Whole Exome Sequencing to Predict Response to Hypomethylating Agents in MDS

Abstract 1698

Hypomethylating agents decitabine and azacitidine are standard treatments for myelodysplastic syndromes (MDS). In their use, one hopes to rectify cytopenias and prolong survival by retarding further disease progression. However, individual treatment responses vary from complete remission (CR) to complete refractoriness. In general, at least 4 cycles of therapy are administered prior to assessing response. Thus, patients may have prolonged exposure to ineffective therapy, suffering toxicities without clinical benefit, while alternative and potentially more effective treatments are delayed. Currently, there are no reliable phenotypic or mutational markers for predicting response to hypomethylating agents.

Once whole exome sequencing (WES) became available for more routine analysis, we theorized that somatic mutational patterns may help identify patients who would most benefit from these drugs, thereby maximizing response rate by rational patient selection. To pursue this hypothesis, we screened a cohort of 168 patients with MDS who received either azacitidine or decitabine for the presence of somatic mutations. Only those who received sufficient therapy, i.e., completed at least 4 cycles, were selected for outcome analysis. Targeted Sanger sequencing, including a panel of up to 19 genes frequently affected by somatic mutations was performed. For a representative subset of 26 patients (this subset is expanding) of whom there were 15 responders and 11 non-responders, mutational analysis was performed by WES to select target genes for further analysis. WES utilizes paired DNA (tumor vs. CD3+ lymphocytes) to produce raw sequence reads aligned using Burrows-Wheeler Aligner (BWA). Variants are detected using the Broad Institute's Best Practice Variant Detection GATK toolkit.

Median age was 68 years (range, 55–85), 50% were female, and MDS subtypes were as follows: RA/RCUD/RARS 13%, RCMD 16%, RAEB-1/2 20%, MDS/MPN & CMML-1/2 31%, and sAML 20%. Response was assessed using IWG 2006 criteria at 4 and 7 months after therapy initiation. Overall response was 48%; rate of CR (including marrow/cytogenetic CR) was 28%, any HI 20%, SD 22%, and no response 29%. The cohort was then dichotomized into “responders” and “non-responders,” with responders classified as those achieving CR or any HI. Baseline patient characteristics were similar between both groups, including average age at treatment initiation, disease subtypes, proportion of abnormal/complex karyotypes, and presence of common cytogenetic aberrations. Overall, the most frequently mutated genes include TET2/IDH1/IDH2, SRSF2, ASXL1, SF3B1, RUNX1, EZH2/EED/SUZ12, SETBP1, CBL, and PPIAF2. The highest rate of refractoriness was noted in mutants of TET2/IDH1/IDH2 (67%), SF3B1 (67%), U2AF1/2 (67%). We also identified several genes whose mutants were few but associated exclusively with refractory disease (100%), including KIT, ZRSR2, PRPF8, LUC7L2.

We next applied a recursive partitioning algorithm to construct a decision tree for identifying the most pivotal mutations associated with response: we found mutant CBL and PPFIA2 to be strongly associated with response, whereas mutant U2AF1/2, SF3B1 and PRPF8 were strongly associated with refractoriness. Our final approach was to dichotomize the cohort by the presence/absence of each mutation/group of mutations, and response within mutant vs. wild type cases was compared. Among refractory cases, TET2/IDH1/IDH2 (26%) and SF3B1 (17%) were most frequently mutated; among responders, mutations in RUNX1 (19% vs. 4%]), CBL (14% vs. 0%), SRSF2 (23% vs. 9%), and SETBP1 (18% vs. 4%) were most frequent. When multiple genes were combined in “either-or” fashion, mutation in TET2, SF3B1, PRPF8, or LUCL71 was significantly associated with refractoriness (52%, p=.0287), whereas mutations of RUNX1, CBL, SRSF2, SETBP1, or PPFIA2 mutation was significantly associated with response (86%, p=.0001).

Mutational patterns appear to predict response to standard hypomethylating agents. Identification of the most predictive genes could guide development of molecular maker-based selection of patients for hypomethylating agent therapy, but will require ongoing analysis and additional prospective testing for validation.