Validation of multiplex immunofluorescence and digital image analysis for programmed death-ligand 1 expression and immune cell assessment in non-small cell lung cancer: comparison with conventional immunohistochemistry

Aims This study aimed to validate the application of combined multiplex immunofluorescence (mIF) and digital image analysis (DIA) in formalin-fixed and paraffin-embedded tissues for the quantitative assessment of programmed death-ligand 1(PD-L1) and immune cells (ICs) in non-small cell lung cancer (NSCLC). Methods Fifty resected samples of NSCLC were sequentially stained with a DNA-tagged mIF (panel including PD-L1, CKpan, CD8, CD68 and 4 ',6-diamidino-2-phenylindole (DAPI)) and conventional immunohistochemistry (cIHC). The assessment of cell density and consistency of tumour proportion score (TPS) via DIA were compared with those by pathologists. Results A strong correlation in the cell population of immune markers was obtained between mIF and cIHC (for PD-L1: R=0.9304, CKpan: R=0.8231, CD8: R=0.9314 and CD68: R=0.8366) within 95% limits of agreement. The continuous TPS calculated using mIF was highly consistent with the IHC staining results which were evaluated by pathologists (R=0.9362). However, in the comparison of TPS using interval variables, a poor agreement was obtained at a cut-off of 1% (kappa=0.197), whereas excellent agreement was achieved at cut-offs of 50% (kappa=0.908) and 5% (kappa=0.823). DIA on mIF showed that PD-L1 commonly colocalised with CD68(+) macrophages and CD8(+) cytotoxic cells were closer to PD-L1(-)/CK+ tumour cells (TCs) than to PD-L1(+)/CK+ TCs in spatial distribution. Conclusions A combination of mIF and DIA is useful for the quantification of PD-L1 expression and IC populations in NSCLC. Further validation of TPS at a cut-off of 1% and assay harmonisation is essential for translating this method in a diagnostic setting.

Automated summary

The study validated the application of combined mIF and DIA for the quantitative assessment of PD-L1 expression and IC populations in NSCLC, demonstrating a high consistency between the two methods, with differences in consistency observed at different cut-off points.