While the role of clonal evolution during leukemia development and therapy has been a focus for a number of avenues of research, our abilities to deduce the clonal composition of individual samples have been limited by the use of bulk samples. Leukemia populations of unknown heterogeneity are typically described by extracting nucleic acids from the entire sample, including cells of unrelated lineages, any residual normal cells, and those cells constituting the leukemic population. Then bioinformatic approaches are employed to reconstruct the identities of the clones involved using feature frequencies in sequence data from the large scale sequencing done (e.g., whole exome or methylation). By definition, this type of approach, while realistic to perform on a large number of clinical samples, must rely on a basic set of assumptions regarding the limits of the type of genomic features that can exist in a sample. For whole exome sequencing data, assumptions include that mutations occur in cells heterozygously and that mutations which cluster together based on observed allele frequencies in the entire DNA sample occur concurrently in the same cells. To describe sub-clones within a population, a typical assumption is that mutations that occur at lower frequencies are sub-clones of the more abundant clone rather than a unique population. It is entirely possible that these assumptions are biologically appropriate for certain loci. However, for those loci for which the evolutionary patterns do not match this model (mutations occur sequentially and are retained throughout the subsequent lineage), we will inevitably come to an incorrect conclusion if this type of model is applied.

In order to best begin testing bioinformatic methods to describe clonal structure and identify the presence of evolution, we have begun experiments mixing AML cell lines with distinct immunologic and mutational characteristics. Initial flow cytometric validation of perspective cell lines along with the molecular characterization of each cell type and verification of comparable growth rates in a normalized medium in vitro has been performed. Subsequent mixing of these cell lines and application of drugs specifically targeting one line or another provides a valuable opportunity to create controlled in vitro clonal evolution scenarios in which we can validate bulk, sorted fractions, and single cell methods and the required computational frameworks for each.

As an example clonal framework, the cell lines were mixed in equal concentrations and allowed to compete in culture for 2 days with or without the presence of crenolanib, to which only one cell line is sensitive. The results of analyzing bulk and sorted fractions showed that we can effectively remove one “clone” in this in vitro poly-clonal leukemia model with crenolanib and that the observed flow cytometric frequency data reflect that seen in the genotyping data knowing the mutational distribution in the sample. However, if the allele frequency data is analyzed via the typical bulk sequencing assumptions, the reconstructed clonal structure is actually incorrect. If this were a clinical sample, we would conclude that initially the leukemia had one predominant clone with one mutation, and a subclone which had acquired 3 additional mutations. The true clonal structure was two clones with two known mutations each, (one mutation can be carried as a homozygous mutant), and one clone with no known mutations. With the addition of therapy, one would conclude that the subclone was reduced in frequency when in reality one clone was nearly completely removed, but two remained untouched.

We have created a simple system to which we can apply different conditions to the in vitro culture, generate different types of molecular data, and test methods of analyzing those data to better understand how to approach bulk sample analyses for clonal evolution.


No relevant conflicts of interest to declare.

Author notes


Asterisk with author names denotes non-ASH members.

Sign in via your Institution