5-hydroxymethylcytosine loss is associated with poor prognosis for patients with WHO grade II diffuse astrocytomas

Currently, the reliable prognostic biomarkers for WHO grade II diffuse astrocytomas (DA) are still limited. We investigated the relations between the level of 5-Hydroxymethylcytosine (5hmC), an oxidated production of 5-methylcytosine (5mC) by the ten eleven translocated (TET) enzymes, and clinicopathological features of glioma patients. With an identified anti-5hmC antibody, we performed immunohistochemistry in 287 glioma cases. We detected that 5hmC variably reduced in most gliomas and 5hmC reduction was closely associated with higher pathological grades and shortened survival of glioma patients. In multivariate analysis, 5hmC had no independent prognostic value in the entire patient cohort. However, multivariate analysis within subtypes of gliomas revealed that 5hmC was still a prognostic marker confined to DA. In addition, we detected that IDH1 mutation by DNA sequencing was associated with favorable survival within DA. Lastly, we detected that the combination of 5hmC/KI67 was a useful prognostic marker for restratification of DA.

Human gliomas are a heterogeneous group of tumors and traditionally classified into various subtypes and grades mostly based on their microscopic characteristics for therapeutic decision-making 1,2 . However, histopathological criteria usually unavoidablely cause subjectively diagnostic interobserver variability 3,4 . Moveover, classification based on microscopic characteristics rather than molecular pathogenesis of gliomas limits the adequate assessment of prognosis and appropriate planning of treatment. For these regards, "ISN-Haarlem" guidelines recently proposed to define diagnostic entities as narrowly as possible and to include applicable molecular data to come up with a more objective and reproducible "integrated diagnosis" for glioma classification 5 . For example, molecular biomarkers isocitrate dehydrogenase (IDH1/IDH2) mutation and 1p/19q codeletion were proposed to resolve oligoastrocytoma as either oligodendroglioma or astrocytoma 6 . In addition, IDH1/IDH2 mutation, 1p/19q codeletion, TP53 mutation and MGMT promoter methylation were used for prognostic modeling and stratification into molecularly determined treatment groups 5,[7][8][9][10] . However, some questions remain ambiguous. For instance, within WHO grade II diffuse astrocytomas (DA) the prognostic relevance of the molecular markers has remained debate [10][11][12][13][14][15][16][17][18] (Supplementary Tables S1 and S2). Therefore, more reliable molecular markers for predicting the course of disease and outcome of gliomas are still needed.
DNA methylation at the 5-carbon position of cytosine (5mC) is the most extensively studied epigenetic modification in human cancer 19 . In 2009, breakthrough studies indicated that 5mC can be converted to 5-hydroxymethylcytosine (5hmC) by the ten eleven translocated (TET) enzymes 20,21 . HPLC-MS analysis and immunohistochemistry revealed that 5hmC is present with highest level in central nervous system 22 . Subsequent studies indicated that 5hmC is not merely serving as an intermediate of DNA demethylation, but also acts as a stable epigenetic marker 23 . Meanwhile, abundant evidence detected that 5hmC globally decreased in most human malignancies, including gliomas [24][25][26][27][28][29][30][31] . Initially, 5hmC loss in gliomas was proposed to be related with IDH1/IDH2 mutations 26 . However, subsequently numerous trials from larger clinical samples argued against this claim 25,[29][30][31] . It was interesting that 5hmC loss were suggested to be prognostic for malignant gliomas (World Health Organization grade III or IV) 29 . Due to small sample and lack detailed information about management and adjuvant treatment in this study 29 , much more work needs to verify the prognostic value of 5hmC in gliomas. Here, we performed immunohistochemical investigation in 287 glioma cases with a well identified homemade anti-5hmC antibody. The results showed that 5hmC was an prognostic marker confined to DA but not grade III or IV glioma patients. Moreover, we detected that IDH1 mutation by DNA sequencing and the combination of 5hmC/KI67 was associated with prognosis of DA respectively.

