Molecular Profiling of Decitabine Response in MDS and AML Patients

The underlying molecular effects of decitabine treatment in MDS and AML patients are poorly understood. To systematically uncover the global genomic and transcriptomic alterations induced by decitabine in vivo, we evaluated a series of primary bone marrowf samples from patients treated with decitabine 20 mg/m² on days 1-10 in 28 day cycles (NCT01687400). Samples were evaluated with: bulk-RNA sequencing (Day 0 & 10, n=29), single-cell RNA sequencing (Day 0 & 10, n=4), whole-genome bisulfite sequencing (WGBS, Day 0, 10, & 28, n=29), and exome sequencing (Day 0 & 28 n=44). Cases were selected for RNA-Seq and WGBS based on adequate sample availability, and founding clone mutations that were present in at least 75% of the cells in each sample on day 0.

As expected, interpatient heterogeneity dominated global transcriptional variance across the bulk-RNA sequencing dataset, due to the unique mutational and cellular heterogeneity of each sample. We identified 638 significantly differentially expressed genes (DEGs) between day 0 and day 10. MSigDB-geneset analysis identified significant pathway enrichment of the day 10 upregulated genesets for myeloid maturation, immune activation, and cell cycle regulation, while down-regulated genesets were enriched for erythroid pathways, oxidative phosphorylation, cellular senescence, and metabolic processes (FDR < 0.05). Upregulated genes were strongly associated with increased expression in external datasets of flow-sorted mature myeloid cells (i.e. Bloodspot), whereas downregulated genes correlated with decreased expression in mature populations, further suggesting that decitabine induces myeloid maturation in AML and MDS patients. Application of the Telescope computational pipeline identified increased overall expression of endogenous retroviruses (ERVs) at day 10 (p < 0.01). Interpatient heterogeneity was again evident, and ERV activation was limited to 18 ERV families (FDR <0.05), which included HERVH, HERV3, HERV9, HERV30, LTR19, and HERVE. Across the dataset, overall ERV expression did not correlate significantly with GSEA-identified transcriptional signatures of inflammatory response or interferon signatures, although rare outlier cases could be identified with strong correlations (but only borderline significance). No clinical features were significantly associated with RNA-Seq transcriptional signatures (response, survival, TP53 mutation status, blast counts, ERV family activation, etc.). Single-cell RNA sequencing analysis detected induction of interferon signatures and inhibition of erythroid signatures by decitabine within subsets of predominantly malignant bone marrow cells (CD3e-CD19-), corroborating the findings from bulk-RNA sequencing. Whole-genome bisulfite sequencing analysis revealed global hypomethylation at day 10, with incomplete re-methylation by day 28 in nearly all cases. Hypomethylation and remethylation patterns were uniform across the genome, and indifferent to local genomic context. Similar hypomethylation signatures were noted at promoters, transcriptional start sites, DNase I hypersensitivity sites, CCCTC binding factor (CTCF) sites, and transcription factor binding sites, suggesting that decitabine-induced hypomethylation effects were not modified by local genomic contexts, and were not consistently associated with transcriptional changes. Of the 44 cases evaluated with serial exomes, 21 had an absolute reduction in bone marrow blasts of at least 10% on day 28, and could be evaluated for differentiation effects. Of these, all but one exhibited greater morphologic blast clearance than founding clone clearance, suggesting the persistence of malignant cells that no longer displayed blast morphology, consistent with decitabine-induced maturation of malignant cells.

In summary, this study provides insights into in vivo mechanisms of decitabine activity in AML and MDS patients, which includes global hypomethylation, myeloid maturation, induction of an inflammatory response, activation of specific ERV families, and strong interpatient heterogeneity. However, no specific transcriptional or epigenetic biomarkers emerged that could serve as effective predictors of decitabine responses.