CTCF demarcates chicken embryonic α-globin gene autonomous silencing and contributes to adult stage-specific gene expression

Abstract

Genomic loci composed of more than one gene are frequently subjected to differential gene expression, with the chicken α-globin domain being a clear example. In the present study we aim to understand the globin switching mechanisms responsible for the epigenetic silencing of the embryonic π gene and the transcriptional activation of the adult α^D and α^A genes at the genomic domain level. In early stages, we describe a physical contact between the embryonic π gene and the distal 3′ enhancer that is lost later during development. We show that such a level of regulation is achieved through the establishment of a DNA hypermethylation sub-domain that includes the embryonic gene and the adjacent genomic sequences. The multifunctional CCCTCC-binding factor (CTCF), which is located upstream of the α^D gene promoter, delimits this sub-domain and creates a transition between the inactive sub-domain and the active sub-domain, which includes the adult α^D gene. In avian-transformed erythroblast HD3 cells that are induced to differentiate, we found active DNA demethylation of the adult α^D promoter, coincident with the incorporation of 5-hydroxymethylcytosine (5hmC) and concomitant with adult gene transcriptional activation. These results suggest that autonomous silencing of the embryonic π gene is needed to facilitate an optimal topological conformation of the domain. This model proposes that CTCF is contributing to a specific chromatin configuration that is necessary for differential α-globin gene expression during development.

Introduction

Compartmentalization of the eukaryotic genome is one of the initial regulatory steps for gene expression. Subsequently, interdependent chromatin remodeling is needed to define sub-domains to allow highly specific regulation of gene expression at a local level.^1,2 To address these aspects of gene regulation in more detail, we adopted the chicken α-globin gene domain as a paradigm to understand the influence of chromatin domain structure on gene expression (Fig. 1A).³ For years we have been interested in understanding genomic domain-level mechanisms of differential α-globin gene expression, not only during organism development but also during erythroid differentiation.

Figure 1. Different DNA methylation patterns of the chicken α-globin locus during development. (A) Differential α-globin gene expression profiles during chicken development.³ (B) Scheme of the α-globin domain showing the location of intergenic CpG-islands, the α-MRE locus control region and the anti-sense cgthba/C16orf35 transcript. Dotted lines define the genomic region analyzed by bisulfite transformation. In the lower panel we show the DNA methylation distribution of the 5- and 10-days RBCs (5dRBCs and 10RBCs, respectively). Open circles correspond to unmethylated CpGs and black circles represent methylated CpGs. Only the 3′ end of the embryonic π body gene was analyzed. The empty triangles indicate the methylation transition zone.

Hemoglobin switching represents the transition from embryonic to fetal and adult globin peptide synthesis in order to adjust oxygen requirements in the organism during development. This is accomplished through coordinated and differentially regulated α- and β-globin gene expression (Fig. 1A). This process depends on a complex interplay among gene promoters, distal regulatory elements, nuclear dynamics, and the epigenetic milieu.^4-6

Unlike the chicken β-globin gene domain, the α-globin domain lies in a constitutive open chromatin context on minichromosome 14 and contains the embryonic π gene and adult α^D and α^A genes (Fig. 1B). The α-globin locus covers a genomic region of around 50 kilobases (kb). On its 5′ non-coding region there are several DNase I hypersensitive sites, including the locus control region known as the α-major regulatory element (α-MRE), followed by the αEHS-1.4 element, which is a CTCF-dependent insulator.⁷ This insulator is located within an intron of the C16orf35 gene antisense transcript,^8,9 between the α-MRE element and the α-globin genes (Fig. 1B). Further downstream, 4 kb upstream of the π gene, there is a CpG island that corresponds to the promoter element of the C16orf35 antisense transcript (Fig. 1B).

Besides three α-globin genes (π, α^D, and α^A) and their promoters, the domain contains several regulatory elements located on its 3′ non-coding region. In particular, there is an erythroid-specific 3′ enhancer located at a distance of 0.9 kb after the α^A gene.¹⁰ As the promoters of the 3 genes, the 3′ enhancer activity depends on the binding of the erythroid-specific transcription factor GATA-1, other transcription factors and its chromatin structure.¹⁰

In the field, it has been suggested that hemoglobin switching is a process that involves, in part, autonomous silencing of the embryonic gene promoter and competition for proximal and distal regulatory elements.^3,4,11 Unfortunately, there is scarce experimental evidence describing the mechanisms for such autonomous silencing. Previous work from our research group suggests there is autonomous epigenetic repression of the embryonic π gene.³ We have shown that the embryonic π gene promoter is DNA methylated and bound by MeCP2, and the histones are mainly deacetylated in order to induce embryonic gene silencing, which then allows regulated transcriptional activation of the adult chicken α-globin genes late in development.³ In our previous studies, we have described the regulatory action of an enhancer, located in the 3′ non-coding region of the locus, over the three chicken α-globin genes.^3,10,12 One conclusion of those studies is that the autonomous silencing of the embryonic π gene also contributes to the selectivity of the 3′ enhancer during development. Here, we decided to address the mechanisms responsible for the establishment of the autonomous silencing.

