Table S1. Clinical characteristics. Table S2. RNA-seq gene function list. Table S3. Gene function analysis training set (Proportion test). Table S4. Gene function analysis training and test set combined (Proportion test). Table S5. Gene function analysis test set (Proportion test). Table S6. Accuracy for 10 proliferation immune-related genes. Table S7. Immunohistochemical assessment of Ki-67 positivity. Table S8. Disease control rate in cold tumors by proliferation status. (XLSX 82 kb)

Resistance to immune checkpoint inhibitors (ICIs) has been linked to local immunosuppression independent of major ICI targets (e.g., PD-1). Clinical experience with response prediction based on PD-L1 expression suggests that other factors influence sensitivity to ICIs in non-small cell lung cancer (NSCLC) patients.

PD-L1 immunohistochemistry (IHC) has been traditionally used for predicting clinical responses to immune checkpoint inhibitors (ICIs). However, there are at least 4 different assays and antibodies used for PD-L1 IHC, each developed with a different ICI. We set to test if next generation RNA sequencing (RNA-seq) is a robust method to determine PD-L1 mRNA expression levels and furthermore, efficacy of predicting response to ICIs as compared to routinely used, standardized IHC procedures.

Background Immune checkpoint inhibitors (ICIs) have changed the clinical management of melanoma. However, not all patients respond, and current biomarkers including PD-L1 and mutational burden show incomplete predictive performance. The clinical validity and utility of complex biomarkers have not been studied in melanoma. Methods Cutaneous metastatic melanoma patients at eight institutions were evaluated for PD-L1 expression, CD8+ T-cell infiltration pattern, mutational burden, and 394 immune transcript expression. PD-L1 IHC and mutational burden were assessed for association with overall survival (OS) in 94 patients treated prior to ICI approval by the FDA (historical-controls), and in 137 patients treated with ICIs. Unsupervised analysis revealed distinct immune-clusters with separate response rates. This comprehensive immune profiling data were then integrated to generate a continuous Response Score (RS) based upon response criteria (RECIST v.1.1). RS was developed using a single institution training cohort (n = 48) and subsequently tested in a separate eight institution validation cohort (n = 29) to mimic a real-world clinical scenario. Results PD-L1 positivity ≥1% correlated with response and OS in ICI-treated patients, but demonstrated limited predictive performance. High mutational burden was associated with response in ICI-treated patients, but not with OS. Comprehensive immune profiling using RS demonstrated higher sensitivity (72.2%) compared to PD-L1 IHC (34.25%) and tumor mutational burden (32.5%), but with similar specificity. Conclusions In this study, the response score derived from comprehensive immune profiling in a limited melanoma cohort showed improved predictive performance as compared to PD-L1 IHC and tumor mutational burden.