Identification of anti-5hmC antibody.
To evaluate specificity of anti-5hmC antibody generated by our lab, we firstly performed dot-blot analysis. The result showed that the rabbit polyclonal anti-5hmC antibody specifically recognized 5hmC instead of other bases (Fig. 1A). West-blot and immunoprecipitation also confirmed this result (Fig. 1B-D). Immunofluorescence analysis showed strong 5hmC staining was in the nucleus of Tet1 transfected cells. Concomitantly, the level of 5mC decreased in Tet1 transfected cells (Fig. 1E). Therefore, these results strongly demonstrated the anti-5hmC antibody had high specificity to recognize 5hmC.
Relations between 5hmC reduction and clinicopathological features. 5hmC level was analyzed by immunohistochemistry (IHC) in 287 glioma cases. In glioma tissues some samples had strong staining while other had weak or no staining, and cells in the same sample also had different degrees of staining intensity ( Fig. 2A). To facilitate the analysis of immunohistochemical results, specific nuclear immunoreactivity was scored using a 9-point scale on the basis of the product of staining intensity (no staining = 0, weak staining = 1, moderate staining = 2, strong staining = 3), and staining extent (% of positive cells; < 5% = 0, 5%-30% = 1, 30%-60% = 2, > 60% = 3) 32 . To facilitate statistical analysis, we divided the samples into 2 groups according to staining scores. Group 1 had no or weak staining with the scores of 0 to 3. Group 2 had moderate and strong staining with the scores from 4 to 9. The relations between 5hmC level and clinicopathological features were summarized in Table 2. On Chi square analysis, the 5hmC reduction (scores 0-3) were significantly associated with the following variables: over age 40 (P < 0.001), high pathological grades (grade III and IV, P < 0.001), vital status (death, P < 0.001), lower KPS scores (< 80, P = 0.018) and larger tumor size (≥ 50 cm 3 , P = 0.015). Nonsignificant variables included gender (P = 0.422), extent of resection (P = 0.225) and adjuvant treatment (P = 0.544). The average scores in grade I, II, III, and IV for 5hmC were 4.167 ± 0.524, 4.839 ± 0.235, 2.362 ± 0.310 and 2.156 ± 0.310 respectively (Fig. 2B). Spearman correlation analysis showed there was a significantly inverse correlation between the pathological grades of gliomas and the 5hmC scores (r = − 0.407, P < 0.001).
5hmC level on patient survival in total glioma cases. To determine the relations between 5hmC level and the survival of the glioma patients, we firstly divided the samples into seven groups according to the 5hmC score 0, 1, 2, 3, 4, 6 and 9. Kaplan-Meier survival analysis revealed low 5hmC scores strongly correlated with poor prognosis while high 5hmC scores correlated with better survival ( Fig. 3A; χ 2 = 40.570, P < 0.001). Next, we divided glioma patients into two groups. The cases with scores of 0 to 3 were defined as group 1, while patients with scores from 4 to 9 group 2. On Kaplan-Meier survival analysis, 5hmC reduction group statistically significantly correlated with poorer survival of patients, while 5hmC positive group with the better prognosis (Fig. 3B, χ 2 = 31.109, P < 0.001). In addition, as previous reports 1,2 , higher pathological grade correlated with worse prognosis in total cases (Fig. 3C, χ 2 = 124.243, P < 0.001).
