Background Lung cancer, the worldwide leading cause of cancer-related deaths, is expected to account for over 127,000 deaths in the United States in 2023. Advances in the management of NSCLC are focused on effective patient selection for anti-PD-(L)1 immune checkpoint inhibition therapy, with multiple immunohistochemical (IHC) biomarkers in clinical use today. However, The Blueprint PD-L1 IHC Assay Comparison Project demonstrated comparable tumor cells staining, variable immune cell staining, and a lack of interchangeability of the assay results. In this study, we demonstrate the use of PictorLabs virtual staining as a pathologist’s interpretive aid; rendering virtual H&E, CD45, and pan cytokeratin stains from the same tissue section in which PD-L1 IHC staining is performed.

Methods Using a standard slide scanner (AxioScan.Z1, Zeiss), four-channel autofluorescence images were captured from unstained sections (4um thick) of NSCLC. The virtual staining was performed by three deep neural networks trained as a conditional GAN using accurately co-registered patches of paired images and output WSIs of virtual H&E, CD45 (LCA), and pan cytokeratin (AE1/AE3). The slides were decoverslipped and stained with VENTANA PD-L1 assay (SP263, Roche). Image registration and viewing was performed on the PictorLabs Deepstain platform.

Results The PictorLabs technology successfully created whole slide image virtual stains for H&E, CD45, and pan cytokeratin. Two board-certified pathologists confirmed the consistency and accuracy of the virtual stains compared to traditional staining methods. The Deepstain platform allowed for simultaneous viewing, annotation, and manual scoring of PD-L1 staining across various cellular subpopulations, within the context of morphology provided by the virtual H&E result.

Conclusions This study demonstrates the feasibility of using PictorLabs virtual staining to provide morphological, spatial, and cell phenotypic context to IHC assays. This ability to overlay various stains for simultaneous viewing offers a significant advancement in the analysis of the tumor microenvironment alongside predictive biomarkers. This approach has the potential to enhance the diagnostic process and opens new avenues for the development of treatment strategies for NSCLC, potentially leading to more immunologically refined methods to combat this deadly disease.