The histone variant H2A.X is a regulator of the epithelial–mesenchymal transition

The epithelial–mesenchymal transition (EMT), considered essential for metastatic cancer, has been a focus of much research, but important questions remain. Here, we show that silencing or removing H2A.X, a histone H2A variant involved in cellular DNA repair and robust growth, induces mesenchymal-like characteristics including activation of EMT transcription factors, Slug and ZEB1, in HCT116 human colon cancer cells. Ectopic H2A.X re-expression partially reverses these changes, as does silencing Slug and ZEB1. In an experimental metastasis model, the HCT116 parental and H2A.X-null cells exhibit a similar metastatic behaviour, but the cells with re-expressed H2A.X are substantially more metastatic. We surmise that H2A.X re-expression leads to partial EMT reversal and increases robustness in the HCT116 cells, permitting them to both form tumours and to metastasize. In a human adenocarcinoma panel, H2A.X levels correlate inversely with Slug and ZEB1 levels. Together, these results point to H2A.X as a regulator of EMT.

T he spread of cancerous cells to distant organs is a longstanding obstacle to successful cancer treatment. The process by which cells from primary tumours acquire the ability to form distant tumours involves the loss of cell-to-cell adhesion as well as the disruption of the apicobasal polarity, and the transition to a cell type with a more spindle-like morphology 1 . Such changes enable the cells to invade the extracellular matrix 2 . This reversible physiological process is known as the epithelialmesenchymal transition (EMT or MET in reverse). The molecular mechanisms underlying EMT include decreased expression of a set of epithelial genes with the concomitant activation of a set of mesenchymal genes, the expression of matrix metalloproteinases markers and the formation of lamellipodia, filopodia and invadopodia 3,4 . At distant sites, some mesenchymal cells may be involved in the establishment of tumours 2,5,6 in a process thought to require at least partial re-acquisition of epithelial characteristics.
Changes in chromatin configuration have emerged as key to EMT-related transcription factor regulation 1,7-10 , but some of these changes still call for further characterization. While the four nucleosome histone families provide equal numbers of molecules to the nucleosome, several of the families include multiple variants, whose stoichiometry can vary due to cell type and growth state among other factors [11][12][13] . Altered expression of variants in several histone families, including H2A has been associated with cancer 14 . Recently, it has been reported that histone H2A variant macroH2A is a critical component of chromatin that suppresses the progression of melanoma 15 .
Histone H2A.X also belongs to the histone H2A family. Like other histone variants, H2A.X is highly conserved among species and achieves critical cellular functions beyond those fulfilled by canonical H2As. H2A.X plays essential roles in DNA doublestrand break repair and genome stability, and is classified as a tumour suppressor. As with other H2A variants, the relative amount of H2A.X varies among cell lines 16,17 . How this variation may affect the transcription regulation of other genes remains poorly investigated.
While comparing growth characteristics of H2A.X-null cells with parental lines, we observed that the null cells exhibited elevated levels of migration and invasion, characteristic of the EMT transition. Given these observations and the increasing evidence for the role of other histone variants in the regulation of gene transcription 18,19 and cancer progression 16,20,21 , we hypothesized that the downregulation of histone variant H2A.X may contribute to the alteration of chromatin configuration and induce changes in cancer gene expression.
Our novel findings provide evidence that H2A.X depletion activates the EMT programme in at least some human colorectal adenocarcinoma cells. The loss of H2A.X was strongly correlated with the EMT-inducing transcription factors Slug and ZEB1 in these cells. These correlations were substantiated by the observations that the silencing of Slug and ZEB1 abrogated the mesenchymal phenotype exhibited by H2A.Xdepleted cells. Most importantly, restored expression of H2A.X at least partially reversed the EMT programme induced by H2A.X loss. H2A.X-deficient cells are proliferation defective, and sensitive to environmental and genotoxic stresses 20,22 ; characteristics which may counteract their increased invasiveness and account for the lack of enhanced metastasis in vivo compared with parental cells. However, in the H2A.X revertants, proliferation is enhanced, but sufficient invasiveness may remain to result in elevated numbers of metastatic lung foci. Taken together, our results demonstrate that H2A.X may be a novel regulator of the EMT programme and suggest a role for H2A.X in cancer progression and metastasis.

