Deep Digital Sequencing Identifies an AML Subclone with Enhanced in Vitro and in Vivo Growth Properties Associated with Disease Relapse

Abstract 407

Acute myeloid leukemia (AML) is a biologically heterogenous malignancy of hematopoietic cells. All AML samples are comprised of a founding clone and usually one or more subclones that are derived from the founding clone; subclones can gain or lose mutations as they evolve from the founding clone, and often become dominant at relapse¹. The clonal architecture of an AML sample can be identified using single nucleotide variants (SNVs) that cluster according to discrete variant allele frequencies (VAFs). To accurately identify clusters with common VAFs, deep digital sequencing must be performed using all of the SNVs present in each genome (hundreds of events). In this report, we studied the subclonal architecture of AML samples from UPN 452198, collected from a 55-year old woman with normal karyotype acute monocytic leukemia (FAB M5) with a high peripheral WBC count (72,700/mm³) at presentation. This bone marrow sample contained a founding clone and 3 subclones at presentation. The SNVs of the founding clone had a mean VAF of 46.4% (i.e. heterozygous mutations found in 92.8% of the cells in the bone marrow sample), including DNMT3A R882H and NPM1 W288 frameshift mutations. The mean VAFs of Subclones 1, 2 and 3 were 31.2%, 12.0% and 2.4%, respectively, and they contained all of the variants in the founding clone, along with additional variants, most notably FLT3 D835H and IDH1 R132H mutations in Subclone 1. The tumor at relapse consisted entirely of Subclone 3, which also contained 42 relapse-specific variants. We designed an oligonucleotide capture reagent to track all 118 de novo and relapse-specific variants, and obtained deep read counts (mean coverage per site: 412 reads/SNV) on the de novo AML sample under different experimental and biological conditions, as follows: 1) We showed that peripheral blood and bone marrow leukemia samples obtained at the same time had nearly identical clonal architectures. We verified this correlation using 4 additional AML samples, suggesting that clonal architecture is preserved in the peripheral blood for many AML samples. 2) We flow-sorted the leukemic peripheral blood sample into blasts, monocytes, and lymphocytes based on side-scatter characteristics and expression of CD45 and CD33. The founding clone and all three subclones were detected in the monocyte population, which was the predominant leukemic cell population in the peripheral blood. By flow cytometry, blasts comprised only 3.3% of the cells, but were strongly enriched for variants in Subclone 3 (mean VAF in sorted blasts 33.9% versus 3.0% in unsorted peripheral blood, p<0.001). Purified lymphocytes, in contrast, contained no leukemia-specific variants, implying that the founding clone for this sample did not contribute to lymphopoiesis. 3) We tested the growth properties of subclones in the de novo sample in vitro and in vivo. We injected 1 million cells from the de novo AML sample into 3 immunodeficient NSG mice, and harvested human AML blasts (co-expression of human CD45 and CD33) 14 weeks later. Although two mice engrafted with the founding clone and Subclone 1 (which comprised the vast majority of the cells in the sample), the 3^rd mouse had a tumor composed entirely of the relapse-specific Subclone 3 (which accounted for only 2.4% of the variants in the de novo sample), suggesting that this subclone had a significant growth or engraftment advantage in vivo. In support of this observation, de novo AML cells were strongly enriched for Subclone 3 when grown in the presence of hematopoietic cytokines (SCF, IL3, IL6, TPO and FLT3L) for 7 days on HS27 stroma (VAF at day 7–19.7%; p<0.001) or MS5 stroma (VAF at day 7–22.8%; p<0.001), implying that this clone also had a strong in vitro growth advantage. In summary, a small subclone of AML cells at presentation — that was known to eventually contribute to relapse — had unique growth properties that were revealed using deep digital sequencing of all variants. This approach has allowed us to dissect the evolutionary history of AML clones, and to define their relationship to other hematopoietic cells in a given sample. Similar studies on additional AML samples should allow us to define the mutational profiles of subclones that are destined to contribute to relapse. This data will be essential for improving therapeutic approaches for AML patients.