Global Histone Modification Patterns as Prognostic Markers to Classify Glioma Patients

Transcription

1 Research Article Global Histone Modification Patterns as Prognostic Markers to Classify Glioma Patients Cancer Epidemiology, Biomarkers & Prevention Bo-lin Liu 1, Jin-xiang Cheng 1, Xiang Zhang 1, Rui Wang 2, Wei Zhang 1, Hong Lin 1, Xian Xiao 1, Sang Cai 1, Xiao-yan Chen 1,3, and Hong Cheng 3 Abstract Background: An altered pattern of epigenetic modifications is central to the development and progression of various tumors. We studied epigenetic changes involving multiple modifications of histones to better predict prognosis of glioma patients. Methods: Immunohistochemistry was done to investigate global histone modification expression of histone 3 lysine 4 dimethylation (H3K4diMe), histone 4 arginine 3 monomethylation (H4R3monoMe), histone 4 lysine 20 trimethylation (H4K20triMe), and acetylation of histone 3 lysine 9 (H3K9Ac), histone 3 lysine 18 (H3K18Ac), histone 4 lysine 12 (H4K12Ac), and histone 4 lysine 16 (H4K16Ac) in resected tumor samples of 230 glioma patients. Data were analyzed using a recursive partitioning analysis (RPA). Results: RPA classified the patients into 10 distinct prognostic groups based on WHO grade, histology, and histone modifications: H3K9Ac (<88% or 88% tumor cells), H3K4diMe (<64% or 64% tumor cells), H3K18Ac (<74% or 74% tumor cells), and H4K20triMe (<75% or 75% tumor cells). The 10 groups were associated with significantly different progression-free (P < ) and overall survival (P < ). Cox proportional hazards models including age, sex, WHO grade, histology, extent of tumor resection, Karnofsky performance status score, and RPA groups retained age and RPA groups as the sole independent factors significantly influencing overall survival. For progression-free survival, RPA grouping was the only independent prognostic factor. Conclusions: Multiple histone modifications seem to have prognostic relevance in glioma. Impact: Further evaluation of histone modifications as prognostic markers of treatment and predictors of chemotherapy response using histone deacetylase inhibitors is warranted. Cancer Epidemiol Biomarkers Prev; 19(11); AACR. Introduction Approximately 22,500 new cases of primary brain tumors occur annually in the United States, accounting for 1.4% of all tumors and 2.3% of cancer-related deaths (1). Glioma represents >70% of all brain tumors, including low-grade glioma (WHO grade 1 and 2) and highgrade glioma (WHO grade 3 and 4; refs. 2, 3). Despite recent advances in surgery, radiotherapy, and chemotherapy, survival of glioma patients remains poor. The 5-year survival rates of low-grade glioma are 30% to 70% depending on histology (4), and the median survival Authors' Affiliations: 1 Department of Neurosurgery, Xijing Institute of Clinical Neuroscience, Xijing Hospital and Departments of 2 Health Statistics and 3 Pathology, Fourth Military Medical University, Xi'an, Shaanxi Province, People's Republic of China Note: B. Liu and J. Cheng contributed equally to this work. Corresponding Author: Xiang Zhang, Department of Neurosurgery, Xijing Institute of Clinical Neuroscience, Xijing Hospital, Fourth Military Medical University, West Changle Road, No. 127, Xi'an, Shaanxi Province , People's Republic of China. Phone: ; Fax: xzhang@fmmu.edu.cn doi: / EPI American Association for Cancer is only 12 to 15 months for the most frequent and malignant glioblastoma (5). Although WHO grade is the most important prognostic factor for glioma patients, there are broad differences within grades, which may be caused by different tumor-specific biological behavior (5, 6). Thus, the identification of biomarkers might help to assess more precisely the prognosis and to address more clearly the use of adjuvant therapy. Besides genetic alterations, epigenetic modifications are critical to the development and progression of cancer (7). Epigenetics was defined as heritable changes in gene expression that are not a result of any alteration in the DNA sequence (8), whose mechanisms include but are not limited to aberrant DNA methylation, histone modification, and micrornas. The best-known epigenetic marker is DNA methylation. Hypermethylation of the CpG island promoter can induce silencing of genes affecting the cell cycle, DNA repair, metabolism of carcinogens, cell-to-cell interaction, apoptosis, and angiogenesis, all of which may occur at different stages in cancer development and interact with genetic lesions (9, 10). In addition to DNA promoter hypermethylation, global alterations of histone modification patterns have the potential to affect the structure and integrity of the 2888 Cancer Epidemiol Biomarkers Prev; 19(11) November 2010

