5-Hydroxymethylcytosine correlates with epigenetic regulatory mutations, but may not have prognostic value in predicting survival in normal karyotype acute myeloid leukemia

Stem cells display remarkably high levels of 5-hydroxymethylcytosine (5hmC). Both TET2 and IDH1/2 mutations can impair the production of 5hmC, thus decreasing 5hmC levels. TET2 or IDH1/2 mutations are commonly observed in acute myeloid leukemia (AML). However, the implications of 5hmC on survival in normal karyotype AML patients have not been fully evaluated. The 5hmC levels were analyzed in 375 patients using ELISA. The levels of 5hmC in DNA samples were converted to a log scale for the analysis and correlations with TET2 and/or IDH1/2 mutations were evaluated. The median 5hmC level was 0.065% (range 0.001–0.999). Mutation rates were 13.1% for TET2mut, 6.7% for IDH1mut, and 13.9% for IDH2mut. The prevalence of TET2 and/or IDH1/2 was 33.1% (124/375). TET2 and IDH1/2 mutated patients had significantly lower levels of log(5hmC) compared with patients without TET2 or IDH1/2 mutations (p<0.001). With a median follow-up of 55.5 months (range, 0.7–179.8), there was no significant difference in overall survival, event-free survival, and relapse risk according to TET2mut or IDH1/2mut (all, p>0.05). To identify its prognostic value, we sub-classified the levels of 5hmC into tertiles for 5hmC values. However, there was no significant association between the categories of 5hmC levels and survival or relapse risk (all p>0.05). Patients with TET2 or IDH1/2 mutations had lower levels of 5hmC. The 5hmC levels may not be predictive of survival in patients with normal karyotype AML.


INTRODUCTION
DNA methylation regulates the expression of specific genes and therefore plays a critical role in development, contributing to normal cellular differentiation, genomic stability, X-chromosome inactivation, and genomic imprinting [1]. The balance between methylation and demethylation is controlled by several proteins and cofactors. This balance is frequently deregulated in cancer, leading to aberrant methylation patterns [2]. The regulation of DNA hydroxymethylation is mediated by several factors including proteins from the TET family, which is responsible for the formation of 5-hydroxymethylcytosine (5hmC) [3]. TET proteins require α-ketoglutarate (αKG) as a co-substrate, which is produced by the family of isocitrate dehydrogenase (IDH) proteins [4]. IDH proteins catalyze the oxidative decarboxylation of isocitrate to αKG, which is an intermediate step in the tricarboxylic acid cycle [5,6].
Several genes that influence hydroxymethylation are mutated in cancer. For example, mutations that disrupt TET expression or activity can affect the normal level of 5hmC. In addition, this level may be affected by changes in the expression of either genes coding for products that bind 5hmC or genes involved in demethylation [2].
5hmC may function as an intermediate in demethylation and is known to be a transcriptional activator [16]. However, the prognostic implications of 5hmC have not been comprehensively evaluated in patients with NK-AML, especially in those affected by mutations. Additionally, the clinical significance of 5hmC levels has not been fully evaluated in patients with NK-AML. Herein, we evaluated the correlation of 5hmC levels with mutant alleles, as well as the significance of 5hmC levels in the context of survival and relapse risk.

RESULTS 5hmC level and somatic mutations
The characteristics of these 375 patients are summarized in Table 1. The median 5hmC level was 0.065% (0.001-1.000). The levels of 5hmC deviated from the standard normal distribution and were therefore reanalyzed using a log scale.
The prevalence rates of other mutations were: FLT3-ITD pos , 27

