Heterogeneous programmed death-ligand 1 expression in gastric cancer: comparison of tissue microarrays and whole sections

Background Programmed death-ligand 1 (PD-L1) expression determines the eligibility for anti-PD-1 treatment in patients with advanced gastric cancer, but evidence indicates that PD-L1 staining is heterogeneous. Patients who are ineligible for radical surgery could be tested for PD-L1 expression with biopsy staining, but it is unclear if a small biopsy is representative of the PD-L1 status of the whole tumor. The aim of our study was to determine how many biopsy specimens are needed to accurately reflect the objective status of PD-L1 expression in whole sections. Methods We built tissue microarrays (TMAs) as substitutes for core biopsies, collecting 6 cores per case from 152 gastric cancer specimens. All of the slides and TMAs underwent PD-L1 immunohistochemical staining, and PD-L1 expression in at least 1% of tumor cells or immune cells was defined as positive. Results It was necessary to randomly select multiple cores from TMAs to reach a suitable agreement rate (> 90%) and Cohen’s κ value (> 0.8) between TMAs and whole sections. We defined the PD-L1 staining status from the whole section as the standard. The evaluation of five randomly selected cores from TMAs agreed well with the evaluation of whole sections. The sensitivity, specificity and the area under the curve (AUC) of the receiver-operating characteristic (ROC) were 0.93, 0.92, and 0.922 (95% confidence interval (CI) 0.863–0.982), respectively. Conclusions We conclude that PD-L1 expression among TMA samples had different degrees of relevance to the corresponding surgical specimens, which indicates that at least five biopsies might be necessary to characterize patients taking anti-PD-1 treatment.