2 Histone Modifications Predict Glioma Prognosis genome and to disrupt normal patterns of gene expression, which may also contribute to carcinogenesis (9, 11). These modifications occur in different histone proteins, histone variants, and histone residues; involve different chemical groups; and have different degrees of methylation (11). Recently, the global pattern of histone modification was identified by different study groups as a predictor of the risk of recurrence in prostate cancer patients (12), a predictor of prognosis in non smallcelllungcancer patients (13), and a predictor of both the recurrence of cancer and independent prognostic factor in gastric adenocarcinoma patients (14). We studied the pattern of global histone modification as a potential prognostic marker in resected glioma tissues. Immunohistochemistry was done to evaluate the global histone modification expression level of histone 3 lysine 4 dimethylation (H3K4diMe), histone 4 arginine 3 monomethylation (H4R3monoMe), histone 4 lysine 20 trimethylation (H4K20triMe), and acetylation of histone 3 lysine 9 (H3K9Ac), histone 3 lysine 18 (H3K18Ac), histone 4 lysine 12 (H4K12Ac), and histone 4 lysine 16 (H4K16Ac; refs. 12, 15). The relationship between multiple histone modifications and patient prognosis was analyzed by a recursive partitioning analysis (RPA), with overall survival (OS) as the primary end point. Materials and Methods Patients Patients hospitalized at the Department of Neurosurgery of Xijing Hospital (Xi'an, People's Republic of China) between January 2003 and December 2007 were included in our study if they met the following criteria: (a) age older than 14 years, (b) with pathologically confirmed glioma (WHO grades 1-4), (c) underwent radical neurosurgery of primary tumor, (d) without other cancer or fatal diseases, (e) not lost to follow-up, and (f) informed consent was obtained. Eligible glioblastoma patients received stereotactic fractionated radiosurgery to a median total dose of 55 Gy and a median of three courses of i.v. carmustine (100 mg/m 2 ) given at 4-week intervals postoperatively; eligible patients with WHO grade 3 tumors (anaplastic astrocytoma, anaplastic oligoastrocytoma, and anaplastic oligodendroglioma) received stereotactic fractionated radiosurgery to a median total dose of 36 Gy postoperatively. Major clinical and biological characteristics of the qualified 230 patients are summarized in Table 1. The histopathologic features and WHO grade of the tumor specimens were determined according to the WHO 2007 classification (3). Local institutional review board approval was obtained to use archived material for research purposes. Immunohistochemistry Immunohistochemistry was done as previously described (13). Briefly, specific rabbit polyclonal antibodies were used to detect global expression of seven different histone modifications, including H3K4diMe, H4R3- monome, H4K20triMe, H3K9Ac, H3K18Ac, H4K12Ac, and H4K16Ac (Abcam), using standard two-step indirect immunohistochemical staining method. Following deparaffinization, rehydration, quench of endogenous peroxidase, antigen retrieval, and blockage of nonspecific protein binding, primary antibodies were applied overnight as follows: H3K4diMe, 1:200; H4R3- monome, 1:100; H4K20triMe, 1:400; H3K9Ac, 1:100; H3K18Ac, 1:200; H4K12Ac, 1:150; H4K16Ac,1:100. Detection was accomplished using the DAKO Envision System, followed by chromogen detection with diaminobenzidine, and then counterstaining with Harris' hematoxylin, dehydration, and mounting. Negative controls were identical whole-tissue section stained without the primary antibody. A semiquantitative assessment was done independently by two observers (B.L. and S.C.) blinded to the clinicopathologic information. Either intermediate or strong nuclear staining was considered positive, and, regardless of the staining intensity, the score of each whole-tissue section was based on the proportion of positive tumor cells (range, 0-100%); the score of each sample was the mean value of three continuous whole-tissue sections per antibody. Statistical analysis Follow-up data were updated