Abstract PS1-06: Intratumor tumor-infiltrating lymphocytes (iTILs) and spatial distribution pattern of stromal TILs (sTILs) evaluated by HALO AI are prognostic indicators of triple-negative breast cancer

Abstract Introduction: A high density of tumor-infiltrating lymphocytes (TILs) has been recognized as an indicator of favorable prognosis of triple-negative breast cancers (TNBCs), and measurement of TILs infiltrating in the stroma of peripheral zone of cancer tissue are considered to be of the optimal clinical significance. However, some studies on other cancer types have indicated that TIL densities in both stromata of peripheral and central zones and in tumor cell nests were associated with prognosis. In this study, we evaluated the density and spatial distribution pattern of TILs manually and with an artificial intelligence (AI) modality and investigated their prognostic implication. Materials and methods: We evaluated 70 surgically resected TNBC specimens from patients who did not receive neoadjuvant chemotherapy. CD3+ TILs and CD8+ TILs were detected by immunohistochemistry. The manual evaluation was performed by counting the total number of CD3+- and CD8+- TILs in peripheral stroma (psTILs), in central stroma (csTILs), and in tumor cell nests (iTILs) in five high-power fields (HPFs) on photomicrographic images. The evaluation with AI was performed by calculating densities of these TILs in 10 tiles (one tile: 500 x 500 μm square) after deep-learning of annotated areas of tumor cell nests, stroma, and other regions on digital slide glasses with HALO AI (Indica Labs, NM, U.S.A.). With AI, the peripheral zone was defined as the region centered on the boundary separating the host tissue from the cancer tissue with a width of 1 mm, and the central zone as the tissue inside the peripheral zone. psTILs, csTILs, and both peripheral and central iTILs (piTILs and ciTILs) were measured. With reference to receiver operating characteristic curves, the optimal cut-off values between high and low TILs were determined to discriminate clinical outcome of the patients. Spatial distribution patterns of TILs were also defined: infiltrated (high psTILs/high csTILs) and non-infiltrated patterns. Recurrence-free survival and overall survival analyses were performed with Cox proportional hazards model analyses. Results: Median follow-up duration was 9.4 years, and 14 patients suffered recurrences and 13 patients died. In manual measurements, cut-off values in totally five HPFs were 1000, 300, 700, 300, 10, and 5 in CD3+ psTILs, CD8+ psTILs, CD3+ csTILs, CD8+ csTILs, CD3+ iTILs, and CD8+ iTILs, respectively. In AI measurements, cut-off values in totally 10 tiles were 1,500, 1,000, 500, 250, 30, and 15 in CD3+ psTILs, CD8+ psTILs, CD3+ csTILs, CD8+ csTILs, CD3+ ciTILs, and CD8+ ciTILs, respectively. By Cox univariate analyses, low CD3+ iTILs and low CD8+ iTILs with AI measurements showed higher risks [recurrence: hazard ratio(HR) 3.11, 95% CI 1.01-9.59, P = 0.047 for CD3+ ciTILs; HR 4.30, 95% CI 1.18-15.7, P = 0.027 for CD8+ ciTILs; HR 5.88, 95% CI 1.30-26.6, P = 0.021 for CD3+ csTILs; death: HR 3.57, 95% CI 1.16-11.0, P = 0.026 for CD3+ ciTILs; HR 4.30, 95% CI 1.18-15.7, P = 0.027; HR 2.23, 95% CI 0.99-10.5, P = 0.052; HR 3.98, 95% CI 1.09-14.5, P = 0.036 for CD3+ csTILs], but other psTILs and csTILs did not show prognostic significance. With regard to spatial distribution of CD3+ TILs, non-infiltrated patterns showed significantly worse prognoses than infiltrated pattern (recurrence: HR 8.52, 95% CI 1.11-65.7, P = 0.039; death: HR 8.98, 95% CI 1.16-69.3, P = 0.035). Similar results were shown by manual evaluations. The CD3+ ciTILs, CD8+ ciTILs, CD3+ csTILs, and spatial distribution pattern with AI measurements were significantly or nearly significant prognostic values independently of patient age, pT, pN, stage, and adjuvant chemotherapy. Conclusions: Evaluation of TILs not only in peripheral stroma but also in central stroma and tumor cell nests appeared important in TNBCs. iTILs and spatial distribution pattern of TILs might be applicable to discern patients who are eligible for dose de-escalation from those who require standard perioperative systemic therapy to TNBC. Citation Format: Makiko Fukumura-Koga, Takanori Watanabe, Keita Kouzu, Kimi Kato, Kosuke Miyai, Tamio Yamasaki, Takahiro Einama, Hideki Ueno, Yoji Kishi, Kimiya Sato, and Hitoshi Tsuda. Intratumor tumor-infiltrating lymphocytes (iTILs) and spatial distribution pattern of stromal TILs (sTILs) evaluated by HALO AI are prognostic indicators of triple-negative breast cancer [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr PS1-06.