Abstract PS4-01-30: Identification and characterization of distinct chemoimmunomodulatory states in breast cancer via unsupervised transcriptomic analysis of pre- and post-treatment specimens

Abstract Background: PD-1 immune checkpoint inhibitors (ICIs) have reshaped cancer treatment paradigms. While PD-1 ICIs displays monotherapy efficacy in some tumor types, others, such as breast cancer (BC), require a synergistic combination of ICIs and chemotherapy for clinical benefit. Beyond cytotoxicity, chemotherapy synergize with ICIs via chemoimmunomodulation (CIM), which comprise the immunomodulatory signals elicited by chemotherapy. While many studies have examined molecular features influencing chemoresistance, few have focused on the molecular features that influence CIM, despite the overlap between CIM and the hallmarks of response to PD-1 ICIs. This paucity may be driven by inadequate means for phenotypically delineating CIM and presents a challenge to selection of the most synergistic agents for PD-1 ICIs for a given patient. Methods: We leveraged transcriptomic data from 20 pre-/post-treatment specimens obtained from BC patients (N = 2 reps/sample) treated with neoadjuvant chemotherapy, including anthracycline, cyclophosphamide, taxanes, and platinating agent, to identify CIM induction profiles. CIM induction was assessed using Δ log2(TPM+1) values for 377 genes obtained from gene sets and literature related to autophagy, lysosomes, cellular stress, and immune function. We developed a novel iterative k-means-based unsupervised clustering approach to identify discrete CIM induction states using CIM gene induction values. Overrepresentation analysis was used to identify transcriptomic programs induced in each CIM induction states. Differential gene expression (DEG) analysis and inferential statistics were used on matched pre-treatment specimens to evaluate baseline differences between states. The abundances of immune cell types were deconvoluted using CIBERSORT. Results: Our iterative unsupervised approached identified two discrete states: one with high (H-CIM; N = 10) and one with low CIM (L-CIM; N = 10) induction. Only one replicate from patients was separately clustered. Stability, robusticity, and the binary nature of the clusters were confirmed using standard cluster evaluation metrics. There were no significant differences in clinical characteristics or treatments between states. The H-CIM state had induction of antigen presentation, upregulation of T-cell populations, and phagocytosis transcriptomic programs while the L-CIM group was primarily characterized by the induction of MYC-governed reactive oxygen species (ROS) detoxification and proteostasis programs (FDR-adjusted P (adjP) < 0.05). Baseline DEG analysis identified upregulated immunosuppressive features such as S100A9 (Fold Change (FC) = 8.9; adjP = 0.04) and HP (FC = 21.2; adjP = 0.008) in the L-CIM group which was supported by a significantly higher baseline amount of M2 macrophages (P < 0.001). Proteostasis safeguards SEC61G (FC = 3.3; adjP = 0.04) and SEC31B (FC =2.6; adjP = 0.002) were also upregulated in the L-CIM state suggesting a predisposition for protection against cellular stress. Interestingly, there was no significant difference in baseline MYC expression between the L-CIM and H-CIM group (P = 0.34). Conclusion: Our unsupervised approach enabled classification of distinct CIM induction states using paired pre- and post-treatment transcriptomic analysis, allowing interrogation of CIM in a manner not captured via traditional chemoresistance studies alone. We uncover underappreciated molecular programs, like proteostasis and ROS detoxification, that may underlie blunted CIM induction in BC. Future work will focus on characterizing baseline molecular features that prompt drug-specific CIM induction states, with the goal of informing and guiding the personalization of chemoimmunotherapy in breast cancer. Citation Format: M. O. Gbadamosi, I. Lopes de Lima, K. H. Streeks, E. Molchan, M. S. Makarem, K. L. Coleman. Identification and characterization of distinct chemoimmunomodulatory states in breast cancer via unsupervised transcriptomic analysis of pre- and post-treatment specimens [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS4-01-30.