in February The OS was calculated from the date of surgery to date of death as a result of any cause. Progression-free survival (PFS) was calculated from the date of surgery to date of documented tumor recurrence or further growth of residual tumor or death from any cause. Patients who were alive at the date of the last follow-up were censored on that date plus 1 day. Probability of survival was estimated using the Kaplan-Meier method. Differences in survival were tested by means of the log-rank test. To test whether variables differed across groups, the χ 2 test or Fisher's exact test was used according to test condition. A Cox proportional hazards model including age, sex, WHO grade, histology, extent of tumor resection, Karnofsky performance status (KPS) score, and RPA groups was done to establish independent factor(s) for survival. Statistical significance was defined as P < All of the tests are two-sided. Statistical analysis was done using the SPSS software package, version 16.0 (SPSS, Inc.). An RPA (16) was done to test how histone modifications might influence prognosis as previously described (13). Briefly, a classification tree is constructed to the procedure for categorizing samples based on a series of sequential decisions. The whole data set served as the root of the classification tree. At each stage, a single predictor is used to split remaining samples into two nodes. The most well-known prognostic factor (for glioma, which is WHO grade) was firstly used as predefined split points to divide the cohort, followed by histology, which is another well-accepted prognostic factor, and then the histone modifications were applied as predictors to split Cancer Epidemiol Biomarkers Prev; 19(11) November

3 Liu et al. Table 1. Clinical characteristics of study sample WHO grade n Sex (M/F) Histology/ WHO grade n Median age (y) KPS ( 80/<80) /4 A/ (17-38) 7/ /40 A/ (14-74) 54/18 5/5 OA/ (14-51) 8/2 O/ (28-44) 10/ /21 AA/ (18-71) 24/18 AOA/ (21-68) 6/0 AO/ (30-37) 4/ /38 pgbm/ (18-72) 25/21 sgbm/ (22-68) 19/9 (Continued on the following page) remaining samples into additional nodes. The cutoff value is selected to maximize the likelihood ratio χ 2 statistic for the test so as to make samples above and below the cutoff value as different as possible with respect to classification. The cutoff values were calculated by a SAS program, which could run continuous likelihood ratio χ 2 tests until the most significant P value (which is the equivalent of the largest χ 2 statistic) occurred, when the patients' survival of the two nodes reached the maximal difference (confirmed by continuous log-rank tests). In the current study, when histone modifications were applied as predictors, all seven histone modifications and the expression level (from 0% to 100%) of them were substituted in the SAS program to determine the most significant one and its expression level. This splitting procedure continues until the data in each node are sufficiently well discriminated (which means when applying further continuous likelihood ratio χ 2 tests, a P < 0.05 could never be achieved) or until there are too few data in any node to support additional analysis. The RPA was carried out by the SAS software package, version 9.0 (SAS Institute, Inc.). Results Immunohistochemical study of H3K4diMe, H4R3monoMe, H4K20triMe, H3K9Ac, H3K18Ac, H4K12Ac, and H4K16Ac There were 230 patients who met the inclusion criteria and thus were included in this study. Major clinical and biological characteristics of these patients are summarized in Table 1. The expressions of H3K4diMe, H4R3- monome, H4K20triMe, H3K9Ac, H3K18Ac, H4K12Ac, and H4K16Ac in gliomas were nuclear rather than diffuse, sometimes with a low level of heterogeneity (Fig. 1A-G). The number of tumor cells stained positive by anti-h3k4dime, anti-h4r3monome, anti-h4k20trime, anti-h3k9ac, anti-h3k18ac, anti-h4k12ac, and anti- H4K16Ac ranged from 0% to 100% (mean, 60%, 61%, 70%, 70%, 76%, 62%, and 64%, respectively; Fig. 1H-N). The overall median expression of H3K4diMe, H4R3- monome, H4K20triMe, H3K9Ac, H3K18Ac, H4K12Ac, and H4K16Ac was 63%, 60%, 75%, 74%, 78%, 65%, and 68%, respectively. For each single tumor sample, the changes in expressions of H3K4diMe, H4R3monoMe, H3K9Ac, H3K18Ac, H4K12Ac, and H4K16Ac were all correlated with each other (Pearson's correlation coefficient, r ranged from to 0.528; P 0.028), with the exception of H4K20triMe when