In the present study, we systematically analyzed the DNA methylation status of the embryonic π gene, its promoter and the 3′ intergenic sequences up to the adult α^D promoter region in different developmental stages and during erythroid cell differentiation. Based on our previously published genome-wide distribution of the multifunctional nuclear factor CTCF, we found a recognition site in proximity of the adult α^D promoter. A series of mutant constructs stably integrated in an erythroid cell line seem to contribute to the formation of a stage-specific DNA methylation sub-domain. These observations further support our model in which delimited chromatin structural changes are responsible for directing the autonomous silencing of the embryonic π gene, which subsequently allows activation of the adult α^D and α^A genes late in development and during definitive erythropoiesis. A model is discussed based on the possible role of CTCF in defining the sub-domain chromatin transition, in addition to its possible role in long distance interactions between the regulatory elements of the locus.

Results

Differential DNA methylation of the embryonic π globin gene and surrounding genomic sequences during development

Based on our previous observations that the embryonic π gene promoter is not DNA methylated in 5-days-old chicken embryos (5dRBCs) but is methylated in 8dRBCs and 10dRBCs,³ we asked whether such methylation is restricted to the promoter or if it can be expanded, in particular, toward the adult α^D gene (Fig. 1B). This aim is further supported by the fact that DNA methylation can be propagated along the genome.^13-15 We performed sodium bisulfite transformation of genomic DNA and sequencing from 5dRBCs (where only the embryonic π gene is expressed) in comparison to 10dRBCs (where the adult α^D and α^A genes are expressed) (Fig. 1A and B). We found that the promoter regions of the π and α^D genes and the intergenic sequences in 5dRBCs are unmethylated, with the exception of two characteristic CpGs that are located 540 bp from the transcription start site (TSS) of the α^D gene (Fig. 1B). In contrast, in 10dRBCs, the π promoter and the intergenic region between the π and α^D genes are heavily methylated (Fig. 1B). However, the most appealing observation is the presence of a notorious transition between a DNA methylated and non-methylated state delimited by 2 constitutively methylated CpG-dinucleotides (C^mpG). This result supports a well-defined DNA methylation sub-domain (Fig. 1B, see empty triangles) coincident with the silencing of the embryonic π gene and the transcription of the adult α^D and α^A genes.

Partial DNA demethylation during erythroid differentiation

In order to further understand the role of DNA methylation during erythroid differentiation we took advantage of HD3 cells, which are AEV-transformed chicken pro-erythroblast cells corresponding to chicken adult erythroid cells arrested at early stages of differentiation.¹⁶ Importantly, these cells do not express globin genes, even though they posses a permissive globin domain chromatin structure.^10,17 Interestingly, these cells can specifically transcribe the adult α^D and α^A genes after induction of terminal erythroid differentiation (Fig. S1A). With this inducible system we were able to examine the DNA methylation distribution during the transition from the inactive state to transcriptional activation. Thus, we compared the DNA methylation patterns of genomic DNA from HD3 cells and differentiated HD3 cells (Fig. 2A). To induce HD3 differentiation (dif-HD3) we cultured the cells in the presence of the Iso-H-7 inducer at 42°C for 48 h or exclusively at 42 °C (dif-HD3/42 °C) (Fig. S1A).¹⁷ Total RNA and genomic DNA were then isolated to verify the induction efficiency through the transcriptional activation of the adult α^D and α^A genes (Fig. 2B).^3,17 We found that the entire region is DNA hypermethylated in undifferentiated conditions (Fig. 2A). However, once the HD3 cells were induced to terminally differentiate and α^D and α^A gene expression was initiated, we observed a localized DNA demethylation of the α^D promoter region (69.1% vs. 42.7%) (Fig. 2A). This is confirmed by a control genomic region located in the upstream intergenic region that does not show any reduction in DNA methylation (Fig. 2A). Supporting these data, intermediate levels of induction reached only by a shift in temperature to 42 °C (Fig. 2B, dif-HD3/42 °C; Fig. S1A) clearly show intermediate levels of α^D gene expression and promoter DNA demethylation (58.2%) in comparison to the levels seen upon robust induction conditions (dif-HD3). Interestingly, we notice that the position of individual CpGs in the α^D promoter that suffer a pronounced DNA demethylation under differentiation conditions are located in proximity to the binding sites for GATA-1 and Sp1 transcription factors (Fig. 3).³

Figure 2. DNA demethylation and gain of the 5hmC modification correlates with adult α^D gene transcriptional activation. (A) Analysis of the DNA methylation status in the transformed avian HD3 cell line in non-induced (HD3) and differentiation-induced (dif-HD3; Iso-H-7 and 42 °C) conditions. For comparative purposes an intermediate degree of induction is incorporated (see dif-HD3/42 °C; induction only with 42 °C). The percentage of the CpG methylation in the promoter region of the α^D gene is shown. The empty triangle indicates the methylation transition zone as in (Fig. 1B). (B) Quantitative real-time PCR showing the differentiation efficiency through the adult α^D gene transcription activation. (C) Quantitative 5-hydroxymethylcytosine-DNA immunoprecipitation was assessed using an antibody against 5hmC and primers located in the α^D promoter. As controls we used bisulfite transformation to test the chicken B-lymphocyte DT40 cell line, where the α^D promoter is DNA hypermethylated, and also 6C2 pre-erythroblast cells (Fig. S1B). Values higher than 1 (discontinuous line) are considered real enrichments. All the error bars were calculated from two independent experiments.