To further evaluate the prognostic quality of 5hmC level, univariate and multivariate Cox regression analyses were performed. Univariate analysis (Table 3) confirmed the association of low 5hmC level with shorter patient survival time (P < 0.001). Additionally, the following variables were also significantly associated with poorer overall survival: aged over 40 at diagnosis (P < 0.001), higher pathological grades (P < 0.001), worse KPS (P = 0.026), larger tumor size (P = 0.014), lack adjuvant treatment after surgery (P = 0.032) and subtotal resection (P = 0.049). Nonsignificant variables included gender (P = 0.068) and tumor locations (P = 0.155). However, when the data were analyzed in the multivariate Cox regression model (Table 3), low 5hmC level was not found to have independent prognostic power (P = 0.720), while higher pathological grades (P < 0.001), lack adjuvant treatment (P = 0.009) and extent of resection (P = 0.017) remained statistically significant. In an alternative multivariate model exclusive of pathological grades, a strong interaction was found between 5hmC level and pathological grades (Table 3). 5hmC level on patient survival in subtypes of gliomas. Given the significant interaction between pathological grades and 5hmC level, we next analyzed the relations between 5hmC level and the patient survival within subtypes of gliomas. Kaplan-Meier survival analysis revealed low 5hmC scores were strongly associated with unfavorable survival within DA (Fig. 3D, χ 2 = 14.788, P < 0.001) but not OG (Fig. 3E, χ 2 = 0.997, To compare the prognostic value of 5hmC with other well-known prognostic markers within DA, we examined the level of IDH1, p53 and KI67 by immunohistochemistry in these tumors (Fig. 4A). Available data for IDH1, P53 and KI67 were 77, 69 and 71 cases respectively. 52% (40 of 77) of DA cases showed IDH1-R132H positive immunostaining. Immunodetection of P53-positive or KI67-positive cell nuclei ranged up to 80% or 30% respectively. According to the 10% cutoff, P53-positive cases were detected in 46% (32 of 69) of DA patients. With cutoff as 4%, KI67-positive cases were 35% (25 of 71) of DA cases. We also examined the mutation status of IDH1/2 by direct DNA sequencing and detected 64% (49/76) of DA bear heterozygous IDH1 mutations (Fig. 4B). The predominant amino acid sequence alteration in IDH1 was R132H accounting for 98% (48/49) of the detected mutations. Only one case was found to bear IDH1-R132C mutation (Fig. 4B). We did not detect codon 172 of IDH2 mutation at present study (data not shown). Of 70 samples by both DNA sequencing and immunohistochemistry assays, 45 cases were identified as carrying IDH1 mutation by DNA sequencing. However, only 78% (35/45) DA with IDH1 mutation by DNA sequencing was positive for IDH1-R132H antibody. The three cases with positive immunoreactivity for IDH1-R132H antibody did not show IDH1/2 mutation by DNA sequencing.
The relations 5hmC level on IDH1 immunostaining signaling, IDH1 mutated status, P53 and KI67 label index and patient clinicopathological features were outlined in Table 4. Chi square analyses confirmed significant association between 5hmC level and vital status (P < 0.001). Additionally, although the significant correlation between 5hmC level and IDH1 mutation was reached (P = 0.036), most cases with low 5hmC scores were detected in the IDH1 wild type cases (61%, 30 of 49). We investigated the relationship between IDH1 mutation status and 5hmC levels. Surprisingly, we detected that 5hmC reduction seem to be associated with IDH1 wild type versus mutation cases (Fig. 4C). Thus, this result was consistent with previous conclusions that 5hmC reduction didn't result from IDH1 mutation 25,29,30 .
