CTCF counter-regulates cardiomyocyte development and maturation programs in the embryonic heart

Cardiac progenitors are specified early in development and progressively differentiate and mature into fully functional cardiomyocytes. This process is controlled by an extensively studied transcriptional program. However, the regulatory events coordinating the progression of such program from development to maturation are largely unknown. Here, we show that the genome organizer CTCF is essential for cardiogenesis and that it mediates genomic interactions to coordinate cardiomyocyte differentiation and maturation in the developing heart. Inactivation of Ctcf in cardiac progenitor cells and their derivatives in vivo during development caused severe cardiac defects and death at embryonic day 12.5. Genome wide expression analysis in Ctcf mutant hearts revealed that genes controlling mitochondrial function and protein production, required for cardiomyocyte maturation, were upregulated. However, mitochondria from mutant cardiomyocytes do not mature properly. In contrast, multiple development regulatory genes near predicted heart enhancers, including genes in the IrxA cluster, were downregulated in Ctcf mutants, suggesting that CTCF promotes cardiomyocyte differentiation by facilitating enhancer-promoter interactions. Accordingly, loss of CTCF disrupts gene expression and chromatin interactions as shown by chromatin conformation capture followed by deep sequencing. Furthermore, CRISPR-mediated deletion of an intergenic CTCF site within the IrxA cluster alters gene expression in the developing heart. Thus, CTCF mediates local regulatory interactions to coordinate transcriptional programs controlling transitions in morphology and function during heart development.


Introduction
The coordinated deployment of genetic programs during lineage commitment and differentiation is a hallmark of developmental processes.Cell specification and maturation are coordinated by controlled activation and repression of specific gene expression programs.In the heart, the first functional organ in the embryo, activation of a core set of cardiogenic transcription factors controls specification of cardiac progenitor cells [1].Shortly after, high expression of genes encoding sarcomeric components defines the contractile cardiomyocyte as early as embryonic day (E) 8.5.Cardiomyocytes then mature by further sarcomere assembly [2], and increased mitochondrial biogenesis [3,4], and finally exit the cell cycle and become binucleated at early postnatal stages [5].
Even though the genes and regulatory networks controlling morphogenesis and function in the heart are well characterized [6], the events that coordinate the progression from differentiation to maturation are not understood.Recent studies using both mouse and human pluripotent cells have revealed that epigenomic landscapes and chromatin signatures dynamically change during cardiomyocyte differentiation [7,8], suggesting that chromatin structure might control cardiogenesis.Chromatin conformational changes allow physical interaction of distal regulatory elements in the genome.However, the chromatin interactions controlling expression of cardiac development and maturation are poorly understood.
The study of genome function during the last decade [9,10] has provided an initial understanding of how functional elements scattered throughout the genome act coordinately to control gene activity.The advent of tools to analyze interactions between distal regions of chromatin [11] has allowed detailed mapping of the three-dimensional genome structure [12,13] and its organization in distinct regulatory domains [14].However, how these domains are established, and their function in gene expression regulation are poorly understood.CTCF (CCCTC-binding factor) is one of the best described architectural proteins with a role in chromatin structure organization.Through sequence specific binding to DNA, CTCF acts as a barrier for heterochromatin spreading, establishes boundaries between adjacent topologically associating domains (TADs), defines insulator elements that block enhancer activity on promoters, and contributes to enhancer-promoter interactions [15,16].Loss of function studies using knock-out and knock-down approaches have shown that CTCF is essential in early embryo development [17][18][19].Conditional deletion of Ctcf in different developmental contexts leads to defects in cell cycle progression [17], increased apoptosis [20,21], and its deletion in postmitotic neurons leads to decreases in the expression of clustered protocadherin genes [22].Yet, we still do not fully understand how CTCF controls chromatin structure to coordinate gene expression.
Here, we have studied how CTCF regulates gene expression in the context of the developing mammalian heart.We deleted Ctcf in a population of cardiac progenitor cells, which results in cardiac malformations and embryonic death.Analysis of global transcriptional changes preceding morphological defects caused by loss of CTCF showed downregulation of the cardiac developmental program, and concomitant upregulation of programs involved in cardiomyocyte maturation.This suggests that Ctcf deletion causes a premature arrest of cardiac development and concomitantly promotes cardiomyocyte maturation.Thus, CTCF mediates local chromatin interactions to coordinate transcriptional programs that control developmental transitions in the heart.

