Dynamic levels of extracellular vesicle PD-L1 and complementary radiomics for the prediction of the response to immune checkpoint inhibitors in lung cancer patients.

e21144 Background: Immune-checkpoint inhibitors (ICIs) have revolutionized the therapeutic landscape of lung cancer patients. However, its low treatment efficacy is still an issue and the current standard of care tissue PD-L1 presents high variability. Liquid biopsy is a promising tool in the discovery of biomarkers in body fluids. In particular, extracellular vesicles (EVs) can present PD-L1 in their membranes, playing a role in the inhibition of the anti-tumor immune response. Likewise, TGF-β is crucial in the immune response found in the circulation and into EVs. On the other hand, radiomics analysis of conventional imaging provides information about tumor heterogeneity and immune response. Hence, we aimed to evaluate the predictive role of circulating biomarkers in lung cancer patients undergoing ICIs and the additional value of radiomics data. Methods: This is a retrospective analysis of 30 advanced/metastatic non-small lung cancer patients treated with ICIs. Plasma samples were collected at baseline and at 8 weeks during treatment, matching the first response evaluation. Patients with complete, partial response, or stable disease were classified as responders and those with progressive disease as non-responders following RECIST v1.1. EVs were isolated from plasma by ultracentrifugation and PD-L1 expression was revealed by immunoblot. Circulating and EV levels of TGF-β were analyzed by ELISA. Additionally, 400 radiomics features from target and non-target lesions were analyzed to evaluate response in 24 patients according to RECIST v1.1 and irRECIST. Robustness of predictive models was validated by ridge penalty and leave-one-out cross-validation. Results: The analysis of the dynamics of EV PD-L1 during treatment identified increased levels in non-responders in comparison to responders ( p= 0.012), while tissue PD-L1 levels were not associated to the response ( p= 0.585). The predictive model for EV PD-L1 reported a high accuracy with an area under the curve (AUC) = 77%, 91.7% sensitivity, and 61.1% specificity. The combination with radiomics, improved the accuracy and reported an AUC = 83%, 82% sensitivity, and 77% specificity. Additionally, the analysis of the association of the circulating biomarkers with the outcome revealed that increasing dynamics of EV PD-L1 and high baseline EV TGF-β were associated with shorter progression-free survival and overall survival, outperforming the circulating levels of TGF-β. Conclusions: This pilot study demonstrated that EV PD-L1 could serve as better predictive factor than tissue PD-L1 for the stratification of lung cancer patients undergoing ICIs and that it could be complemented with radiomics. Moreover, EV levels of PD-L1 and TGF-β have potential for the stratification and prognosis of patients treated with ICIs and their novel combinations with TGF-β blockade.