To further test the 5hmC prognostic value for DA patients, we performed univariate and multivariate analyses using a Cox proportional hazards model. Consistent with Kaplan-Meier and Chi square analysis, univariate and multivariate analysis verified low 5hmC level was associated with unfavorable survival of DA patients ( Table 5, P < 0.001,P = 0.013). In addition, univariate Cox regression analyses also revealed that DA patients with either positive immunoreactivity for IDH1-R132H antibody (P = 0.030) or IDH1 mutation by DNA sequencing (P = 0.001) had a better survival (Table 5). However, multivariate analyses revealed that IDH1 mutation by DNA sequencing (P = 0.021) versus IDH1 positive immunstaining (P = 0.254) had an prognostic value for DA patients ( Table 5). In addition, we found that neither P53 nor KI67 harbored prognostic power by uni-or multivariate analysis (Table 5).   KI67 (Fig. 5F, X 2 = 6.582, P = 0.086). Surprisingly, the group for best assessment of prognosis and restratification of DA was 5hmC/KI67 combinations (

Discussion
The majority of the DA exhibits a relatively good prognosis. However, some DA show unexpectedly aggressive clinical course leading to early patient death 10 . This creates a diagnostic dilemma for routine histopathology. The heterogeneous clinical courses of DA may be associated with different epigenetic and genetic abnormalities 33 . Therefore, molecular markers would be useful for the accurate restratification of these tumors and provide help for prognostication or therapeutic decision making. However, currently, available molecular markers are limited. Epigenetic modifications play crucial roles in normal development and frequently alter during carcinogenesis 33 . 5hmC was a new detected DNA modification and the knowledge about its roles in gliomas remained limited. With an identified homemade anti-5hmC antibody, we detected that 5hmC variably reduced in most gliomas and the reduction was closely associated with higher pathological grades. These findings were consistent with previous reports 24,25,[28][29][30] . In clinical practice, it still remains a challenge for pathologists to histologically define AA and DA based on morphological features of anaplasia, such as mitotic activity and microvascular proliferation, for poor interobserver agreement 34 . IDH1/2 mutation has been accepted as a favorable prognostic biomarker for gliomas 5 . However, IDH1/2 mutated status can't be used to differentiate AA from DA because both tumors have high rates of IDH1/2 mutation. In the present study, we detected 5hmC level decreased more severely in grades III/IV versus I/II gliomas (Fig. 2B). This means 5hmC loss may be associated with anaplastic progression and can be used to stratify low-grade gliomas (I/II) from anaplastic gliomas (III/IV). In addition, we also detected that 5hmC dramatically reduced in AA and GBM compared to DA (Fig. 2C). Therefore, 5hmC loss may be a usefully marker for differentiating AA from DA.
Apart from the finding that 5hmC loss was a marker for anaplastic progression of gliomas, we also detected that 5hmC reduction was closely associated with shortened survival of glioma patients. However, in multivariate analysis, 5hmC had no independent prognostic value in the entire patient cohort. Alterative multivariate model found that pathological grade was the major interacting factor (Table 3). Further analysis within subtypes of gliomas revealed that 5hmC was still a prognostic marker confined to diffuse astrocytomas WHO II (Fig. 3D, Table 5). Inconsistent with Orr et al. report 29 , we didn't find prognostic relevance between 5hmC level and AA or GBM (Fig. 3F, H). The discrepancy may be associated with sample size (only 12 cases of AA with prognostic data in Orr et al. study), evaluation of 5hmC level methods, patient races or choice of treatment. Although Orr et al. had 52 adult GBM for evaluating prognostic relevance with 5hmC level, the short survival time (mean 15 months) and most cases without detected 5hmC may limited the 5hmC prognostic value within GBM. In addition, only 10  cases with prognostic data in Orr's study may account for their failed detection of the relevance between 5hmC level and DA. Many studies have confirmed an association of IDH1 mutations with favorable outcome for patients with malignant gliomas (WHO III/IV) 7,35-37 . However, prognostic value of IDH1 mutations for low-grade gliomas (LGG, WHO II), especial DA, was subject to debate 11,14,17,38,39 (Supplementary Table S1 and S2). For example, in a largest series IDH1 was of no prognostic value for 360 patients suffering from LGG, in which 186 was oligodendroglial tumors and 174 astrocytic tumors 17 . A recent Cancer Genome Atlas Research showed IDH mutations had significant survival prediction in lower grade gliomas (WHO II and III), including oligodendroglioma, oligoastrocytoma and astrocytoma. However, the report didn't particularly analyze prognostic value of IDH1 mutations within DA (WHO II) 38 . Ahmadi et al. performed IDH1/2 mutation assays in a series of 100 DAs patients and detected no survival benefit of IDH1 mutations in these patients 11 . However, by immunohistochemistry, two groups recently independently found positive IDH1 staining still hold prognostic relevance with DA patients 14,39 . In the present study, both univariate and multivariate Cox regression analyses confirmed that IDH1 mutation by DNA sequencing was associated with better survival for DA patients (Table 5). Therefore, our results suggested that IDH1 mutation was still a favorable prognostic marker for DA.
