Quantitative comparison of PD-L1 immuno-histochemical assays in hepatocellular carcinoma: The Blueprint-HCC study.
91 Background: Programmed cell death ligand 1 (PD-L1) expression by immunohistochemistry (IHC) enriches for responses to PD-1/PD-L1 inhibitors, however its role as a predictive biomarker in hepatocellular carcinoma (HCC) is inconclusive, with no consensus on any particular assay. We evaluated the performance of 4 different PD-L1 detection assays previously published in landmark clinical studies. Methods: PD-L1 IHC was performed on 4 serial sections from tissue microarray (TMA) blocks containing 100 archival cases of HCC that included tumour and surrounding non-tumorous tissue. Antibody clones E1LN3, 28-8, 22c3, SP263 were compared on the basis of percentage and intensity of staining in malignant cells (M) to generate an H-score (range 0-300). Immune cells infiltrating (ICI) and at the periphery, in non-tumorous tissue (ICP) were scored on a 4-tier system (0-3). Results: Patients were 76% males, 20% HCV-positive, 64% cirrhotic with a median age of 67 years. Median tumour size was 4 cm, 70% of patients had T1-T2 tumours and 48% were of grade 2. The proportion of PD-L1 positive cases according to M-ICI-ICP pattern was 2-6-2% for E1LN3, 10-18-19% for 28-8, 9-22-18% for 22c3 and 5-14-13% for SP263. Pairwise comparison of M H-scores revealed heterogeneity across antibodies, with highest concordance between E1L3N/SP263 (R2 = 0.95), E1L3N/22c3 (R2 = 0.65), 22c3/SP263 (R2 = 0.66) and increasing discordance for 28-8/22c3 (R2 = 0.44), E1L3N/28-8 (R2 = 0.29), and 28-8/SP263 (R2 = 0.26). Detection of PD-L1-positive immune infiltrates using a semi-quantitative scoring system revealed significantly different scores in pairwise non-parametric comparisons of ICI (p < 0.05) but not ICP (p > 0.05 for chi-square test). Conclusions: In the Blueprint-HCC study we demonstrated that quantification of PD-L1 protein levels in tumour cells, intra-tumoural and peri-tumoural infiltrate is characterised by inter-assay discordance in HCC. This has profound implications in the clinical development of predictive correlates of efficacy to immunotherapy in HCC. Sources of such discordance should be explored.