DNA methylome analysis identifies epigenetic silencing of FHIT as a determining factor for radiosensitivity in oral cancer: an outcome-predicting and treatment-implicating study.

Radioresistance is still an emerging problem for radiotherapy of oral cancer. Aberrant epigenetic alterations play an important role in cancer development, yet the role of such alterations in radioresistance of oral cancer is not fully explored. Using a methylation microarray, we identified promoter hypermethylation of FHIT (fragile histidine triad) in radioresistant OML1-R cells, established from hypo-fractionated irradiation of parental OML1 radiosensitive oral cancer cells. Further analysis confirmed that transcriptional repression of FHIT was due to promoter hypermethylation, H3K27me3 and overexpression of methyltransferase EZH2 in OML1-R cells. Epigenetic interventions or depletion of EZH2 restored FHIT expression. Ectopic expression of FHIT inhibited tumor growth in both in vitro and in vivo models, while also resensitizing radioresistant cancer cells to irradiation, by restoring Chk2 phosphorylation and G2/M arrest. Clinically, promoter hypermethylation of FHIT inversely correlated with its expression and independently predicted both locoregional control and overall survival in 40 match-paired oral cancer patient samples. Further in vivo therapeutic experiments confirmed that inhibition of DNA methylation significantly resensitized radioresistant oral cancer cell xenograft tumors. These results show that epigenetic silencing of FHIT contributes partially to radioresistance and predicts clinical outcomes in irradiated oral cancer. The radiosensitizing effect of epigenetic interventions warrants further clinical investigation.


INTRODUCTION
Radiotherapy is an important modality for the treatment of oral cancer. After radical surgery, postoperative radiotherapy (with or without chemotherapy) is recommended for patients with pathologically adverse features, such as pT3-4, pN1-3, extracapsular nodal extension, and positive surgical margins [1]. However, a major obstacle is the development of radioresistant cancer cells, eventually leading to locoregional recurrences, and the molecular mechanism that underlies radioresistance is not fully understood.
Unlike genetic events, epigenetic modifications do not involve changes of DNA sequences, but do have profound effects on gene promoter activity [2][3][4]. Two types of epigenetic modifications are crucial in cancers. First, DNA methylation of promoter "CpG islands" of tumor suppressor genes, resulting in their transcriptional repression, is frequently observed in cancers [5][6][7]. Second, specific post-translational modifications of chromatin proteins, such as EZH2-mediated trimethylation of histone 3 lysine 27 (H3K27me3), are known transcriptional silencing gene "marks" [8][9][10]. Interestingly, EZH2bound region of the genes marked by H3K27me3 are able to recruit DNA methyltransferases (DNMTs) for DNA methylation, also resulting in epigenetic silencing of tumor suppressor genes [11].
We and others have previously demonstrated that aberrant promoter hypermethylation of tumor suppressor genes is an important cancer-specific event useful for predicting prognosis in several human cancers [12][13][14][15]. For example, we have previously identified promoter hypermethylation of DAPK to be predictive of locoregional control in head and neck cancers [16]. However the role of aberrant epigenetic alterations in radioresistance of oral cancer is not fully explored.
In the current study, we established an in vitro model to investigate the role of aberrant epigenetic modifications in the development of oral cancer radioresistance. Using a methylation microarray, we show that promoter hypermethylation of Fragile histidine triad (FHIT) facilitates radioresistance after massive irradiation in oral cancer cells, and that this event has in vitro and in vivo prognostic value for demarcating possible radioresensitization of this deadly disease by epigenetic interventions.

