Abstract 1514: Significance assessment of mutations in 944 MDS patients using publicly available variant databases and mutation impact prediction software

Abstract Introduction: In 2015 the FDA issued a call to the public to receive feedback on FDA's regulatory approaches to diagnostic tests using next generation sequencing technology. For the clinical performance of such tests one of the proposals was to use community-derived databases to classify variants, especially ClinVar, conceived as a clinically grade database. To test this we here applied ultra-deep sequencing and subsequent mutation profiling in patients with myelodysplastic syndomes (MDS). Aim: Investigate the performance of public databases and mutation impact prediction software for the interpretation and distinct classification of variants of a well-characterized MDS mutation dataset (Haferlach et al, Leukemia 2014). Patients and Methods: A total of 944 patients with various MDS subtypes were screened for gene mutations in 104 known/putative genes relevant to MDS using targeted deep-sequencing (Illumina, San Diego, CA). For this assessment the following databases were used: ClinVar (release 2015-11), COSMIC (v73) and dbSNP (v142). Additionally, mutations were computationally tested for their severity of impact on protein level using PolyPhen-2 and SIFT. Results: In total, 845/944 patients (89.5%) harbored at least one mutation (median, 3 per patient; range, 0-12). A total of 2764 variants were called, among them 1,608 being distinct in 96 genes. Assessment was conducted by submitting positional information of each mutation to the database/prediction software. ClinVar yielded information for 141 (9%) of the mutations, with TP53 being the best-characterized gene out of 34, comprising of 33 entries (23%). Querying COSMIC yielded information for 671 (42%) mutations in 61 genes, with a subset of them being particularly well characterized (TET2, TP53, DNMT3A, and ASXL1). 255/1608 variants were listed in dbSNP. In the majority of instances, no global minor allele frequency (MAF, frequency of occurrence in the population of a variant base) is given and the validation status lists only a single submitter, indicating poorer reliability. Additionally, we analyzed the mutations with two tools (PolyPhen-2, SIFT) to predict the possible impact of an amino acid substitution on the structure and function of human proteins. Results were available for 1137/1608 mutations. In 72% (820/1137) results were concordant, but surprisingly in 28% (317/1137) instances, results were contradicting, leaving them non-interpretable based on the combined use of both tools. Conclusion: 1) Assessment demonstrates that current methods for variant interpretation using publicly available databases have to be improved for the characterization of mutations in patients with myeloid neoplasms. 2) So far COSMIC seems to outperform ClinVar. 3) Tools for novel mutations (no record in any databases) seem to perform well in quite a few instances, but a consensus of multiple tools is needed due to contradicting results. Citation Format: Niroshan Nadarajah, Manja Meggendorfer, Wolfgang Kern, Claudia Haferlach, Torsten Haferlach. Significance assessment of mutations in 944 MDS patients using publicly available variant databases and mutation impact prediction software. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 1514.