Artificial intelligence–powered tumor-infiltrating lymphocytes analyzer to reveal distinct immune landscapes in breast cancer by molecular subtype and HER2 score

• Published in: Journal of clinical oncology Vol. 41; no. 16_suppl; p. 1049
• Main Authors: Shin, Seunghwan, Cho, Soo Youn, Cho, Eun Yoon, Kim, So-Woon, Jung, Minsun, Song, Seung-Geun, Lim, Yoojoo, Cho, Soo Ick, Kim, Sukjun, Park, Gahee, Song, Heon, Ma, Minuk, Yoo, Donggeun, Ock, Chan-Young
• Format: Journal Article
• Language: English
• Published: 01.06.2023
• ISSN: 0732-183X; 1527-7755
• DOI: 10.1200/JCO.2023.41.16_suppl.1049

1049 Background: Tumor-infiltrating lymphocytes (TIL) reflect an ongoing anti-tumor immune response in the tumor microenvironment (TME) and have been studied as a predictive as well as prognostic biomarker in breast cancer (BC). The immune profile differs depending on BC molecular subtypes, with triple-negative BC (TNBC) and HER2-positive BC more frequently infiltrated by higher numbers of TIL than hormone receptor (HR)-positive tumors. In this study, we analyze TIL and immune phenotype by molecular subtype in BC and explore differences based on HER2 scores using an AI-powered TIL analyzer. Methods: A total of 1,973 cases of BC with molecular subtype, HER2 immunohistochemistry (IHC) or in situ hybridization (ISH) results, and TIL analysis from H&E slides from TCGA (n=1021), Samsung Medical Center (SMC, n=663), and KyungHee Medical Center (KHMC, n=289) were included in the analysis. The density of TIL were spatially analyzed from H&E WSI, and the immune phenotype (IP) was analyzed for each 1 mm2-sized patch. Tumor microenvironment (TME) was categorized into three types: inflamed (TIL distributed intratumorally), Immune-excluded (TIL excluded, out of cancer stroma), immune-desert (scant TIL in TME), and the score for each IP was calculated as the percentage of the total patches in WSI. Results: TNBC showed the highest TIL density in both cancer epithelium (214.5 ± 299.7/mm2) and stroma (1133.3 ± 1245.9/mm2), and a high percentage of inflamed IP (35.7%), IS (28.4 ± 28.1) than other subtypes (inflamed IP: 17.5%, IS 16.2±20.2, both p<0.001). HER2-positive BCs showed higher sTIL density (904.0 ± 901.7/mm2) compared to HR- positive BC (616.7 ± 794.1/mm2) but were associated with a higher percentage of immune-excluded IP (71.2%) and IES (57.6 ± 24.6). HR-positive BC had a high percentage of immune-desert IP (30.4%). In both HR-positive and negative BCs, higher IES was observed as the HER2 score increased. HER2-positive BC significantly showed a higher proportion of cancer stroma in the cancer region from WSI (Stromal ratio, 0.682 ± 0.149 vs. 0.578 ± 0.176, p<0.001). Conclusions: Different TIL densities and IP were seen in BC for each molecular subtype, pointing to a distinct immune landscape. HER2 score was associated with immune-excluded IP despite having higher TIL density, which should be considered for immunotherapy in BC.