Results and Discussion
H2A.X regulates EMT and colon cancer metastasis signalling. We observed that when cultures of the human colon cancer line HCT116 were made deficient in histone H2A.X, they lost their epithelial shape, became more mesenchymal-like (Fig. 1a), and more invasive (Fig. 1b). These findings suggested a possible role for histone H2A.X in EMT. We then performed a genome-wide differential gene expression analysis comparing H2A.X-deficient and control HCT116 cells (Full data available at http:// www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=mdcbmqikhdcvt sp&acc=GSE75444). Pathway analysis of the differentially expressed genes using Ingenuity Pathway Analysis (IPA) revealed a significant enrichment for multiple signalling pathways involved in cancer, with the EMT and the colorectal cancer metastasis signalling pathways ranking among the top three (Fig. 1c). The output from IPA using the 939 differentially expressed genes revealed 85 genes involved in cell movement (Fig. 1d.). Of these, 18 genes were common to three cell movement-associated phenotypes, including cell invasion, migration and metastasis (Fig. 1d); 13 genes known to be essential for mesenchymal traits increased in expression, while 5 epithelial genes exhibited decreased expression levels (Fig. 1e). Real-time PCR (RT-PCR) confirmed that silencing H2A.X promoted greatly increased expression of several key mesenchymal genes, including SNAI2 or Slug, ZEB1, vimentin (VIM), THBS1, versican (VCAN) and TGFB2, coupled with decreased expression of a key epithelial gene, CDH1 (Fig. 1f). Two of these genes, Slug and ZEB1, encode EMT-inducing transcription factors. Thus, our results point to Slug and ZEB1 as key EMT-related transcription factors regulated by H2A.X in colon cancer cells.
One possible explanation for these results is that the observed phenomena are not due specifically to H2A.X removal, but to a disruption in the orderly chromatin structure, which is manifested by EMT-like changes. If this is the case, then removal of other H2A variants might lead to the same EMT-like changes. One variant, macroH2A, has been reported to be involved in cancer progression and metastasis 15 , and EMT is considered essential for cancer metastasis 1,23 . However, when we investigated whether macroH2A.1 or H2A.Z downregulation in HCT116 cells led to EMT-like changes, little, if any, change in Slug or ZEB1 or the other genes examined was found ( Supplementary Fig. 1a-d). Likewise, to investigate whether any transient changes in the pool of the nucleosomal H2A.X might result in the activation of EMT, we conducted protein-and transcript-level analysis of EMT markers in cells overexpressing H2A.X for 5 days. Data indicate that H2A.X overexpression did not substantially affect the expression level of the EMT markers examined ( Supplementary  Fig. 2a,b), whereas a slight activation of the EMT programme was observed in cells transiently silenced for H2A.X within the same period of time ( Supplementary Fig. 2c). Thus, we conclude that EMT programme initiation in HCT116 cells can likely be attributed specifically to H2A.X silencing.
Slug and ZEB1 mediate H2A.X loss-induced EMT. The EMT programme is orchestrated by a set of pleiotropically acting transcription factors, including Snail1/2 (refs 24,25), ZEB1/2 (refs 26,27), Twist1/2 (refs 28,29), FOXC2 (ref. 30), among others 3 . It is widely accepted that the E-box-binding zinc finger transcription factors Slug and ZEB1 contribute to the EMT in primary epithelial cancers 31 ; and a Slug-driven transcriptional activation of ZEB1 has been proposed for a cooperative regulation of the EMT programme in melanoma 32 .
