Cell state transitions drive the evolution of disease progression in B-lymphoblastic leukemia

Cancer stem cells (CSCs) are hypothesized to promote tumor progression through innate chemoresistance and self-renewal. CSCs reside in the CD34 ⁺ /CD38 ^- immunophenotypic subpopulation of acute myeloid leukemia (AML); correspondingly, the size of this subpopulation has a strong negative impact on overall survival. Isolation of CSCs from B-lymphoblastic leukemia (B-ALL) has proven difficult, and the cells of interest apparently are not isolated to the CD34 ⁺ /CD38 ^- compartment. This may be explained, in part, by temporal variations of CD34 and CD38 expression which result in stochastic cell state transitions (e.g., from CD34 ⁺ /CD38 ⁺ to CD34 ⁺ /CD38 ^- ). We present a mathematical model of these transitions and correlate salient findings with BCR::ABL1 status, minimal residual disease (MRD), and relapse in adult B-ALL. As the CSC hypothesis is well supported in AML, we focus on transitions to and from the hematopoietic stem cell compartment (CD34 ⁺ /CD38 ^- ), which our findings suggest can be estimated from peripheral blood or bone marrow samples. Additionally, we find that BCR::ABL1 positive patient samples are associated with high transition rates into the CD34 ⁺ /CD38 ^- compartment, including self-renewal. In contrast, BCR::ABL1 negative patient samples have low CD34 ⁺ /CD38 ^- self-renewal rates and either high CD34 ⁺ /CD38 ⁺ or CD34 ^- /CD38 ⁺ incoming rates. High CD34 ⁺ /CD38 ^- self renewal is also associated with MRD post-induction chemotherapy. We find a lack of observable changes in cell state transitions between diagnosis and relapse specimens. Our analysis provide evidence of de-differentiation transitions to a CD34+/CD38-stem cell-like immunophenotype in leukemia samples collected from B-ALL patients, and the tendency for these transitions is especially strong for B-ALL with BCR::ABL1. The modeling framework used here is a novel, useful tool to infer both prognosis and genotype from routine flow cytometry data.