Currently, IDH1, TP53 and KI67 are routinely used molecular markers for assisting prognostic decision 5,10,40-42 (Supplementary Table S1). To test the prognostic value of combination of 5hmC, IDH1, P53 or KI67, we stochastically divided patients into subgrokups with two molecular combinations in DA. It is interesting that the combination of 5hmC/KI67 harbored the best value for assessment of prognosis and restratification of DA. (Fig. 5, Table 6). The reasons for 5hmC/KI67 but not other combinations in best predicting DA patient survival are unknown. One possible reason is that 5hmC/KI67 is a combination of cell differentiation (5hmC) and proliferation (KI67) markers. Several lines demonstrated that differentiated cells instead of stem cells harbor high 5hmC level 24,29 . In addition, an inverse relationship between 5hmC levels and cell proliferation was detected in proliferating or differentiated cells 25 . Functionally, double-knockout of Tet1 and Tet2 resulted in reduced 5hmC level and delayed brain development 43 . Therefore, the complementary combination of 5hmC and KI67 may be reason for their best assessment of prognosis and restratification of DA. It is interesting that the emerging data for superior prognosis of ATRX/IDH co-mutant diffuse astrocytomas was recently addressed in some reports 38 . Further work needs to compare prognostic value between 5hmC/KI67 and ATRX/IDH within DA.
Previously, some reports suggested IDH1 mutations might account for 5hmC reduction in gliomas by means of the presumed role of 2-hydroxyglutarate as an inhibitor of TET oxidases 26 . However, this suggestion was challenged by other observations that 5hmC reduction not associated with IDH1 mutations 25,29,30 . In the present study, we detected a higher versus lower level of 5hmC in IDH1 mutated DA compared to IDH1 wild type tumors (Fig. 4C). Therefore, our results didn't support 5hmC reduction was associated with IDH1 mutations. This conclusion was further supported by the observation that 5hmC loss was associated with poorer prognosis (Fig. 3A,B,D), while IDH1 mutations correlated with better survival in DA (Table 5). Since 5hmC can be converted from 5mC by TET enzymes 20,21 , 5hmC loss may be associated with 5mC reduction in malignancies. However, Kraus et al. observed that 5hmC level was unrelated to 5mC values by isotope-based liquid chromatography mass spectrometry assays 28 . Therefore, 5hmC decrease in cancer cells may primarily result from the alterations of TET genes. Indeed, one of TET family genes, TET2, was detected with high mutation rates in some hematologic malignancies, which simultaneously had aberrant levels of 5hmC in their genomes 44,45 . Some other evidence showed that 5hmC loss was associated with decreased expression or nuclear exclusion of TETs proteins 29,30 . Much more work need to confirm the relations between TETs alteration and 5hmC reduction in gliomas.