Cardiac specific deletion of Ctcf leads to embryonic lethality
To address the role of CTCF during development, we deleted Ctcf in cardiac progenitor cells and their derivatives by using a floxed allele [17] and the Nkx2.5-Credriver, which starts acting as early as E7.5-E8.0 in cardiomyocyte precursors in the cardiac crescent [23].E10.5 or E11.5 mutant (Ctcf fl/fl ;Nkx2.5-Cre)embryos appeared normal and showed no gross morphological alterations in the heart (Fig 1A -1D).At E12.5, mutant embryos presented pericardial edema (Fig 1E ) and the cardiac chambers did not expand properly (Fig 1F).Histological examination showed no defects in E9.5 mutant hearts as compared to controls (Fig 1G and 1H).In E10.5 mutants, the four chambers and the atrioventricular canal formed properly, although the interventricular septum appeared slightly disorganized (Fig 1I and 1J).This defect was exacerbated by E11.5, when thinning of the myocardial wall was also evident (Fig 1K and 1L).No mutant embryos were recovered beyond E12.5 (S1 Table ).Control Nkx2.5-Cre and compound Ctcf fl/+ ; Nkx2.5-Cre heterozygotes showed normal morphology at these stages (S1 Fig).
To understand the effect of Ctcf deletion in the developing heart, we determined the time point when CTCF protein was lost in cardiomyocytes.We performed co-immunostaining for CTCF and the cardiomyocyte marker TNNT2 at different stages of heart development (S2 Fig) .In E9.5 mutant hearts, 46% of cardiomyocytes still had detectable nuclear CTCF, albeit at lower levels than in controls, in which all nuclei were double positive (S2A and S2B Fig; S2 Table).However, by E10.5 we were not able to identify any cardiomyocyte expressing CTCF in mutant hearts, although expression was present in endocardium (S2C and S2D Fig) .The same pattern was observed at E11.5 (S2E and S2F Fig) .Persistent CTCF protein in cardiomyocytes at E9.5, despite that Nkx2.5-Cre is active from E7.5-E8.0[23], could be explained by protein long half-life [24,25].It has been recently shown that 15% of CTCF is sufficient for proper function [26].Therefore, the remaining protein we observe can explain why morphological defects are not detected until E10.5.
We assessed whether cardiac defects were due to decreased cell proliferation or increased apoptosis.Numbers of cardiomyocytes positive for TUNEL staining (S3A-S3F, S3M and S3O Fig; S3 Table) or phosphorylated histone H3 (S3G-S3L, S3N and S3P Fig; S3 Table ) were not altered in mutant hearts at E10.5 and E11.5, indicating that apoptosis and proliferation were unaffected.Accordingly, we did not observe a difference in the number of cardiomyocytes between mutants and controls at these stages (S3 Table ).Our results differ from others showing that CTCF loss in other developmental systems cause increased apoptosis [20], suggesting

Transcriptional changes in the developing heart caused by Ctcf deletion
To understand the function of CTCF as a transcriptional regulator in the developing heart, we analyzed the global effects of CTCF loss on gene transcription.We performed RNA-seq analysis on hearts homozygous for cardiac-specific deletion of Ctcf (Ctcf fl/fl ;Nkx2.5-Cre);heterozygotes, with deletion of only one allele (Ctcf fl/+ ;Nkx2.5-Cre);and control heterozygotes for the floxed allele but not carrying the Nkx2.5-Credriver (Ctcf fl/+ ).To identify the earliest transcriptional effects of Ctcf deletion this analysis was performed at E10.5, when we first observed a complete loss of CTCF in the mutants.Comparison between homozygous Ctcfdeleted hearts and heterozygotes or controls yielded approximately 2,000 differentially expressed genes in each case, of which roughly half were upregulated and half downregulated (S4 Table ).Interestingly, comparison between heterozygotes and controls returned only 24 differentially expressed genes, including Ctcf itself and11 pseudogenes (S4 Table ).This suggests that, at least in the developing heart [27], one functional Ctcf allele is sufficient for correct regulation of gene expression.
Gene-ontology analysis [28] of all genes differentially expressed upon Ctcf deletion showed enrichment in terms related to developmental processes, including heart development, contractile fibers, translation and mitochondria (Fig 2A ; S5 and S6 Tables).When we analyzed upregulated and downregulated genes separately, we found a clear distinction in the functional categories enriched in each case.Numerous upregulated genes were enriched in categories related to translation and mitochondrial function.In contrast, the downregulated genes were over represented in categories related to heart development and the sarcomere (S5 and S6 Tables).
Detailed analysis revealed that more than 300 genes related to mitochondrial function were up-and down regulated in Ctcf mutant hearts.Many of such genes encode subunits of mitochondrial respiratory complexes I, III, IV and V/ATP synthase, Sdhd in Complex II, and the large and small mitochondrial ribosome subunits (S4A Fig; S6 Table).This could suggest that CTCF controls expression of core transcriptional regulators of the mitochondrial gene program.However, none of these factors, such as PGC-1α, ERRs, or NRF-1/2 [29], where dysregulated in mutant hearts (S4 Table ).Other upregulated genes are involved in translation and encode most cytoplasmic ribosomal proteins, various initiation and elongation factors, and members of the spliceosomal complex (S4B Fig; S6 Table ).