Thus, the finding that H2A.X-deficient cells exhibited substantially increased Slug and ZEB1 expression raised the question of whether these transcription factors mediated H2A. X-regulated EMT. This was examined utilizing siRNAs to silence Slug and ZEB1 in the H2A.X-deficient cells. Expression of epithelial proteins including E-cadherin and b-catenin returned to near control levels as measured by immunoblotting (Fig. 2a) and immunofluorescence (Fig. 2b). However, although Slug and ZEB1 appeared to be fully silenced (Fig. 2a), and although the expression levels of the three examined mesenchymal genes, VIM, VCAN and integrin beta-4 (ITGB4), were decreased to varying degrees, they remained well above the values in the control cells ( Fig. 2a and Supplementary Fig. 3). Independent silencing of Slug and ZEB1 suggested that these transcription factors may cooperate to regulate EMT in H2A.X-deficient cells, as the co-silencing of Slug and ZEB1 resulted in a greater reversal of the expression of the examined EMT markers than silencing each separately ( Supplementary Fig. 4). In addition, silencing of ZEB1 alone seemed to partially affect the transcript level of Slug, further suggesting a link between Slug and ZEB1. Importantly, however, while silencing Slug and ZEB1 failed to lower the expression of mesenchymal factors to control levels, it was sufficient to restore the epithelial phenotype of H2A.X-deficient HCT116 cells (Fig. 2c). These observations indicate that the Slug and ZEB1 transcription factors are essential in mediating the EMT phenotype driven by H2A.X loss.

Number of cells
To confirm these observations, we generated H2A.X knockout HCT116 cells using the CRISPR/Cas9 procedure 33 . These H2A.X knockout HCT116 cells exhibited greatly increased expression of the mesenchymal genes including Slug, ZEB1, VIM, THBS1, VCAN and TGFB2, coupled with decreased expression of the epithelial gene CDH1 ( Supplementary Fig. 5a), confirming our results using short hairpin RNA. H2A.X-null (H2A.X-KO) cells exhibited decreased protein levels of E-cadherin and b-catenin, coupled with increased ITGB4 levels similar to those found in the shH2A.X HCT116 cultures (Fig. 2d). Likewise, H2A.X depletion using CRISPR/Cas9 system in an additional colorectal cancer cell line, HCT15, resulted in the induction of the EMT programme and enhanced invasion, further confirming a broader effect of H2AX loss (Supplementary Fig. 5b-e). In addition, these results show that decreasing Slug and ZEB1 expression led to a partial reversal of the EMT induced by H2A.X removal, with the epithelial markers E-cadherin and b-catenin returning to near control levels but without large decreases of the mesenchymal marker ITGB4 (Fig. 2d,e and Supplementary Fig. 6).
As H2A.X is primarily known for its role in DNA damage repair 16,34 , the observation that H2A.X appeared to affect the EMT raises the question as to whether H2A.X directly influences the transcriptional regulation of Slug and ZEB1 or whether accumulation of DNA damage resulting from H2A.X deficiency indirectly activates Slug and ZEB1, perhaps through DNA damage pathway kinases such as ATR and ATM. Recently, ATM-mediated stabilization of ZEB1 was reported to link EMT to radioresistance in vivo as well as in vitro 35 . However, inactivation of ATM with KU55933 for 3 days in HCT116 cells with downregulated H2A.X did not result in altered ZEB1 protein levels ( Supplementary Fig. 7), suggesting that H2A.X-deficiencyinduced ZEB1 accumulation most likely involves a different mechanism.
