Novel Automated Quantitative Analysis (AQUA) Algorithms for Reproducible Assessment of PD1-PD-L1 Interaction and Immune Cell Subsets by Multiplex Fluorescence Immunohistochemistry in Cancer Immunotherapy Trials

摘要

PROBLEM STATEMENT: While promising clinical responses were observed in cancer patients with programmed death-1 (PD-1) receptor-targeted antibodies, their widespread use will require reliable clinical diagnostics to improve risk-benefit and facilitate reimbursement. Despite regulatory approvals, the utility of PD-L1 immunohistochemistry (IHC) tests is hotly debated due to their imperfect correlation with clinical response and subjective interpretation. Hence, there is a strong desire to develop innovative diagnostic solutions that utilize routine clinical specimens. As a first step in this direction, we developed novel AQUA image analysis algorithms that not only accurately and reproducibly quantify predictive biomarkers of immunotherapies (e.g., tumor-infiltrating lymphocytes (TIL) PD-L1, PD-1) but also their interaction densities (e.g., PD-L1 and PD-1) by multiplexed fluorescent immunohistochemistry (FIHC) on a single formalin-fixed, paraffin-embedded (FFPE) section employing an automated high-throughput workflow suitable for a diagnostic laboratory. METHOD: Our high-throughput clinical trial laboratory workflow consists of automated stainers (Biocare), Vectra 2 Image Acquisition platform (Perkin Elmer) and AQUA Image Analysis software that enables objective, quantitative, and standardized biomarker assessments. Using this workflow, we developed multiple clinical trial assays for quantitation of regulatory T cells (CD25/FoxP3), exhausted T cells (CD3/PD1), TILs (CD4/CD8; CD56/CD16), and immune checkpoint inhibitors (e.g., PD1/PD-L1) in different tumor indications (e.g., non-small cell lung cancer, gastric, diffuse large B-cell lymphoma, and melanoma). Multiplexing was achieved by sequential application and removal of same-species primary antibodies via microwave treatments. Fluorescent images were acquired on the Vectra Imaging system that acquires tissue regions of interest based on pattern recognition algorithms. Vectra images were analyzed by proprietary AQUA analysis algorithms, which remove the subjectivity of the traditional scoring system and provide more continuous and reproducible measures of protein expression in terms of biomarker intensity, percent positivity, and receptor-ligand interactions. All assays were evaluated in multiple analytical runs and patient specimens to verify reproducibility and the prevalence of biomarkers prior to clinical trial application. RESULTS: We successfully assessed the accuracy of AQUA analysis algorithms by correlating the frequencies of regulatory T cells (66% CD25+ and 24% FoxP3+) in artificially stimulated whole blood by flow cytometry (66% CD25+ and 22% FoxP3+); furthermore, the prevalence of CD25/FoxP3+ T cells ranged from 1%-10% in archival lung, gastric, and melanoma tissue specimens. In contrast, a broader range of expression, 10%-50% was observed for CD4+ and CD8+ in sequential sections of these tumor specimens. Greatest dynamic range of expression was observed for PD-L1 (1%-90%) and PD1 (1%-30%) in NSCLC patients (n=38) tested and, most importantly, a subset of these patients exhibited high levels of receptor-ligand interaction reminiscent of immune suppression. Furthermore, differential distribution of exhausted T cells (CD3+/PD1+) were observed in primary (low levels) versus secondary sites (high levels) in diffuse large B cell lymphoma patients (n=43). CONCLUSION: Availability of a broad range of automated, sensitive, quantitative, and reproducible multiplex FIHC assays with unique reportables (receptor-ligand interaction score) should aid in reliable verification of predictive biomarkers in immunotherapy trials and implementation of this automated work flow in a diagnostic laboratory setting.