Domain-specific PD-L1 protein measurement in non-small cell lung cancer (NSCLC).

Abstract

8064 Background: PD-L1 protein expression is being explored as a companion predictive test for anti-PD-1 therapies. Studies from our group, however, reveal limitations of available antibodies and indicate that different immunohistochemical (IHC) methods yield discordant results. Differential results have been seen when comparing antibodies targeting the extracellular (EC) versus intracellular (IC) domains. Here, we studied the specificity of anti-PD-L1 antibodies using quantitative fluorescence (QIF) and the performance of two of them in NSCLC. Methods: We measured PD-L1 protein expression using antibody clones E1L3N (IC) and E1J2J (EC) in 509 NSCLC tumor samples represented in two tissue microarrays (YTMA79, n=204 and YTMA250, n=305). Antibody specificity was tested using QIF in formalin-fixed paraffin-embedded (FFPE) samples from human placenta and Mel624 transfectants. PD-L1 signal was determined in the tumor compartment using the AQUAmethod. In YTMA79, scores were compared with previous data using 5H1 antibody (EC). Reproducibility was assessed by staining serial cuts, and the correlation coefficient (R2) was calculated. For survival analysis, PD-L1 signal was used as a continuous score, or binarized with cutoffs defined by X-tile. Results: Of eight antibodies tested, only three (5H1, E1L3N, and E1J2J) validated. The serial section reproducibility was high for 5H1 (R2>0.9), and lower for E1L3N and E1J2J (R2 range 0.54-0.93). The regression coefficient of scores and dynamic range of the antibodies were dissimilar. The EC domain antibodies E1J2J and 5H1 showed no correlation with the IC antibody E1L3N. There were trends toward association with better outcome with each antibody, but the level of significance was a function of the experimental cutpoints and is only considered exploratory. Conclusions: Evaluation of PD-L1expression in NSCLC samples using validated antibodies targeting different protein domains produced discordant results. This could be due to different antibody affinities, cross reactivity, or distinct target epitopes. Evaluation of the predictive value of these antibodies for anti-cancer immunotherapies is currently underway.