Genomic features of genes dysregulated in the CTCF-deficient heart
CTCF organizes chromatin structure and contributes to the establishment of regulatory domains in the genome [30].We found that genes misregulated in Ctcf mutant hearts do not cluster in specific genomic regions (S5A Fig), suggesting that CTCF does not control gene expression in large genomic regulatory domains similarly to what has been recently shown in embryonic stem cells [26].
We next examined CTCF chromatin binding near genes whose expression changed upon Ctcf deletion in the heart by mapping the distance between the transcriptional start site (TSS) and the nearest CTCF ChIP-seq peak obtained from published datasets on adult 8 week hearts [10].We found that the up-and downregulated genes are closer to a CTCF binding site than genes whose expression does not change upon Ctcf deletion (S5B Fig) .Arbitrarily analyzing 10 or 20 kb windows based on the above distribution, we found that TSS of down-and up-regulated genes are surrounded by CTCF binding sites more frequently than genes whose expression did not change in mutants (Fig 2B).These CTCF binding sites are conserved across multiple tissues, and we did not observe enrichment for heart-specific CTCF peaks near genes deregulated in the mutant.These results suggest that CTCF regulates gene expression mainly by binding nearby genomic regions, possibly by mediating local chromatin interactions.
CTCF could define gene regulatory domains by shielding genes from the influence of nearby enhancers or by facilitating enhancer-promoter interactions [15].In the first scenario, Ctcf loss would lead to upregulation of gene expression and to downregulation in the second.To distinguish between these two possibilities, we determined the distance between dysregulated genes and the nearest heart enhancer [10].We found that downregulated, but not upregulated, genes are significantly closer to a heart enhancer than expressed genes with no change in Ctcf mutants (S5B Fig) .Again, this pattern is preserved when analysis is restricted to a 10 or 20 kb window surrounding the TSS of differentially expressed genes (Fig 2B).As downregulated genes are enriched in developmental regulators, these results suggest that CTCF promotes enhancer-promoter interactions in genes controlling cardiac progenitor establishment and differentiation.
The previous results suggest specific genomic features of up-or downregulated genes in Ctcf mutant hearts.To further explore these features, we analyzed in more detail the distribution of CTCF binding peaks on the same dataset used above surrounding the TSS of dysregulated genes.Up-and downregulated genes showed enrichment in CTCF binding immediately upstream of the TSS, and the binding signal was higher than that in genes whose expression did not change in mutants (Fig 2C), which agrees with the previous result (Fig 2B).However, when we analyzed only the dysregulated genes belonging to development, or mitochondria and translation categories (S6 Table ) we observed a clear difference.Mitochondrial and translation genes showed increased CTCF binding near the TSS, but in developmental genes CTCF binding spread over more distal regions (Fig 2C).We searched promoter-proximal and distal sites for de novo and known sequence motifs to address the possibility that the presence of CTCF sites or other motifs would underlie the differences in distribution between categories.However, we only identified binding motifs for CTCF itself (S7 Table ).
Our results suggests that Ctcf binds near the TSS to repress genes acting in mitochondria and regulating translation, both crucial for cardiomyocyte maturation [29].In contrast, CTCF binding to genomic regions more distal to TSS promotes expression of genes located near heart enhancers and controlling cardiac development.), such as Grp75, a key stress chaperone regulating mitochondrial protein translocation, folding and functions [31]; and Tfam, the transcriptional regulator of mitochondrial DNA [32], was comparable between control and mutant hearts (Fig 3B ).

Mitochondria do not mature properly in Ctcf mutant cardiomyocytes
We asked whether upregulation of OXPHOS components favors functional assembly of respiratory complexes and supercomplexes on the mitochondrial inner membrane.Bluenative gel electrophoresis revealed assembled complexes I, IV and V, and supercomplex I+III 2 in both control and mutant hearts at E10.5 and E11.5 (Fig 3C).The mitochondrial voltagedependent anion channel (Vdac), whose assembly is independent of respiratory complexes and supercomplexes, and whose encoding gene (Vdac1-3) levels did not change in mutants (S4 Table ), was used as loading control.In control hearts, Complex I, both in the free form I or in the form of I+III 2 , and Complex V substantially increased from E10.5 to E11.5 (Fig 3C and  3D), consistent with maturation of mitochondrial OXPHOS [4].In contrast, levels of I+III 2 and I did not increase from E10.5 to E11.5 in Ctcf mutant hearts.Complex V was similarly increased between E10.5 and E11.5 in control and Ctcf mutant hearts (Fig 3C and 3D).This agrees with maturation/stabilization of the electron transport chain components being regulated independently of their production [33].Our results suggest that despite increased transcription of subunits of CI and CIII that may lead to more complexes and supercomplexes assembled at the mitochondrial inner membrane, maturation of the respiratory chain is blunted in the Ctcf mutant heart.
Transmission electron microscopy (TEM) analysis revealed immature, but overall normal mitochondria with healthy cristae packaging in Ctcf mutant cardiomyocytes at E10.5 (Fig 3E).Mitochondria in control E11.5 cardiomyocytes have a more electro-dense matrix containing more cristae than E10.5, and are embedded in newly assembled sarcomere.This ultrastructure change agrees with blue native gel electrophoresis data and is consistent with cardiomyocyte maturation occurring between E10.5 and E11.5 [3,4].Mitochondria in Ctcf mutant cardiomyocytes at E11. TEM analysis also revealed that E10.5 mutant, but not control, cardiomyocytes have long and continuous sarcomeres, which were also visible at E11.5 (Fig 3E).Quantification on images of cardiomyocytes from E10.5 hearts immunostained for α-actinin revealed comparable numbers of sarcomeric Z-bands between control and Ctcf mutants (S6C and S6D Fig) .These results suggest that sarcomere assembly, but not sarcomeric component synthesis, is premature in the Ctcf mutant embryonic heart.