The effect of H2A.X on Slug and ZEB1 promoter activities was found to be a 2.44-and 2.70-fold activation, respectively, using shH2A.X, and a corresponding 6.44-and 7.24-fold activation with the H2A.X knockout (Fig. 3a). By contrast, the promoter activity of GAPDH, a gene known not to be involved in EMT, remained relatively unchanged. However, H2A.X overexpression did not lead to changes in Slug and ZEB1 promoter activity ( Supplementary Fig. 8a,b). Chromatin immunoprecipitation (ChIP) of H2A.X and H3K9ac, a histone mark of active promoters (H3K9ac) [36][37][38] , revealed that reduction of H2A.X in the Slug and ZEB1 promoters (Fig. 3b) was strongly correlated with an accumulation of H3K9ac at the same locus (Fig. 3c). Similar results were obtained with an additional active transcription mark (H3K27ac) 39,40 (Supplementary Fig. 8c). Thus, we surmise that H2A.X removal from the Slug and ZEB1 promoters may enhance their transcriptional activation through chromatin relaxation, subsequently leading to induction of the EMT programme (Fig. 3d). However, how this chromatin relaxation specifically occurs in EMT-related transcription factors loci still remains unclear.
Restored expression of H2A.X induces partial EMT reversal. H2A.X expression was restored in the knockout HCT116 cells to examine whether the characteristics of the parental HCT116 cells would be restored. Revertant clones with parental H2A.X levels exhibited growth rates similar to those observed in the parental cells, while clones with lower H2A.X levels exhibited slower proliferation rates ( Supplementary Fig. 9a-c). Interestingly, however, a quantitative assessment of key EMT protein and transcript levels revealed that the re-expression of normal levels of H2A.X did not completely restore the EMT markers to their original levels. Epithelial markers, including E-cadherin and b-catenin, increased in expression to levels close to those in the parental cells. Mesenchymal markers, including Slug, ZEB1 and ITGB4, decreased in expression, but the levels of these markers remained substantially higher than the parental levels ( Fig. 4a,b). These findings are consistent with the revertants exhibiting a morphology intermediate between the H2A.X KO and H2A.X parental cells ( Fig. 4c and Supplementary Fig. 9d).
H2A.X re-expression promotes metastatic colonization. Since cancer metastasis may involve the ability to undergo both EMT and MET 2,3 , cells that have both epithelial and mesenchymal characteristics may have increased metastatic potential. To investigate how the presence, removal and reintroduction of H2A.X impacts the metastatic colonization potential of cancer cells, we injected parental, H2A.X knockout and H2A.X revertant cells into the tail veins of immuno-compromised mice and analysed the number of macroscopic lung metastases after 4 weeks (see 'Methods' section). While the parental and knockout cells resulted in a similar number of metastatic lung foci (Fig. 4d,e), the cells with restored expression of H2A.X resulted in a 2.5-fold increase in the number of macroscopic metastases (Fig. 4d,e).
These results indicate that re-introduction of H2A.X into HCT116 H2A.X-null cells induced an incomplete MET, and enhanced proliferative ability coupled with some retained mesenchymal characteristics. These observations are consistent with published reports supporting the requirement for MET and cell proliferation in establishing metastatic colonization 28,[41][42][43][44] . When the H2A.X revertant cells were exposed to ionizing radiation, they exhibited levels of DNA damage (Fig. 5a,b), DNA repair capacity (Fig. 5c,d and Supplementary Fig. 10) and cell survival (Fig. 5e,f) similar to those of the parental cells, observations consistent with proposal that the survival of mesenchymal cells in the circulation requires resistance to environmental and genotoxic stresses 1,45 . Collectively, our observations suggest that restoration of H2A.X endows the H2A.X knockout cells with enhanced survival ability while still permitting some mesenchymal characteristics, thus leading to higher potential for metastatic colonization.
