31 Dynamic change of PD-L1 expression on extracellular vesicles predicts response to immune-checkpoint inhibitors in non-small cell lung cancer patients

BackgroundImmune-checkpoint inhibitors (ICIs) have revolutionized the treatment of advanced/metastatic non-small cell lung cancer patients (NSCLC), however, only a small subset of patients derives clinical benefit.1–3 To date, PD-L1 immunohistochemical evaluation is the gold-standard assay and the only approved biomarker, but associated with several limitations due to technical and biological factors such as spatial and temporal tumor heterogeneity.4 5 In this context, liquid biopsies emerge as novel powerful tools that could allow the non-invasive real-time characterization of the tumor PD-L1 status. In particular, extracellular vesicles (EVs), defined as cell-derived double-membrane structures involved in cell communication, hold strong potential as tissue surrogates. Recent studies have suggested that EV PD-L1 could stratify melanoma patients receiving ICIs, but none has showed the predictive value of this biomarker in NSCLC patients.6 7 We hypothesize that EV PD-L1 cargo can serve to stratify the response to ICIs in NSCLC patients.MethodsThis study enrolled advanced/metastatic NSCLC patients receiving ICI treatment. Plasma samples were obtained at baseline (T1) and at 8 weeks (T2) during the first response evaluation. Patients were classified as responders when showing partial, stable or complete response or as non-responders when manifesting progressive disease following RECIST v1.1.8 Plasma EVs were isolated by standard serial ultracentrifugation methods and characterized according to ISEV recommendations.9 10 Tissue PD-L1 expression was measured by immunohistochemistry while EV PD-L1 expression was measured by immunoblot. A predictive model was created by logistic-regression and a bootstrap corrected ROC curve to validate the results.ResultsPaired plasma samples from 21 patients were analyzed. PD-L1 tissue expression was not correlated with treatment response (p=0.394) nor matched the baseline EV PD-L1 levels (p=0.337) (figure 1.A). However, the dynamics of EV PD-L1 (T1-T2) correlated with the treatment response, observing an increase of PD-L1 expression in non-responders and a decrease or stable levels in responders (p=0.043) (figure 1.B). The predictive model reported an AUC=0.85, 90% CI=0.72–0.97, with 74.2% sensitivity and 73.5% specificity (figure 1.C). Moreover, the increase of EV PD-L1 was associated with shorter overall survival (HR=4.34, p=0.037) and shorter progression-free survival (HR=5.06, p=0.025) (figure 1 D & E).ConclusionsOur prelim