Differential methylation analysis between OML1-P and OML1-R cells
To explore the role of epigenetic modifications in the onset of radioresistant oral cancer, we developed a radioresistant oral cancer cell (OML1-R) subline, from parental OML1-P cell, using a hypo-fractionated irradiation protocol (5Gy by 10 fractions; Fig. 1A). A single test fraction of 10-Gy irradiation confirmed the established radioresistance of OML1-R cells, as compared to the parental cells (P < 0.0001; Fig. 1B, C).
To identify genes differentially methylated in radioresistant cells, OML1-R and the parental cells were compared by methylation analysis using Illumium 27K methylation BeadChip microarrays. Of approximately 300 probes found to be significantly hypermethylated in OML1-R cells (Fig. 1D, red spots), and filtering for probes with initial β values of <0.5 in the parental cells, 180 probes were found significantly hypermethylated (Supplementary Table S2). Subjecting this set of differentially methylated genes to ontology analysis by DAVID [17] revealed several significantly enriched biological processes (Supplementary Table S3).
Aberrant DNA methylation has been previously reported in cellular non-response to ionizing radiation, particularly for genes involved in cell cycle control, DNA repair, and apoptosis [18]. In this regard, genes significantly enriched in the corresponding biological processes (purine nucleoside metabolic process and DNA metabolic process), based on our gene ontology analysis, were further screened. One of the targets, FHIT, a gene that regulates G2/M checkpoint and apoptosis was then selected for further analysis. FHIT was also selected based on its location in a fragile chromosome site (3p13.2) that would likely be damaged by ionizing irradiation [19].

FHIT is epigenetically silenced in OML1-R cells
To validate our microarray result, we performed methylated-binding DNA (MBD) capture coupled to real time PCR (MBDcap-PCR), in addition to bisulphite pyrosequencing, of the promoter region of FHIT ( Fig.  2A). Both MBDcap-PCR (Fig. 2B) and bisulphite pyrosequcening (Fig. 2C) confirmed higher FHIT promoter methylation in OML1-R than in OML1-P cells. More importantly, FHIT promoter hypermethylation associated with down-regulation of its mRNA and protein (Fig. 2D). These phenomenon may be attributed to an increased expression of DNMT3a, 3b but not DNMT 1 in OML1-R cells (Fig. S2).
We also conducted H3K27me3 ChIP-PCR to examine the histone chromatin status of the promoter region of FHIT in OML1-R and the parental cells. As expected, OML1-R cells possessed higher H3K27me3 levels around the promoter region of the FHIT gene (Fig.  2E). The enrichment of this repressive histone mark may be due to overexpression of the histone methyltransferase, EZH2 in OML-1R cells (Fig. 2F).
EZH2 is a key component of the Polycomb repressive complex 2 (PRC2) and is involved in transcriptional repression [20]. Previous studies www.impactjournals.com/oncotarget demonstrated that EZH2-mediated H3K27me3 correlates tightly with DNA methylation [11,[21][22] or at least in a regional-dependent manner [23]. However, contradictory evidences also suggested that H3K27me3 and DNA methylation are mutually exclusive [24][25]. To examine the role of EZH2 in the epigenetic silencing of FHIT, we depleted EZH2 in OML1-R cells ( Fig 3A). As expected, lentiviral knock-down of EZH2 resulted in a re-expression of FHIT in OML1-R cells (Fig 3B). This re-expression is accompanied by a relaxation of chromatin as demonstrated by about 2-fold increase of H3K4me3 and 2-fold decrease of H3K27me3 in the promoter region of FHIT ( Fig 3C). Importantly, EZH2-depleted cells also showed a 10%-decrease of FHIT promoter methylation as demonstrated by bisulphite pyrosequencing (Fig 3D). These results suggested that promoter methylation of FHIT is partially controlled by EZH2 in the promoter region of FHIT. schema for establishing a radioresistant oral cancer cell subline. One day before irradiation (R1 -1 , R2 -1 …etc), cells were seeded onto 10cm cell culture plates. On the following day (R1, R2…R10), a fraction of 5-Gy irradiation was delivered using a 6-MV linear accelerator. The cells were allowed to recover for 5-7 days before another round of irradiation. This process was repeated for up to 10 fractions (R10) until a total dose of 50 Gy was obtained. The irradiations from R2 to R10 are indicated by dash line. (B) Radiation stress test in parental OML1-P and radioresistant OML1-R cells using a single 10-Gy irradiation. Cells were allowed to recover for several days and then stained with 0.4% crystal violet (C) Quantitative analysis of the radiation stress test described in B. Stained cells were lysed and measured by a spectrophotometer, and the relative number of cells were expressed as OD 580 . Data are expressed as means ± SD (n=3). (D) Volcano plot showing methylation differences and P values of each probe between OML1-R and their parental OML1-P cells. A total of 24,000 gene loci were compared. Of these, 330 probes (red) demonstrated an increased β value ≥ 0.2 and P <0.05, in OML1-R cells.

