Epigenetic dysregulation by nickel through repressive chromatin domain disruption

Significance

Histone modifications associated with gene silencing typically mark large contiguous regions of the genome forming repressive chromatin domain structures. Since the repressive domains exist in close proximity to active regions, maintenance of domain structure is critically important. This study shows that nickel, a nonmutagenic carcinogen, can disrupt histone H3 lysine 9 dimethylation (H3K9me2) domain structures genome-wide, resulting in spreading of H3K9me2 marks into the active regions, which is associated with gene silencing. Our results suggest inhibition of DNA binding of the insulator protein CCCTC-binding factor (CTCF) at the H3K9me2 domain boundaries as a potential reason for H3K9me2 domain disruption. These findings have major implications in understanding chromatin dynamics and the consequences of chromatin domain disruption during pathogenesis.

Abstract

Investigations into the genomic landscape of histone modifications in heterochromatic regions have revealed histone H3 lysine 9 dimethylation (H3K9me2) to be important for differentiation and maintaining cell identity. H3K9me2 is associated with gene silencing and is organized into large repressive domains that exist in close proximity to active genes, indicating the importance of maintenance of proper domain structure. Here we show that nickel, a nonmutagenic environmental carcinogen, disrupted H3K9me2 domains, resulting in the spreading of H3K9me2 into active regions, which was associated with gene silencing. We found weak CCCTC-binding factor (CTCF)-binding sites and reduced CTCF binding at the Ni-disrupted H3K9me2 domain boundaries, suggesting a loss of CTCF-mediated insulation function as a potential reason for domain disruption and spreading. We furthermore show that euchromatin islands, local regions of active chromatin within large H3K9me2 domains, can protect genes from H3K9me2-spreading–associated gene silencing. These results have major implications in understanding H3K9me2 dynamics and the consequences of chromatin domain disruption during pathogenesis.

Investigations into the histone modification landscape of eukaryotic genomes have revealed organization of the chromatin into functionally distinct active and silent domains (1–4). Although active histone modifications typically display a punctate pattern, the silencing histone modifications such as histone H3 lysine 9 dimethylation (H3K9me2) and H3K27me3 mark large regions of the genome-forming repressive domains (1, 2, 5, 6). Organization of active and silent chromatin regions into compartments has direct implications on the establishment of gene expression patterns and cell-type specificity (3, 7).

Recent evidence indicates that the H3K9me2 domains are highly dynamic and important for differentiation and maintaining cell identity (8–10). Although H3K9me2 marks large domains of mostly transcriptionally silent regions of the genome, small windows of active chromatin have been characterized within the large heterochromatin domains, termed euchromatin islands (EIs) (9, 11). These EIs have been shown to be enriched for active histone modifications, DNase I hypersensitive (HS) sites, and CCCTC-binding factor (CTCF) binding (11). Existence of H3K9me2 domains in close proximity to active genes suggests that delimitation of these domains is critical in maintaining cellular identity. Spreading of H3K9me2 beyond its domain boundaries is associated with gene silencing at the chicken β-globin HS4 locus (12). Moreover, discrete local changes in H3K9me2 are associated with corresponding gene expression changes during cellular differentiation (9), further emphasizing the importance of maintaining the structural integrity of the H3K9me2 domains for cell-type specificity. However, how H3K9me2 domains are maintained and the consequences of aberrant domain disruption are poorly understood. An important clue toward understanding the H3K9me2 domains comes from studies that investigated silencing of a transgene following the exposure of cells to nickel (13, 14). Nickel compounds are environmental carcinogens that cause a multitude of human health risks including allergic dermatitis, bronchitis, pulmonary fibrosis, pulmonary edema, diseases of the kidney and cardiovascular system, and lung and nasal cancers (15, 16). Despite the conclusive health risks of Ni compounds, the underlying mechanisms are not well understood because their mutagenic potential is very low and does not correlate with its potent toxicity (17). Recent evidence suggests that Ni could alter transcriptional regulation through epigenetic alterations (18–20).

