Prospective multi-institutional evaluation of pathologist assessment of PD-L1 assays for patient selection in triple negative breast cancer

Programmed death ligand 1 (PD-L1) expression by tumor-infiltrating lymphocytes (TILs) and tumor cells in breast cancer has been reported, but the relationships between PD-L1 expression by TIL, carcinoma cells, and other immunologic features of the breast tumor microenvironment remain unclear. We therefore evaluated the interrelationships between tumor cell surface and TIL PD-L1 expression, lymphocyte subpopulations, and patterns of immune cell infiltration in cohorts of treatment-naive, primary breast cancers (PBCs) (n = 45) and matched PBC and metastatic breast cancers (MBC) (n = 26). Seventy-eight percent of untreated PBCs contained PD-L1(+) TILs, but only 21% had PD-L1(+) carcinoma cells. Carcinoma PD-L1 expression localized to the tumor invasive front and was associated with high tumor grade (P = .04). Eighty-nine percent of PD-L1(+) carcinomas contained brisk TIL infiltrates, compared to only 24% of PD-L1(-) carcinomas; this included CD3(+) (P = .02), CD4(+) (P = .04), CD8(+) (P = .002), and FoxP3(+) T cells (P = .02). PD-L1(+) PBCs were more likely to contain PD-L1(+) TIL than PD-L1(-) PBCs (P = .04). Peripheral lymphoid aggregates were present in 100% of PD-L1(+) compared to 41% of PD-L1(-) PBC (P < .001). No patient with PD-L1(+) PBC developed distant recurrence, compared to 15% of patients with PD-L1(-) PBC. For the matched PBC and MBC cohort, 2 patients (8%) had PD-L1(+) tumors, with 1 case concordant and 1 case discordant for carcinoma PD-L1 expression in the PBC and MBC. Our data support PD-L1 expression by tumor cells as a biomarker of active breast tumor immunity and programmed death 1 blockade as a therapeutic strategy for breast cancer.

Different clones, protocol conditions, instruments, and scoring/readout methods may pose challenges in introducing different PD-L1 assays for immunotherapy. The diagnostic accuracy of using different PD-L1 assays interchangeably for various purposes is unknown. The primary objective of this meta-analysis was to address PD-L1 assay interchangeability based on assay diagnostic accuracy for established clinical uses/purposes. A systematic search of the MEDLINE database using PubMed platform was conducted using “PD-L1” as a search term for 01/01/2015 to 31/08/2018, with limitations “English” and “human”. 2,515 abstracts were reviewed to select for original contributions only. 57 studies on comparison of two or more PD-L1 assays were fully reviewed. 22 publications were selected for meta-analysis. Additional data were requested from authors of 20/22 studies in order to enable the meta-analysis. Modified GRADE and QUADAS-2 criteria were used for grading published evidence and designing data abstraction templates for extraction by reviewers. PRISMA was used to guide reporting of systematic review and meta-analysis and STARD 2015 for reporting diagnostic accuracy study. CLSI EP12-A2 was used to guide test comparisons. Data were pooled using random-effects model. The main outcome measure was diagnostic accuracy of various PD-L1 assays. The 22 included studies provided 376 2×2 contingency tables for analyses. Results of our study suggest that, when the testing laboratory is not able to use an Food and Drug Administration-approved companion diagnostic(s) for PD-L1 assessment for its specific clinical purpose(s), it is better to develop a properly validated laboratory developed test for the same purpose(s) as the original PD-L1 Food and Drug Administration-approved immunohistochemistry companion diagnostic, than to replace the original PD-L1 Food and Drug Administration-approved immunohistochemistry companion diagnostic with a another PD-L1 Food and Drug Administration-approved companion diagnostic that was developed for a different purpose.