FHIT is epigenetically silenced in a panel of oral cancer cell lines
We then explored the correlation between FHIT expression and promoter methylation in a primary human oral keratinocyte (HOK) and four other oral cancer cell lines (OCSL, SCC25, SAS, and SCC4). As expected, similar to OML1-R cells, FHIT expression (Fig. 4B) demonstrated a tight inverse correlation with promoter methylation of FHIT (Fig. 4C) in those cells. Taken together, these results strongly suggest that epigenetic modifications contribute to the process of radioresistance via down-regulation of FHIT in oral cancer cells.

Overexpression of FHIT restores radiosensitivity in OML1-R cells
Next, we want to investigate the role of FHIT in radiosensitivity. OML1-R cells were transfected with an FHIT-overexpressing cDNA myc-tag plasmid (Fig. 5A). Compared to control cells, OML1-R cells overexpressed with FHIT formed fewer colonies after a single fraction of 10-Gy irradiation, supporting a role for FHIT in radiosensitivity (Fig. 5B, C). Moreover, overexpression of FHIT significantly induced apoptosis in OML1-R cells after irradiation ( Fig. 5D and S3). Finally, FHIT overexpression restored phosphorylation of Chk2 (  together, these results suggest that FHIT may mediate radiosensitivity by inducing apoptosis and activation of a G2/M checkpoint.

Expression of FHIT is associated with radiosensitivity in oral cancer cells
The above experiments demonstrated that epigenetic silencing of FHIT may contribute to the radioresistance in OML1-R cells, while restoring FHIT expression facilitates radiosensitivity. To avoid the possibility of cell line effects, we investigated the radiosensitivity in another two oral cancer cell lines (SAS, SCC25), both of which endogenously underexpress FHIT (Fig. 4B). Interestingly, both SAS and SCC25 cells demonstrated an intrinsic radioresistance (Fig 6B, D, Control), while ectopic FHIT expression significantly increased the radiosensitivity in these two cells ( Fig. 6A-D).
We also performed a reciprocal experiment of knocking down FHIT in OSCL cells, which intrinsically highly express FHIT (Fig. 6E). Consistent with a role in radiosensitivity, FHIT knockdown slightly enhanced OSCL cell radioresistance (Fig. 6F). These results further suggest a role for FHIT in oral cancer radiosensitivity.

Epigenetic silencing of FHIT is correlated with poor clinical outcome in oral cancer patients
To further examine a clinical role for FHIT in oral cancer progression, we investigated its promoter methylation and expression in 40 match-paired, paraffinembedded oral cancer patient samples (Fig. S1, Table 1 and 2). Results from bisulphite pyrosequencing revealed a higher median methylation level in high-staged patients than that of low-staged patients (P = 0.0143; Fig. 7A). Importantly, patients with a negative staining of FHIT demonstrated a significantly higher promoter methylation of FHIT than that of patients with positive staining of FHIT (P = 0.0395, Fig. 7B). Moreover, patients with a higher FHIT methylation (>10%, n=22/40) showed a lower locoregional control rate than that of patients with lower methylation (P = 0.038; Fig. 7C). Remarkably, this predicting ability further translated into overall survival (P = 0.024; Fig. 7D). To exclude the effect of chemotherapy, we also performed a subgroup analysis in patients treated with post-operative radiotherapy alone (n=19). However, due to the small sample size, we only observed a trend that patients with higher FHIT methylation tend to have a 2-fold lower locoregional control rate than that of patients with lower methylation (survival%: high vs low methylation, 33.9% vs 66.7%, P = 0.0998). Moreover, consistent with methylation results, patients lacking FHIT expression also showed a lower locoregional control rate than that of patients expressing FHIT (P = 0.046; Fig. 7E, Fig. S5).
We further explored the outcome-predicting role of three independent factors (promoter methylation, stage, and very close surgical margin). Both Kaplan-Meier (Fig.  S6) and Cox regression analysis (Table S4) demonstrated that patients with any two of these three factors or all three factors had poorer clinical outcomes than those with only one or no risk factor.

