Usefulness of EHR-Linked Biobanks in Variant Classification for Myeloid Neoplasms

Unlike next generation sequencing panels (NGS) for solid tumors, NGS for hematopoietic myeloproliferative neoplasms (MPN) usually lack paired non-neoplastic specimens for comparison to confirm somatic vs. germline nature of detected mutations. Current variant classification guidelines for oncogenicity are based on a phenotype-first model, where the variants of interest are observed from ascertained affected patients and classified based on population frequency, functional data, hotspots, and computational evidence. The high prevalence of clonal hematopoiesis and recurrent mutations make population frequency threshold less effective (5% or 1%) in filtering out less common circulating germline variants. As a genotype-first approach in this study, we used data from the All of Us Research Program, an electronic health records (EHR)-linked genomic biobank with ~200k participants in version 7 from diverse ancestry and socioeconomic backgrounds, to evaluate potential driver mutations for myeloid neoplasms reported in the literature.

First, we used a known oncogenic mutation JAK2 p.V617F (rs77375493-T, gnomAD allele frequency [AF]=2.9e-4) as a positive example, although it can be harbored in reportedly healthy individuals. Among the 226 participants with this variant detected in All of Us with EHR data available, 134 (59.3%) had recorded diagnoses of myeloid neoplasms. The variant allele fraction (VAF) of these participants (derived from allelic depth [AD]) spread variably from 10% (detection limit by genomic sequencing) to over 90%. In the gnomAD Age Distribution Chart, the carrier-to-all ratios were much higher in individuals over 50 years old.

We then tested variants that had been reported as (likely) pathogenic in the literature. For the JAK2 p.R564L (rs368927897-T, AF=6.6e-5) and p.R564Q (-A, rarer) variants in Exon 13, the VAF ranged from 0.32 to 0.65 and centered around 0.5. Fewer than 5% of those with these variants had diagnostic codes for essential thrombocythemia (ET) with a highest recorded platelet count of 594e9/L. A phenome-wide association study (PheWAS, All by All dataset) on participants of predicted European ancestry (EUR) did not show significant associations with hematopoietic phenotypes. Therefore, JAK2 p.R564L and Q appear to be mostly germline variants with limited, if any, association with MPN.

For the FLT3 p.V591I (rs773466203-T, AF=2.5e-5) variant in the juxtamembrane domain, ≤20 participants carried this variant, and none had reported acute myeloid leukemia or other myeloid neoplasms. The VAF of these participants ranged from 0.40 to 0.64. Therefore, our data suggests FLT3 p.V591I is a germline variant and should not be treated as other oncogenic mutations in this gene in the setting of myeloid neoplasms.

Of the 626 All of Us participants with the SH2B3 p.E208Q (rs202080221-C, AF=6.6e-4) variant, 482 also had EHR data available. The VAF of these participants were centered to 50% (26-73%) or 100%. Fewer than 5% of the participants had diagnostic codes for any myeloid neoplasms. The PheWAS (All by All dataset) on EUR showed significant associations with “Other nonthrombocytopenic purpura” but not myeloid neoplasms. Therefore, SH2B3 p.E208Q is a relatively common germline variant with limited, if any, association with myeloid neoplasms.

In this study, we demonstrated that population-circulating non-oncogenic germline variants in driver genes for hematopoietic neoplasms can have similar or lower AF than recurrent oncogenic mutations. Providers should be very cautious in or avoid using aforementioned variants (other than JAK2 p.V617F) to classify myeloid neoplasms or as therapeutic targets. We propose that EHR-linked biobank data and evidence of somatic mutations should be incorporated into variant oncogenicity classification for hematopoietic neoplasms.

No relevant conflicts of interest to declare.