Protein profiles predict treatment responses of chronic lymphocytic leukemia to the PI3K inhibitor umbralisib

The management of chronic lymphocytic leukemia (CLL) has significantly improved with targeted therapy, but given the suboptimal response in a subset of patients predictive biomarkers are needed. Functional phenotyping of CLL cells of patients receiving the PI3K inhibitor umbralisib found that the treatment led to significant changes in (phospho)protein levels, including AKT, in responders but not in non-responders. Furthermore, the proportion of cytotoxic natural killer cells increased at the end of the treatment in responders but not in non-responders, suggesting a role in the antitumor response. To identify molecular predictors of response, the authors used baseline levels of 30 (phospho)proteins as input features for a machine learning model, which achieved significant prediction accuracy in cross-validation and maintained its predictive power in a cohort of patients receiving umbralisib plus the BTK inhibitor acalabrutinib. Drug sensitivity profiling of baseline CLL cells suggested that PI3K plus Bcl-2 inhibitors are effective in umbralisib non-responders. The authors conclude that functional phenotyping reveals differential cellular responses to umbralisib treatment in responders and non-responders, predicts treatment response of individual CLL patients, and suggests alternative treatment option for non-responders.