State-transition Modeling of Blood Transcriptome Predicts Disease Evolution and Treatment Response in Chronic Myeloid Leukemia

Abstract

Chronic myeloid leukemia (CML) is initiated and initially maintained solely by the fusion gene BCR-ABL, encoding a multifaceted chimeric kinase targeted in the clinic with tyrosine kinase inhibitors (TKIs) TKIs are effective in inducing long-term remission, but are also frequently not curative. Thus, CML is an ideal system to test our hypothesis that transcriptome-based state-transition models accurately predict cancer evolution and treatment response. To test our hypothesis, we collected time-sequential blood samples from tetracycline-off (Tet-Off) BCR-ABL-inducible transgenic mice and wild-type controls. Using the time-series bulk RNA-seq analysis to capture a system-wide view of distinct disease states, we identified a single principal component that constructed a CML state-space with a three-well BCR-ABL leukemogenic potential landscape. The potential stable critical points defined observable disease states. Early states were characterized by anti-CML genes opposing leukemic transformation; late states were characterized by pro-CML genes. Genes with expression patterns shaped similarly to the potential landscape were identified as disease transition drivers. Re-introduction of tetracyclines to silence the BCR/ABL gene returned diseased mice transcriptomes to a stable state near to health, without reaching it, suggesting partly irreversible transformation changes. TKI treatment only reverted the diseased mice transcriptomes to an earlier disease state, without approaching health; disease relapse occurred soon after treatment completion. Using only the earliest time-point as initial conditions, our parametrized state-transition models accurately predicted both disease progression and treatment response, supporting this as a potentially valuable approach to time clinical intervention even before phenotypic changes become detectable.