compared respectively with the former six modifications. H3K4diMe staining (Spearman's correlation coefficient, r = 0.328; P = ), H4R3monoMe (r = 0.193; P = 0.039), H3K9Ac (r = 0.200; P = 0.032), and H4K12Ac (r = 0.314; P = ) had low positive correlations with increasing WHO grades. No additional significant relationship was observed between the modifications and patient clinical characteristics. Global histone modification patterns and survival The median follow-up was 22.1 months. In the panel of 230 patients, 138 patients were dead at the time of this analysis, and 92 were alive, including 17 patients alive with tumor recurrence. The median OS and PFS of the whole population were 21.9 months (95% CI, months) and 19.4 months (95% CI, months), respectively. The survival data of different WHO grades and histologies are summarized in Table 1. RPA classified patients into 10 distinct prognostic groups (Fig. 2; Table 2). The first prognostic node was based on tumor grade, with those with low-grade glioma (WHO grades 1 and 2) having a longer survival than those with high-grade glioma (WHO grades 3 and 4). Patients with low-grade glioma were divided into two additional prognostic groups based on histology. Lowgrade glioma patients with oligodendroglioma or oligoastrocytoma had a better prognosis than those with astrocytoma. No other variable could segregate patients with oligodendroglioma or oligoastrocytoma (group 1). In contrast, patients with astrocytoma were classified into two separate groups by a third node based on acetylation 2890 Cancer Epidemiol Biomarkers Prev; 19(11) November 2010 Cancer Epidemiology, Biomarkers & Prevention

4 Histone Modifications Predict Glioma Prognosis Table 1. Clinical characteristics of study sample (Cont'd) Extent of tumor resection (total/subtotal/partial) Median PFS (mo) No tumor regrowth Median OS (mo) Alive at LO 4/0/ ( ) 6/ ( ) 6/10 56/8/ ( ) 44/ ( ) 50/72 8/2/ ( ) 10/ ( ) 10/10 10/0/ ( ) 10/ ( ) 10/12 27/6/ ( ) 16/ ( ) 6/42* 5/1/ ( ) 4/ ( ) 4/6 4/0/ ( ) 4/ ( ) 2/4 29/15/2 9.3 ( ) 12/ ( ) 2/46 21/1/6 6.9 ( ) 12/ ( ) 2/28* Abbreviations:M,male;F,female;LO,lastobservation;A,astrocytoma;OA,oligoastrocytoma;O,oligodendroglioma;AA, anaplastic astrocytoma; AOA, anaplastic oligoastrocytoma; AO, anaplastic oligodendroglioma; pgbm, primary glioblastoma; sgbm, secondary glioblastoma. *Death of one patient not tumor related. of H3K9. Patients whose tumors expressed H3K9Ac in <88% of tumor cells (group 3) had a worse survival compared with patients whose tumors had at least 88% of cells expressing H3K9Ac (group 2). Patients with high-grade glioma were further divided into WHO grade 3 and 4 groups by a second node of WHOgrade.WithinWHOgrade3patients,RPAprovided a third node on the basis of histology. Again, patients with oligodendrocytic components (anaplastic oligodendroglioma and anaplastic oligoastrocytoma), which had a better prognosis, could not be segregated (group 4), whereas patients with anaplastic astrocytoma were split into two groups by a fourth node based on dimethylation of H3K4. Patients whose tumors expressed H3K4diMe in <64% of tumor cells (group 6) had a worse survival compared with patients whose tumors had at least 64% of cells expressing H3K4diMe (group 5). RPA also provided WHO grade 4 patients with a third node, which divided them into primary and secondary glioblastoma groups, with the former having a better prognosis. Furthermore, RPA incorporated histone modifications as a fourth node for WHO grade 4 patients. Acetylation of H3K18 significantly influenced survival of primary glioblastoma patients, with a greater survival for patients whose tumor expressed lower levels (<74% of tumor cells) of H3K18Ac (group 7 versus group 8). Meanwhile, trimethylation of H4K20 significantly influenced survival of secondary glioblastoma patients, with a greater survival for patients whose tumor expressed higher levels ( 75% of tumor cells) of H4K20triMe (group 9 versus group 10). The 10 groups defined the terminal classification of the 230 patients and were associated with significant different PFS (P < ) and OS (P < ; Table 2; Fig. 3). Cox proportional hazards models including age, sex, WHO grade, histology, extent of tumor resection, KPS, and RPA groups were analyzed to evaluate independent factors for OS and PFS, respectively. Sex, WHO grade, histology, extent of tumor resection, and KPS were not retained by the model. Conversely, age