Platelet transcriptome yields progressive markers in chronic myeloproliferative neoplasms and identifies putative targets of therapy

Abstract Predicting disease natural history remains a particularly challenging endeavor in chronic degenerative disorders and cancer, thus limiting early detection, risk stratification, and preventive interventions. Here, profiling the spectrum of chronic myeloproliferative neoplasms (MPNs), as a model, we identify the blood platelet transcriptome as a proxy strategy for highly sensitive progression biomarkers that also enable prediction via machine learning algorithms. Using RNA sequencing (RNA-seq), we derive disease-relevant gene expression in purified platelets from 120 peripheral blood samples constituting two independently collected and mutually validating patient cohorts of the three MPN subtypes – essential thrombocythemia, ET (n=24), polycythemia vera, PV (n=33), and primary or post ET/PV secondary myelofibrosis, MF (n=42), as well as healthy donors (n=21). The MPN platelet transcriptome discriminates each clinical phenotype and reveals an incremental molecular reprogramming that is independent of patient driver mutation status or therapy. Leveraging this dataset, in particular the progressive expression gradient noted across MPN, we develop a machine learning model (Lasso-penalized regression) predictive of the advanced subtype MF at high accuracy under two conditions of validation: i) temporal internal (AUC-ROC of 0.96) and ii) geographic external cohorts (AUC-ROC of 0.86). Lasso-derived signatures offer a robust core set of Highlights Leveraging two independent and mutually validating MPN patient cohorts, we identify progressive transcriptomic markers that also enable validated prediction in MPNs. Our platelet RNA-Seq data identifies impaired protein homeostasis as prominent in MPN progression and offers putative targets of therapy.