5hmC level inversely correlates with adverse clinical factors of treatment outcomes
We analyzed the levels of 5hmC according to the clinical factors with adverse outcomes. Older age (r= −0.151, p=0.027), high WBC count (r= −0.195, p<0.001), high blast percentage in bone marrow (r= −0.129, p=0.013), and high peripheral blast counts (r= −0.183, p<0.001) were inversely correlated with a low 5hmC level on a log scale. The log(5hmC) levels were not significantly correlated with gender (p=0.372) or CR achievement (p=0.807). In the sub-classification of the 5hmC values into tertiles, the high 5hmC group was younger and had a lower WBC count than the low or intermediate group. The low 5hmC group had a higher bone marrow blast percent than the intermediate or high 5hmC group (Table 1). NPM1 mut , DNMT3A mut and ASXL1 mut were observed more frequently in the low 5hmC groups, while CEBPA dm was more frequent in the high 5hmC group (p <0.05) ( Table 1).
2) The p-values refer to comparisons between groups with TET2 wild vs. TET2 mut .
The results of multivariate analysis are shown in Table 3. The results of NPM1 mut and receiving allogeneic SCT were favorable, while FLT3-ITD pos and DNMT3A mut were poor risk factors for OS. The results of NPM1 mut , CEBPA dm , and receiving allogeneic SCT were favorable, while FLT3-ITD pos and DNMT3A mut were poor risk factors, for EFS. In addition, age (≥65 years), FLT3-ITD pos , and ASXL1 mut were significantly associated with a higher relapse risk, while NPM1 mut , CEBPA dm , and receiving allogeneic SCT reduced the relapse risk. There were no clinically significant differences in OS, EFS, and relapse risk according to the levels of 5hmC in multivariate analysis (all p>0.05).
We sub-analyzed the significance of levels of 5hmC according to each clinically significant mutational status. The levels of 5hmC did not influence the survival or relapse risk in each mutational status, even including that of TET2 or IDH1/2 (Supplementary Figure 2 and  Supplementary Table 3).

DISCUSSION
We examined the prognostic implications and clinical significance of changes to 5hmC levels as a result of TET2 or IDH1/2 mutations. The log(5hmC) levels were found to be inversely correlated with age, white blood cell (WBC) count, and the percentage of blasts in bone marrow. Patients with TET2 or IDH1/2 mutations had significantly lower levels of log(5hmC) than patients without any TET2 or IDH1/2 mutations. However, when we sub-categorized the levels of 5hmC into tertiles, we found that low, intermediate, or high levels did not influence the achievement of CR. Furthermore, relapse risk, EFS, and OS were not found to be significantly different among the patients of any sub-group.
TET2 and IDH1/2 genes are important for regulating DNA methylation. Disruptions to their normal functions (i.e. alterations to DNA methylation) have been observed in several cancers. For example, in another study, it was demonstrated that AML patients with TET2 or IDH 1/2 mutations displayed decreased levels of 5hmC [12,[17][18][19]. The altered patterns of covalent cytosine modifications point to the potential for novel diagnostic, prognostic, and therapeutic applications. The level of 2-hydroxyglutarate in leukemic cells of AML patients (at the time of diagnosis) is suggested as an excellent surrogate marker for IDH1/2 mutations [16]. TET2 haploinsufficiency may contribute to abnormal myeloid transformation [7]. TET2 mutation occurs early in leukemogenesis, suggesting its significance in the onset and progression of hematologic malignancies [20]. Given that the levels of 5hmC indicate  the presence of TET2 or IDH1/2 mutations, we speculate that 5hmC could also be a predictive marker for the prognosis and detection of TET2 or IDH1/2 mutations. The significance of 5hmC levels in AML is unclear. Kroeze et al. previously attempted to demonstrate the effect of 5hmC levels on AML [19]. Specifically, they showed that the levels of 5hmC during CR were normalized to those levels seen in healthy bone marrow and peripheral blood. This indicated that aberrant levels of 5hmC at diagnosis were an intrinsic property of leukemic cells [19]. Interestingly, they also demonstrated that high levels of 5hmC were associated with a poor prognosis (as per a multivariate analysis), along with variable results of the effects of TET2 or IDH1/2 mutations on survival. Since the group with high levels of 5hmC was small (8.2%, 17/206), the results cannot be generalized to all AML patients. In our cohort, there was no significant difference in survival according to the levels of 5hmC. We tried to sub-classify the 5hmC value using R partitioning to find the valuable cut-off range but, we did not find a significant cut-off value for levels of 5hmC in OS, EFS, and relapse risk. Therefore, we sub-classified the patients into low, intermediate, and high level tertile groups of 5hmC to determine clinical significance. The difference between our findings and those of previous studies can likely be explained by the higher proportion of patients >60 years of age (32.3%); furthermore, the effects of cytogenetic abnormalities on the levels of 5hmC were not considered in our study. The levels of 5hmC decline with aging, so the inclusion of older patients could influence the results of 5hmC [21]. We observed similar results in the NK-AML population with respect to the inverse correlation between log(5hmC) levels and age. WBC count and the percentage of blasts in bone marrow were also negatively correlated with log(5hmC) levels. In our previous study, TET2 mutations were associated with older age and a high WBC count. This suggests that patients with high WBC counts have more TET2 mutations, which might explain the correlation [13]. Our study shows that the 5hmC levels positively correlated with CEBPA dm , but negatively correlated with NPM1 mut , DNMT3A mut and ASXL1 mut . To our knowledge, no report has directly explained the relationship of 5hmC levels and CEBPA dm , NPM1 mut , DNMT3A mut , and ASXL1 mut . Previously, we showed that TET2 mut is mutually exclusive with CEBPA dm , but has a positive correlation with NPM1 mut [13]. DNMT3A mut and CEBPA dm are also mutually exclusive, whereas DNMT3A mut was positively correlated with NPM1 mut [22]. We speculate that mutations of epigenetic modifying genes influence the associated mutational status of the 5hmC groups.
Our results have significant clinical relevance because the study population was restricted, exclusively, to patients with NK-AML. Furthermore, all patients received induction chemotherapy, while any untreated patients were excluded. However, this study had a methodological limitation given that Sanger's fluorescent dideoxynucleotide chain termination sequencing analysis has a detection sensitivity of approximately 10% of mutant alleles. Recently, a novel technology for next-generation sequencing was developed. This technology recognizes unique sequences, provides depth of coverage and accuracy of sequencing [23], and offers a powerful tool that may become integral to resolving correlation clonal dynamics with levels of 5hmC. In this study, we did not demonstrate the prognostic significance of 5hmC levels. However, TET2 mutations have been shown to predict the response of patients with myelodysplastic syndrome to hypomethylating agents [24]. Hypomethylating agents are used in the standard treatment of elderly patients with AML who are not eligible for standard induction therapy [25,26]. The investigation of 5hmC levels in such a population could be a good prognostic marker for prediction of the response to hypomethylating agents and to tailor therapies and assess responses to anticancer drugs [12].
In summary, TET2 or IDH1/2 mutated patients had lower levels of 5hmC. Apart from affecting the methylation status of DNA, other processes may be influenced by altered levels of 5hmC in patients with NK-AML and TET2 wild and IDH1/2 wild . 5hmC may not have prognostic value for predicting survival or relapse risk in patients with NK-AML who have been treated with intensive induction therapy.

