Mutational Spectrum of Myelodysplastic Syndrome Malignancies Revealed by Whole Exome Sequencing

Abstract 307

Whole-exome (WES) sequencing revealed tremendous mutational heterogeneity in leukemia. While WES can be applied for discovery, it also has potential as a diagnostic tool that can overcome the shortcomings of current methods. We theorized that, in addition to mutation discovery, systematic application of WES in MDS may reveal distinct mutational patterns allowing for new molecular classification.

We performed WES in 116 paired exomes, including MDS (n=57), MDS/MPN (n=36), and sAML (n=23). We also included comparative analysis with pAML (N=202; TCGA), and other publicly available data for a total of 333 exomes; 10 patients were studied serially. Paired DNA (marrow/CD3+ cells) was subjected to WES, sequence-aligned by BW Aligner, and variants detected via GATK pipeline (Broad Institute). We used defined criteria to minimize false-positives: P<.001 tumor/control, alterations ≥10% of total tumor reads, <25% in germline, >5% prevalence, and not found in ex/internal SNP databases. This narrowed the spectrum to 645 mutations (54 genes) for analysis with clinical/phenotypic correlations. Mutations were isolated or grouped by pathway, e.g., PRC2, cohesin complex, plexins and dyneins, etc. In MDS, examples of prevalent mutations include SF3B1 (14%), DNMT3A (11%) and U2AF1/2 (9%). In MDS/MPN: TET2 (36%), SRSF2 (22%) and ASXL1 (19%) and SETBP1 (6%); in sAML: NRAS/RAS (16%), RUNX1 (16%) and cohesin mutations (12%), in contrast to pAML with mutational spectrum dominated by FLT3, DNMT3A, NMP1 or SMC3/1A (cohesin complex). The exome panel did not cover 20% patients, suggesting that their pathogenesis may be related to less recurrent events (613 candidates: 2^nd screening phase). When mutational spectrum of sAML vs pAML were compared, mutants of SF3B1 (7 vs 1%, P=.04), BCOR (7 vs 1%, P=.04), CDH11/23 (13 vs. 1%, P=.003), FMN2 (7 vs. 1%, P=.04), PPFIA2 (7% vs 0%, P=.01), SPTAN1 (7% vs 0%, P=.01) and VPS8 (7 vs 0%, P=.017) were more frequent in sAML while DNMT3A and NPM1 were less common. Analysis of MDS/MPN revealed mutations in PRC2 (2 vs 11%, P=.05), SRSF2 (5 vs. 22%, P=.010) and TET2 (3 vs. 33%, P<.001) more frequent than in MDS. Mutations in SF3B1 were more recurrent in low/Int-1 IPSS categories compared to Int-2/high/sAML (21 vs. 3%, P=.01), in which mutations in N/KRAS (0 vs. 14%, P=.01) and TP53 (0 vs. 14%, P=.01) were more frequent. Functional group comparisons revealed that lesions in epigenetic (56 vs 23%, P=.001) and signal transduction genes (36 vs 9%, P=.001) were more prevalent in MDS/MPN compared to MDS in which they accumulated according to risk (high vs low: 36 vs 5%, P=.001 or 52% in pAML). Spliceosomal mutations were overrepresented in MDS/MPN vs MDS (58 vs 37%, P=.031), in sAML vs pAML (23 vs 9%, P=.032), and in low risk vs high risk cases (45 vs 22%, P=.02). Cytoskeleton organization gene mutations were overrepresented in sAML vs pAML (39 vs 13%, P=.001). TSG were more frequent in high-risk vs low-risk MDS (30 vs 5%, P=.003). Moreover, TET2 mutations coincided with SRSF2 and PRC2 mutations (P<.001 and P=.010); DNMT3 mutations with SF3B1 and BCOR (P=.04 and P=.004); SRSF2 with ASXL1 (P=.017); RUNX1 with cohesin and BCOR (P=.003 and P=.04), CBL mutations with PRPF8 and ASXL1 (P=.04 or P=.003); TP53 with PRPF8 (P=.04).

After analyzing survival impact of individual mutations, functional groups, cytogenetic category and clinical parameters, we found TP53, ETV6, PRPF8, FMN2, UMODL1, KIT, GATA2, complex karyotype and chr. 5 anomalies had a prognostic impact on OS. However, in multivariate analyses, the first variable to stratify our cohort was, as expected, the diagnosis subtype (HR 2.2, P<.001), but also mutations in PRPF8 (HR 5.4, P=.004). In MDS and grouped MDS/MPN, significant variables included KIT (HR 12, P=.022) and TP53 mutations (HR 3.6, P=.045). Apart from traditional analyses, we also applied a recursive partitioning algorithm to construct an unbiased survival tree encompassing every mutation: e.g., PRPF8, CSMD1, U2AF2, IDH2, PPFIA2, SF3B1 and NRAS showed the highest difference in OS with this method.

In sum, mutational spectrum of myeloid neoplasms can be assessed with WES. The pattern of frequency and concurrence in each diagnostic subtype differs substantially, a feature that can be exploited diagnostically. Despite heterogeneity, mutations and their combinations can be found to categorize patients and serve as prognostic markers. Analysis of additional cases is ongoing and will be presented at the meeting.