Figure 3. DNA methylation overview at individual CpG-dinucleotides in the α^D promoter. (A) A comparative percentage of DNA methylation is shown in the graph. The data are generated from the results in Figure 2A. HD3 cells represent undifferentiated erythrocytes, dif-HD3/42 °C which corresponds to the intermediate condition of differentiation induction, and dif-HD3 represents the highest differentiation induction condition. At the bottom, the distribution and position of the CpGs in the α^D promoter is indicated. Interestingly, CpGs located in positions 3, 4, 6, and 8 are subjected to most significant levels of DNA demethylation. (B) DNA nucleotide sequence of the α^D promoter region analyzed by sodium bisulfite conversion and sequencing. The CpGs that are preferentially demethylated (positions 3, 4, 6, and 8) coincide with a DNA segment where a tandem GATA-1 binding-site and one Sp1 binding-site are found.³

Since there is growing evidence supporting active enzymatic reactions capable of erasing or modifying pre-existing DNA methylation patterns,¹⁸ we first investigated the possibility of active DNA demethylation over the α^D promoter region. We analyzed the presence and abundance of the 5-hydroxymethylcytosine (5hmC) modification in RBCs and chicken cell lines (Fig. S1C). Using a specific antibody against 5hmC, we detected a modest recognition in chicken erythroid and non-erythroid cells compared with mouse embryonic stem cell genomic DNA, which was used as positive control (Fig. S1C).¹⁹ Then, we performed a quantitative 5hmC-DNA immunoprecipitation (hMeDIP) analysis (Fig. S1D). We found that 5hmC is already present in early erythroid differentiation stages as seen in the pre-erythroblast 6C2 cells and the undifferentiated HD3 cells (Fig. 2C). These results are further supported by no enrichment of 5hmC and DNA hypermethylation in the non-erythroid chicken lymphoid DT40 cell line (Fig. 2C; Fig. S1B). Therefore, we observe an erythroid-specific gain in 5hmC that is coincident with the activation of the α^D gene (Fig. 2C). We do not believe that DNA demethylation of the α^D gene promoter has a direct effect in its robust transcription activation. Different levels of regulation are probably involved including the activation of specific-signal transduction cascades, the synthesis of stage-specific transcription factors, the recruitment of chromatin remodelers and the silencing of the embryonic π gene. This finding suggests an indirect correlation between DNA demethylation and transcriptional activation of the adult α^D gene once HD3 cells are induced to differentiate. Taking these results together, we conclude that DNA demethylation is, to some extent, coincident with adult gene transcriptional activation. Further investigations are needed to determine a direct link between DNA demethylation and transcriptional activation of α-globin adult genes.

Active histone marks and CTCF enrichment at the DNA methylation transition zone

Next, we focused our study on understanding the chromatin features of the DNA methylation transition zone (Fig. 1B, see empty triangle). We assessed the enrichment of a series of histone marks, the presence of the H2A.Z histone variant, and the nuclear factor CTCF. For CTCF, we used an antibody that we generated to recognize the N-terminal of chicken CTCF, and we took advantage of our previously published ChIP-seq series of experiments using immunoprecipitated chromatin from RBCs isolated from chicken embryos at 5 and 10 days of development.²⁰ In this report, we found more than 5,000 CTCF binding sites were identified in 5dRBCs, and more than 21,000 in 10dRBCs,²⁰ among them a novel CTCF-binding site that was found 190 bp upstream of the α^D gene TSS (Fig. 4A; Fig. 5A). Based on this novel CTCF location, in the present study we decided to validate its presence by semiquantitative duplex PCR using primers extending from the CTCF site and associated histone marks (Fig. 4B; Fig. S2). The results show that CTCF is bound to the upstream limit of the chicken α^D promoter in 5dRBCs and 10dRBCs, but absent in HD3 cells, consistent with the DNA hypermethylation seen in this cell line. Interestingly, there is no enrichment of negative histone marks in any condition (H3K9me3 and H3K27me3; Fig. S3), and this observation is in agreement with the data published by others.²¹ Instead, we observed a constant incorporation of active histone marks like H3ac, H4ac, and H3K4me2 (Fig. 4B).