Loss of CTCF disrupts the cardiac developmental program
The set of genes downregulated in Ctcf mutant hearts that are enriched in heart development functional categories include key transcription factors and members of several signaling pathways controlling cardiac development (Fig 4A).In situ hybridization confirmed downregulation of the transcription factors Nkx2-5 and Hopx in the Ctcf mutant heart at E10.5.(Fig 4B and 4C).This analysis also showed reduced expression of Nppa to a more restricted domain in the left ventricle, and loss of Pitx2 expression in the right ventricle (Fig 4D and 4E).
Our RNA-seq showed reciprocal up and downregulation of genes from the Tnnt1/Tnni3 and Tnni2/Tnnt3 troponin clusters (Fig 4F and 4G).Accordingly, in situ hybridization revealed strong downregulation of Tnnt1 and Tnni2, and upregulation of Tnni3 in atria and ventricles and of Tnnt3 in atria in the Ctcf mutant heart (Fig 4H -4K).These changes in the expression pattern of troponin genes, which are arranged in clusters in the genome, suggest that CTCF coordinates their expression during development.
Irx4, which encodes a transcription factor critical for heart development [34,35] was downregulated in Ctcf mutants.We analyzed this gene and its genomic context in more detail as a means to understand the mechanisms through which CTCF regulates transcription in vivo.Irx4 forms part of the 1.5 Mb IrxA cluster, which also contains the related Irx1 and Irx2 genes [36].All three IrxA genes are expressed in the developing heart, in distinct but partially overlapping patterns.Whereas Irx1 and Irx2 express at low levels in the interventricular septum, Irx4 is strongly expressed throughout the ventricles [37].Previous studies have shown that the IrxA cluster is regulated through long-distance gene-specific interactions [38].Furthermore, our RNA-seq analysis revealed significant upregulation of the three genes closest to Irx4 outside the IrxA cluster: Ndufs6, encoding a subunit of mitochondrial Complex I; Mrpl36, encoding a mitochondrial ribosomal protein; and Lpcat1, involved in phospholipid metabolism To confirm and extend these observations, we examined the expression of the three genes in the IrxA cluster, and of their immediate neighbor Ndufs6, in control and mutant hearts by in situ hybridization.Expression of Irx4 in mutant hearts was indistinguishable from controls at E9.5 (Fig 5B), but expression levels were strongly reduced in mutants at E10.5 (Fig 5C ) and E11.5 (Fig 5D and 5J).The RNA-seq analysis showed that Irx1 and Irx2 expression was not significantly different in mutants.However, in situ hybridization showed that their expression domains expanded from the interventricular septum to the adjacent trabecular myocardium in the right and left ventricle of Ctcf mutants (Fig 5E , 5F, 5H and 5I).In control hearts, Ndufs6 was expressed ubiquitously but with higher intensity in the ventricles, similar to Irx4, but its expression was subtly increased in mutant hearts (Fig 5G).Together, these observations suggest that loss of CTCF leads to overall dysregulation of the IrxA cluster and its neighboring genes, perhaps through modification of its 3D structure.