Patients and treatment
This study included patients diagnosed with NK-AML at seven participating institutions between October 1998 and September 2012. Out of 407 patients screened for this study, 375 DNA samples were available to evaluate the levels of 5hmC [13]. The median patient age was 52 (range 15-83) years and the subjects included 190 males (Supplementary Table 1). Most cases were AML not otherwise specified (88.8%), with some therapy-relapsed AML (2.9%) or AML with myelodysplasia-related changes (8.3%). Patients had received induction chemotherapy using a standard protocol [a 3-day course of anthracycline with a simultaneous 7-day course of cytosine arabinoside (Ara-C) or N 4 -behenoyl-1-b-d-arabinofuranosylcytosine (BHAC)] [13]. Idarubicin was administered daily at a dose of 12 mg/m 2 or daunorubicin was administered at a dose of 60 mg/m 2 on three consecutive days. Ara-C was administered daily at a dose of 100 mg/m 2 and BHAC at a dose of 300 mg/m 2 on seven consecutive days. In all, 210 patients were treated with idarubicin + Ara-C, 71 patients were treated with idarubicin + BHAC, and 94 patients received daunorubicin + Ara-C induction chemotherapy. Of 375 patients, 260 (69.3%) achieved complete remission (CR) after first induction chemotherapy. Seventy-Two patients received second induction chemotherapy (44 patients with first induction regimen, 13 patients with mitoxantrone based induction and, 15 patients with fludarabine based induction) and 43 patients achieved CR after second induction chemotherapy. Six out of 17 patients achieved CR after third induction chemotherapy. Patients who achieved CR received consolidation chemotherapy with or without allogeneic SCT, depending on the availability of an HLA-matched donor (related or unrelated). Genetic factors such as FLT3-ITD or NPM1 mutation were not considered when deciding whether to perform allogeneic SCT for consolidative treatment. Written informed consent was obtained from all subjects for the genetic analysis of samples taken at the time of diagnosis. The study was approved by the Institutional Review Board of Chonnam National University Hwasun Hospital, South Korea.