Figure 4. In vivo association of the CTCF nuclear factor and histone marks over the adult α^D gene promoter. (A) Location of the new CTCF binding site determined in a previous publication by chromatin immunoprecipitation and massive sequencing (ChIP-seq);²⁰ black squares represent CTCF in vivo occupancy (May 2006, WUGSC 2.1/galGal3 Assembly; UCSC, Genome Browser). (B) Semi-quantitative and comparative ChIP assays validating the in vivo binding of CTCF and the incorporation of histone post-translational modifications at the α^D gene promoter. (C) ChIP experiments comparing non-induced HD3 cells (HD3) and differentiation-induced cells (dif-HD3). The same primers are used in (B) and (C). Asterisks indicate experiments not done. Values higher than 1 represent real enrichments (discontinuous line). The error bars were calculated from two or more independent experiments.

Figure 5. CTCF is a key component for α^D gene promoter function. (A) Schematic representation and nucleotide sequence of the putative CTCF binding motifs in the π - α^D intergenic region. Using the Logo generated from the chicken CTCF ChIP-seq experiments we identified a potential binding motif located 190 bp upstream of the α^D-TSS (CTCF-1, black square 1).²⁰ Additionally, we looked for other potential binding sites using bioinformatics tools (MathInspector) and we found another CTCF-binding motif (CTCF-2, black square 2). Both binding sites are shown and for the CTCF-1 site the published Logo is included for comparison purposes. The vertical lines show the distribution of CpG-dinucleotides along the studied region. The gray triangles indicate the sequences cloned for the transgene experiments (see below). We incorporated the core α^D promoter (white box in schemes) containing the putative CTCF-binding sites (black boxes inside; αD and αDIG) driving the expression of the GFP gene (for details see Fig. S4). (B) We performed deletions of the proximal (αΔC1) and distal (αΔC2) CTCF-binding sites. 5′-IGr corresponds to a DNA segment from the upstream intergenic region (938 bp). (C) We introduced the 3′ enhancer (enh) downstream of the GFP gene in each construct described in (B). The discontinuous line corresponds to the average (%) cell fluorescence in (B). (D) ChIP over transgene constructs at day 35 of culture using primers over the α^D gene promoter (forward) and the GFP gene (reverse). (E) ChIP for transgene as in (D) in the constructs that harbor the 3′ enhancer.

In order to verify the chromatin status of the region in induced HD3 cells, we performed a comparative ChIP from HD3 cells and induced HD3 cells (Fig. 4C). In undifferentiated HD3 cells there is no enrichment of H3ac and a modest increment of H4ac. Furthermore, there is no binding of CTCF. These observations are in accordance with the DNA hypermethylation status of the region and the lack of α^D gene expression. In contrast, under HD3 cells differentiation conditions (dif-HD3) there is a gain of open histone chromatin marks and binding of CTCF, favoring the idea of barrier element formation. Such a configuration may delimit the spreading of DNA methylation and it defines a chromatin sub-domain necessary for the autonomous silencing of the embryonic π gene and the expression of the adult genes in terminally differentiated erythroid cells.

The absence of the CTCF site in the transition zone induces rapid epigenetic silencing in stably transformed cells

Based on its location and the results obtained up to this point, we decided to demonstrate the barrier function of this CTCF motif located upstream of the α^D gene promoter. Reducing the relative abundance of CTCF in HD3 cells was an alternative that we discarded based on the fact that CTCF is bound in vivo at least to four different sites along the locus, rendering the interpretation of the results very difficult.^7,20,22 In addition, RNA interference experiments are inefficient in the non-dividing 5dRBCs and 10dRBCs. Thus, to address this aim we generated a series of constructs that incorporate or remove the different CTCF-binding sites (Fig. 5). On the basis of the canonical CTCF motif that emerged from the published ChIP-seq data and bioinformatics prediction,²⁰ we analyzed the genomic sequence in more detail and found two potential binding motifs about 190 bp upstream of the α^D gene TSS (Fig. 5A).²⁰ Derived from this prediction and the size of CTCF-binding sites, we designed two different deletions that excluded each predicted motif (C1: proximal site and C2: distal site) (Fig. S4). It is worth mentioning that none of the previously characterized binding sites for transcription factors required for α^D promoter activity were removed.^3,12 These constructs were stably transfected in HD3 cells and pools were selected and maintained in continuous cell culture. GFP expression was evaluated by GFP fluorescence emission by flow cytometry (FACS), revealing that the loss of the proximal CTCF site (αΔC1) is associated with a rapid and robust decay of transgene expression (Fig. 5B). Next, we performed a ChIP assay comparing the intact transgene (αDIG) and the transgene lacking the proximal CTCF site (αΔC1), which resulted in loss of CTCF binding, a drastic decrease in histone H3 and H4 acetylation and a decrease in GFP expression after 35 days of continuous cell culture (Fig. 5D). From these data we predict that the CTCF-1 motif is the preferential site responsible for such activity. These results support our model in which CTCF is creating a barrier against silencing chromatin signals. Nevertheless, based on the rapid loss of transgene activity in the αΔC1 construct, we cannot rule out the possibility that CTCF is also having a transcriptional activation function on the α^D gene promoter (Fig. 5B; Day 1).