CTCF regulates local chromatin structure of the IrxA cluster
To uncover the function of CTCF in regulating the IrxA chromatin structure we performed chromosome conformation capture followed by deep sequencing (4C-seq) [39], using as viewpoints the promoters of Irx4 and Ndufs6 in E11.5 control and homozygous mutant (Ctcf fl/fl ; Nkx2.5-Cre) hearts (Fig 6A and 6B; S7 Fig) .In controls, the promoter of Irx4 interacted strongly with the Irx2 promoter and with specific CTCF binding sites located upstream and downstream in the Irx2/Irx4 and Irx4/Ndufs6 intergenic regions, respectively (asterisks in Fig 6A ).The CTCF binding sites were previously identified by ChIP-seq in E14.5 and 8-week hearts [10]).In mutants, the interaction of the Irx4 promoter with the Irx2/Irx4 intergenic CTCF site was lost, and new interactions appeared upstream and 350 kb downstream in the Clptm1l locus (Fig 6B).The Ndufs6 promoter established interactions with the promoters of Irx4, Lpcat1 and Clptm1l as well as with the two intergenic CTCF sites in control hearts.In CTCF mutants, interactions of the Ndufs6 promoter with both CTCF binding sites and the Irx4 promoter were lost, and only contacts with the promoters of the downstream genes Lpcat1 and Clptm1l were maintained (Fig 6B ).In controls, the Irx2/Irx4 intergenic CTCF site interacts upstream and downstream of and extended region spanning Irx4, Ndufs6 and Mrpl36, and the downstream Clptm1l locus (Fig 6B).Interaction between the CTCF site and the Irx4 promoter was greatly reduced in Ctcf mutants, reciprocating the pattern observed when using Irx4 as viewpoint.Furthermore, novel interactions appeared that extend to Irx2 (Fig 6B).Overall, these results suggest that CTCF plays a central role in organizing chromatin in the Irx4 regulatory domain by mediating the interaction of several CTCFbound sites with various promoters in the region.Removal of CTCF from the developing heart re-structures the local 3D organization of the extended IrxA cluster; this results in loss of the interaction between Irx4 and Ndufs6 with flanking CTCF sites.These results also suggest that CTCF limits chromatin interactions domains, as its loss causes an expansion of local contacts.
To unambiguously demonstrate that the CTCF binding site located in the Irx2/Irx4 intergenic region is necessary for proper expression of Irx4, we generated a mouse line in which we deleted such CTCF binding site using the CRISPR/Cas9 system [40] (S8 Fig) .Homozygote mice for the deletion are viable and fertile, as expected since Irx4 mutant mice are also viable in homozygosity and only show mild hypertrophy and compromised contractility as adults [34].We analyzed the expression of Irx4 by in situ hybridization in E10.5 embryos from this line.Irx4 was slightly reduced in the ventricles of the mutant line as compared with controls (Fig 6C and 6D).Irx4 was also ectopically expressed in the oral-esophageal region (Fig 6E and  6F), in which Irx1 is normally expressed (Fig 6G).Therefore, deletion of the Irx2/Irx4 intergenic CTCF site leads to expansion of the expression domain of Irx4, perhaps by allowing Irx1 regulatory elements to contact and activate Irx4.We have thus shown how CTCF is critical to maintain the correct chromatin structure across the IrxA cluster and neighboring genes, and that specific CTCF binding sites are essential for the proper regulation of gene expression.