Epigenetic interventions enhance radiosensitivity, in vivo
Next, we tested the role of FHIT in tumor growth and radiosensitivity in an in vivo xenograft mouse model. Compared to empty vector control, FHIT -overexpressing OML1-R cells exhibited significantly less tumor growth than mice injected with mock-transfected OML1-R cells (Fig. 8A, B), thus confirming high tumor suppression by FHIT, in vivo.
Since FHIT was epigenetically silenced in OML1-R cells, and radioresistance correlated with aberrant epigenetic changes in vitro, we examined the possibility of in vivo epigenetic interventions to resensitize OML1-R cells to irradiation. Mice were subcutaneously injected with OML1-R cells, followed by treatment of radiotherapy (RT, 4-Gy by 2 fractions) , i.p administration of 5-Aza or concurrent 5-Aza and RT. Significant reduction in tumor volumes was observed in mice treated with combined 5-Aza and RT when compared with 5-Aza or RT alone (Fig. 8C, D) demonstrating a synergistic effect and potential treatment direction of combining demethylating www.impactjournals.com/oncotarget agents with RT.

DISCUSSION
In the therapy of oral cancer, after massive irradiation, it is believed that radioresistant cancer cells harbored in the surgical bed are the likely sources of locoregional recurrence [27]. Moreover, post-irradiation epigenetic changes have been suggested to play a significant role in eliciting radioresistance of irradiated cancer cells [18,[28][29]. These phenomena make "epimutations" attractive bio-targets for both predicting clinical outcomes and enhancing treatment efficacy. However, the role of epigenetic changes in radioresistant oral cancer has not been fully explored.
In the present study, we used a microarray-based strategy to identify genes differentially methylated in an in vitro model of acquired oral cancer radioresistance. One particularly interesting microarray target, FHIT (fragile histidine triad), was found epigenetically silenced in radioresistant cells. Moreover, restoring expression of epigenetically repressed FHIT resensitized radioresistant oral cancer cells to ionizing radiation. Remarkably, epigenetic interventions using DNMT inhibitors significantly enhanced the radiosensitivity of resistant oral cancer cells, in vivo. Our results together with previous observations [30][31] suggest that concomitant treatment of DNA-demethylating agents and radiotherapy is a new treatment direction for this aggressive malignancy.    FHIT has been reported to be involved in several tumor-suppressive processes, such as enhancing apoptosis [32] and inhibiting tumor growth [33]. On the other hand, reduced FHIT expression has been correlated with increased genome instability [34], decreased DNA damage-induced cell killing [35], and solid tumor progression [36]. However, the role of FHIT in the radioresistance of human cancers is less explored. Herein, we observed a role of FHIT in initiating radioresistance in irradiated oral cancer cells. In vivo and human sample studies validated both therapeutic and predictive efficacy, warranting further clinical trials to confirm these clinical roles.
Ionizing radiation has been reported to induce ATM-Chk2-dependent checkpoint signaling followed by a G2/M cell cycle arrest [37][38][39], believed to drive sublethally damaged cancer cells toward apoptosis [40][41][42]. As a result, loss of pChk2 may facilitate cancer cells' escape from checkpoint-dependent apoptosis, leading to a high level of radioresistance [43][44]. Restoring FHIT expression has been observed to restore pChk2 activity and then re-radiosensitization of oral cancer cells [45]. Our data confirmed this observation and re-emphasized the role of reduced apoptosis in the development of radioresistance.