In the transgenic gpt+ Chinese hamster V79-derived G12 cells, silencing of the bacterial xanthine-guanine phosphoribosyltransferase (gpt) transgene following Ni exposure is associated with an increase in H3K9me2 and a decrease in H3 and H4 acetylation at the promoter and coding regions (13, 14, 21). Intriguingly, Ni-induced silencing of gpt and increased H3K9me2 levels occurred only when the transgene was integrated near a heterochromatic region (13, 14, 21). This suggests that Ni could induce H3K9me2 spreading. Therefore, we reasoned that studying the effects of Ni exposure on the epigenome could provide an opportunity to understand the dynamics of H3K9me2 domain maintenance and the mechanistic basis of potential domain disruption and aberrant gene silencing.

In this study, we characterized the changes in the global profiles of several active and repressive histone modifications in the Ni-exposed human lung cells. Our data showed that Ni exposure triggered H3K9me2 domain disruption, resulting in spreading of H3K9me2 into the active regions, which was associated with gene silencing. We found weak CTCF-binding sites and reduced CTCF binding at the Ni-disrupted H3K9me2 domain boundaries. Moreover, CTCF knockdown resulted in H3K9me2 spreading. These results suggest a loss of CTCF-mediated insulation function as a potential reason for domain disruption and spreading. Because epigenetic alterations are believed to be the basis of Ni-induced cancers (14, 16), our finding of H3K9me2 domain disruption by Ni and a potential role for CTCF in maintaining H3K9me2 domains will have profound implications in understanding the process of carcinogenesis.

Results

Ni Exposure Leads to Genome-Wide H3K9me2 Domain Spreading.

To investigate the Ni-induced epigenetic changes, we treated immortalized noncancerous human bronchial epithelial BEAS-2B cells to a noncytotoxic concentration of 500 μM NiCl₂ for 72 h, recapitulating the concentration used in several earlier studies that investigated the impact of Ni exposure on BEAS-2B cells (22–24). Western blotting analysis showed significant increase in the total levels of the repressive marks H3K9me2 and H3K27me3 (Fig. 1A). To identify the genes that are differentially expressed due to Ni exposure, we performed RNA-Seq analysis. We then examined the levels of H3K9me2 and H3K27me3 at candidate Ni-downregulated gene promoters using chromatin immunoprecipitation (ChIP) analysis (Fig. 1 B and C). Although we found appreciable increase in the levels of H3K9me2 (Fig. 1C), we did not find any changes in the levels of H3K27me3 (Fig. S1A). To investigate the effect of Ni exposure on the genome-wide distribution of repressive marks, we mapped H3K9me2 and H3K27me3 in untreated and Ni-treated cells, using ChIP-Seq (Fig. 1 D and E and Fig. S1 B and C). We noted a striking reorganization of H3K9me2 domains in Ni-treated cells with spreading of H3K9me2 into previously non-H3K9me2 regions (Fig. 1E). Interestingly, even though the total levels of H3K27me3 significantly increased in Ni-exposed cells (Fig. 1A), the domains of H3K27me3 did not substantially change (Fig. S1 B–D).

Fig. 1.

H3K9me2 domains spread upon NiCl₂ treatment. (A) Western blotting analysis showing H3K9me2 and H3K27me3 levels in BEAS-2B cells exposed to 500 μM NiCl₂ for 72 h. (B) qPCR analysis showing downregulation of candidate gene expression in Ni-treated cells. GAPDH was used as internal control. (C) ChIP-qPCR analysis showing increase in H3K9me2 levels at candidate downregulated gene promoters. (D) Workflow for H3K9me2 genome-wide profiling and domain calling. (E) Gene expression (RNA-Seq) and H3K9me2 (ChIP-Seq) profiles at the tumor suppressor CTGF gene locus in the untreated and Ni-treated BEAS-2B cells. In the untreated cells, depletion of H3K9me2 indicates active chromatin. Nickel exposure resulted in H3K9me2 spreading into the CTGF genic region (box), which was associated with downregulation of gene expression. The blue and red bars represent H3K9me2 domains [RSEG analysis (25)] in the control and Ni-exposed cells, respectively. (Inset) ChIP-qPCR analysis showing increase in H3K9me2 levels at CTGF promoter following Ni exposure. (F) Cartoon representing the H3K9me2 domain spreading upon Ni exposure. Spreading domains are defined as domains that are larger in Ni-treated cells, with the Ni domain overlapping at least 90% of the control domain. (G) Size distribution of H3K9me2 domains in untreated and Ni-treated cells for all domains and spreading domains. A total of 1,514 of the H3K9me2 domains in control cells (blue) increase in size after nickel exposure (red), indicating domain spreading. In addition, a number of small domains combine following Ni exposure, as indicated by the decrease in the number of domains in the Ni-exposed cells. For qPCR, statistical significance was evaluated using t test: **P < 0.001; ***P < 0.0001.

