Abstract P5-04-04: The spatial interactions between FOXP3+ Tregs, CD8+ cytotoxic T cells and tumor cells predict response to therapy and prognosis in inflammatory breast cancer

Abstract

Abstract Background: Inflammatory breast cancer (IBC) is a rare and aggressive type of locally advanced breast cancer. A 79-gene signature, reported by our lab, is shaped by specific immune response programs and discriminates between IBC and non-IBC (nIBC). However, it remains an enigma how infiltrating immune cells are able to determine the IBC phenotype. Furthermore, the presence of immune cells like FOXP3+ Tregs and CD8+ cytotoxic T cells is associated with outcome in proliferative subtypes of breast cancer and the interaction between these cells plays a role in the functional immune response. Therefore, we aimed to assess the spatial associations between immune cells in IBC. Additionally, we used deep-learning to examine interactions between cancer and immune cells. Methodology: Immunostainings (Hematoxylin-DAB, H-DAB) were done according to well-validated protocols for CD8 (cytotoxic T-cells), FOXP3 (Tregs) and CD163 (tumor-associated macrophages, TAMs) in a large population of 134 IBC patients. All slides were digitalized and evaluated using VISIOPHARM® software, allowing virtual multiplexing. We quantified the number of DAB+ immune cells and each positive immune cell was located using XY coordinates. Spatial co-localization was examined using statistics developed for ecological studies based on point pattern and quadrant analysis. TILs were scored according to the TIL working group guidelines on H&E slides. Tumor cell coordinates were collected using a deep-learning algorithm applied to the CD8 stained slide. To perform deep-learning, we aligned two consecutive slides: one PanCK stained slide and one slide stained for CD8 (H-DAB). Using virtual multiplexing and the PanCK staining, we determined the tumor regions on the H-DAB stained slide. Subsequently, 18 images were incorporated to train the algorithm with more than 150.000 iterations (Deeplabv3+), after which the algorithm was evaluated in a test set of 12 images, approved and applied to all images to locate the tumor cells. Results: Most of the patients presented with a hormone receptor (HR) positive carcinoma (60.6%, n= 82/132). The presence of distant disease, HR status and TIL score were associated with overall survival (OS), but the density of the different immune cells or the CD8/FOXP3 ratio was not. However, using an effector index we demonstrated that patients with more FOXP3+ cells in a radius of 30 μm surrounding a CD8+ cell had a significant worse outcome (Median OS: 2.7 vs. 6.3 years, P= .01) and this remained significant in a multivariate model (HR: 2.85, P< .001). Complete pathological response (pCR) after neo-adjuvant chemotherapy was achieved by 28.7% (n= 27/94) of the patients with initially localized disease. Infiltration with CD8+ T cells (P= .005), TAMs (P= .008) and TILs (P= .002) predicted pCR, but a likelihood ratio test showed no difference between a model using CD8+ cells, TAMs or TILs. Interestingly, pCR was less often achieved in patients that had colocalization of FOXP3+ cells near the tumor cells (P= .003). This colocalization was determined using a Morista-Horn index and independent of the number of Tregs, CD8+ cells or TILs. The deep-learning algorithm had a mean dice coefficient in the test set of 0.83, indicating a good overlap between the tumor area determined by the AI on the CD8 slides and the area on the PanCK stained slides. Conclusions: We have created a deep learning algorithm that adequately detects IBC tumor cells on H-DAB stained images and show, in a large cohort of 132 patients, that the negative impact of Tregs appears to depend on the spatial arrangement. While solely the number of Tregs is not associated with pCR or OS, patients with FOXP3+ Tregs that cluster together near CD8+ cytotoxic T cells had a worse outcome and pCR was achieved more often in patients with less Tregs near the tumor cells. Citation Format: Christophe Van Berckelaer, Charlotte Rypens, Steven Van Laere, Koen Marien, Pieter-Jan van Dam, Peter Vermeulen, Luc Dirix, Mark Kockx, Cecile Colpaert, Peter van Dam. The spatial interactions between FOXP3+ Tregs, CD8+ cytotoxic T cells and tumor cells predict response to therapy and prognosis in inflammatory breast cancer [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr P5-04-04.