Association of PD-L1 expression by immunohistochemistry and gene microarray with molecular subtypes of ovarian tumors

Abstract

Identifying patients who respond to immune checkpoint blockade (ICB) is a significant challenge in oncology. While PD-L1 expression by immunohistochemistry (IHC) is the current diagnostic gold standard for patient selection, it nevertheless does not capture all patients who may respond to ICB. Recent gene expression studies in high-grade serous ovarian carcinoma have defined an immunoreactive molecular subtype that has a measurable difference in patient survival compared with non-immunoreactive subtypes, but no studies have yet demonstrated its impact on predicting response to ICB. As a step toward establishing the predictive value of gene expression classifiers in ICB, we assessed the relationship between PD-L1 IHC and molecular subtypes of ovarian epithelial cancer. This was done by analyzing a total of 93 tissue specimens from patients with stage III and IV disease, and comparing PD-L1 IHC with gene expression by Agilent microarrays using TCGA-defined subtypes. We showed that ovarian tumors with elevated IHC PD-L1 expression are most strongly associated with immunoreactive subtype as compared with other molecular subtypes, reaching statistical significance against differentiated (Dunn's test, 33.39, p = 0.0003) and mesenchymal (39.63, p < 0.0001) subtypes. Comparing PD-L1 scoring with CPS vs. TPS showed similar trends, but with stronger correlation strength when using CPS (Kruskal–Wallis, H = 27.52, p < 0.0001), as opposed to TPS (H = 25.04, p < 0.0001). Interestingly, while PD-L1 gene expression by microarray was significantly increased in the immunoreactive subtype (H = 20.25, p = 0.0002), it showed a positive but relatively poor correlation to IHC. Overall, the results demonstrate potential value in use of the molecular classifier to select patients for ICB, pending further studies that assess its ability to predict treatment outcomes. In the future, integration of cellular, protein, and genomic biomarkers in the tumor and tumor microenvironment may improve current methods of predicting treatment response.