Summarizing, our data suggested that the 5hmC level, IDH1 mutation and 5hmC/KI67 combination harbor the value for assessment of prognosis of DA. Some limitations existed in this study. These data are derived from an unselected single-center collective. The sample size for each entity was not large, the design was in retrospective, and choice of treatment was not standardized. Thus, the prognostic value of 5hmC level, IDH1 mutation and 5hmC/KI67 combination in DA needs further verify.  Production and examination of anti-5hmC antibody. To examine 5hmC level in glioma tissues, we produced polyclonal rabbit anti-5hmC antibody. The specificity of the anti-5hmC antibody was examined by dot blot, western blot, immunoprecipitation and immunofluorescence. In brief, each nucleoside was conjugated to ovalbumin and quantified by mass spectrometrical analysis. Equal conjugated bases were spotted onto nitrocellulose filter membrane and reacted with the 5hmC (1:1000) or 5mC (1:500, Calbiochem) antibody. For western blot, 5mC or 5hmC conjugated with ovalbumin was mixed with 293T cell lysates and separated by SDS-PAGE and immunoblotted. For immunoprecipitation assay, a 367-bp DNA fragment spanning 4605333-4605699 in mouse chromosome 10 was PCR-amplified with 5mC or 5hmC in place of cytodine. 8.7 ng PCR products was mixed with 5μ g sonicated genomic DNA of 293T cells. Then 2.5 μ g of 5hmC antibody was added to the denatured DNA for precipitation. IP-DNA was extracted for quantitative reverse transcriptase PCR or regular PCR. To test whether the antibody could be used for immunofluorescence analysis, we transfected cDNA encoding the Tet1 catalytic domain into 293T cells by mean of that Tet1 can convert 5mC to 5hmC. Subsequently, immunofluorescence was performed with fluorescence-conjugated secondary antibody (1:500; Santa Cruz) as described.
Immunohistochemistry (IHC) analysis. Immunohistochemistry for 5hmC was performed in tissue sections of glioma samples as described previously 32    blocked with goat serum to reduce nonspecific binding and then incubated with primary 5hmC (1:1000 dilution) antibody overnight at 4 °C. On the following day, the peroxidase-conjugated secondary antibody (1:500 dilution; Santa Cruz) was incubated for 1 hour at room temperature. Diaminobenzidine (DAB) substrate was used for detection and hematoxylin was used for counterstaining. The samples were then mounted for visualization. The cells with brown nuclei were considered positively stained. The level of 5hmC staining was accessed independently by 3 pathologists. Immunohistochemistries for IDH1-R132H (H09, Dianova, Hamburg, Germany; dilution 1:100), P53 (DO-7, Dako, Carpinteria,CA, USA; dilution 1:100) and KI67 (MIB-1, Dako, Glostrup, Denmark; dilution 1:50) were performed as described above exclusive the step of DNA denature by HCl. Each slide stained for IDH1-R132H, P53 and KI-67 was individually reviewed and scored by 3 independent observers. Microscopic areas with highest labeling intensity were chosen for calculation. The p53 or KI-67 labeling index (LI) was defined as the percentage of immunoreactive tumor cell nuclei. In each case either at least 1000 tumor cell nuclei were counted were examined. For statistical analysis, cutoff value of LI for P53 and KI67 was 10% and 4% respectively according to previous reports 41,42,46 . PCR amplification and genes sequencing. For IDH1 and IDH2 mutations assays, DNA extraction from formalin fixed paraffin embedded tissue was used. Tumor content of at least 80% was histologically determined for each sample used for DNA extraction. Nucleic acid extraction was performed by standard procedures. 150 ng of genomic DNA was used for PCR amplification in a total volume of 50 μ l. The primer sequences for IDH1 and IDH2 and PCR amplification conditions were described previously 47 . IDH1 codon 132 and IDH2 codon 172 were analyzed by direct sequencing. If results were ambiguous the IDH1 or IDH2 were amplified by use of a different set of primers as described previously 47 .

Groups Subgroups
Mean time (months) of OS (95% CI) P-value Statistical Analysis. We used SPSS 20.0 (SPSS Inc. Chicago, IL, USA) for the statistical analysis. The associations between 5hmC level and clinicopathological features were compared by using a Chi square test or Fisher's exact test. The nonparametric Spearman correlation was used to analyze the relationship between pathological grades of glioma and 5hmC levels. Overall survival (OS) was defined as the interval between the date of diagnosis and the date of death or the last known follow-up. OS data were censored at the date of last follow-up, if the patient was still alive. Kaplan-Meier survival analysis was used to present the relationship between patient survival and 5hmC levels. Survival differences were analyzed by the log-rank test. Univariate and Multivariate analyses were performed using a Cox proportional hazards model to identify independent prognostic factors. P-values were all 2-sided and used as significance threshold less than 0.05.