Discussion
Recent years have seen substantial advances in our understanding of the relationship between chromatin structure and gene expression.It is now clear that the spatial organization of the genome sets constraints that determine how different functional elements (promoters, enhancers, and boundaries) interact with one another [14].Nevertheless, we still do not fully understand how this 3D structure is maintained or the role played by chromatin-bound factors in this process.In this study we have explored how one of these factors, CTCF, regulates gene expression and genome structure, by analyzing the effects of its loss during the development of the mammalian heart.
Deletion of Ctcf in the developing heart rapidly leads to cardiac defects and embryonic death.However, transcriptomic analysis shows that loss of CTCF does not lead to dysregulation of gene expression across large chromosomal domains; rather, changes appear to be local, as described in other developmental settings [20].Upregulated and downregulated genes are both more likely to have CTCF binding sites in their vicinity, but only downregulated genes are closer to heart enhancers [10].Downregulated genes include major regulators of the cardiac developmental program, strongly suggesting that CTCF facilitates enhancer-promoter interactions for these genes in a tissue-specific fashion.In contrast, upregulated genes are highly enriched for genes involved in mitochondrial function and protein translation, which interestingly show higher levels of CTCF binding close to their promoters.Cardiomyocyte maturation involves an increase in the demand for energy and protein production, which is accompanied by increased transcription of mitochondrial and ribosomal genes [8].Cardiomyocytes lacking CTCF prematurely activate these programs, but fail to maintain functional mitochondria despite the increase in transcription.Therefore, we observe a premature maturation of cardiomyocytes lacking CTCF, accompanied by the shutting down of developmental and patterning processes.Nevertheless, there is a lack of coordination of this precocious differentiation, leading to embryonic lethality at E12.5.
To gain insight into the relationship between CTCF, genome structure and the regulation of gene expression, we analyzed the IrxA gene complex.This complex is a paradigm of gene clustering [36], with genes expressed in overlapping but distinct domains in the developing heart [37] and separated into distinct structural and regulatory domains [38].We observed changes in the levels and pattern of IrxA gene expression in Ctcf mutants, and interestingly detected similar changes in unrelated genes neighboring Irx4.This strongly suggests that the changes in the extended IrxA cluster are caused by the loss of CTCF binding in the region, and are not a secondary effect of changes in other genes.
Analysis by 4C-seq in wild type and mutant hearts showed that when CTCF is lost, the Irx4 promoter forms fewer contacts with CTCF-flanking sites and gains interactions with regions situated outside of its regulatory domain.Furthermore, the promoter of the mitochondrial Ndufs6 gene also losses interactions with these CTCF sites.These changes in promoter interactions in Ctcf mutants were mirrored by 4C analysis of the Irx2/Irx4 intergenic CTCF site.Consequently, deletion of this CTCF site leads to reduced Irx4 cardiac expression accompanied by gain of expression in novel territories.Therefore, in this context CTCF is acting both as an insulator to separate adjacent regulatory domains, and as a facilitator of promoter-enhancer interactions to ensure proper gene expression within these domains.
The structure we have defined for Irx4 is reminiscent of the recently described superenhancer domains (SD), where regions of a few hundred kilobases located between CTCF sites are organized as insulated neighborhoods within large-scale topological domains [41].The domain we identify would contain tissue specific enhancers together with their target genes.As is the case with SDs, loss of CTCF leads to downregulation of the gene central to this domain (Irx4) and upregulation of flanking genes (Irx1 and Irx2 on one side; Ndufs6, Mrpl36, and Lpcat1 on the other).
Overall, our analysis suggests that chromatin structure is relatively stable during development and that the deletion of CTCF does not cause a marked disassembly of this organization, in line with recent reports [42].However, local intra-TAD domains loops [16,43] are affected, leading to dysregulation of genes important in the cardiac developmental program as we observe for the IrxA genes.
More puzzling is the observation that only a limited set of genes is altered by Ctcf deletion in the heart.CTCF binds to thousands of sites throughout the genome, most of which are shared between different tissues and cell lines [10,12,44].An essential role for CTCF in determining and maintaining chromatin structure and organization was therefore anticipated [30].Accordingly, constitutive loss of CTCF results in very early embryonic death [17][18][19] and its selective deletion in different developmental contexts leads to profound defects in the targeted organ or tissue, usually through increased apoptosis [20][21][22].However, in all of these cases the same pattern is observed: despite wide distribution of CTCF binding, only a fraction of expressed genes is dysregulated.Furthermore, there is little overlap among genes regulated by CTCF in different systems, indicating a context-specific role of this factor.In the developing heart, we observe the concomitant upregulation and downregulation of maturation and developmental programs, suggesting that the role of CTCF here is to maintain the coordination of expression transitions.
In this scenario, only genes subject to dynamic regulation at the time of Ctcf deletion would show changes in expression.The description of different models of neural-specific deletion of Ctcf is compatible with this interpretation.When deleted in precursors during neural development, CTCF loss leads to apoptosis and subsequent death [21,22].However, deletion in postnatal neurons results in long-term survival of Ctcf mutant neurons, but activity-induced changes of gene expression are altered [45].Together, these observations suggest that CTCF, and possibly local chromatin structure, are not necessary for basal gene activity but essential for dynamic transitions in expression.

Ethics statement
Mice were bred in the core animal facility in the Centro Nacional de Investigaciones Cardiovasculares in accordance with national and European legislation.All procedures were approved by the CNIC Committee of Animal Welfare and by the Madrid Autonomous Government Regional Ministry of the Environment and Territorial Organisation (reference number PROEX 196/14).

Mouse strains
The Ctcf floxed allele and Nkx2.5-Cre line have been previously described [17,23].Primers used for genotyping are detailed in S8 Table .Ctcf fl/+ or Ctcf fl/fl embryos were used as controls.Mice were bred in the core animal facility in the Centro Nacional de Investigaciones Cardiovasculares in accordance with national and European legislation.Some of the experiments were performed in Toronto, and they were approved by the Toronto Centre for Phenogenomics Animal Care Committee.