and RPA groups were found to influence OS significantly and independently. For PFS, age was no longer retained, with RPA groups being the only independent prognostic factor (Table 3). Discussion An altered pattern of epigenetic modifications is central to many common human diseases, including cancer (7,9,11).Epigeneticmodificationsmayaffectseveral steps of tumor cell biology, and may have a prognostic value in several cancer patients such as prostate cancer, breast cancer, and leukemia (17). Glioma tumor tissues typically contain both neoplastic and stromal tissues, which lead to their histologic heterogeneity and thus the discrepancies between pathologic grades and clinical outcomes. Genetic and epigenetic studies potentially allow for better classification of these tumors and separation of the tumors into different prognostic groups. The contribution of epigenetic changes to glioma carcinogenesis has been broadly studied in terms of aberrant promoter methylation induced gene silencing (9, 11, 18). Besides glioma, aberrant DNA methylation has been investigated in carcinogenesis and cancer development in various types of solid cancers and hematologic malignancies, and the potential targeting of DNA methylation as a therapeutic approach for cancer treatment is being evaluated (7, 9, 11). In comparison with DNA methylation, other mechanisms of epigenetic regulation, such as multiple modifications of histones, are less well characterized, especially for clinical purposes. Seligson et al. (12) first reported the global pattern of histone modifications, which may serve as predictors of clinical outcome that lower levels Cancer Epidemiol Biomarkers Prev; 19(11) November

5 Liu et al. Figure 1. Immunohistochemical analysis of glioma samples. Representative examples of positively stained sections (original magnification, 40) for H3K4diMe (A), H4R3monoMe (B), H4K20triMe (C), H3K9Ac (D), H3K18Ac (E), H4K12Ac (F), and H4K16Ac (G) from different patients. Distribution of staining for H3K4diMe (H), H4R3monoMe (I), H4K20triMe (J), H3K9Ac (K), H3K18Ac (L), H4K12Ac (M), and H4K16Ac (N) antibodies across all the 230 tumor samples. The Y axis represents the number of samples showing positive staining for the indicated percentage of cells stained (X axis), scored under high-powered field Cancer Epidemiol Biomarkers Prev; 19(11) November 2010 Cancer Epidemiology, Biomarkers & Prevention

6 Histone Modifications Predict Glioma Prognosis Figure 2. RPA results individualizing 10 different prognostic groups among the 230 samples from glioma patients who underwent resection included in our study. Each node, where the branches of the RPA tree bifurcate, divides patients according to whether the value of a specific feature (predictor) is above or below a selected cutoff value. The first node is represented by the tumor grade. In low-grade glioma patients, histologic subtype provides the second node, and histone modifications (e.g., the percentage of cells stained positively for H3K9Ac) provide the third node. High-grade glioma patients were further divided into WHO grade 3 and 4 ones. Histologic subtype and pathogenesis provide the third node, respectively, and histone modifications of H3K4diMe, H3K18Ac, or H4K20triMe provide the fourth node. of H3K18Ac, H3K9Ac, H3K4diMe, H4K12Ac, and H4R3diMe predicted higher risk of recurrence in prostate cancer patients by an unsupervised Random Forest clustering. Later on, Barlesi et al. (13) used RPA to classify resected non small cell lung cancer patients into seven distinct prognostic groups, which were marked with different tumor-node-metastasis stage, histology, and global histone modification patterns. Park et al. (14) further showed that some specific modifications could be predictors for the risk of recurrence and prognosis in Table 2. Median PFS and OS of the different groups of glioma patients established by RPA Group n Median PFS, mo (95% CI)* Median OS, mo (95% CI)* ( ) 52.4 ( ) ( ) 57.6 ( ) ( ) 32.0 ( ) ( ) 28.5 ( ) ( ) 18.6 ( ) ( ) 11.3 ( ) ( ) 14 ( ) ( ) 10.1 ( ) ( ) 9.7 ( ) ( ) 6.5 ( ) *P < Mean survival. Cancer Epidemiol Biomarkers Prev; 19(11) November

