Targeted Ultra-Deep High Accuracy Sequencing of Pre-Treatment AML Reveals a Diversity of Mutational Phenotypes and Evidence of Preexisting Relapse-Associated Subclones

Human tumors contain large numbers of clonal, subclonal and random mutations. Clonal mutations, present in 30% or more of the cells, are classified as either drivers that promote proliferation or passengers of unknown or to be determined function. Subclonal and random mutations that are present in a small subset of tumor cells, prior to chemotherapy, could serve as a reservoir for the emergence of drug resistance. Even in the presence of clinical complete remissions in AML, subclonal mutations could be present in a small number of cells that selectively expand during chemotherapy and result in relapse and death. Next-generation DNA sequencing now makes it feasible to monitor the frequency of different subclonal mutations in patients as tumors evolve over time, raising the possibility of personalizing treatment by anticipating the mutations that signal relapse in time to prevent clinical recurrence of AML. We have used Duplex DNA Sequencing (DS) to detect very low-frequency subclonal and random mutations in AML during relapse and prior to treatment. While whole genome sequencing (WGS) provides extensive data on the clonal distribution of mutations in AML, it lacks sufficient accuracy to identify subclonal mutations when they are present at frequencies less than 5%. In contrast, DS focuses on a limited number of target genes at the level of single DNA molecules. Mutations are scored only if they are present at the same position in both strands of the same DNA molecule and are complementary, resulting in sequencing accuracy that is more than 1000-fold greater than that of routine next-generation DNA sequencing. Using DNA from 12 treatment-naïve AML samples and 2 normal bone marrow samples, we first targeted the exons that encode the catalytic domains of the five major human replicative polymerases. We detected both synonymous and non-synonymous mutations in most of the targeted genes. Mutations in the two major human replicative DNA polymerases have been recently identified in adenocarcinomas of the colon, and mutations in the proof-reading domain of DNA polymerase epsilon result in the highest reported point mutation frequencies so far reported in any cancer. Until now, mutations in DNA polymerases have not been described in AML. Presumably a similar or higher subclonal mutation load exists in the coding regions of other genes in AML and in genes found in other tumors.

In order to follow the course of mutation accumulation in AML after treatment and leading to relapse, we used capture hybridization that was designed to enrich for 15 genes previously reported to be mutated in AML. We identified multiple subclonal and random mutations in many of these genes. In one relapse sample, we identified a mutation in NRAS that was present in 32% of the cells. The same mutation was detected by DS in 1% of the cells from the same patient prior to treatment, which is well below the signal threshold of WGS. These initial studies demonstrate the feasibility of using DS to define the changes that occur during and after treatment, and suggest the use of DS to determine mutations that impart drug resistance. The findings from the DNA polymerase capture offers evidence that abundant non-synonymous mutations are present in treatment-naïve AML, implying that the seeds of treatment resistance may already have taken root by the time of diagnosis.