One intriguing issue is whether the DNA repair-associated motif of H2A.X, phosphorylatable Ser139 residue, is required for the ability to induce EMT reversal. To test this, we generated H2A.X knockout cells expressing a mutant H2A.X (H2A.X  S139A) that is unable to detect DNA damage and is strongly associated with defective cell proliferation ( Supplementary  Fig. 11a-c). Surprisingly, the EMT reversal is impaired in H2A.X mutant revertants, just as in the H2A.X-null cells ( Supplementary Fig. 11d, e). These results suggest that the DNA repair function of H2A.X may be required for its regulation of EMT-related genes. However, how S139 phosphorylation may control the direct transcriptional regulation of genes to which H2A.X-containng nucleosomes are bound still remains unclear. Several clinical studies have pointed out that metastases arising from carcinomas exhibit a well-differentiated epithelial phenotype 46,47 . Also, in an experimental metastasis model, both primary tumours and metastatic nodules were found to express similar levels of key EMT markers 28 . In line with these findings, we compared the transcript levels of Slug, ZEB1 and H2A.X in a panel of 18 primary colorectal cancers and 18 matched liver metastases, using the publicly available data sets from Gene Expression Omnibus, GEO (GSE14297; ref. 48). The data indicate that both primary tumours and metastatic nodules displayed similar levels of Slug, ZEB1 and H2A.X ( Supplementary Fig. 12), further supporting the idea that EMT markers are similar in the metastases and their primary carcinomas. Furthermore, Slug,   ZEB1 and H2A.X display similar transcript levels in normal and primary tumour tissues from the same data sets, suggesting that these genes are not involved in the early stages of tumorigenesis. However, as the panel is very limited with only seven specimens from normal colon tissues, follow-up experiments with a large cohort will be needed.
H2A.X correlates with Slug and ZEB1 in human adenocarcinoma.
Colorectal cancer (CRC) is the second-and third-most common cancer in women and men, respectively worldwide 49,50 . To investigate whether the role of H2A.X in the regulation of EMT-inducing transcription factors may provide clues for any clinical correlation, we conducted an expression level comparison of H2A.X, Slug and ZEB1 in human colon adenocarcinoma samples available in the Cancer Genome Atlas (TCGA) data sets.
The data indicate that the transcript levels of H2A.X, Slug and ZEB1 are significantly correlated across all the colorectal carcinoma samples, based on the Pearson correlation coefficient (H2A.X and ZEB1, r ¼ À 0.516, Po0.0001; H2A.X and Slug, r ¼ À 0.287, Po0.0001, using the non-parametric Mann-Whitney t-test). No correlation was found with Snail1, an EMT-transcription factor found to be near the cut-off value of the fold-change level ( À 1.52506) in the microarray data sets (Fig. 6a-c, Supplementary Fig. 13). Furthermore, utilizing the NCI60 genome-wide expression data set (http://discover.nci.nih.gov/cellminer/), we found a strong negative correlation between H2A.X and Slug/ZEB1 expression in four of the seven colon cancer cell lines (Fig. 6d). Collectively, these observations indicate that Slug and ZEB1 may be selectively regulated by H2A.X in colon cancer, and further suggest that H2A.X may regulate EMT-inducing transcription factors during colon cancer progression.
From the physiological standpoint, EMT is a process that involves regulation by a set of factors ranging from alterations in cellular metabolism to contextual microenvironmental signals including inflammatory cytokines and oncogenic stressors among others [51][52][53] . Transforming growth factor beta (TGF-b) is one of these factors 54,55 . Genes profiles from Gene Expression Omnibus, GEO (GSE17708; ref. 56), indicate that the H2A.X gene is downregulated in the A549 lung adenocarcinoma cell line in a model of TGF-b-induced EMT. Hence, it could be anticipated that changes in the control of TGF-b regulation may be essential in maintaining H2A.X levels in vivo; and this provides a clue for the physiological conditions that may link H2A.X to EMT. On the other hand, additional hits including changes in H2A.X gene regulation during the cell cycle may also be critical for the role of H2A.X in EMT in vivo. For instance, it is known that H2A.X is synthesized in replication-dependent manner to ensure the presence of sufficient H2AX molecules in the replicating  genome for efficient DSB detection 16,57 . Changes in the H2A.X levels may then occur during the cell cycle progression, and this raises the possibility that such variations could subsequently affect the EMT programme. Nevertheless, further investigation is needed to delineate well-defined physiological environments that may underlie the link between H2A.X level and the regulation of EMT genes in vivo.
