Exploring expression levels of HER2, HER3, MET, Claudin18.2, and MUC16 across 16 cancer types using an artificial intelligence-powered immunohistochemistry analyzer.

Abstract

3135 Background: Quantifying tumor-associated antigen (TAA) expression levels on tumor cells (TC) from immunohistochemistry (IHC) stained slides has been widely used to predict response to novel anti-tumor agents. However, obtaining the precise quantification of TAA expression levels from various cancer types is still challenging. To quantify five targets of interest, we applied an artificial intelligence (AI)-powered analyzer to 16 cancer types. Methods: AI-powered whole slide image (WSI) analyzers were developed using IHC positive or negative tumor cells (TC) from PD-L1 22C3 pharmDx and HER2 4B5 datasets, annotated by pathologists. The universal model was trained by IHC-positive or negative TCs from all datasets, while the second model was trained by the intensity of positive (from 1+ to 3+) or negative TCs from the HER2 dataset. Both AI models were applied to the pan-cancer tissue microarray dataset (n = 1,370) with MUC16, CLDN18.2, MET, HER3, and HER2 IHC staining, respectively. The expression levels of those TAAs were calculated with tumor proportion score (TPS) or ASCO/CAP scoring criteria for HER2 in 16 cancer types including biliary tract, breast, colorectum, esophagus, head and neck, kidney, liver, lung, ovary, bladder, pancreas, prostate, stomach, thyroid, cervical, and endometrial cancer. Results: Based on the TPS of 1% or more, the positivity of MUC16 and CLDN18.2 were most frequently observed in ovary (51.6%) and kidney (24.2%), respectively, and MET and HER3 in colorectum (63.8%; 40.5%). The five most highly expressed tumors and their frequency of TPS≥1% are shown. HER2 3+ tumor cells were most highly expressed in the breast, followed by the bladder. Conclusions: This study describes the TAA expression levels in various tumors using an AI-powered IHC analyzer. This method could be useful for exploring target cancer types and predicting response to novel TAA targeted agents. [Table: see text]