PD-L1 and PD-L2 Expression in Cervical Cancer: Regulation and Biomarker Potential

PD-1/PD-L1 immune checkpoint inhibitors show potential for cervical cancer treatment. However, low response rates suggest that patient selection based on PD-L1 protein expression is not optimal. Here, we evaluated different PD-L1 detection methods and studied transcriptional regulation of PD-L1/PD-L2 expression by The Cancer Genome Atlas (TCGA) mRNAseq analysis. First, we determined the copy number of the PD-L1/PD-L2 locus by fluorescence in situ hybridization (FISH), PD-L1 mRNA expression by RNA in situ hybridization (RNAish), and PD-L1/PD-L2 protein expression by immunohistochemistry (IHC) on tissue microarrays (TMA) containing a cohort of 60 patients. Additionally, distribution of PD-L1/PD-L2 was visualized based on flow cytometry analysis of single-cell suspensions (n=10). PD-L1/PD-L2 locus amplification was rare (2%). PD-L1 mRNA expression in tumor cells was detected in 56% of cases, while 41% expressed PD-L1 protein. Discordant scores for PD-L1 protein expression on tumor cells between TMA-cores from one tumor were observed in 27% of cases. Interestingly, with RNAish, PD-L1 heterogeneity was observed in only 11% of the cases. PD-L2 protein expression was found in 53% of the cases. PD-L1 mRNA and protein expression on tumor cells were strongly correlated (p < 0.001). PD-L1 and PD-L2 protein expression showed no correlation on tumor cells (p = 0.837), but a strong correlation on cells in stroma (p < 0.001). Co-expression of PD-L1 and PD-L2 on myeloid populations was also observed with flow cytometry based t-SNE-analysis. PD-L1 and PD-L2 TCGA transcript levels correlated strongly, and were both strongly correlated with interferon gamma (IFNG) transcript levels (p < 0.0001). Importantly, patients with high PD-L1/PD-L2/IFNG transcript levels had a survival advantage over patients with high PD-L1/PD-L2 and low IFNG expression. Based on these findings, we conclude that PD-L1/PD-L2 expression in cervical cancer is mainly associated with interferon induction and not gene amplification, which makes the latter an unsuitable biomarker. The highly heterogeneous PD-L1 and PD-L2 expression patterns suggest IHC unreliable for patient selection. RNAish, in conjunction with interferon signaling evaluation, seem promising for immune checkpoint detection. These results warrant further investigation into their prognostic and predictive potential.