AI-powered analysis of the tumor immune microenvironment in lung squamous cell carcinomas with and without neoadjuvant therapy.

Abstract

e20017 Background: Chemotherapy (CT) and radiotherapy (RT) are known modulators of the local and systemic immune response with stimulating and inhibiting effects. Neoadjuvant therapy, which is frequently applied in lung squamous cell carcinomas (LUSC), permits assessment of CT and RT effects on the tumor immune microenvironment (TIME). We used an AI-based approach to perform analyses of immune cell compositions in tumor samples of neoadjuvantly treated and untreated LUSC. Methods: Digitized H&E whole slide images of 19 untreated LUSC (naiveLUSC) and 19 matched LUSC neoadjuvantly treated with CT, RT, or a combination (neoLUSC) were analyzed. Aignostics' algorithms for nucleus detection as well as foundation model RudolfV-based cell classification (carcinoma cell, epithelial cell, fibroblast, lymphocyte, plasma cell, macrophage, granulocyte, other) and tissue segmentation (carcinoma, stroma, vessel, epithelium, blood, necrosis, other) models were applied to all slides and partially refined on the dataset. Model performance was visually validated by a board-certified pathologist. Whole tumor regions (WTRs) were manually outlined to focus analyses on tumor-relevant areas. Spatial TIME characteristics were computed for the WTR from the combined model predictions and analyzed via Mann-Whitney-U-tests. Results: Tissue segmentation analyses revealed no differences for carcinoma, tumor-associated stroma, or necrosis extent within the WTR, except for a trend towards a decrease of relative vessel area for neoLUSC compared to naiveLUSC samples (0.226 % ± 0.039 (standard error of the mean, SEM), vs. 0.301 % ± 0.034, p=0.058). Tumor infiltrating immune cells were separately assessed in carcinoma areas and tumor-associated stroma areas within the WTR. Macrophages and granulocytes did not significantly differ between naiveLUSC and neoLUSC. In contrast, immune cells of lymphoid lineage were significantly decreased in the neoLUSC group with a lower lymphocyte density (0.00107 cells/µm 2 ± 0.00020 (SEM) vs. 0.00067 cells/µm 2 ± 0.00023 (SEM), p=0.033) and percentage (15.48 % ± 1.90 (SEM) versus 11.62 % ± 2.66 (SEM), p=0.047) in the tumor-associated stroma. The most pronounced difference was observed for plasma cells with a decrease in density (0.00101 cells/µm 2 ± 0.00016 (SEM) vs. 0.00049 cells/µm 2 ± 0.00012 (SEM), p=0.006) and percentage (14.84 % ± 1.63 (SEM) versus 9.95 % ± 1.96 (SEM), p=0.015) in the tumor-associated stroma of neoLUSC samples. Conclusions: Our results indicate that CT and RT contribute to a restriction of anti-tumor immune response by decreasing lymphocytes and plasma cells in the tumor-associated stroma including potential hindrance of immune cell recruitment due to less vascularization. Next steps include an in-depth analysis of spatial effects on the TIME and the impact on prognosis to further dissect the interactions of CT, RT and TIME in LUSC.