Use of causal modeling to uncover cell-cell communication dynamics in the tumor microenvironment during anti-PD-1 therapy in breast cancer patients.

Abstract

e14585 Background: Immune checkpoint blockade (ICB) targeting PD-1/PD-L1 is crucial in treating breast cancer. Despite clinical success, the mechanisms by which ICB therapies reshape the tumor microenvironment (TME) and enhance anti-tumor activity remain unclear. Recent studies reveal that anti-PD-1 therapies broadly alter TME cells beyond PD-1+ T cells, highlighting extensive cell-cell communication (CCC) as a key factor. Understanding CCC changes in response to anti-PD-1 therapy can illuminate its mechanisms of action (MOA) and TME dynamics. This study applies causal inference methodology to investigate CCC between T cells and non-T cells in the TME of breast cancer patients receiving anti-PD-1 therapy. Methods: We analyzed single-cell RNA-seq data from 31 breast cancer patients pre- and on-treatment with anti-PD-1 (pembrolizumab) reported by Bassez et al., 2021. We identified differentially expressed genes (DEGs) across cell types. We applied the instrumental variable method to identify causal relationships between signals in T-cell and non-T-cell DEGs to uncover treatment-induced CCC. We further searched for ligand-receptor pairs mediating CCC and identified gene expression modules (GEMs) regulated by these ligand-receptors. Finally, we constructed a CCC network from T to non-T cells. Results: Anti-PD1 treatment induced broad gene expression changes in diverse cell populations. Major T cell pathways influenced by anti-PD-1 therapy include NF-κB, interferon-γ and interleukins. CD4 + and CD8 + exhausted cells, expressing high levels of PDCD1 (encoding PD-1), exhibited distinct activated pathways. We identified the CCC network from T cells to non-T cells via ligand-receptor interactions. For example, anti-PD1 treatment activated cellular stress, apoptosis, and pro-inflammatory cytokine signaling pathways in T cells, altering the expression of a GEM that included TSC22D3 and TXNIP . This initiated signaling to myeloid cells via RPS19 - C5AR1 , leading to NF-κB activation. CD4+ exhausted T cells primarily signaled to monocytes, enhancing helper functions to recruit and activate monocytes, thereby promoting immune regulation and amplification. In contrast, CD8+ exhausted T cells primarily interacted with macrophages, intensifying cytotoxic responses that facilitated effective tumor-cell killing. Conclusions: Our analyses provide insights into the dynamic interplay of cells within the TMEs during anti-PD-1 therapy. These findings could facilitate the identification of new biomarkers for predicting heterogeneous treatment responses to anti-PD-1 regimens, potentially enhancing the design and customization of immunotherapeutic strategies for breast cancer patients. Finally, this study demonstrated the utility of causal inference methodologies for mechanistic studies of CCC.