Abstract 5443: Pathomics reveals the molecular and immune evolution from lung preneoplasia to invasive adenocarcinoma

Abstract

Abstract Introduction: Atypical adenomatous hyperplasia (AAH) is the only recognized preneoplasia of lung adenocarcinoma, which can progress to adenocarcinoma in situ (AIS), minimally invasive adenocarcinoma (MIA) and eventually to invasive adenocarcinoma (ADC). A more complete understanding of the early carcinogenesis of lung cancer is critical for lung cancer early detection and interception. However, studying these lung cancer precursors is challenging because these lesions are often insufficient for molecular and immune profiling. Artificial intelligence (AI)-based studies on H&E histopathology images, termed pathomics, have achieved substantial progress in revealing heterogeneous phenotypic characteristics of various cancers. However, pathomics on lung precancerous progression and their correlation with genomic features remain underexplored. Methods: We curated FFPE H&E slides from two ethnic groups, including the Caucasian cohort containing 46 lesions with 170 regions of interest (ROI) (74 AAH, 10 AIS, 21 MIA, and 65 ADC) and the Asian cohort containing 128 lesions with 369 ROIs (59 AAH, 84 AIS, 77 MIA, and 149 ADC). We adopted the expert-in-the-loop strategy to develop a deep learning pipeline to segment and annotate the cells within ROI into three categories: epithelial, lymphocyte, and other. Next, we measured the ratio and density of epithelial cells and lymphocytes inside each ROI as pathomics features. Finally, we interrogated ROI-level features and examined their correlation with molecular and immune features in the Asian cohort. Results: We observed a progressive increase in the ratio and density of epithelial cells and a progressive decrease in the ratio and density of lymphocytes defined by the AI model from AAH to AIS, MIA, and ADC, consistent with the same trends defined by T cell receptor (TCR) sequencing and multiplex immunofluorescence (mIF). When correlating pathomics features with molecular/immune features, the epithelial cell ratio exhibited prominent positive correlations with the frequency of allelic imbalance (rho=0.588, p=4.71e-13) and nonsynonymous mutation burden (rho=0.453, p=1.03e-7). In contrast, the lymphocyte ratio showed a notable negative correlation with copy number variation burden (rho=-0.412, p=1.61e-6). Conclusion: Employing AI tools to analyze HE images of lung precancerous lesions, we revealed that molecular and immune evolution during early lung carcinogenesis is consistent with the results from complicated, time-consuming, and expensive molecular/immune profiling, which requires a large number of tissue specimens, highlighting the potential of pathomics in the study of cancer biology, particularly in diseases having limited tissue specimens. Citation Format: Pingjun Chen, Frank Rojas, Xin Hu, Junya Fujimoto, Alejandra Serrano, Bo Zhu, Lingzhi Hong, Rukhmini Bandyopadhyay, Muhammad Aminu, Maliazurina B. Saad, Morteza Salehjahromi, Sheeba J. Sujit, Neda Kalhor, Harvey I. Pass, Andre L. Moreira, Ignacio I. Wistuba, Don L. Gibbons, John V. Heymach, Luisa M. Solis Soto, Jianjun Zhang, Jia Wu. Pathomics reveals the molecular and immune evolution from lung preneoplasia to invasive adenocarcinoma. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5443.