Chang-Chieh Hsu

Only a subset of patients with non–small cell lung cancer (NSCLC) benefits from immune checkpoint inhibitors (ICIs). Tissue PD-L1 has limited accuracy in predicting ICI response and requires invasive biopsies, highlighting the urgent need for reliable, minimally invasive biomarkers. Extracellular vesicle (EV) PD-L1 is a promising liquid biopsy marker, but its predictive performance is confounded by PD-L1 derived from nontumor cells. We developed a tumor-specific EV-based assay to predict ICI response. Tumor-derived EVs (TEX) were selectively captured using the tumor-specific Thomsen–Friedenreich glycoantigen α (TF-Ag-α), and PD-L1 levels in TEX expressing TF-Ag-α (i.e., TEXTF-Ag-α PD-L1) were measured in sera from 91 patients with NSCLC receiving anti–PD-1 therapy (40 responders and 51 nonresponders). Predictive performance was evaluated at pretreatment baseline, at the end of the first treatment cycle, and via fold changes (FC). At pretreatment baseline, TEXTF-Ag-α PD-L1 showed moderate predictive ability [area under the curve (AUC): 0.76; sensitivity: 0.97; specificity: 0.56]. By the end of the first treatment cycle, responders exhibited a ∼4.18-fold increase in TEXTF-Ag-α PD-L1 expression, whereas nonresponders showed an approximate 20% decrease. Dynamic FCs after the first treatment cycle substantially improved predictive performance, achieving an AUC of 0.93, sensitivity of 0.95, and specificity of 0.80 for distinguishing responders from nonresponders, outperforming both total EV PD-L1 and tissue PD-L1. These results demonstrated that selective TEX capture using TF-Ag-α significantly enhanced the predictive value of EV PD-L1. Dynamic changes in TEXTF-Ag-α PD-L1 provided a highly accurate, minimally invasive biomarker for early assessment of anti–PD-1 response, supporting tumor-specific EV-based liquid biopsy for precision immunotherapy. Significance:Reliable biomarkers for predicting ICI response remain lacking. We developed a tumor-specific liquid biopsy assay that selectively captures TEX via TF-Ag-α, enabling accurate PD-L1 measurement. By reducing interference from nontumor EVs, the assay achieved 93% predictive accuracy, outperforming tissue PD-L1 and supporting precision immunotherapy.

Detection of TEXTF-Ag-α PD-L1 using the Exo-PROS assay. A, Sensing mechanism of the Exo-PROS assay. B, Representative SPR curve for detecting TEX TF-Ag-α in the serum sample of a patient with stage IV NSCLC. A.U., arbitrary unit.

Evaluation of TEX TF-Ag-α and TEXTF-Ag-α PD-L1 for predicting anti–PD-1 therapy response in a discovery cohort of patients with NSCLC (15 responders and 14 nonresponders). TEX TF-Ag-α and TEXTF-Ag-α PD-L1 expression at pretreatment baseline (A and B), at the end of the first treatment cycle (E and F), and their FCs (I–L). ROC analysis of their predictive performance at baseline (C and D), at the end of the first treatment cycle (G and H), and FCs (M and N). *, P < 0.05; **, P < 0.01; ***, P < 0.001. NS, not statistically significant. A.U., arbitrary unit.

Evaluation of TEX LG3BP and TEXLG3BP PD-L1 for predicting anti–PD-1 therapy response in a discovery cohort of patients with NSCLC (15 responders and 14 nonresponders). TEX LG3BP and TEXLG3BP PD-L1 expression at pretreatment baseline (A and B), at the end of the first treatment cycle (E and F), and their FCs (I–L). ROC analysis of their predictive performance at baseline (C and D), at the end of the first treatment cycle (G and H), and FCs (M and N). *, P < 0.05. NS, not statistically significant. A.U., arbitrary unit.

Figure S1. Evaluation of TEX TF-Ag-α and TEXTF-Ag-α PD-L1 combined biomarkers for predicting response to anti-PD-1 therapy in a discovery cohort of NSCLC patients.

Figure S2. Evaluation of TEX LG3BP and TEXLG3BP PD-L1 combined biomarkers for predicting response to anti-PD-1 therapy in a discovery cohort of NSCLC patients.

Predictive performance of total EV PD-L1 and tissue PD-L1 for response to anti–PD-1 immunotherapy (n = 25 responders and n = 37 nonresponders). Total EV PD-L1 expression at the pretreatment baseline (A), at the end of the first treatment cycle (C), and their FCs (E and F). Corresponding ROC analyses for distinguishing responders from nonresponders in B, D, and G, respectively. Tissue PD-L1 TPS (H) and corresponding ROC analysis (I) for distinguishing responders from nonresponders. **, P < 0.01; ****, P < 0.0001. NS, not statistically significant. A.U., arbitrary unit.

Evaluation of TEX TF-Ag-α and TEXTF-Ag-α PD-L1 for predicting anti–PD-1 therapy response in a validation cohort of patients with NSCLC (25 responders and 37 nonresponders). TEX TF-Ag-α and TEXTF-Ag-α PD-L1 expression at pretreatment baseline (A and B), at the end of the first treatment cycle (E and F), and their FCs (I–L). ROC analysis of their predictive performance at baseline (C and D), at the end of the first treatment cycle (G and H), and FCs (M and N). Kaplan–Meier survival analysis showed that patients with TEXTF-Ag-α PD-L1 FC >1.59 had substantially longer PFS (O) and OS (P) compared with patients with TEXTF-Ag-α PD-L1 FC ≤1.59 (P < 0.001). *, P < 0.05; **, P < 0.01; ****, P < 0.0001. NS, not statistically significant. A.U., arbitrary unit.

Figure S3. Evaluation of TEX TF-Ag-α and TEXTF-Ag-α PD-L1 combined biomarkers for predicting response to anti-PD-1 therapy in a validation cohort of NSCLC patients.

Figure S4. Comparison of predictive performance of TEXTF-Ag-α PD-L1, total EV PD-L1 and tissue PD-L1 for response to anti-PD-1 immunotherapy in the validation cohort.