We then set out to characterize all of the regions of the genome that displayed H3K9me2 domains that increased in size following Ni treatment, which we refer to as “spreading domains” (Fig. 1 F and G and SI Materials and Methods). After saturation analysis of the ChIP-Seq reads to ensure sufficient coverage (SI Materials and Methods and Fig. S2), we called all domains in both control (Ctrl) and Ni-treated (Ni) cells using RSEG software (25). To validate H3K9me2 ChIP-Seq data and to optimize H3K9me2 domain calling, we selected several loci and performed ChIP-quantitative PCR (qPCR) analysis on untreated and Ni-treated cells (Fig. 1D and SI Materials and Methods). ChIP-qPCR analysis strongly correlated with the ChIP-Seq data (Fig. S3). Furthermore, using a peptide immunoprecipitation (IP) assay, we evaluated the specificity and affinity of the antibody used in this study for performing H3K9me2 ChIP-Seq (SI Materials and Methods) (26). The antibody showed moderate affinity with a submicromolar K_D value and outstanding specificity with almost no cross-reactivity to other histone modification marks tested (Fig. S4), suggesting the high quality of the antibody.

We identified 2,461 control domains and 1,298 Ni domains, ranging in size from 1 kb to >1 Mb (Fig. 1G). A total of 61% (1,514/2,461) of the domains in untreated control cells increased in size upon Ni exposure, indicating spreading of these domains beyond their normal boundaries (Fig. 1G and Fig. S5). The total number of spreading domains reduced to a third of the domains detected in control cells (562/1,514) due to spreading domains combining to form larger domains (Fig. 1G and Fig. S5).

Genes in H3K9me2-Spreading Domains Are Generally Downregulated.

As shown in Fig. 1 B and E and Fig. S3, MAP2K3, DKK1, TAGLN, and CTGF were strikingly downregulated upon Ni exposure, and this loss of gene expression was associated with increase in promoter H3K9me2 levels, consistent with the earlier studies that showed association of H3K9me2 spreading with decreased gene expression (27–30). To examine this on a genome-wide scale, we stratified all of the genes into those in H3K9me2-spreading domains (3,930) and genes that are not in H3K9me2 in either control or Ni-exposed cells (8,096) (Fig. 2A). Examination of the RNA-Seq fragments per kilobase of exon per million fragments mapped (FPKM) values for genes in each set revealed that genes in the spreading domains were significantly downregulated compared with genes not in H3K9me2 domains (Fig. 2B; Wilcoxon rank-sum test, P = 1.729e-13). Despite this, we also detected a number of genes that were not downregulated, leading us to further investigate this phenomenon. It was recently reported that H3K4me3 and H3K9ac marks could exist within stable H3K9me2 domains, forming EIs (9, 11). To investigate the potential presence of EIs at the nondownregulated H3K9me2-spreading genes, we mapped the genome-wide distribution of H3K9ac in control and Ni-exposed cells using ChIP-Seq. We examined the ratio of H3K9ac ChIP-Seq reads in Ni-treated cells compared with control cells in promoter regions of genes that were stratified by expression (Fig. 2C). Our analysis indicated that the upregulated H3K9me2-spreading genes were marked with higher levels of H3K9ac in promoter regions (Fig. 2C; Kolmogorov–Smirnov test on H3K9ac read numbers in 4-kb windows results in P < 2.2e-16 for upregulated vs. downregulated genes), suggesting the potential presence of EIs at these loci.

Fig. 2.

H3K9me2 spreading is associated with downregulation of gene expression. (A) Cartoon representing the two categories of genes used for evaluating the impact of H3K9me2 spreading. Blue arrow: genes that were incorporated into H3K9me2 domains upon Ni exposure; green arrow: genes that were not located within H3K9me2 domains. (B) Comparison of the gene expression (FPKM) of the two categories of genes. The genes in the spreading domains display the greatest downregulation. (C) The ratio of H3K9ac signal at genes in different expression groups indicates that genes that are upregulated despite H3K9me2 spreading have higher H3K9ac signals compared with the downregulated genes.

