Stem Cell-Derived Gene Expression Scores Predict Survival and Blastic Transformation in Myelofibrosis

Abstract

Purpose. Myelofibrosis (MF) is the most severe myeloproliferative neoplasm (MPN) where there remains a need for improved risk stratification methods to better inform patient management. Since MF is a stem cell driven disease and stem cell informed transcriptomic information has been shown to be prognostic across other clinical settings we sought to use this information to generate novel transcriptomic-based risk stratification models that could complement current approaches. Patients and Methods. We identified 358 MF patients from the MPN registry at the Princess Margaret Cancer Centre (ClinicalTrials.gov Identifier: NCT02760238) from whom peripheral blood mononuclear cells were collected and clinical data was available. We randomly split our cohort into a 250-patient training set and a 108-patient test set to train and validate prognostic models, respectively. Results. Within the training set we used repeated nested cross validation together with LASSO regression from various starting gene sets and found that the best prognostic models were consistently derived from transcriptomic variation among MF stem cells. From this gene set we trained our final model, a 24-gene weighted expression score (termed, MPN24) that is prognostic for overall survival in MF patients. Importantly, MPN24 was validated in the test set patients. MPN24 captures unique prognostic information to current risk stratification models such as DIPSS, MIPSS70 and the Genomic-Personalized Risk scores. Therefore, we present a novel 3-tier risk stratification approach that integrates DIPSS and MPN24 to more effectively risk stratify MF patients. Finally, from MPN24 we derived a 13-gene subsignature (termed, MPN13) from the training set patients that was validated to predict time-to-transformation in the test set patients. Conclusions. Transcriptomic information informed by MF stem cells offer novel and unique prognostic potential in MF that significantly complements current approaches. Future work will be needed to validate the robustness of the approach in external cohorts and identify how patient management can be optimized with these novel transcriptomic biomarkers.