With the intention to differentiate CTCF activity from classical transcription factor and barrier element activity, we repeated the same experiment but this time incorporating the 3′ α-globin enhancer in our transgenes.^3,10 The rationale behind this experiment is that in the absence of CTCF, the enhancer may compensate by trans-activating the transgene through the formation of an optimal chromatin configuration. This prediction is further supported by published data demonstrating that enhancers can positively influence the chromatin structure of a transgene and favor their trans-activation potential.^23,24 However, with the incorporation of the 3′ enhancer there was between 40–60% transgene expression reduction in the αΔC1-enh transgene, but the overall expression was maintained in the other constructs and over time (Fig. 5C). The absence of CTCF binding was corroborated by ChIP analysis (Fig. 5E). In agreement with our prediction the absence of the CTCF site did not affect promoter functionality and the enhancer retained its trans-activation potential. Of note, and based on the ChIP-seq data,²⁰ we validated another CTCF site in the 3′ enhancer (Fig. 7 and B). Thus, the presence of CTCF at both the α^D promoter and the 3′ enhancer favors a robust and sustained transgene expression by the establishment of a permissive chromatin configuration. Therefore, these results also suggest that CTCF is acting as a chromatin remodeler but at this point we are not able to discard a role of CTCF as a classical transcriptional factor. In any case, both activities cannot be excluded.

Figure 7. Physical contacts between the 3′ enhancer and the π and α^D genes. (A) The scheme shows the distribution of the 4-base cutter DpnII restriction enzyme over the α-globin genes and the 3′ enhancer (3′-Enh; black boxes represent gene exons). The arrowheads correspond to the test-primers and the asterisk marks the anchor-primer. (B) CTCF occupancy over the 3′-Enh region analyzed by ChIP. (C) 3C experiments performed in 5dRBCs and 10dRBCs. The cross-linking frequency was calculated from two independent experiments (Table S2) and associated controls are shown in detail; see Figure S7. The error bars were calculated from two or more independent experiments.

Gain of DNA methylation in the α^D gene promoter upon CTCF binding site removal

To further assess the role of CTCF and on the basis of its capacity to nucleate histone acetylases and methylases, we incubated the cells expressing the different transgenes with the histone deacetylase inhibitor, trichostatin-A (TSA), in the absence or the presence of the 3′ enhancer (Fig. 6A). Interestingly, we found that in the absence of the enhancer there was no transgene reactivation in the αΔC1 constructs. In contrast, when the enhancer was incorporated in the transgene, we observed a more robust transgene expression in response to TSA (Fig. 6A). Thus, the presence of CTCF is probably necessary to recruit histone modifiers such as HATs and HMTs to the α^D promoter region.

Figure 6. Transgene reactivation assays and DNA methylation analysis. (A) TSA-treatment at day 35 of culture. (B) DNA methylation analysis over the transgene. The percentage at individual CpG-dinucleotides of the transgene comparing day 1 and day 25 of continuous cell culture is shown. Interestingly, CpGs located in the vicinity of the deleted CTCF binding-site gain preferentially DNA methylation. The error bars were calculated from two independent experiments.

To further explore the contribution of CTCF in demarcating the DNA methylation sub-domain, we surveyed the DNA methylation of the α^D gene promoter in the transgenes both with and without the CTCF-binding sequence (αDIG vs. αΔC1 and αΔC1-enh). We found a gain in DNA methylation after 25 days of continuous cell culture in the absence of the CTCF-binding sequence (Fig. 6B; Fig. S5). It is worth mentioning that at Day 1 no DNA methylation was found in any of the three transgenes analyzed (Fig. 6B). This is consistent with the low expression levels of the αΔC1 transgene (Fig. 5B), and is in agreement with previous published data demonstrating that transgene silencing begins with histone deacetylation followed by gain in DNA methylation.²⁵ In addition, even in the presence of the 3′ enhancer (αΔC1-enh), we observed an increment in DNA methylation concomitant with a partial reduction of transgene expression (Fig. 5C; compare αDIG-enh with αΔC1-enh). This result was confirmed when transgenic cells where treated with the DNA methylation inhibitor 5-aza-2’-deoxycytidine (5-azadC), which induced reactivation of transgene expression in the presence of the 3′ enhancer and absence of the CTCF-1 binding site (Fig. S6; αΔC1-enh).

Altogether, this series of experiments is in agreement with a model whereby the CTCF site located 190 bp from the α^D gene TSS represents a barrier element that counteracts the propagation of silencing chromatin, in particular, DNA methylation. In conjunction, all these factors facilitate the establishment of a chromatin configuration that allows the specific and regulated transcriptional activation of the adult α^D gene.