EIs Protect Genes from H3K9me2-Spreading–Associated Silencing.

To systematically analyze the potential roles of EIs in preventing H3K9me2-spreading–associated gene silencing, we searched for H3K9ac peaks (see SI Materials and Methods for details) that were unperturbed in H3K9me2-spreading domains, as exemplified by the NEO1 promoter (Fig. S6). Closer inspection of the promoter revealed depletion of H3K9me2 immediately surrounding the region marked by H3K9ac (Fig. S6). We then classified EIs across the genome in both control and Ni-treated cells, as H3K9ac peaks within H3K9me2 domains, resulting in 8,485 EIs in control cells and 10,715 EIs in Ni-treated cells with 5,615 EIs in H3K9me2-spreading domains in Ni-treated cells. Examining H3K4me3, H3K9ac, and H3K9me2 profiles around the transcription start sites for genes in H3K9me2-spreading domains with and without EIs indicated that genes in EIs within spreading domains have H3K4me3 in addition to H3K9ac marks and have drastically reduced H3K9me2 signals (Fig. 3A). Furthermore, genes within EIs in spreading domains are protected from Ni silencing compared with genes without EIs in spreading domains (Wilcoxon rank-sum test of FPKM distributions with P ≤ 0.01; see SI Materials and Methods for details). These results suggest that EIs can function to prevent H3K9me2-spreading–associated silencing, analogous to their characterized function in stem cell differentiation (11). To further investigate the role of EIs in preventing gene silencing, we tabulated the genomic distributions of EIs in both control and Ni-treated cells (Fig. 3B). We found that the distribution of EIs in control cells was biased toward intergenic regions vs. promoters whereas in Ni-treated cells the distribution was biased toward promoters vs. intergenic regions (Fig. S7; Fisher’s exact test P < 2.2e-16). These results indicate that H3K9me2 spreading preferentially disrupted EIs in the intergenic regions.

Fig. 3.

Euchromatin islands protect genes from H3K9me2-spreading–associated silencing. (A) H3K4me3, H3K9ac, and H3K9me2 profiles in Ni-treated cells for genes in EIs in spreading domains and for genes not in EIs in spreading domains. (B) Genes in EIs are protected from silencing by H3K9me2 spreading.

Disrupted H3K9me2 Domain Boundaries Are Associated with Weaker CTCF-Binding Sites.

Several studies have demonstrated enrichment of CTCF at heterochromatin domain boundaries (3, 31). To investigate the potential role of CTCF in maintaining H3K9me2 boundaries in Ni-exposed cells, we separated H3K9me2 domains into those that are maintained after Ni treatment and those that are disrupted (SI Materials and Methods). We then evaluated the best position weight matrix (PWM) score of the CTCF motif [TRANSFAC M00129 (32)] at the domain boundaries. Disrupted boundaries had a significantly lower maximum PWM score compared with the nondisrupted boundaries (Fig. 4A; t test, P = 0.0088), suggesting that stronger CTCF-binding sites could prevent H3K9me2 domain disruption and spreading. To investigate this in detail, we mapped CTCF-binding sites in control and Ni-exposed cells using ChIP-Seq. Analysis of the ChIP-Seq profiles at disrupted and maintained H3K9me2 boundaries confirmed the presence of stronger CTCF signals at the maintained boundaries (Fig. 4B). To examine if the presence of the canonical CTCF motif was correlated with the retention of CTCF binding upon Ni exposure, we stratified CTCF ChIP-Seq peaks into two categories: (i) CTCF sites that were retained upon Ni exposure and (ii) CTCF sites that were lost (SI Materials and Methods). Evaluation of the prevalence of the canonical CTCF motif in these two categories showed that the retained CTCF sites were much more likely to have the canonical CTCF motif compared with the CTCF sites that were lost upon Ni exposure (Fig. 4C). Interestingly, de novo motif discovery analysis (SI Materials and Methods) at disrupted CTCF peaks indicated the presence of a CTCF motif that was previously characterized to be associated with weaker CTCF-binding compared with the canonical CTCF site (Fig. 4D) (33). Furthermore, evaluating the levels of H3K9me2 at the lost and retained CTCF sites demonstrated that the lost CTCF sites do indeed have enrichment of H3K9me2 relative to retained sites (Fig. 4E). These results suggest that Ni could potentially affect DNA binding of CTCF with the effect being more prevalent at the weaker CTCF-binding sequences.