7 Liu et al. Figure 3. PFS (A) and (B) OS of the 10 different groups of glioma patients who underwent resection established by RPA. gastric adenocarcinoma patients. In our study, the pattern of expression and biological relevance of several histone modifications were analyzed for the first time in glioma patients who had undergone resection. To assess the potential prognostic use of histone modification profiles in a more widespread manner, we evaluated five modifications of H3K9Ac, H3K18Ac, H3K4diMe, H4K12Ac, and H4R3monoMe almost the same as in Seligson et al.'s study (12), and modifications of H4K16Ac and H4K20triMe, which are the two most common histone modification changes in human neoplasia (9, 15). The results suggested that histone H3K4diMe, H4K20triMe, H3K9Ac, and H3K18Ac may have an effect on OS and PFS of glioma patients, implying that more attention should be paid to epigenetic modifications in glioma carcinogenesis, especially global histone modification patterns. However, in comparison with previous reports that found that lower levels of H3K18Ac in prostate, lung, kidney, and breast cancer patients predict poor survival (12, 19, 20), we had reached an opposite finding that lower levels of H3K18Ac in glioblastoma patients predict better survival, which is in accordance with Tzao et al.'s results in esophageal carcinoma patients (21). This interesting phenomenon indicated that even if the same histone modification profiles could predict opposite prognosis in different cancer types, and that histone modification patterns may possess a tissue- and/or time-specific feature of heterogeneity. In addition, histone modifications are posttranslational covalent modifications (22); therefore, like most previous studies evaluating histone modifications, we had only detected the expressions of the modifications in protein level by immunohistochemistry but had not checked the expressions of the specific histones that were modified in mrna level, nor the relationship between immunohistochemical expression and mrna expression. This should be improved in future studies because the mrna level regulation may play an important role in influencing both the quantity and function of the histones that would be modified posttranslationally. There are several methodologies to investigate the potential prognostic value of clinical and biological features, and we chose RPA for the same reasons indicated by Barlesi et al. (13), that RPA is suitable for analyzing variables intricately connected and that the procedure of RPA is concise and the result is easy to follow (23, 24). The validity of this method is supported by using WHO grade, a well-known prognostic factor in glioma patients, as the first node. Therefore, the histone modifications retained by RPA could be regarded as potentially valuable prognostic factors. However, how the expression of particular genes or tumor behavior is modulated by specific epigenetic alterations or combination of alterations remains unclear. One emerging model, called histone code hypothesis, assumes that specific combinations of histone modifications may confer the overall expression status of a region of chromatin (22), which should be further investigated in tumors characterized by different patterns of histone modifications. Additionally, one limitation of the RPA method is that it requires a single predictor at each node for splitting; therefore, the combination of two or more histone modifications could not severed as a predictor in the RPA procedure. However, as indicated in the results, several histone modifications were correlated, and in the current study, only the most significant ones were retained by RPA. In the future studies, more statistical strategies can be applied because that summation of the individual histone modifications may predict survival better than individual markers. Our efforts to identify multiple histone modifications in the tumors may not only provide additional prognostic information but also contribute to clinical decisions after further verification. Management decisions, such as timing of intervention, extent of surgical resection, 2894 Cancer Epidemiol Biomarkers Prev; 19(11) November 2010 Cancer Epidemiology, Biomarkers & Prevention

