalized versus input and then versus the average of control regions (TNF +4.9 kb, Myc 220 kb and IL6).Error bars represent SD from three independent qPCR replicates.Data are representative of two different experiments.(F) Nucleosome mapping at the Egr1 locus performed with MNase-treated chromatin from RAW264.7 cells treated with LPS for the indicated time points in minutes.Horizontal axis indicates primers used for the real-time PCR depending on the distance from the TSS in kb.Data are normalized versus input and then versus the control region Egr1 22.7 kb.Horizontal axis indicates primers used for the real-time PCR.Secondary y-axis (right) represents paired t-test probabilities of four independent experiments formatted as a percentage and shown as black dots comparing 0 ′ with 30 ′ LPS treatment.Only P , 0.05 are shown in the figures.P , 0.05 are also obtained for the same primers when comparing 30 ′ with 60 ′ or 120 ′ but not when comparing 0 ′ with 60 ′ or 120 ′ .

Figure 7
Figure 7 Transient CTCF eviction mediated by LPS is accompanied by H2A.Z depletion and antisense transcription.(A-D) ChIP performed with RAW264.7 cells treated with LPS for the indicated time points in minutes, with anti-RNAPII S2p (A), anti-histone H3 (B), anti-histone H2A.Z (C), and anti-H3K9ac (D) antibodies.Horizontal axis indicates primers used for the real-time PCR.Data are normalized versus input and then versus the control region Myc 220 kb.Error bars represent SD from three independent qPCR replicates.Data are representative of two different experiments.(E) At the Egr1 locus, CTCF prevents cryptic transcription by stabilizing the paused RNAPII and potentially represses transcription, for example, by preventing enhancer-promoter communication.After CTCF depletion, the association between the RNAPII polymerase and the DNA