Automated Tumour Recognition and Digital Pathology Scoring Unravels New Role for PD-L1 in Predicting Good Outcome in ER-/HER2+ Breast Cancer

Abstract

The role of PD-L1 as a prognostic and predictive biomarker is an area of great interest. However, there is a lack of consensus on how to deliver PD-L1 as a clinical biomarker. At the heart of this conundrum is the subjective scoring of PD-L1 IHC in most studies to date. Current standard scoring systems involve separation of epithelial and inflammatory cells and find clinical significance in different percentages of expression, e.g., above or below 1%. Clearly, an objective, reproducible and accurate approach to PD-L1 scoring would bring a degree of necessary consistency to this landscape. Using a systematic comparison of technologies and the application of QuPath, a digital pathology platform, we show that high PD-L1 expression is associated with improved clinical outcome in Triple Negative breast cancer in the context of standard of care (SoC) chemotherapy, consistent with previous findings. In addition, we demonstrate for the first time that high PD-L1 expression is also associated with better outcome in ER- disease as a whole including HER2+ breast cancer. We demonstrate the influence of antibody choice on quantification and clinical impact with the Ventana antibody (SP142) providing the most robust assay in our hands. Through sampling different regions of the tumour, we show that tumour rich regions display the greatest range of PD-L1 expression and this has the most clinical significance compared to stroma and lymphoid rich areas. Furthermore, we observe that both inflammatory and epithelial PD-L1 expression are associated with improved survival in the context of chemotherapy. Moreover, as seen with PD-L1 inhibitor studies, a low threshold of PD-L1 expression stratifies patient outcome. This emphasises the importance of using digital pathology and precise biomarker quantitation to achieve accurate and reproducible scores that can discriminate low PD-L1 expression.