Fig. 4.

Ni-disrupted H3K9me2 domain boundaries have weak CTCF-binding sites. (A) PWM scores of the CTCF-binding sequence motifs at the nondisrupted H3K9me2 domain boundaries (blue) are significantly higher than those of the motifs at the Ni-disrupted boundaries (red), indicating the disruption of boundaries with weaker CTCF motifs (t test, P = 0.0088). (B) CTCF ChIP-Seq signals are stronger at the nondisrupted boundaries (green) compared with the Ni-disrupted boundaries (red). (C) CTCF-binding sites that are retained after Ni exposure possess canonical CTCF-binding motif whereas the sites that are lost after Ni exposure did not. (D) CTCF-binding sites that were lost upon Ni exposure possessed DNA sequence motifs that were similar to the weak downstream motif described by Nakahashi et al. (33) Reproduced with permission from Nakahashi et al. (33). (E) H3K9me2 signals at disrupted and retained CTCF-binding sites following Ni treatment. Disrupted CTCF sites have substantially greater H3K9me2 levels before and after Ni treatment.

Nickel Inhibits DNA Binding of CTCF in Vitro.

Ni-induced reduction in CTCF binding at the sites containing weak CTCF-binding sequences suggests that Ni could potentially interfere with the DNA-binding ability of CTCF (Fig. S8 A and B). Zinc (Zn) finger structures are important targets of toxic metal compounds. Several metals have been proposed to exert toxicity through their ability to bind Zn finger proteins (34). To examine the potential of Ni to bind CTCF, we performed in silico molecular docking analysis between Ni²⁺ and CTCF (SI Materials and Methods). Our analysis showed the binding affinity for Ni²⁺ to CTCF to be −24.0 kcal/mol. The Ni²⁺-binding affinity of the known Ni-interacting Zn finger protein XPA was −6.93 kcal/mol, indicating a very strong affinity of Ni²⁺ to CTCF.

To investigate if Ni could interfere with CTCF DNA binding in vitro, we performed electrophoretic mobility shift assay (EMSA). As a CTCF-binding sequence, we used a DNA fragment containing the well-characterized CTCF-binding site at the APP locus (Fig. S9A) (35). When incubated with BEAS-2B nuclear extracts, retardation in the migration of the wild-type probe due to the formation of a protein/DNA complex was observed (Fig. 5A). To test the effect of Ni exposure on CTCF DNA binding, we incubated the nuclear extracts with various concentrations of NiCl₂ before addition of the probe. Increasing concentrations of NiCl₂ progressively affected the gel retardation (Fig. 5B), suggesting concentration-dependent inhibition of CTCF/DNA complex formation by NiCl₂. Because we found H3K9me2 spreading to be more prevalent in the boundaries containing weak CTCF sites (Fig. 4), we next asked if the CTCF binding is more prone to Ni-induced disruption at the weak binding sites compared with the strong sites. To answer this question, we designed oligonucleotides encompassing two strong (Fig. S9 B and C) and two weak (Fig. S9 D and E) CTCF-binding sites and performed EMSA analysis. Lower concentrations of NiCl₂ could inhibit CTCF binding at the weak binding sites (Fig. 5D) compared with the strong sites (Fig. 5C), suggesting that the strength of the CTCF-binding site is inversely correlated with the Ni-mediated inhibition of its DNA binding.

Fig. 5.

