The spatial localization of CD163+ tumor-associated macrophages predicts prognosis and response to therapy in inflammatory breast cancer

3086 Background: The mechanisms contributing to the aggressive biology of inflammatory breast cancer (IBC) are under investigation. A specific immune response seems to be an important driver, but the functional role of infiltrating immune cells in IBC remains unclear. Tumor-associated macrophages (TAMs) are associated with worse outcome, while CD8+ cytotoxic T cells demonstrate anti-tumor properties in breast cancer. In this study, we assessed spatial associations between CD163+ TAMs, CD8+ cells and cancer cells in IBC, using deep-learning and ecological statistics. Methods: We collected clinicopathological variables, evaluated PDL1-positivity (SP142, Ventana) and scored TILs according to the TIL working group guidelines on H&E slides for 144 IBC patients. Immunostainings for CD8 and CD163 (Hematoxylin-DAB) were done according to validated protocols. All slides were digitized, underwent virtual multiplexing and were evaluated in Visiopharm to quantify the number of DAB+ immune cells. Each immune cell was located using XY coordinates and spatial interactions were examined using a Morisita Horn Index (MHI). Tumor cell coordinates were collected using a deep-learning algorithm applied to the CD8-stained slide. This algorithm was trained in 18 images with more than 150.000 iterations (Deeplabv3+). Results: Complete pathological response (pCR) after neo-adjuvant chemotherapy was achieved by 30.6% (n= 30/98) of the patients with initially localized disease. Besides PDL1-postivity ( P= .03), infiltration with CD8+ T cells ( P= .02) and TAMs ( P= .01) also predicted pCR. However, a likelihood ratio test showed no difference between a model using CD8+ cells, TAMs or TILs. Interestingly, the colocalization of CD163+ and CD8+ cells (MHI >0.83) was associated with pCR (P= .01) and remained significant in a multivariate model (OR: 3.18; 95% CI: 1.04 – 10.6; P= .05) including TIL score, PDL1-positivity and hormone receptor (HR) status. Furthermore, a shorter disease-free survival (DFS) was associated with HR- status, no pCR and the colocalization of TAMs near tumor cells (HR: 3.3; 95% CI: 1.6 – 7.1; P= .002) in a multivariate model. The density of TAMs was not associated with outcome. Conclusions: The impact of TAMs on clinical outcome appears to depend on the spatial arrangement. The number of TAMs solely was not associated with outcome, but patients with more TAMs in proximity of the tumor cells had a worse DFS. Surprisingly, the clustering of TAMs near CD8+ cells was associated with pCR independent of the number of TAMs or TILs.