In summary, our study provides novel evidence that H2A.X loss activates the EMT programme in two human colon cancer cell lines, making them more prone to invasion. However, H2A.Xdeficient cells failed to exhibit enhanced metastatic colonization potential in vivo, perhaps because they have reduced proliferative ability, and are vulnerable to genotoxic stress. Restoration of H2A.X promotes partial EMT reversal, endowing cells with enhanced metastatic colonization ability, by enhancing their proliferative capacity and damage resilience, while maintaining their invasive properties (Fig. 6e,f). This may explain, in part, the lack of difference in H2A.X expression seen between the primary tumours and the matched liver metastases in human samples.

Methods
Cell culture and plasmid transient transfection. The human colorectal carcinoma cell lines HCT116 and HCT15 (parental lines obtained from ATCC and kept at the Frederick National Laboratory for Cancer Research, DTP, DCTD, NCI Tumor Repository), were grown at 37°C with 5% CO 2 in DMEM (Invitrogen, Carlsbad, CA, USA), supplemented with 10% fetal bovine serum (FBS, Atlanta Biologicals, GA, USA). The cells were used as models to study the changes associated with epithelial-mesenchymal transition. In addition, the cells were authenticated using short tandem repeat Identifiler, and tested for mycoplasma contamination. All media were supplemented with 2 mM glutamine, penicillin and streptomycin (Invitrogen).  Luciferase assay. The promoter luciferase reporter assay was performed as previously described 59  low-melting agarose to make a final agarose concentration of 0.85%. The mixture was spread over the sample area on FLARE slides (Trevigen, Gaithersburg, MD, USA). After lysis, the slides were transferred into alkali solution (pH 413) for 1 h at room temperature in the dark, then electrophoresis was performed (0.7 V cm À 1 per 300 mA) for 1 h at room temperature in the dark. After electrophoresis, the slides were washed three times for a total of 20 min with water. Then, the slides were washed in chilled 70% alcohol for 15 min. Each slide was stained with 50 ml of 1:10,000 diluted SyBr green solution (Trevigen, Gaithersburg MD); and then air-dried overnight in the dark. The individual cells with or without comet were visualized using an Olympus fluorescence microscope. The analysis was done by scoring at least 50 comets per sample using the CometScore software package (Tritek Corp., Sumerduck VA, USA). The results are shown as means ± s.e.
Gene deletion using CRISPR/Cas9 system. H2A.X gene (H2AFX) deletion was performed using CRISPR/Cas9 system as previously described [63][64][65] . Briefly, the H2AFX-specific guide RNA sequence was chosen from http://crispr.mit.edu database; and inserted into the pX330 vector (Addgene, Cambridge, MA, USA). For the donor plasmid, the left and the right homology arms were PCR-amplified from human primary fibroblasts genomic DNA. Both the arms along with antibiotic resistance ORF (hygromycin) were inserted between BamHI and EcoRI sites of the PCR2.1 vector (Life Techonologies; Grand Island, NY, USA).
Experimental lung metastasis model. All animal procedures were performed according to protocols approved by NCI-Frederick Animal Care and Use Committee (ACUC). Tail vein injection was performed as previously described 66,67 . Four to six-week-old male athymic nude mice (Charles River Laboratories, Frederick, MD, USA) were injected with HCT116 parental cells (WT), H2A.X knockout cells (KO) and H2A.X revertant cells (KO þ H2A.X) via the lateral tail vein using 25-gauge needles and followed up for lung metastases burden. In brief, 5 Â 10 6 cells suspended in 500 ml PBS were injected into the tail vein of each mouse. After 4 weeks, the animals were killed and examined macroscopically and microscopically for the presence of metastases.
Statistical analysis. Statistical analyses were performed using GraphPad Prism 6 software (GraphPad Software) and Microsoft Excel 2010. Parametric data were analysed using a two-tailed t-test. A value of Po0.05 was considered statistically significant. Data are presented as mean ± s.d.