Nickel inhibits CTCF DNA binding. (A) EMSA showing inhibition of DNA binding of CTCF by NiCl₂ at APP locus. The protein/DNA complex formed (lane 1) was competed by a 100-fold excess of the unlabeled wild-type probe (WT) (lane 3), but not the mutant probe (M) (lane 2). CTCF antibody supershifted the complex (lane 5), whereas the preimmune serum did not (lane 4). (B) NiCl₂ in the EMSA reaction mix inhibited binding of CTCF to the APP CTCF-binding site in a concentration-dependent manner, as seen by the loss of the shifted band. Although higher concentrations of NiCl₂ in the EMSA reaction mix were required to disrupt CTCF binding at the strong binding sites (C), CTCF binding at the weaker binding sites were disrupted by relatively lower concentrations of NiCl₂ (D). Arrow indicates CTCF/DNA complex in all lanes in panels A–D. (E) Western blot analysis showing CTCF depletion in BEAS-2B cells infected with lentiviral shCTCF constructs. β-Actin was used as loading control. (F) Gene expression analysis (qPCR) of candidate genes in CTCF knockdown cells and cells treated with 500 μM NiCl₂ for 72 h. GAPDH was used as internal control. mRNA levels of control cells were normalized to 1. (G) ChIP-qPCR analysis showing H3K9me2 levels at downregulated gene promoters in CTCF knockdown cells and cells treated with 500 μM NiCl₂ for 72 h. Fold enrichment of H3K9me2 levels over IgG were calculated and the H3K9me2 levels of control cells were normalized to 1. For both gene expression and ChIP analysis, untreated cells and cells infected with nonspecific sequence containing shControl were used as controls for Ni treatment and CTCF knockdown, respectively. Statistical significance was evaluated using t test: *P < 0.05; **P < 0.001; ***P < 0.0001.

To obtain further insight into the role of CTCF at H3K9me2 domain boundaries, we knocked down CTCF in BEAS-2B cells (Fig. 5E). To investigate whether CTCF depletion affected gene expression, we examined the mRNA levels of genes located near the H3K9me2 domain boundaries. We observed a reduction in the expression of several genes (Fig. 5F). ChIP analysis at the promoter regions of these genes showed increase in the levels of H3K9me2 at ANXA1, DKK1, and ID3 promoters (Fig. 5G), suggesting spreading of H3K9me2 in CTCF knockdown cells. These results support a role for CTCF in maintaining H3K9me2 domains. We then compared the gene expression and H3K9me2 alterations between CTCF knockdown cells and Ni-exposed cells. Surprisingly, both the decrease in the gene expression (Fig. 5F) and the increase in H3K9me2 levels (Fig. 5G) were larger in Ni-exposed cells compared with the CTCF knockdown cells. Previous studies have shown that the JmjC domain H3K9me2 demethylases could be inhibited by Ni, resulting in an increase in the total levels of H3K9me2 (19, 36, 37). Taken together, our results suggest that, although depletion of CTCF could initiate spreading of H3K9me2 and gene silencing, the effect is more robust in Ni-exposed cells, where reduction in CTCF binding could be accompanied by a reduction in demethylase activity, thus augmenting H3K9me2 spreading and gene silencing (Fig. 5 F and G). Interestingly, MAP2K3 gene expression level was maintained in CTCF knockdown cells with no changes in H3K9me2 levels. However, it was significantly downregulated by nickel, accompanied by an increase in H3K9me2 (Fig. 5 F and G). This suggests that mere loss of CTCF binding is insufficient for H3K9me2 spreading into the MAP2K3 promoter and its downregulation and that other mechanisms such as the inhibition of demethylases are required.

Discussion

In this study, we used nickel exposure to understand the dynamics of H3K9me2 domains. Ni is a nonmutagenic carcinogen. Whereas Ni affects several aspects of cellular regulation (17, 23, 38), epigenetic alterations induced by Ni play an important role in inducing gene expression changes (16, 21). Because H3K9me2 is organized into large repressive domains, Ni silencing of a transgene occurring only if it is positioned close to the heterochromatic locus (13) suggested disruption of H3K9me2 domain structure. We used this phenomenon to understand the dynamics of H3K9me2 domains on a genome-wide scale. Our results show that Ni can induce H3K9me2, which corresponded with the downregulation of gene expression. This suggests breakdown of H3K9me2 chromatin domain structure during Ni-induced gene silencing. Local discrete changes in H3K9me2, associated with corresponding gene expression changes during neuronal differentiation of embryonic stem cells (9), suggested a role for H3K9me2 in fine-tuning gene expression by functioning as a regulatory switch. Ni-induced spreading of H3K9me2 is reminiscent of the local discrete changes, suggesting that misregulation of this regulatory switch results in altered gene expression. Furthermore, several earlier studies have suggested ectopic spreading of H3K9me2 to be an important signal for transcriptional silencing (27–30, 39). However, the alternate possibility of gene silencing resulting in spreading of H3K9me2 in a subset of the spreading domains cannot be ruled out. Nevertheless, H3K9me2 spreading into active regions remains a significant event because this could play a role in stably silencing the genes, given that H3K9me2 could be a precursor to DNA methylation and long-term gene silencing (40, 41).

