Landscape of DNA Methylation and Genetic Profiles in 291 Patients with Myelodysplastic Syndromes

DNA hypermethylation has long been implicated in the pathogenesis of myelodysplastic syndromes (MDS) and also highlighted by the frequent efficacy of demethylating agents to this disease. Meanwhile, recent genetic studies in MDS have revealed high frequency of somatic mutations involving epigenetic regulators, suggesting a causative link between gene mutations and epigenetic alterations in MDS. The accumulation of genetic and epigenetic alterations promotes tumorigenesis, hypomethylating agents such as Azacitidine exert their therapeutic effect through inhibition of DNA methylation. However, the relationship between patterns of epigenetic phenotypes and mutations, as well as their impact on therapy, has not been clarified.

To address this issue, we performed genome-wide DNA methylation profiling (Infinium 450K) in combination with targeted-deep sequencing of 104 genes for somatic mutations in 291 patients with MDS. Beta-mixture quantile normalization was performed for correcting probe design bias in Illumina Infinium 450k DNA methylation data. Of the >480,000 probes on the methylation chip, we selected probes using the following steps: (i) probes annotated with "Promotor_Associated" or "Promoter_Associated_Cell_type_specific; (ii) probes designed in "Island", "N_Shore" or "S_Shore"; (iii) removing probes designed on the X and Y chormosomes; (iv) removing probes with >10% of missing value. Consensus clustering was performed utilizing the hierarchical clustering based on Ward and Pearson correlation algorithms with 1000 iterations on the top 0.5% (2,000) of probes showing high variation by median absolute deviation across the dataset using Bioconductor package Consensus cluster plus. The number of cluster was determined by relative change in area under cumulative distribution function curve by consensus clustering.

Unsupervised clustering analysis of DNA methylation revealed 3 subtypes of MDS, M1-M3, showing discrete methylation profiles with characteristic gene mutations and cytogenetics. The M1 subtype (n=121) showed a high frequency of SF3B1 mutations, exhibiting the best clinical outcome, whereas the M2 subtype (n=106), characterized by frequent ASXL1, TP53 mutations and high-risk cytogenetics, showed the shortest overall survival with the hazard ratios of 3.4 (95% CI:1.9-6.0) and 2.2 (95% CI:1.2-4.0) compared to M1 and M3, respectively. Finally, the M3 subtype (n=64) was highly enriched (70% of cases) for biallelic alterations of TET2 and showed the highest level of CpG island methylation and showed an intermediate survival.

In the current cohort, we had 47 patients who were treated with demethylating agents, including 11 responders and 36 non-responders. When DNA methylation status at diagnosis was evaluated in terms of response to demethylating agents, we identified 54 differentiated methylated genes showing >20% difference in mean methylation levels between responders and non-responders (q < 0.1). Twenty-five genes more methylated in responders were enriched in functional pathways such as chemokine receptor and genes with EGF-like domain, whereas 29 less methylated gene in responders were in the gene set related to regulation of cell proliferation. Genetic alterations were also assessed how they affected treatment responses. In responders, TET2 mutated patients tended to more frequently respond (45% vs 34%), whereas patients with IDH1/2 and DNMT3A mutations were less frequently altered (0% vs 14%, 9% vs 14%) in responders, compared in non-responders.

In conclusion, our combined genetic and methylation analysis unmasked previously unrecognized associations between gene mutations and DNA methylation, suggesting a causative link in between. We identified correlations between genetic/epigenetic profiles and the response to demethylating agents, which however, needs further investigation to clarify the mechanism of and predict response to demethylation agents in MDS.