Gene mutation analyses
Cryopreserved bone marrow or peripheral blood samples taken at diagnosis were archived. Genomic DNA was extracted using QIAamp DNA blood mini-kits (QIAGEN, Valencia, CA, USA), as per the manufacturer's protocol. Mutation analyses were performed using Sanger sequencing and polymerase chain reaction (PCR). TET2, FLT3-ITD, and NPM1 mutation testing was performed as described previously [27,28]. TET2 missense mutations were included in the analysis only when they were located within one of two evolutionarily conserved domains (amino acids 1,104-1,478 or 1,845-2,002) and identical TET2 mutations in both alleles were defined as homozygous [29,30]. The DNMT3A, WT1, NRAS, ASXL2, and IDH1/2 mutation analysis was performed as reported previously [31,32] [29][30][31][32][33][34][35]. The CEBPA, FAT1, DNAH11, and GATA2 were amplified by genomic PCR; overlapping PCR products covering the entire coding sequence were generated and sequenced using the PCR primers in Supplementary Table 2. Amplification featured initial denaturation at 95°C for 5 min, followed by 40 cycles of 94°C for 30 s, 62°C for 30 s, and 72°C for 1 min, and a final 10-min extension at 72°C. The amplification products were sequenced on an ABI 3100 platform using a cycle sequencing kit (BigDye Terminator; Applied Biosystems, Foster City, CA, USA).

Level of 5hmC
The 5hmC levels were measured in 375 DNA samples using Quest 5hmC ™ DNA ELISA Kits (Zymo Research, Boston, MA, USA). DNA (100 ng) from each patient or control was used for quantification. Biochemical assays were performed according to the manufacturer's recommendations and samples were read in a spectrophotometer at 450 nm. Absolute quantification was performed using the concentrations of positive control samples by the standard curve method. The amount of 5hmC was calculated from 100-ng samples of singlestranded DNA and is reported as a percentage (%).

Response and survival endpoints
The definition of CR followed reported criteria [36]. Relapse risk was defined as the time from achieving remission to the date of relapse in all patients who achieved CR, considering the competing risk of death without relapse. Non-relapse mortality (NRM) was defined as death occurring in the absence of relapse. EFS was defined as the time from commencing induction chemotherapy to the date of death from any cause, relapse, or non-achievement of CR, whichever occurred first. OS was defined as the time from the start of induction chemotherapy to the date of last followup, or death from any cause. Patients undergoing allogeneic HSCT were not censored at the time of transplantation.

Statistical analyses
First, we tried to sub-classify the 5hmC value using R partitioning to find the valuable cut-off range. However, we could not find a significant cut-off value for 5hmC (p>0.05). Therefore, we simply sub-classified the 5hmC values into tertiles. Descriptive statistics are presented as frequencies (%) and medians (with ranges) for categorical and continuous variables, respectively. The clinical characteristics and treatment outcomes of each mutation were compared with each other. The χ 2 test was used to compare the differences in categorical data distributions, while the Wilcoxon rank-sum test was used to evaluate the differences between continuous variables. ANOVA was used to evaluate the differences between continuous variables and compare three groups for statistical significance. Pearson and Spearman correlation coefficients were used to calculate the correlations between two continuous variables. EFS and OS were calculated using the Kaplan-Meier method; the differences among groups were compared using the log-rank test and Cox's proportional hazard model for univariate and multivariate analyses, respectively. Age, WBC count, peripheral blast count, receiving allogeneic SCT, frequencies of NPM1 mut , CEBPA mut (double), FLT3-ITD pos , DNMT3A mut , and ASXL1 mut , as well as the 5hmC level were included in the final multivariate model. Covariates with parameters that were significant in univariate analyses were included in the multivariate analysis. We also included the 5hmC level in the multivariate analysis to identify its clinical significance. The incidences of relapse and NRM were calculated using a cumulative incidence method that considered competing risks. Gray's test and the Fine-Gray test were used for univariate and multivariate comparisons, respectively [37]. P-values < 0.05 were considered to be significant. Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using a predetermined value for reference risk unity. All statistical analyses were performed using SPSS ver. 21.0 (SPSS, Chicago, IL, USA) and EZR software in 'R' language (available at http://www.jichi.ac.jp/saitamasct/SaitamaHP.files/statmedEN.html) [37].