Stage-specific chromatin loop formation between the embryonic π gene promoter region and the 3′ enhancer

Up to this point, we cannot exclude stage-specific three-dimensional contacts between the gene promoters, the 3′ enhancer and other regulatory components of the locus.²⁶ The spatial organization of the chicken α-globin locus has already been established during erythroid differentiation (HD3 cells) and only during late development, revealing the differential formation of chromatin hubs.¹⁷

With all this in mind we asked whether there is a physical interaction between the 3′ enhancer and the embryonic π promoter region, an aspect that has not been previously explored (Fig. 7A). To address this question, we performed a chromosome conformation capture (3C) assay in 5dRBCs vs. 10dRBCs (Fig. 7; Fig. S7).^27,28 In agreement with the data presented here, we found a physical association between the embryonic promoter region and the 3′ enhancer only in 5dRBCs (Fig. 7C). No significant association was seen in 10dRBCs, when the promoter and surrounding regions of the π gene are DNA hypermethylated and the α^D gene is transcriptionally active. Of note, such interactions do not seem entirely dependent on CTCF, as suggested for other loci,²⁹ since we have not been able to demonstrate CTCF binding around the π gene (Fig. S8). In contrast, we found CTCF binding at the 3′ enhancer in 5dRBCs and 10dRBCs (Fig. 7A and B). Of note, we found that the 3′ enhancer is constitutively contacting the α^D promoter region even though there is no trans-activation (like in the 5dRBCs) (Fig. 7). This is in agreement with the previously proposed role of the α^D promoter region as a constitutive anchor point for the α-globin domain chromatin hub.^17,21 In conclusion, even though the locus is in a general open chromatin conformation there is a preference for the 3′ enhancer to interact with the π promoter region in early stages of development.

Discussion

The genomic distribution and its topological orientation within the nucleus hierarchically represent one of the first levels of gene regulation. On the local scale the genome requires another degree of sophistication with the definition of transcriptionally active domains and sub-domains that are shielded from neighboring signals or even genes. During development we observed the establishment of a DNA methylation sub-domain, which is demarcated by the binding of CTCF and other transcription factors.³

Concerning the protection against DNA methylation expansion, our data are consistent with the observation that a sub-set of CTCF-binding sites found in the vicinity of tumor suppressor genes and microRNAs prevent epigenetic silencing.^14,30,31 Furthermore, the presence of CTCF and other nuclear factors upstream of the α^D gene TSS establishes that DNA-binding factors are involved in creating an unmethylated state even in the absence of transcription, as is the case in 5dRBCs (Fig. 3B; Fig. 4B).^3,15,32,33 Similarly, this type of barrier function that prevents DNA methylation was described in association with the activity of the VEZF1 nuclear factor in the chicken β-globin cHS4 insulator elements.³³ Interestingly, mutation of the VEZF1 binding site in stably transfected constructs demonstrated the spreading of DNA methylation over the β^A promoter and transgene silencing.^33,34 The stable lines that we generated with the deletion of the CTCF site support an equivalent role for CTCF, even though the CTCF partners and/or post-translational modifications associated with such a particular role remain to be determined.⁷

The enrichment of the histone mark H3K4me2 in 5dRBCs and 10dRBCs is in accordance with the published data demonstrating that hypomethylated CpG-islands show elevated levels of H3K4me2, even in the absence of transcription. In fact, it has been demonstrated that the H3K4me2 modification occurs uniformly on all types of CpG-island promoters, supporting the concept that this in an inherent feature of CpG-islands.³⁵ Furthermore, it has also been shown that broad histone H3 hyperacetylation occurs in CpG-islands.³⁵ Together, we consider CTCF and its associated co-factors and chromatin remodelers to be responsible for the formation of an active chromatin state which precludes the propagation of DNA methylation. A remarkable aspect of the genomic region under study is its capacity to be DNA demethylated through the gain of 5hmC in response to cellular differentiation signals (dif-HD3 cells). Interestingly, a regular distribution of 5hmC around CTCF binding sites has recently been reported in mouse embryonic stem cells.³⁶ This observation implies that CTCF and/or its associated partners are involved in the establishment of a 5hmC profile in a regulated manner to achieve regular and optimal nucleosome positioning.³⁶ Thus, future investigations should address the DNA demethylation of gene promoters in response to cellular differentiation and the role, if any, of CTCF in this process.

An alternative mechanism for CTCF could be related to the work of Caiafa and collaborators showing that the poly(ADP-ribose) polymerase 1 (PARP-1) is self-poly(ADP-ribosyl)ated in the presence of CTCF.³⁷ Such an interaction generates a CTCF-PARP-1-Dnmt1 complex that interferes with the activity of the Dnmt1 DNA methyltransferase. Interestingly, CTCF dissociation abolishes such interactions and favors DNA hypermethylation in the vicinity of the CTCF sites.³⁸ Studies from our group in the chicken α-globin locus demonstrated that CTCF activity can be modulated by poly(ADP-ribosyl)ation.⁷ Together, we conclude that the activity of CTCF located in proximity to the α^D gene promoter can be, in part, regulated by poly(ADP-ribosyl)ation and/or by blocking DNA methylation through its interaction with PARP-1.

A complementary aspect of the regulation and structure of the chicken α-globin domain is its three-dimensional organization. One of the most relevant aspects of such a spatial organization is that the promoter region of the adult α^D gene is apparently a key contact point, which is coincident with the newly identified CTCF-binding site.^17,20 In such a scenario, one possible model is that the embryonic π region can be excluded from the hub, in part, by its autonomous silencing in later stages of development. This view is consistent with the one described for the mouse β-globin locus.³⁹ Furthermore, and as suggested previously,⁷ the combination of molecular allies and post-translational modifications may also induce a differential selectivity in the function of the CTCF sites located along the domain. At this point, we cannot discard the complementary intervention of non-coding RNAs in such multi-step regulation.