Epigenetic silencing of FHIT has been observed in several human cancers [46][47][48][49]. In this study, high promoter methylation of FHIT (>10%) can be observed in 55% (  methylation specific PCR (MSP), promoter methylation of FHIT was only observed in 1.8% of 53 cases of oral and oropharyngeal cancer [50]. This discrepancy may be due to a technical difference such that a quantitative bisulphite pyrosequencing was applied in our study. It is also interesting to point out that there is a group of patients with a relatively low methylation of FHIT (<10%) but also negative staining of FHIT (Fig 7B). Other mechanisms such as genetic alteration may play a role in the downregulation of FHIT in this group of oral cancer patients [51][52][53].
Nevertheless, the present study is the first to demonstrate that epigenetic silencing of FHIT can initiate radioresistance in irradiated human oral cancers cells. FHIT is located in the fragile site of chromosome 3p [19], likely a frequent site of DNA damage following irradiation. As a result, repressive histone marks and DNA methylation eventually accumulate within this region, as previously observed [54][55]. Thus, we may conclude that through recruitment of polycomb repressors (e.g., EZH2) and DNMTs, repeated radiation-induced DNA damages results in epigenetic silencing of FHIT, likely initiating post-irradiation radioresistance.
A more systematic assessment of radiation enhancement by epigenetic interventions has been conducted in head-and-neck cancer cells [31]. However, the role of promoter hypermethylation of FHIT may be underestimated. Based on our data, promoter hypermethylation of FHIT demonstrated dual roles not only in predicting clinical outcomes, but also implicating a treatment target. Thus, further clinical trials to investigate a combined effect of radiotherapy and epigenetic interventions are justified, especially in targeting restoration of FHIT [65].
Unexpected close surgical margin after a radical surgery is frequently encountered in resected oral cancer patients [66][67]. In addition to surgeon effects, unexpected close surgical margins have been recognized partially as an aggressive cancer behavior [68][69]. That is, bioaggressive tumors harboring highly migratory cancer cells demonstrate a higher incidence of unexpected close surgical margins than that of bio-indolent tumors [16]. Clinical data supported this biological reasoning. Patients with close surgical margins demonstrate a higher risk of cancer recurrence and poorer survival than patients with free surgical margin [67,[70][71][72]. However, close margin alone does not self-sufficiently guide post-operative adjuvants [1,66,[73][74], this clinical discrepancy reveals a possibly mixed population in these patients. Therefore, exploring a predictive biomarker is valuable for further patient stratification. We have previously confirmed this point of view by identifying promoter hypermethylation of DAPK as a predictive factor for locoregional control [16], and here, we further explored another potential prognostic marker, i,e., promoter hypermethylation of FHIT, as a useful predictor of both locoregional control and overall survival, in irradiated oral cancer patients.
In conclusion, we have identified FHIT as epigenetically silenced by DNA methylation and histone modification in radioresistant oral cancer cells. Epigenetically or ectopically restoring FHIT suppresses tumor growth and enhances radiosensitivity in oral cancer cells. In conjunction with pathological factors, FHIT promoter hypermethylation demonstrates a promising role in predicting clinical outcomes, stratifying high-risk patients, and implicating new potential treatment targets. The radiosensitizing effect of DNMT and/or EZH2 inhibitors warrants further investigation in randomized clinical trials.