8 Histone Modifications Predict Glioma Prognosis timing and dose of radiotherapy, and long-term benefits versus risks of chemotherapy, have been controversial about the optimal therapy in gliomas (4, 5). Although WHO grade is the most important prognostic factor for glioma patients, the broad differences of treatment response and survival within grades suggest that the aggressiveness of treatment may not be decided simply by grade. Based on our results, histone modifications may be considered in the attempt to better define prognostic subgroups by further validation. An interesting observation that RPA group 7 (belongs to WHO grade 4) had a better survival than group 6 (WHO 3) had further highlighted the discrepancies within WHO grading, which has been noticed by many neuro-oncologists. For those with worse prognosis as indicated by epigenetic profiles, more aggressive adjuvant therapy could be considered. About chemotherapy, to date, the strongest evidence that epigenetic alterations can predict responses to therapy is provided by the methylation-associated silencing of the DNA repair protein MGMT in glioma. MGMT hypermethylation is the best independent predictor of glioma patients' response to alkylating agents (25), including temozolomide (26). Intriguingly, several drugs interacting with histone modifications have been developed in recent years, such as the histone deacetylase inhibitors (27). Several structurally distinct classes of histone deacetylase inhibitors have been developed and tested (28), of which a significant anticancer activity as well as good toleration were reported in several cancer types including glioma (29). The imbalance of histone acetylation in cancer cells provides an experimental basis for these treatments (12, 15), which suggests that the global histone modification patterns, when characterized in more detail, might provide insights into the specificity of these drugs. For instance, the loss of monoacetylated H4K16 can be reversed by a new class of drugs that inhibit sirtuins (30), the specific subclass of histone deacetylase inhibitors that deacetylate H4K16. Numerous additional studies are still ongoing and results are awaited. Based on the existing results (25, 26) including ours, we suggest that histone modifications should be further investigated in combination with DNA methylation as predictors of response to chemotherapy in the future, which may better help to select the most appropriate therapies on the basis of epigenetic profiles of tumors. Moreover, when evaluating new prognostic biomarkers in future studies, those well-accepted biomarkers, such as MGMT for glioblastoma and 1p19q for oligodendroglial tumors, should be incorporated to help validating the results. The detection of tumor molecular signatures in body fluids has implications for the identification of patients with high-risk or preinvasive/early-stage lesions and for monitoring residual disease. Aberrant DNA methylation has been studied in body fluids such as plasma/ serum, urine, sputum, bronchoalveolar lavage fluid, and ductal lavage fluid (11), whereas histone modifications have never been tested in body fluids yet. To this end, Table 3. Cox proportional hazards regressions for PFS and OS Variable PFS OS HR (95% CI) P HR (95% CI) P Group 1 Reference Reference ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) < ( ) < ( ) < ( ) ( ) < ( ) < ( ) < ( ) < ( ) < ( ) < ( ) < Sex: male vs female 0.88 ( ) ( ) Age 50 y 1.29 ( ) ( ) Extent of resection Total resection Reference Reference Subtotal resection 1.32 ( ) ( ) Partial resection 0.96 ( ) ( ) KPS < ( ) ( ) NOTE: Pathology and WHO grade were not retained by the model because of colinearity with RPA group variable (linearly dependent covariables). Cancer Epidemiol Biomarkers Prev; 19(11) November

9 Liu et al. we are planning to do a prospective study on epigenetic alteration status including gene promoter methylation and histone modification in a longitudinal way. In conclusion, our results highlight the prognostic value of epigenetic changes involving multiple histone modifications in glioma, which may help in identifying glioma patients with different prognosis and selecting patients for optimal adjuvant treatments. In addition, our observations provide a rationale to modify chemotherapy with drugs interacting with histone modifications, such as histone deacetylase inhibitors. Disclosure of Potential Conflicts of Interest No potential conflicts of interest were disclosed. Received 04/29/2010; revised 08/03/2010; accepted 08/05/2010; published OnlineFirst 09/14/2010. References 1. Jemal A, Siegel R, Ward E, et al. Cancer statistics, CA Cancer J Clin 2006;56: Ohgaki H. Epidemiology of brain tumors. Methods Mol Biol 2009; 472: Louis DN, Ohgaki H, Wiestler OD, Cavenee WK. World Health Organization classification of tumours of the central nervous