Histology and immunohistochemistry
Whole mount embryos were dissected in cold PBS and imaged using a Nikon SMZ1500 microscope with NIS-Elements BR 4.12.01 imaging software.For sections, embryos were collected in cold PBS and fixed in 4% PFA overnight at 4˚C, dehydrated in an ethanol series, embedded in paraffin, and sectioned at 5 μm for immunostaining amd hematoxylin and eosin, and at 7 μm for in situ hybridization.Sections were observed under an Olympus BX51 microscope and photographed with an Olympus DP71 digital camera.5 μm paraffin sections were incubated with CTCF 1:1500 (Bethyl labs A300-543A) and CT3 1:10 (Hybridoma Bank) or processed for histological analysis by hematoxylin and eosin staining.For TUNEL the Terminal Transferase recombinant kit (Roche 03 333 574 001) and biotin-16-dUTP (Roche 11 093 070 910) were used, together with anti-PH3 antibody 1:200 (Millipore 06-570) and CT3 1:10 (Hybridoma Bank); images were acquired with a Nikon A1R Confocal microscope.
Quantification of TUNEL and PH3 staining were performed with ImageJ software.Positive cells for each antibody were counted in 3-5 sections per heart.Three control and three KO hearts were used.Statistical significance was determined using one-tailed Student´s test.αactinin bands quantification was performed as previously described [3].Quantification of CTCF staining at E9.5 was performed with ImageJ software.Positive cells for CTCF were counted in 5 sections per heart.
For cardiomyocyte purification, individual E10.5 ventricles were dissected in cold PBS and incubated 5 min at 37˚C with 100 μL of 48 mg/mL Collagenase type II and pancreatin 0.025gr/ mL (plus NaCl 0.0085gr/mL) diluted in ADS buffer (5x: 0.034 g/mL NaCl, 0.238 g/mL Hepes, 0.0006 g/mL NH 2 PO 4 , 0.005 g/mL glucose, 0.002 KCl, 0.001 g/mL MgSO 4 ).After mechanical disruption by pipetting, tubes were left 2 min in the hood, liquid phase was separated, mixed with 100microL of cardiomyocyte media (4:1 mixture of of Dulbecco's Modified Eagle Medium (DMEM, 1x Gibco, 41965-039) and medium M199 (Sigma, M2154).This is supplemented with 15% inactivated bovine serum; 5mM HEPES, pH7.4; 2mM L-glutamine; and 1x penicillin-streptomycin [46] and incubated at 37˚C while enzymatic and mechanical disruption was repeated.Both fractions were pooled and centrifuged at 1000 rpm at room temperature.Pellets were resuspended in 300 μL of cardiomyocyte culture media and incubated 45 min in a 24 well plate previously coated with gelatin 1% at 37˚C.Cardiomyocytes were counted and incubated in a new 24 well plate without gelatin for 24 hrs.Cardiomyocytes were fixed with PFA diluted in cardiomyocyte media (4% for α-actinin sarcomeric staining, 3.6% for Tom20 staining) for 10 min at 4˚C, washed 2x with PBS and stored at 4˚C until used.Cardiomyocytes were blocked with BSA, and incubated ON 4˚C with 1:100 Tom20 or α-actinin.

In situ hybridization
In situ hybridization using digoxigenin-labeled probes (Roche 11277073910) was performed as described [47] and developed with anti-dioxigenin-AP (Roche 11093274910) and BM-Purple (Roche 11442 074 001).Probes for Irx1 and Irx2 were previously described [48].Probes for Nkx2.5, Nppa and Pitx2 were kindly provided by Jose Luis de la Pompa (CNIC).All other probes were generated by PCR (primers used are detailed in S8 Table ), adding the sequence of the SP6 promoter at the 5' of forward primers and of the T7 promoter at the 5' of reverse primers.Sections were observed under an Olympus BX51 microscope and photographed with an Olympus DP71 digital camera.

RNA-seq analysis
RNA-seq was conducted on three pools of six E10.5 hearts each from control (Ctcf fl/+ ), heterozygotes (Ctcf fl/+ ;Nkx2.5-Cre)and homozygous mutants (Ctcf fl/fl ;Nkx2.5-Cre).Sequencing was performed by the CNIC Genomics Unit using the GAIIx sequencer, and differential gene expression analysis was performed by the CNIC Bioinformatics Unit using the transcriptome set from Mouse Genome Reference NCBIM37 and Ensembl Gene Build version 65.GO term enrichment was done using DAVID (http://david.abcc.ncifcrf.gov)with cut-offs of minimum 10 genes and an EASE value of 0.0001 [28,49].Distances between transcriptional start sites (TSS) and between these and CTCF binding sites or predicted heart enhancers were analyzed using R. We used ChIP-seq data for CTCF in adult 8-week hearts.Although data is also available for E14.5 embryonic hearts, its quality is much lower (one replicate with 2.8M reads for E14.5; two replicates with over 8M reads each for 8w).Coordinates for CTCF binding sites, and predicted heart enhancers were obtained from the Ren Lab (http://chromosome.sdsc.edu/mouse/download.html)[10]).Heat maps were generated using R and the heatmap.2package.For density plots of CTCF ChIP signal, values +/-2kb relative to the TSS were extracted with bwtools [50] from bigWig files available at ENCODE (ENCSR000CBI), and z-scores were calculated transforming each value by substracting the mean over each chromosome and dividing by the standard deviation.After a smoothing step results were plotted using in-house scripts in R. The observed distances between TSS of gene-sets were compared with the expected proportion calculated from 5,000 groups randomly sampled without replacement taken from all expressed genes.To identify heart-specific CTCF sites, we first merged all sites identified across multiple tissues from the ENCODE data (121,147 sites) and subtracted these from the set of sites bound in heart.This analysis rendered only 1,432 heart-specific peaks that were not enriched near different sets of deregulated genes in the Ctcf mutant.
For known motif enrichment and de novo motif search using Homer [51], we defined the promoter-proximal region as the 200 bp immediately upstream from the TSS of differentialexpressed genes belonging to each of the "development", "translation" and "mitochondria" categories, and the distal region as the next 1.8 kb upstream and the 2 kb downstream of the TSSs.We then identified all CTCF peaks present in these regions for each category, and used the 200 bp sequences surrounding the summit of CTCF peaks as input.As a background, we used 2 kb upstream of the TSSs of genes expressed but unchanged in our RNAseq data that had no CTCF peaks in that region (8,445 genes).As cut-offs we used p-value < 1e-10 for de novo motif search, and a Benjamini adjusted p-value < 0,001 for known motif enrichment.
All sequencing data has been deposited at GEO under accession number GSE77644.

