No abstract

In the past decade, advances in precision oncology have resulted in an increased demand for predictive assays that enable the selection and stratification of patients for treatment. The enormous divergence of signalling and transcriptional networks mediating the crosstalk between cancer, stromal and immune cells complicates the development of functionally relevant biomarkers based on a single gene or protein. However, the result of these complex processes can be uniquely captured in the morphometric features of stained tissue specimens. The possibility of digitizing whole-slide images of tissue has led to the advent of artificial intelligence (AI) and machine learning tools in digital pathology, which enable mining of subvisual morphometric phenotypes and might, ultimately, improve patient management. In this Perspective, we critically evaluate various Al-based computational approaches for digital pathology, focusing on deep neural networks and ‘hand-crafted’ feature-based methodologies. We aim to provide a broad framework for incorporating AI and machine learning tools into clinical oncology, with an emphasis on biomarker development. We discuss some of the challenges relating to the use of AI, including the need for well-curated validation datasets, regulatory approval and fair reimbursement strategies. Finally, we present potential future opportunities for precision oncology.

Recent strategies targeting the interaction of the programmed cell death ligand-1 (PD-L1, B7-H1, CD274) with its receptor, PD-1, resulted in promising activity in early phase clinical trials. In this study, we used various antibodies and in situ mRNA hybridization to measure PD-L1 in non-small cell lung cancer (NSCLC) using a quantitative fluorescence (QIF) approach to determine the frequency of expression and prognostic value in two independent populations. A control tissue microarray (TMA) was constructed using PD-L1-transfected cells, normal human placenta and known PD-L1-positive NSCLC cases. Only one of four antibodies against PD-L1 (5H1) validated for specificity on this TMA. In situ PD-L1 mRNA using the RNAscope method was similarly validated. Two cohorts of NSCLC cases in TMAs including 340 cases from hospitals in Greece and 204 cases from Yale University were assessed. Tumors showed PD-L1 protein expression in 36% (Greek) and 25% (Yale) of the cases. PD-L1 expression was significantly associated with tumor-infiltrating lymphocytes in both cohorts. Patients with PD-L1 (both protein and mRNA) expression above the detection threshold showed statistically significant better outcome in both series (log-rank P ¼ 0.036 and P ¼ 0.027). Multivariate analysis showed that PD-L1 expression was significantly associated with better outcome independent of histology. Measurement of PD-L1 requires specific conditions and some commercial antibodies show lack of specificity. Expression of PD-L1 protein or mRNA is associated with better outcome. Further studies are required to determine the value of this marker in prognosis and prediction of response to treatments targeting this pathway. Non-small cell lung cancer (NSCLC) is the leading cause of cancer-related death in the United States. The overall survival (OS) for metastatic NSCLC is dismal with 5-year survival of o5% and for patients with early stage NSCLC the 5-year survival is o50%. 1 Over the past decade, identification of several oncogenic driver mutations have helped improve the outcomes in certain subtypes of patients with NSCLC. 2 However, a majority of the patients with lung cancer do not have an actionable molecular aberration. Other treatment approaches, such as immune therapies, are being investigated in clinical trials. Programmed cell death-1 (PD-1) pathway is a major immune checkpoint by which tumors suppress lymphocyte function within the tumor microenvironment, and antibody blockade of PD-1 with its ligands (B7-H1/PD-L1 and B7-DC/PD-L2) showed promising activity in several malignancies. 3 In particular, blocking antibodies against PD-1 and PD-L1 have shown clinical activity in NSCLC. 4 Preliminary data suggest that tumor PD-L1 protein expression on human cancers using chromogenic-based immunohistochemistry (IHC) in formalin-fixed paraffinembedded tissue samples (FFPE) may predict clinical response to PD-1/PD-L1 directed therapy. 4,5 There are limited data on the prevalence and the prognostic role of PD-L1 expression in NSCLC. Data from small previously ...

Importance Early phase trials with monoclonal antibodies targeting PD-1/PD-L1 have demonstrated durable clinical responses in patients with NSCLC, however, current assays for the prognostic/predictive role of tumor PD-L1 expression are not standardized with respect to either quantity or distribution of expression. Objective In this study, we demonstrate PD-L1 protein distribution in NSCLC tumors using both conventional immunohistochemistry (IHC) and quantitative immunofluorescence (QIF), and compare results obtained using two different PD-L1 antibodies. Design PD-L1 was measured using two rabbit monoclonal antibodies (E1L3N and SP142) in 49 NSCLC whole tissue sections and a corresponding tissue microarray with the same 49 cases. Mel624 cells stably transfected with PD-L1, as well as Mel624 parental cells and human term placenta were used as controls and for antibody validation. PD-L1 protein expression in tumor and stroma was assessed using chromogenic IHC and the AQUA® method of QIF. Tumor-infiltrating lymphocytes (TILs) were scored in hematoxylin/eosin stained slides using current consensus guidelines. The association between PD-L1 protein expression, TILs, and clinico-pathological features were determined. Setting NSCLC resections were all performed at Yale New Haven Hospital. Participants NSCLC resection cases from 2011–2012 were collected retrospectively from the Yale Thoracic Oncology Program Tissue Bank in Yale Pathology based on tissue availability. Main Outcome Measure PD-L1 expression discordance or heterogeneity using DAB and QIF was the main outcome measure selected prior to performing the study. Results Using chromogenic IHC, both antibodies showed fair to poor concordance. QIF showed that PD-L1 expression using both PD-L1 antibodies was heterogeneous. Using QIF, the scores obtained with E1L3N and SP142 for each tumor were significantly different according to nonparametric-paired test (p <0.001). Assessment of 588 serial section fields of view by QIF showed discordant expression at a frequency of 25%. Expression of PD-L1 using both E1L3N and SP142 was correlated with high TILs (p = 0.007 and p = 0.021). Conclusions Objective determination of PD-L1 protein levels in NSCLC reveals heterogeneity within tumors and prominent inter-assay variability or discordance. This could be due to different antibody affinities, limited specificity, or distinct target epitopes. Efforts to determine the clinical value of these observations are underway.

No abstract