In summary, the results presented here allow the proposal of an integrative model of regulation that incorporates diverse components and the coordination of multiple genetic and epigenetic processes (Fig. 8). Historically, it has been assumed that the chicken α-globin locus is found in a constitutively open chromatin conformation.⁴⁰ Tissue- and stage-specific transcription factors are then needed for the activation of the embryonic π gene in a context in which the rest of the domain remains in an open chromatin structure.³ In such an embryonic context no stage-specific factors are present for the activation of the adult genes. Once switching between embryonic and adult gene expression occurs, CTCF and its co-factors act as a barrier and a nucleation center to shield the silencing of the embryonic π gene and surrounding genomic regions through DNA methylation in an autonomous way (Fig. 8). As a consequence, the embryonic π gene region is excluded from the active chromatin hub. Based on our model, such a well-delimited silencing is needed to allow the transcriptional activation of the adult α^D and α^A genes in later stages of development. A controversy has been raised arguing that there is no silencing in any part of the chicken α-globin locus.²¹ This is based only in the study of active and repressive histone marks but no survey on the DNA methylation status of the regions was assessed.²¹ Our results support combinatorial regulatory events which could be similar to those observed for RARβ2 gene expression, where the XPG endonuclease promotes DNA breaks and DNA demethylation at promoters, including CTCF recruitment and looping, which are events that can also be mediated by specific TAFs.^41,42 This attempt to integrate genetic and epigenetic levels of regulation represents a starting point to address more detailed mechanistic questions.

Figure 8. Autonomous silencing of the embryonic π gene and the derived topological conformation of the α-globin domain. In early development (5dRBCs), the model shows the formation of a hub composed by interactions between the π gene, the α^D promoter region and the 3′ enhancer. In late development (10dRBCs), the embryonic π gene and surrounding genomic regions are DNA methylated with its concomitant dissociation of the hub. At this stage, the physical contact between the α^D gene promoter and the 3′ enhancer is maintained among other previously described interactions.¹⁷

Materials and Methods

Cell culture

RBC primary cultures were obtained from chicken embryos at 5 and 10 days of development (5dRBCs and 10dRBCs, respectively). The cell culture conditions for RBCs and the chicken cell lines (DT40, 6C2, and HD3) were as previously described.³ To induce erythroid differentiation, HD3 cell line were treated with Iso-H-7 (20 μM, Sigma) in HD3 cell medium containing 8% of FBS (Multicell), 2% chicken serum (Gibco), 1% penicillin/streptomycin (Gibco), and 10 mM Hepes pH 8.0 in a 1% CO₂ atmosphere at 42 °C for 48 h. HD3 cell transfections were performed using Lipofectamine plus 2000 (Invitrogen) under antibiotic resistant selection in the presence of 0.8 mg/ml of Geneticin (Calbiochem). Transgene expression was evaluated by flow cytometry as described.³ For drug treatments, Trichostatin-A (TSA, Sigma) and 5-aza-2′-deoxycytidine (5-azadC, Sigma) were used at 2.5 ng/ml and 3 μM, respectively.

Plasmids

The pGαD, pGαDIG, pGΔC1, and pGΔC2 plasmids and their associated versions containing the 3′-enh were generated by PCR amplification of the α^D promoter (Fig. S4), the intergenic region between the π and α^D genes and the 3′-enh using the specific primers listed in Table S1. The PCR DNA fragments were sub-cloned and restriction enzyme-digested for introduction into the pEGFP-1 (Clontech) vector reporter system.

Sodium bisulfite DNA conversion and sequencing

Three μg of genomic DNA from DT40, 6C2, HD3, dif-HD3 cells, 5dRBC, and 10dRBCs were denatured at 95 °C for 5 min, chilled on ice, and incubated with 0.3 M NaOH at 37 °C for 5 min. Freshly-prepared solutions of sodium bisulfite (Sigma), adjusted to pH 5.0 with NaOH, and hydroquinone (Sigma) were added at final concentrations of 1.7 M and 0.5 mM, respectively. DNA solutions were mixed and incubated at 55 °C for 13 h in the dark. Non-reacting bisulfite was removed by column purification (Wizard DNA Clean-Up System; Promega). Purified DNA samples were de-sulfonated with NaOH at a final concentration of 0.3 M, incubated at 37 °C for 15 min followed by ethanol precipitation. DNA fragments of interest were PCR-amplified using the modified primers listed in Table S1, then cloned into the pGEM-T Easy system (Promega), and sequenced using SP6 or T7 sequence primers.