Transmission electron microscopy (TEM)
Embryonic hearts from E10.5 and E11.5 were isolated from the specified genotype (n = 3).Samples were fixed in 2.5% glutaraldehyde (Sigma-Aldrich), 4% formaldehyde (Electron Microscopy Sciences) in 0.1M HEPES buffer for 3-4 h at 4˚C, and processed as previously described [54].Briefly, after buffer washes samples were post-fixed for 1h at room temperature in a 1:1 solution of 2% osmium tetroxide (Electron Microscopy Sciences) and 3% aqueous potassium ferrocyanide (Sigma-Aldrich).Samples were rinsed in distilled water.Tissues were dehydrated through a graded acetone series and embedded in Spurr's Low Viscosity embedding mixture (Electron Microscopy Sciences).Ultrathin sections (60 nm) were then mounted on copper grids and stained with lead citrate.Samples were examined on a JEOL 10-10 electron microscope and analyzed by ImageJ v.1.6.0Software.

4C-sequencing
4C was performed as previously described [55,56].Briefly, hearts were dissected from control and mutant (Ctcf fl/fl ;Nkx2.5-Cre)E11.5 embryos (in order to be able to obtain more starting material).Each heart was minced and passed throw a 70 micrometers cell strainer, crosslinked with 2% PFA, frozen with liquid nitrogen, and stored at -80˚for later processing once genotypes were established.Pools of 45-65 hearts from controls or mutants were lysed and digested first with DpnII (New England Biolabs, cat.no.R0543M) followed by Csp6I (Fermentas, cat.no.ER0211).For all experiments, 100 to 200 ng of the resulting 4C template was used for the subsequent PCR reaction (primers used are detailed in S8 Table ).Sequencing was performed by the CNIC Genomics Unit.All sequencing data has been deposited at GEO under accession number GSE77644.

4C-seq data analysis
For sample comparison between controls and mutants, 4C-seq data was normalized by total weight in a window of 15 Mb surrounding the IrxA cluster.The mean and standard deviation for each fragment were calculated for each group of replicates, and the difference was determined between the means in control and mutants for each viewpoint.Data were compared using the R package edgeR [57], which applies statistical tests based on negative binomial distributions.
For contact estimation, aligned reads were assigned to their corresponding virtual first cutter digested genome fragment.Each fragment end was considered as a captured site (ligated to the viewpoint) if one or more sequences mapped to it starting at the end of the fragment and in the right orientation, facing the center of the fragment.The number of captured sites was summarized per 30 fragment window (max of 60 captured sites per window).The frequency of captured sites per window was used to fit a distance decreasing monotone function and zscores were calculated from its residuals using a modified version of FourCSeq [58].Significant contacts were considered in cases where the z-score was >2 in both replicates and deviated significantly (adjusted p value <0.05) from its normal cumulative distribution in at least one of the replicates.

CRISPR/Cas9 genomic deletions
Two guide RNAs (g5.3-g3;S8 Table ), together with the Cas9 protein with NLS (PNA Bio), were injected into the pronucleus of E0.5 B6/CBA F1 embryos, which were then transferred to CD1 pseudo-pregnant females.Guide RNAs were injected at 25ng-μL, and the Cas9 protein at 30 ng/μL.Embryos from the established line were collected at E10.5 and processed for in situ hybridization and genotyping (primers listed in S8 Table ).

Fig 2 .
Fig 2. Transcriptional changes in Ctcf mutant hearts.(A) Iris plot showing enriched GO terms in differentially expressed genes between control and mutant E10.5 hearts (outside circle), together with changes in expression of individual genes (inner circle; red, upregulated in mutants; blue, downregulated).(B) Genes that are up-or downregulated in Ctcf mutant hearts are more likely to have a CTCF binding site in their vicinity (10 or 20 kb surrounding the transcriptional start site) than genes that are expressed in the embryonic heart but do not change.