Distinct DNA Methylation and Expression Profiles Underlie CMML Responsiveness to Decitabine and Uncover Novel Mechanism of Resistance

Myelodysplastic syndromes (MDS) and the related disorder chronic myelomonocytic leukemia (CMML) are characterized by abnormal DNA hypermethylation and the DNA methyltransferase inhibitors (DMTis) azacytidine and decitabine (DAC) are frequently used as frontline therapy in these patients. However, DMTis are ineffective for ~50% of the patients who must still undergo treatment for at least 6 months before they can be deemed resistant. Therefore, it is of critical importance to identify baseline molecular differences associated with DMTi sensitivity that can help (i) to improve patient risk-stratification at diagnosis and (ii) identify the underlying mechanisms of resistance to these agents. Previous efforts to identify baseline DNA methylation differences at promoter regions between DMTi responders and non-responders have not been successful, so we hypothesized that any potential differences would be located distally from promoter regions. For this purpose we studied 40 CMML patients at diagnosis, all of whom had been uniformly treated with DAC 20 mg/m²/day x 5 days as frontline therapy. After 6 cycles of therapy patients were classified as responders (n=19, hematological improvement or better), or non-responders (n=21, stable or progressive disease). Mutational analysis showed no significant differences in the frequency of mutations in TET2, ASXL1, DNMT3A, RUNX1, TP53, JAK2, KIT, KRAS, EZH2, IDH1/2 and spliceosome genes. Using Enhanced Reduced Representation Bisulfite Sequencing (ERRBS) we analyzed the baseline methylation status at ~3M CpG sites across the genome of 39/40 CMML patients. We identified 158 statistically significant differentially methylated regions (DMRs) (FDR<0.1 and methylation difference ≥25%) between the two groups. DAC-sensitive patients displayed both regions of higher methylation as well as regions with lower methylation compared to DAC-resistant patients. As predicted, DMRs were depleted at promoters (DMRs 9% vs. Background [BG] 21%, p-value: 3.4×10^-5) and CpG islands (DMRs 8% vs. BG 25%, p-value: 1.5×10^-8). Further analysis showed that hypermethylated DMRs were enriched at intronic regions (Hyper DMRs 58% vs. BG 33%, p-value: 3.7×10^-6) while hypomethylated DMRs were enriched at intergenic regions (Hypo DMRs 49% vs. BG 38%, p-value: 2.6×10^-2). Moreover, hypermethylated DMRs were significantly enriched for enhancer regions, and in particular, enhancers located within gene bodies (hyper DMRs 38% vs. BG 18%, p-value: 2.3×10^-5). KEGG pathway analysis showed a significant enrichment of DMRs in the MAPK signaling pathway (FDR<0.01). Next, using a support vector machine algorithm with 10-fold cross validation we were able to develop a classifier capable of predicting response to DAC with high level of accuracy (ROC AUC: 0.99) based solely on the DNA methylation status at diagnosis of 17 genomic regions. Three different random splits of the cohort into training and test sets achieved correct predictions for 85.7%, 89.47%, and 100% of cases, respectively, demonstrating the accuracy and potential utility of such a classifier. Finally, RNA-seq analysis identified 53 differentially expressed genes between responders (n=8) and non-responders (n=6) at diagnosis. Genes implicated in cell cycle and DNA replication were overexpressed in responders. By contrast, very few genes were overexpressed at the time of diagnosis in primary resistant patients. Among these were CXCL4 and CXCL7 which, given their reported contributions to cell cycle arrest and chemoresistance, were tested for their functional roles in DAC resistance. Pre-treatment of normal CD34+ cells for 72 h with 10nM DAC significantly reduced colony formation (p<0.05) but the addition of 50ng/mL of CXCL4 and CXCL7 restored colony formation to that of untreated cells. Moreover, treatment of primary CMML cells with 10 nM DAC for 72h significantly reduced viability of these cells, while concomitant treatment with 50ng/mL of CXCL4 and CXCL7 was sufficient to abrogate this effect. Taken together, our findings demonstrate that (i) specific DNA methylation profiles targeting non-promoter regulatory regions are associated with DAC sensitivity, (ii) these differences can be harnessed for the development of clinical biomarkers predictive of response and (iii) we identified a novel mechanism of resistance to DAC mediated through two chemokines that are exclusively overexpressed in non-responders.