Although spreading of H3K9me2 into promoters was associated with reduction in gene expression, we detected a subset of genes whose expression was not affected by H3K9me2 spreading. Previous studies have characterized short H3K9me2-depleted EIs within large H3K9me2 domains in both pluripotent and differentiated cells (11). Active modifications of H3K4me3 and H3K9ac are hallmarks of EIs (11). Our results suggest that the repressive effect of H3K9me2 could be overcome by the high levels of H3K9ac at the EIs, indicating the existence of an additional level of regulation that potentially could overcome the silencing effect of H3K9me2 spreading. Acetylation maintains an open chromatin structure, allowing transcription factor binding (42). It remains to be seen if potential binding of any transcriptional activator(s) to the strongly acetylated regions has a role in protecting the gene from epigenetic repression.

Interestingly, Ni-disrupted H3K9me2 domain boundaries predominantly contained weaker CTCF-binding motifs and displayed lower CTCF ChIP-Seq signals compared with the nondisrupted domain boundaries. Previously, several studies, including ours, have suggested a role for CTCF in establishing functional expression domains (3, 43, 44). Although loss of CTCF was shown to result in spreading of H3K27me3 at the HoxA locus (45), no evidence of H3K27me3 spreading was detected following CTCF knockdown in Drosophila (43, 46, 47). Therefore, the domain barrier activity of CTCF has remained inconclusive (43, 48). Our results showing disruption of H3K9me2 domain boundaries at weak CTCF-binding sites suggest a role for CTCF in H3K9me2 domain maintenance. Previous studies have shown that Ni interaction with Zn finger proteins results in inhibition of DNA binding and alteration of DNA-binding specificity (49, 50). Interestingly, our results suggest that nickel at noncytotoxic concentrations preferentially inhibits binding of CTCF to the weaker binding sequences (Fig. 5 and Fig. S9), resulting in H3K9me2 domain disruption and spreading. A recent work has indicated a remarkable diversity of DNA sequence motifs for CTCF binding, with specific motifs associated with strength of CTCF binding (33). Intriguingly, we identified a motif at the weak CTCF-binding sites that was previously shown to be associated with weaker CTCF binding (33).

Furthermore, our results show that CTCF depletion results in downregulation of gene expression, accompanied by an increase in the H3K9me2 levels, which is suggestive of H3K9me2 spreading in CTCF knockdown cells (Fig. 5). These results suggest a role for CTCF in H3K9me2 domain barrier function. Interestingly, the levels of gene silencing (Fig. 5F) and enrichment of H3K9me2 (Fig. 5G) were larger in Ni-treated cells compared with the CTCF knockdown cells. This suggests involvement of other mechanism(s) in addition to loss of CTCF binding during Ni-induced gene silencing. H3K9me2 demethylation is catalyzed by JmjC domain histone lysine demethylases (JHDMs), which are members of the dioxygenase superfamily of enzymes containing iron at its catalytic center. The demethylation of lysines by JHDMs occurs by catalyzing the generation of oxidized iron. Ni inhibits this family of enzymes by displacing the iron (19, 36, 37). Previously, loss of H3K9 demethylase LSD1 has been demonstrated to result in spreading of H3K9me2 (51). Therefore, inactivation of H3K9me2 demethylases could be a contributing factor for Ni-induced H3K9me2 spreading and gene silencing.

In conclusion, our studies demonstrate that Ni exposure can inhibit CTCF DNA binding at the weak binding sites. We show that H3K9me2 domain disruption and spreading preferentially occurs at the boundaries containing weak CTCF-binding sites, which is associated with downregulation of gene expression. Because chromatin domain structures form an important basis for the establishment of gene expression profiles (3), the mechanistic insights that we provide on the underlying causes of domain disruption will have profound implications in understanding pathogenesis induced by aberrant alterations to the chromatin structure.

Materials and Methods

Full details are in SI Materials and Methods.

BEAS-2B cells were cultured in Dulbecco's modified Eagle's medium (Cellgro). For Ni treatment, the cells were exposed 500 μM NiCl₂ (Sigma N6136) for 72 h. ChIP-Seq and RNA-Seq data were deposited in the Gene Expression Omnibus (www.ncbi.nlm.nih.gov/geo) under accession no. GSE56053.