Artificial intelligence (AI)-powered spatial analysis of macrophages in tumor microenvironment and its association with interferon-gamma (IFNG) signature and immune phenotype (IP) in pan-cancer dataset

2621 Background: Macrophages are known to play a crucial role in cancer progression, but few studies have reported macrophage spatial distribution and its association with genomic signatures and tumor-infiltrating lymphocytes (TIL). Here, we applied an AI-powered Whole Slide Image (WSI) analyzer in The Cancer Genome Atlas (TCGA) pan-cancer dataset to show correlation between specific localization of macrophage and anti-tumor activity, which is reflected by IFNG signature and IP based on spatial TIL density. Methods: For spatial macrophage analysis in TCGA pan-cancer dataset, we used Lunit SCOPE IO, an AI-powered H&E WSI analyzer, which identifies and quantifies various classes of cells such as tumor cell, macrophage, and TIL within cancer or stroma area. IPs were defined as follows: inflamed IP (IIP) as high intratumoral TIL (iTIL) and stromal TIL (sTIL); immune-excluded IP (IEP) as low iTIL and high sTIL; immune-desert IP (IDP) as low TIL overall. The immune cell profiling and quantification of the IFNG signature score were analyzed by using CIBERSORT and RSEM. Results: In TCGA dataset (N = 7,402), median macrophage density in total TME area (tMP) was 19.9 /mm2 (interquartile range [IQR] 9.0-44.9), and it was highly enriched in lung adenocarcinoma (median 65.1, IQR 34.3-110.0), lung squamous cell carcinoma (61, 32.4-101.3), and head and neck squamous cell carcinoma (39.4, 19.6-77.3). Based on IP classification, tMP was highly enriched in IIP (37.8, 18.7-73.8) compared to IEP (25.5, 13.8-48.6) and IDP (8.9, 4.4-17.1). Interestingly, spatial analysis showed that intratumoral macrophage density (iMP) in IIP was highly enriched compared to the others (mean fold change 2.57) and the difference of stromal macrophage density (sMP) among IPs was relatively smaller (mean fold change x1.85). Classified into four groups based on median values of iMP and sMP, high-iMP/high-sMP group had the highest proportion of IIP (47.3% vs 18.7%, p < 0.001), whereas low-iMP/high-sMP group had the highest one of IEP (54.9% vs 32.3%, p < 0.001) compared to the others, respectively. Moreover, the high-iMP/high-sMP group had the highest level of IFNG signature score among 4 groups (x2.13 times higher, p < 0.001). Notably, both iMP and sMP were significantly correlated with CD68 expression (rho 0.292 and 0.287) but CSF1R expression was more strongly correlated with iMP (rho 0.259) compared to sMP (rho 0.078). Conclusions: IIP, IFNG signature, and CSF1R expression are more strongly correlated with intratumoral macrophage density versus stromal macrophage density, suggesting a relationship between macrophage location and function in the TME warranting further investigation.