Framework to Identify and Prioritize Candidate Inherited Myeloid Malignancy Germline Variants Leveraging the BEAT AML Cohort

Background: Familial predisposition to myeloid malignancies is more common than previously appreciated. 15-20% of acute leukemia patients have at least 1 additional first-degree relative with leukemia predisposition. Germline predisposition to myeloid neoplasms was incorporated in the WHO 2016 classification of myeloid neoplasms and acute leukemia. The clinical guidelines now include testing for inherited susceptibility as a critical element of patient diagnostics. Identification of germline predisposition syndrome can significantly impact treatment decisions, screening of potential sibling donors for allogeneic stem cell transplantation, and patient and family surveillance. Here we developed a framework to guide routine identification of potential pathogenic germline mutations for further characterization and aid in identification of individuals who are at high risk of developing leukemia so that these individual can be closely monitored for disease management and treatment interventions.

Methods: As part of the Beat AML consortium, we performed deep whole-exome sequencing of 424 tumor samples from 378 AML patients with matched skin biopsies (Tyner, et al, Nature 2018). High confidence germline variants called by VarScan2 with Exac and gnoMD frequency of less than 1% and normal variant allele frequency greater than 40%, were annotated if they appeared in a curated list of 244 genes that have been previously associated with hematological malignancies and leukemia predisposition. Putative novel variants were nominated based on review of other databases (Exac, gnoMD, dbSNPOSMIC, ClinVAR, Varsome). The mutants were classified as pathogenic/ likely pathogenic using Clinvar and ACMG/AMP guidelines. In silico curation were made using computational predictions of pathogenicity and functional impact (e.g. Sift, Polyphen, Revel, DANN, CADD) with emphasis on the Revel scores. All patients with available family history data were evaluated for evidence of familial clustering of hematological malignancies in 1^st, 2^nd or 3^rd degree relatives.

Results: The overall frequency of pathogenic and likely pathogenic germline variants using Clinvar and ACMG/AMP criteria is 11.1% (42/378) with 38 variants in 29 leukemia predisposition genes. The most frequent variants were DDX41 (6), CHEK2 (5), and FANC complex (10) with the remaining patients having single observations in a diversity of leukemia predisposition genes. Recognizing the concerns about tumor contamination within the skin biopsy, we validated a subset of variants from 12 cultured stromal samples for matched patients. Overall, in 378 primary AML patients, we identified 55 novel germline variants in 44 leukemia predisposition genes with the potential to be pathogenic by in silico prediction [Revel > 0.5 or DANN > 0.9 or CADD > 19]. Interestingly, evaluation of the family history found 16 out of 172 patients with the available data who have a 1^st degree relative with a hematologic malignancy, 9 patients with an affected 2^nd degree relative, and 5 patients with a 3^rd degree affected relative. All of these patients with familial clustering have germline variants in leukemia predisposition genes. 16 patients with familial clustering have 6 novel variants and 35 reported variants with previously unpredicted clinical significance that show potential to be deleterious by in silico prediction. For these 16 patients with familial clustering, age at the diagnosis of AML ranged from 36-83 years. Ten of these patients have overlapping somatic variants in FLT3 (5), NPM1 (4), DNMT3A (5), and CEBPA (4). The remaining 6 patients have isolated cases of somatic variants in genes such as RAS, IDH2, U2AF1, EZH2, TET2, etc. Overall, novel germline variants are candidates for functional validation to understand their contribution in disease pathogenicity.

Conclusions: Exome sequencing and in silico prediction identified ~11% of AML patients in the Beat AML cohort have pathogenic/likely pathogenic germline variants with leukemia predisposition genes. In the cohort, 9.3% of the patients had evidence of familial clustering of hematological malignancies. Combining identification of novel germline variants in leukemia predisposition genes, with emphasis on those seen in patients with familial clustering with functional validation can improve in silico prediction efforts and help to guide patient monitoring and clinical disease management.