Reproducible evaluation of immune infiltration parameters is needed for the development of reliable biomarkers in the era of cancer immunotherapy. This study investigated inter-observer reproducibility when evaluating tumor-infiltrating lymphocytes (TIL), tertiary lymphoid structures (TLS), PD-1 and PD-L1 positivity on immunohistochemically-stained (IHC) breast cancer (BC) tissues, including detailed analysis of spatial positioning in the tumor microenvironment and expression on individual positive cells. Retrospectively collected archival BC samples (N=441) were dual IHC-stained for CD3 (pan T cells) plus CD20 (pan B cells) for TIL and TLS assessment with a consecutive tissue section dual-IHC stained for PD-1 and PD-L1. The 882 tissue sections were independently read by two immuno-pathologists blinded to the clinical data for specific studies between 2014 and 2017. Five independent study series were used for the current analysis: 1) samples of BC relapses, 2) residual disease in breast after neodjuvant treatment, and 3) three independent series of untreated primary tumors. Intraclass correlation coefficients for evaluations are shown in the table below:

 Relapses Residual disease Untreated primary tumors 
 Series 1: N=88 Series 2: N=20 Series 3: N=85 Series 4: N=113 Series 5: N=135 
 2014 2014 2014-2015 2016 2017 
 ICC ICC ICC ICC ICC 
%Stromal TIL 0.85 [0.76-0.90] 0.71 [0.27-0.89] 0.79 [0.68-0.87] 0.74 [0.62-0.82] 0.77 [0.68-0.83] 
%Intratumoral TIL 0.40 [0.09-0.61] 0.63 [0.08-0.86] 0.73 [0.58-0.82] 0.87 [0.81-0.91] 0.64 [0.51-0.74] 
%Global TIL 0.87 [0.80-0.92] 0.79 [0.47-0.92] 0.83 [0.74-0.89] 0.80 [0.72-0.86] 0.81 [0.73-0.86] 
%CD3 0.83 [0.74-0.89] 0.71 [0.28-0.89] 0.81 [0.70-0.88] 0.72 [0.60-0.80] 0.73 [0.63-0.81] 
%CD20 0.74 [0.62-0.81] 0.81 [0.52-0.92] 0.76 [0.63-0.84] 0.78 [0.68-0.84] 0.78 [0.70-0.84] 
TLS 0.80 [0.70-0.87] 0.78 [0.45-0.91] 0.83 [0.74-0.89] 0.95 [0.93-0.97] 0.97 [0.96-0.98] 
%PD-L1 tumor cells 0.78 [0.66-0.86] 0.56 [-0.12-0.82] 0.82 [0.73-0.89] 0.62 [0.45-0.74] 0.45 [0.25-0.61] 
%PD-L1 stromal cells 0.61 [0.40-0.74] -1.56 [-1.53-0.60] 0.43 [0.12-0.63] 0.61 [0.44-0.73] 0.31 [0.05-0.5] 
%PD-L1 immune cells 0.53 [0.29-0.70] 0.94 [0.84-0.97] 0.29 [-0.09-0.54] 0.38 [0.10-0.57] 0.10 [-0.25-0.35] 
%PD-1 TLS -0.00 [-0.54-0.35] NA 0.72 [0.57-0.82] 0.57 [0.35-0.72] 0.76 [0.62-0.84] 
 Relapses Residual disease Untreated primary tumors 
 Series 1: N=88 Series 2: N=20 Series 3: N=85 Series 4: N=113 Series 5: N=135 
 2014 2014 2014-2015 2016 2017 
 ICC ICC ICC ICC ICC 
%Stromal TIL 0.85 [0.76-0.90] 0.71 [0.27-0.89] 0.79 [0.68-0.87] 0.74 [0.62-0.82] 0.77 [0.68-0.83] 
%Intratumoral TIL 0.40 [0.09-0.61] 0.63 [0.08-0.86] 0.73 [0.58-0.82] 0.87 [0.81-0.91] 0.64 [0.51-0.74] 
%Global TIL 0.87 [0.80-0.92] 0.79 [0.47-0.92] 0.83 [0.74-0.89] 0.80 [0.72-0.86] 0.81 [0.73-0.86] 
%CD3 0.83 [0.74-0.89] 0.71 [0.28-0.89] 0.81 [0.70-0.88] 0.72 [0.60-0.80] 0.73 [0.63-0.81] 
%CD20 0.74 [0.62-0.81] 0.81 [0.52-0.92] 0.76 [0.63-0.84] 0.78 [0.68-0.84] 0.78 [0.70-0.84] 
TLS 0.80 [0.70-0.87] 0.78 [0.45-0.91] 0.83 [0.74-0.89] 0.95 [0.93-0.97] 0.97 [0.96-0.98] 
%PD-L1 tumor cells 0.78 [0.66-0.86] 0.56 [-0.12-0.82] 0.82 [0.73-0.89] 0.62 [0.45-0.74] 0.45 [0.25-0.61] 
%PD-L1 stromal cells 0.61 [0.40-0.74] -1.56 [-1.53-0.60] 0.43 [0.12-0.63] 0.61 [0.44-0.73] 0.31 [0.05-0.5] 
%PD-L1 immune cells 0.53 [0.29-0.70] 0.94 [0.84-0.97] 0.29 [-0.09-0.54] 0.38 [0.10-0.57] 0.10 [-0.25-0.35] 
%PD-1 TLS -0.00 [-0.54-0.35] NA 0.72 [0.57-0.82] 0.57 [0.35-0.72] 0.76 [0.62-0.84] 

Cohen's κ coefficients reveal poor agreement (κ<0.4) for assessment of the categorical global PD-1 and PD-L1 variables (cut-off >1%). Sporadically, fair agreement (κ>0.4) was observed for PD-L1 expressed by immune and tumor cells. Globally, the inter-observer agreement remained constant overtime. Poor reproducibility for categorical global PD-1 and global PD-L1 assessment was observed. On the contrary, pathologists have excellent reproducibility for assessing global TIL and TLS, good/excellent reproducibility for PD-1 in TLS (primary tumors) and for PD-L1 expression on tumor cells which make these of great interest for clinical use.

Citation Format: Cinzia Solinas, Gert Van den Eynden, Alexandre De Wind, François Richard, Anaïs Boisson, Soizic Garaud, Karen Willard-Gallo. Reliability of immune biomarker assessment in breast cancer: A report on interobserver variability from studies at a single institution [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 1624.