Quantitative RT-PCR

RNA isolation was performed as described previously.³ For quantitative RT-PCR the primers used for amplification are listed in Table S1. All primers were intron-spanning. For quantitative real time PCR (qPCR), Platinum Taq DNA polymerase (Invitrogen) was used mixed with SYBR-Green I (Sigma) dissolved in DMSO. The qPCR reactions were performed in a StepOne™ detection system (Applied Biosystems). The relative expression levels of the α^D gene were determined by using the ribosomal 28S mRNA levels as an endogenous reference and the ΔΔCT method was used for normalization.

5-hydroxymethylcytosine DNA immunoprecipitation (hMeDIP)

In order to detect only 5hmC present in the α^D gene promoter and not in the surrounding sequences, the genomic DNAs were first BamHI-digested and then used for hMeDIP experiments. The 5-hydroxymethylcytosine DNA immunoprecipitation (hMeDIP) technique was performed using an anti-5hmC antibody (Active Motif; 39769) as described.⁴³ The technique was validated by Dot-Blot analysis (Fig. S1D). qPCR was performed using the primers listed in Table S1. The 5hmC enrichment was calculated, first by normalizing α^D promoter amplification values over a 5hmC-depleted region (HI, repetitive region in a heterochromatic context),^3,44 and then over the IgG background value. The calculations were performed as follows: Enrichment = antibody(Ct_T − Ct_C) / IgG (Ct_T − Ct_C); where Ct_T = amplification of the test region, and Ct_C = amplification of the negative control region (HI, Fig. S2A).

Identification of in vivo CTCF interacting regions in the chicken genome

The chicken CTCF binding maps previously published were incorporated into the chicken assembly Genome Browser (May 2006, WUGSC 2.1/galGal3 Assembly; UCSC) and the in vivo binding of CTCF was determined at the α-globin locus.²⁰ The accession codes for the ChIP-seq data previously published are the following: GEO: GSM691016 for 5dRBCs and GSM691017 for 10dRBCs.²⁰ The identified CTCF binding sites were validated by quantitative PCR (see below).

Chromatin immunoprecipitation analysis

Chromatin was prepared for immunoprecipitation as previously described.³ The chromatin was immunoprecipitated with anti-CTCF antibody,⁴⁵ anti-Acetyl histone H3 antibody (06-599, Millipore), anti-Acetyl histone H4 antibody (06-866, Millipore), anti-H3K4me2 antibody (07-030, Millipore), anti-H2A.Z antibody (ab18263, Abcam), anti-H3K27me3, and anti-H3K9me3 antibodies (kindly supplied by Dr Thomas Jenuwein, IMP-Freiburg), and anti-Normal rabbit IgG (12-370, Millipore). The purified and recovered DNA was assayed by qPCR using the primers listed in Table S1. For the CTCF and open chromatin marks, the enrichment was calculated over a repetitive sequence region in a heterochromatic context (HI in Fig. S2A).^3,46 In the inverse sense, for close chromatin marks, the enrichment was calculated over a constitutive open region (FII in Fig. S2A).^3,46 The enrichment calculations were performed as follows: Enrichment = antibody(Ct_T − Ct_C) / IgG (Ct_T − Ct_C); where Ct_T = amplification of the test region, and Ct_C = amplification of the negative control region (HI or FII; Fig. S2A). Alternatively, we used semi-quantitative radioactive duplex PCR using the same formula described for the enrichment calculations.

Chromosome conformation capture assay (3C)

The 3C experiments were performed as described with minor modifications.²⁷ In brief, 4 × 10⁷ RBCs were cross-linked with either 1% or 2% formaldehyde for 10 min at room temperature (Fig. S7). The reaction was stopped by adding glycine (125 mM final concentration). The cells were washed with cold-PBS and the nuclei were isolated with incubation in ice-cooled lysis buffer (10 mM Tris (pH 8.0), 10 mM NaCl, 0.2% NP-40, and protease inhibitor mix) for 90 min. The nuclei were dispersed into 1 × 10⁶ aliquots and frozen in liquid nitrogen. The 1% formaldehyde cross-linked nuclei were used for the following steps of the protocol (Fig. S7). In order to determine the frequency of interacting genomic fragments at high resolution, a 1 × 10⁶ aliquot was digested with 1,000 units of the four-cutter DpnII restriction enzyme (New England Biolabs) overnight. Inactivation of the enzyme was performed by incubation at 65 °C for 20 min. The samples were incubated in 7 ml of ligase Buffer (NEB) and 400 units of T4 DNA ligase (NEB) was added at 16 °C for at least 4 h. The cross-linking was reversed by incubation at 65 °C overnight in the presence of Proteinase K and RNase A (Sigma). The DNA was purified with phenol:chloroform:isoamyl alcohol (Sigma), and precipitated with ethanol and glycogen (Roche). To specifically detect the interaction hybrids, the following Taqman probe was designed over the “anchor” site: 5′-6-FAM/ACT CAG CAC /ZEN/CTG GCA GGT TTT AC-3′-IABkFQ (IDT Technologies). The calculations and normalization data are shown in Table S2.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Supplemental Materials

Supplemental materials may be downloaded here:

http://www.landesbioscience.com/journals/epigenetics/article/25472