Multi-omics analysis from the IMMUcan consortium to identify predictors of chemo-immunotherapy response in early-stage triple-negative breast cancer.

Abstract

2588 Background: Combining PD-1/PD-L1 immune checkpoint blockade (IO) with neoadjuvant chemotherapy (CT) is now a standard treatment for early-stage triple-negative breast cancer (eTNBC). However, no validated biomarker is currently used to guide patient selection for CT-IO benefit. We aimed to define baseline tumor–immune–stroma ecosystem states that explain differential responses and could inform future biomarker-driven trial designs. Methods: The IMMUcan consortium prospectively enrolled 422 patients with eTNBC treated with CT alone (n=221) or CT-IO (n=201). Pretreatment FFPE biopsies underwent whole-exome sequencing (n=397), RNA sequencing (n=360), multiplex immunofluorescence (2 panels, n=353/368), and imaging mass cytometry (n=341). Response was assessed as pathological complete response (pCR) versus residual disease (n=400). Clinical, genomic, transcriptomic, and spatial features were analyzed for associations with pCR and treatment interaction. Multi-omics integration was performed using Multi-Omics Factor Analysis (MOFA) to derive latent factors and ecosystem states. Results: Baseline clinicopathological characteristics were balanced between groups, while the pCR rate was higher with CT-IO than CT alone (73.1% vs 54.8%; Δ=20.4%; p<0.01). High tumor grade and tumor-infiltrating lymphocytes predicted pCR in both groups without IO-specific predictive value. Among transcriptomic TNBC subtypes, immunomodulatory, mesenchymal (M), and luminal androgen-receptor (LAR) derived the greatest benefit from CT-IO. Spatial profiling identified as key CT-IO responsive feature the enrichment of activated PD-1⁺GZMB⁺CD8⁺ T cells in proximity to tumor cells. MOFA identified biological axes with predictive value, including organized adaptive immunity, effector functions, immune-memory, stromal/angiogenic barrier, and luminal-metabolic lineage. Organized adaptive immunity predicts CT-IO benefit in M and LAR subtypes (OR: 5.42; 95% CI 0.87-33.6; p interaction = 0.048; and OR: 2.97; 95% CI 0.98-8.9; p interaction = 0.036, respectively). Conversely, LAR tumors characterized by luminal–metabolic lineage showed no benefit from CT-IO. Conclusions: Integration of multi-omics data revealed distinct baseline biological ecosystem states capable of identifying patients most likely to benefit from, or exhibit resistance to, CT-IO. These findings provide a comprehensive framework for designing biomarker-driven therapeutic strategies in next-generation neoadjuvant trials for early-stage TNBC.