Ethical considerations
All experiments involving human samples were conducted in accordance with the Helsinki Declaration of 1975, as revised in 2000. This study was also approved by the Institution Review Board (IRB) of the Buddhist Dalin Tzu Chi Hospital, Chia-Yi, Taiwan (approval number: B09804009). All animal experiments were approved by the Animal Experimentation Ethics Committee of the National Chung Cheng University, Taiwan.

Patient samples
Between Aug 2004 and Dec 2008, we retrospectively selected 40 match-paired oral cancer patient samples based on the distance of surgical margin (Table 1 and Supplementary Fig. S1). Pathological features were prospectively defined at the time of radical surgery by using a checklist, and all of pathological reports were confirmed by two independent pathologists according to the requirement of Taiwan Medical Accreditation on Cancer Center [16]. A cancer case manager and a radiation oncologist independently reviewed all patient records and data discrepancies were resolved by consensus. Locoregional control and overall survival were defined as study end points, as previously described [67,75]. www.impactjournals.com/oncotarget Cancer staging was defined according to the American Joint Committee on Cancer, the 6 th edition [76]. For DNA extraction, paraffin-embedded tissue blocks were re-sectioned according to our IRB protocol. Tumorburden areas that contained >70% cancer cells were microdissected under microscopy.

Establishment of a radioresistant subline by using fractionated irradiations
For studying the role of epigenetics in oral cancer radioresistance, we established a radioresistant subline of the oral cancer cell line OML1, using hypo-fractionated irradiations (see schema in Fig. 1A). Briefly, one day before irradiation, 1 x 10 6 of the parental ("OML1-P") cells were seeded into 10-cm cell culture plates. A fraction of 5-Gy irradiation was then delivered by a 6-MV linear accelerator (Elekta, Sweden). For achieving equivalent dose distributions, two 1.5-cm biomaterial-equivalent boluses were applied on both up-and down-side of the culture plate (i.e., arranged like a sandwich). After irradiation, cells were allowed to recover for 5-7 days. These procedures were then repeated for 10 sessions, up to a total dose of 50 Gy, to obtain the radioresistant cell subline OML1-R.

Microarray-based methylation analysis
Bisulphite-modified DNA was subjected to methylation analysis using an Illumina Infinium Human Methylation27 microarray (Illumina, San Diego, CA). Duplicate experiments were then performed to determine methylation β-values of specific probes, defined as the ratio of the methylated signal intensity to the sum of methylated and unmethylated signal of a probe. A β-value of 0 represented un-methylation while 1.0 represented full methylation. For selection of differentially methylated probes, the following criteria were applied: 1) probes were present in "CpG islands"; 2) probes having a mean β-value of <0.5 (for hypermethylated probes) or >0.5 (for hypomethylated probes) in the control groups (OML1-P cells, as compared to the radioresistant daughter cells); and 3) when comparing mean β-values between experimental groups, probes with both P <0.05 and differential changes of β-value ≥0.2 were obtained.

In vivo tumorigenicity assay
Four 3-week-old athymic nude mice (BALB/cAnN. Cg-Foxn1nu/CrlNarl) were obtained from National Laboratory Animal Center, Taiwan. All mice were kept under specific pathogen-free conditions using laminar airflow racks, with free access to sterilized food and autoclaved water. All experiments were performed under license from the Animal Experimentation Ethics Committee of the National Chung Cheng University. 1 x 10 6 OML1-R cells stably transfected with empty vector (control) or pcDNA3.1/FHIT were re-suspended in 0.1ml 1:1 PBS/ Matrigel (BD Bioscience, San Jose, CA) mixture. The cell suspension was then injected subcutaneously into the flank of each mouse (day 0). Tumor length (L) and width (W) were measured daily with calipers, and tumor volume calculated using the formula (L x W 2 /2). At the end of each experiment, all mice were sacrificed. To test the radiosensitizing effect of 5-Aza, 1 x 10 6 OML1-R cells were injected subcutaneously into the flank of each mouse. When the tumor volume exceeded 500mm 3 , two fractions of 4-Gy irradiation with 5-Aza (at a dose of 0.5mg/kg dissolved in 0.01M DMSO) or vehicle only (DMSO) were delivered IP, as shown in Fig. 6C. Tumor volume was monitored and presented as the tumor volume growth www.impactjournals.com/oncotarget ratios (final volume / initial volume).

Statistical analysis and definitions
SPSS (version 12.0; SPSS Inc., Chicago, IL, USA) was used for statistical analyses, as follows. Kaplan-Meier analysis was used to estimate survival and cancer control rates; the log-rank test was applied to assess curve differences between groups; and Cox proportional hazard regression was used to perform univariate and multivariate analyses for time-to-event endpoints. All time-to-event analyses calculated the time interval from the day of pathological diagnosis to the day of corresponding end events, as previously described [67,75]. A DNA methylation level at 10%, which is the average methylation level in OML1-P cells (Fig. 2C), was used as a cut-off. Independent variables that achieved a statistical significance (P < 0.05) or a statistical trend (P < 0.1 but ≥ 0.05) in univariate analysis were used in multivariate analysis. Comparison of grouped quantitative data was conducted by using student t test. For demarcating the effective size, 95% confidence (95% CI) intervals were provided in conjunction with point-estimated hazard ratios (HRs), in addition to a conventional P value. All tests were two-tailed and considered to be statistically significant when P <0.05.
Additional Methods can be found in supplementary information.