system. Lyon: IARC Press; Duran I, Raizer JJ. Low-grade gliomas: management issues. Expert Rev Anticancer Ther 2007;7:S Wen PY, Kesari S. Malignant gliomas in adults. N Engl J Med 2008; 359: Schiff D, Brown PD, Giannini C. Outcome in adult low-grade glioma: the impact of prognostic factors and treatment. Neurology 2007;69: Lund AH, van Lohuizen M. Epigenetics and cancer. Genes Dev 2004; 18: Holliday R. The inheritance of epigenetic defects. Science 1987;238: Esteller M. Cancer epigenomics: DNA methylomes and histonemodification maps. Nat Rev Genet 2007;8: Herman JG, Baylin SB. Gene silencing in cancer in association with promoter hypermethylation. N Engl J Med 2003;349: Esteller M. Epigenetics in cancer. N Engl J Med 2008;358: Seligson DB, Horvath S, Shi T, et al. Global histone modification patterns predict risk of prostate cancer recurrence. Nature 2005; 435: Barlesi F, Giaccone G, Gallegos-Ruiz MI, et al. Global histone modifications predict prognosis of resected non small-cell lung cancer. J Clin Oncol 2007;25: Park YS, Jin MY, Kim YJ, Yook JH, Kim BS, Jang SJ. The global histone modification pattern correlates with cancer recurrence and overall survival in gastric adenocarcinoma. Ann Surg Oncol 2008; 15: Fraga MF, Ballestar E, Villar-Garea A, et al. Loss of acetylation at Lys16 and trimethylation at Lys20 of histone H4 is a common hallmark of human cancer. Nat Genet 2005;37: Zhang HP, Singer B. Partitioning in the Health Sciences. New York: Springer; Jones PA, Baylin SB. The fundamental role of epigenetic events in cancer. Nat Rev Genet 2002;3: Martinez R, Schackert G. Epigenetic aberrations in malignant gliomas: an open door leading to better understanding and treatment. Epigenetics 2007;2: Seligson DB, Horvath S, McBrian MA, et al. Global levels of histone modifications predict prognosis in different cancers. Am J Pathol 2009;174: Elsheikh SE, Green AR, Rakha EA, et al. Global histone modifications in breast cancer correlate with tumor phenotypes, prognostic factors, and patient outcome. Cancer Res 2009;69: Tzao C, Tung HJ, Jin JS, et al. Prognostic significance of global histone modifications in resected squamous cell carcinoma of the esophagus. Mod Pathol 2009;22: Wang Y, Fischle W, Cheung W, Jacobs S, Khorasanizadeh S, Allis CD. Beyond the double helix: writing and reading the histone code. Novartis Found Symp 2004;259:3 17; discussion 17 21, Young SS, Ge N. Recursive partitioning analysis of complex disease pharmacogenetic studies. I. Motivation and overview. Pharmacogenomics 2005;6: Zaykin DV, Young SS. Large recursive partitioning analysis of complex disease pharmacogenetic studies. II. Statistical considerations. Pharmacogenomics 2005;6: Esteller M, Garcia-Foncillas J, Andion E, et al. Inactivation of the DNA-repair gene MGMT and the clinical response of gliomas to alkylating agents. N Engl J Med 2000;343: Hegi ME, Diserens AC, Gorlia T, et al. MGMT gene silencing and benefit from temozolomide in glioblastoma. N Engl J Med 2005;352: Gilbert J, Gore SD, Herman JG, Carducci MA. The clinical application of targeting cancer through histone acetylation and hypomethylation. Clin Cancer Res 2004;10: Lindemann RK, Gabrielli B, Johnstone RW. Histone-deacetylase inhibitors for the treatment of cancer. Cell Cycle 2004;3: Lukas RV, Boire A, Nicholas MK. Emerging therapies for malignant glioma. Expert Rev Anticancer Ther 2007;7:S Pruitt K, Zinn RL, Ohm JE, et al. Inhibition of SIRT1 reactivates silenced cancer genes without loss of promoter DNA hypermethylation. PLoS Genet 2006;2:e Cancer Epidemiol Biomarkers Prev; 19(11) November 2010 Cancer Epidemiology, Biomarkers & Prevention

10 Global Histone Modification Patterns as Prognostic Markers to Classify Glioma Patients Bo-lin Liu, Jin-xiang Cheng, Xiang Zhang, et al. Cancer Epidemiol Biomarkers Prev 2010;19: Published OnlineFirst September 14, Updated version Access the most recent version of this article at: doi: / epi Cited articles Citing articles This article cites 28 articles, 6 of which you can access for free at: This article has been cited by 1 HighWire-hosted articles. Access the articles at: alerts Sign up to receive free -alerts related to this article or journal. Reprints and Subscriptions Permissions To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at pubs@aacr.org. To request permission to re-use all or part of this article, use this link Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.