Comprehensive multi-omic analyses were performed on the FUSCC-TNBC cohort, integrating whole-exome sequencing, RNA sequencing, and functional validation in vitro and in vivo. Somatic mutation profiling, gene set enrichment analysis (GSEA), and weighted gene co-expression network analysis (WGCNA) were used to define genomic and transcriptomic signatures. A machine learning model using the Mime1 package was applied to derive a senescence-associated prognostic signature (LAR-S) and validation in external cohorts. Immune deconvolution was performed to decipher the tumor microenvironment. Functional assays, patient-derived organoids (PDOs), and TS/V mouse models were used to evaluate therapeutic responses to senescence-modulating agent and immunotherapy combinations.

The LAR subtype was enriched for PIK3CA, PTEN, and ERBB2 kinase domain mutations. Functional studies confirmed ERBB2 variants (e.g., V777L and E698_P699delinsA) as oncogenic drivers conferring sensitivity to neratinib. Transcriptomic analyses revealed a dominant cellular senescence program associated with immune suppression. The LAR-S signature stratified survival across cohorts and predicted immunotherapy resistance. Targeting cellular senescence inhibited LAR subtype organoid growth and when combined with anti-PD-1 therapy synergistically suppressed tumor growth in vivo.

The LAR subtype harbors two therapeutic vulnerabilities: ERBB2 mutation-driven kinase activation; and senescence-mediated immune evasion. The LAR-S signature enables precise patient stratification and supports senescence-targeted and immunotherapy combination strategies as promising approaches for this refractory TNBC subtype.

Integrated scRNA-seq and metabolomic analyses were performed to characterize metabolic alterations in macrophages within the breast cancer tumor microenvironment (TME). According to the identified metabolic vulnerabilities, SLC38A2-overexpressing anti-HER2 CAR-Ms were engineered. