High levels of global hydroxymethylation predict worse overall survival in MDS patients treated with azacitidine

Myelodysplastic syndromes (MDS) are a heterogeneous group of hematological malignancies characterized by cytopenia, dysplasia, and a risk of progressing to acute myeloid leukemia (AML).¹ Using the international prognostic scoring systems (IPSS, IPSS-R, and recently IPSS-M), patients can be categorized into different risk groups for overall and leukemia-free survival.^2-4 In combination with fitness and individual preferences, the therapeutic strategy for each patient is determined.⁵ Currently, the strategies most commonly used are best supportive care (BSC) with or without EPO/G-CSF in lower-risk MDS, lenalidomide (LEN) in patients with a del(5q), or luspatercept in patients with ring sideroblasts/SF3B1 mutations. In higher-risk MDS, hypomethylating agents (HMAs), chemotherapy, and/or stem cell transplantation can be considered. MDS patients carry mutations in genes involved in DNA methylation including TET2 (20%–30%), DNMT3A (10%), and IDH1/2 (5%–10%).⁶ DNMT3A is a DNA methyltransferase that converts cytosine (C) into 5-methylcytosine (5mC). Methylated DNA can in turn be actively demethylated by TET enzymes (including TET2), converting 5mC into 5-hydroxymethylcytosine (5hmC) which is further converted into cytosine by subsequent actions of TET proteins, thymidine DNA glycosylase (TDG), and the base excision repair (BER) pathway. Mutations in TET2 result in defective enzymatic activity and significantly decreased levels of 5hmC. TET proteins need vitamin C, Fe^2+, and alpha-ketoglutarate (α-KG) as cofactors for proper enzymatic activity. The latter is produced by IDH1/2 enzymes. Mutations in IDH1 and IDH2 result in the aberrant production of 2-hydroxyglutarate instead of α-KG, which inhibits TET activity. Therefore, also in IDH1/2 mutated cells, decreased 5hmC levels can be observed.⁷

Cancer cells often show hypermethylation, which may result in silencing of tumor suppressor genes.⁸ The methylation process is reversible and can be influenced by the administration of HMAs like azacitidine (AZA) and decitabine. Both compounds have shown important activity in MDS and AML.⁹ HMAs are analogs of the nucleoside cytidine and they are incorporated into the DNA during DNA replication, inhibiting the DNA methylation process and causing hypomethylation. In addition, 80%–90% of azacitidine is incorporated into the RNA. As not all patients respond to HMAs and the response may take several courses of therapy before an effect becomes apparent,¹⁰ the identification of markers that predict response is warranted. Recently, a set of 39 methylation sites was found significantly different in MDS patients responding to AZA, compared to nonresponders.¹¹ We previously demonstrated that in AML patients receiving high-dose chemotherapy, high 5hmC was an independent prognostic marker for poor overall survival (OS).¹² In this study, we assessed the impact of global 5mC and 5hmC on OS in MDS patients receiving BSC, LEN, or AZA.

To do so, we measured 5mC and 5hmC in 504 MDS patients (demographics in Table S1) and 20 healthy controls. We isolated DNA from bone marrow or peripheral blood samples, collected before treatment. We measured 5mC and 5hmC using high-performance liquid chromatography-tandem mass spectrometry (HPLC-MS/MS), as described.¹² To perform the analyses, we grouped the patients into three cohorts, based on the received treatment (BSC, LEN, or AZA). To assess the impact of 5mC and 5hmC on OS, we divided the cohorts into quartiles based on 5mC and 5hmC levels. In addition, we analyzed the mutational profile using a panel of frequently mutated genes in myeloid malignancies.

Median 5mC levels were comparable between MDS patients and healthy controls, but in MDS patients, the values were distributed across a broader range (Figure 1A; median MDS = 3.8, range = 2.665–4.500; median healthy donors = 3.8, range = 3.600–4.000, p = 0.868). In contrast, the median value of the demethylation intermediate 5hmC was significantly lower in MDS patients compared to healthy controls (Figure 1B; median MDS = 0.023, range = 0–0.061; median healthy controls = 0.030, range = 0.021–0.033, p < 0.001). Overall methylation (5mC) was lower in higher IPSS-R categories (Figure 1C) whereas 5hmC levels were increased in very high-risk patients (Figure 1D). TET2 mutations were found in 31% of the patients, DNMT3A in 13%, and IDH1/2 in 8% (Figure S1 and Table S2), which is in line with previous studies.^{6, 13, 14} 5mC was not significantly influenced by the presence of TET2/IDH mutations or by DNMT3A mutations (Figure 1E). As expected, 5hmC was significantly decreased in patients carrying mutations in TET2 and IDH1/2, compared to patients with wild type TET2 and IDH1/2 (Figure 1F; median TET2/IDH/DNMT3Awt = 0.026; median TET2mut = 0.015, p < 0.001; median IDH1/2mut = 0.016, p < 0.001). Mutations in TET2 or IDH1/2 can co-occur with mutations in DNMT3A. In these patients, the 5hmC levels were comparable to the 5hmC in patients with solitary TET2 or IDH1/2 mutations (median TET2/IDH/DNMT3Amut = 0.018, p < 0.001). The overall 5mC values did not increase significantly in patients with single or more TET2 mutations (Figure 1G); 5hmC levels decreased in the case of a single TET2 mutation, which was further decreased in the case of two mutations (Figure 1H). In patients with three or more mutations (suggestive of the presence of separate TET2 mutated clones), the 5hmC levels did not further decrease. Mutations in TET2 are considered to cause a loss of function,¹⁵ irrespectively of the type of mutation. This was confirmed by the observation that no significant differences were found in 5hmC depending on the type of mutation (frameshift versus nonsense or missense mutations) (data not shown).

As expected, the OS of our cohort was highly influenced by the IPSS-R risk group (Figure S2A; n = 459, p < 0.001). Patients receiving AZA performed worse compared to patients receiving BSC or treated with LEN (Figure S2B; p < 0.001), which is in line with the higher IPSS-R risk patients present in this treatment group (Figure S2C).

It was previously reported that TET2 mutations independently predict better response to HMA treatment,^{16, 17} but no effect was seen on overall survival. In our study, we found an improved OS in TET2 mutated patients receiving AZA (Figure S3A; n = 170, p = 0.021); however, this was not significant in multivariate analysis including IPSS-R (Figure S3B; HR = 0.797, 95% HR = 0.512–1.239, p = 0.313).

To investigate whether pre-treatment 5mC/5hmC levels have an effect on OS in patients receiving different treatment modalities, we divided the cohort based on the treatments received by the patients: BSC (n = 194), LEN (n = 115), and AZA (n = 170); for each cohort, we divided the patients into quartiles based on the levels of 5mC (Figure S4A) and 5hmC (Figure 2A). The 5mC status (from the time of sampling, before the start of the treatment) did not have an impact on the OS of MDS patients who received any of these three different regimes (Figure S4B–D). Our results do not confirm data from a previous study, in which it was reported that the level of 5mC was predictive of overall survival.¹⁸ However, to measure the global methylation, the authors used enzyme-linked immunosorbent assay (ELISA), reported to be less accurate and sensitive compared to HPLC-MS/MS.¹⁹ This difference makes it hard to compare the group of patients defined as low or high 5mC in the two studies. Furthermore, the studied cohort was smaller, and the majority of the patients had lower or intermediate IPPS-R scores, which might have influenced the definition of high/low 5mC and therefore the results. Next, we analyzed the effect of the demethylation intermediate 5hmC on OS. The 5hmC levels did not have an impact on MDS patients who received supportive care or on patients treated with LEN (Figure 2B,C). We also did not observe any impact of 5hmC on the OS of patients treated with LEN, neither when performing the analysis separately in del(5q-) (n = 21) and non del(5q-) patients (n = 82) (data not shown). Interestingly, in patients receiving AZA, high 5hmC levels (≥0.0290) correlated with a significantly worse OS (Figure 2D; p = 0.012) compared to low (≤0.0150) or intermediate 5hmC levels (5-year OS low 5hmC = 27.9%, intermediate 5hmC = 26.8%, and high 5hmC = 11.2%). This could not be explained by a difference in the number of AZA cycles that was received (median = 7 in all groups). Since patients with low and intermediate 5hmC levels did not show a significantly different OS, they were considered as one group in further assessments. In the multivariate analysis, together with the IPSS-R, the effect of the 5hmC was less striking, but the same trend was still observed (Figure 2E; HR = 1.477, 95% CI = 0.977–2.233, p = 0.064).

The exact mechanism behind the difference in response to AZA is not clear, but it can be hypothesized that in patients with low and moderate 5hmC levels, the tumor cells may be dependent on the silencing of specific tumor-suppressor genes by hypermethylation, which may be corrected by hypomethylating agents such as AZA. Conversely, tumor cells with already very active demethylation (high 5hmC) may be transformed in a different manner, being less dependent on the hypermethylation of specific tumor suppressor genes, and therefore less responsiveness to hypomethylating agents. It would be interesting to identify the crucial genomic areas and associated genes and test their methylation status before and during treatment with AZA. We conclude that the pre-treatment, global 5hmC level is a prognostic marker and that lower 5hmC levels can help to identify MDS patients who are more likely to respond to AZA treatment. These results should be confirmed in an independent MDS cohort treated with AZA, as well as in a patient cohort treated with decitabine.

Francesca Tiso, Florentien E. M. in 't Hout, and Joop H. Jansen designed the research, analyzed data, and wrote the paper. Ruth Knops and Arno van Rooij performed the experiments. Arjan A. van deLoosdrecht, Theresia M. Westers, Saskia M. C. Langemeijer, Claude Preudhomme, Nicolas Duployez, Pierre Fenaux, Olivier Kosmider, Didier Bouscary, Lionel Adès, and Michaela Fontenay provided patient material and the clinical data. Aniek O. deGraaf contributed to the sequencing data and all authors discussed the results and commented on the manuscript at all stages.

The authors declare no conflict of interest.

This work was supported by a grant from the Dutch Cancer Society (grant #10813 and grant #2008-4333), the French Health Ministry and Institut National du Cancer (funding numbers are INCa_9290 and INCa-DGOS_5480), and by HOVON89 (EudraCT 2008-002195-10; METC: 2009/50 NL25632.029.08).