Causal modeling reveals cell-cell communication dynamics in the tumor microenvironment during anti-PD-1 therapy in breast cancer patients.

Immune checkpoint blockade (ICB) targeting PD-1/PD-L1 axis has transformed breast cancer treatment, yet how therapy reshapes the tumor microenvironment (TME) through cell-cell communication (CCC) remains unclear. Existing CCC inference methods relying on correlations have difficulty distinguishing genuine signaling from confounded associations. Here, we present a causal inference framework that uses single-cell data and leverages treatment as an instrumental variable to identify genuine CCC networks, referred to as scIVCCC, which infers causal signal transduction across cell types. Applying scIVCCC to single-cell RNA-seq data from 31 breast cancer patients before and after anti-PD-1 therapy, we constructed causal CCC networks linking exhausted T cells to tumor-associated macrophages (TAMs). Our analysis reveals a dual role of T cell-macrophage crosstalk: CD4+ and CD8+ exhausted T cells drive anti-tumor M1-like TAMs activation via TNF-TNFRSF1A, TNFSF14-LTBR, and ICAM1-ITGAL/ITGB2. Conversely, they also induce immunosuppressive M2-like polarization through pathways such as TNF-TNFRSF1B (TNFR2), TNFSF14-TNFRSF14 (HVEM), and RPS19-C5AR1, which likely contribute to therapeutic resistance. Our causal modeling suggests that receptors within these networks, such as C5AR1, TNFR2, and CSF1R, may serve as potential candidates for combination therapies to enhance anti-PD-1 efficacy. Collectively, these findings demonstrate that scIVCCC offers a robust framework for dissecting treatment-induced CCC dynamics and prioritizing actionable targets for clinical translation.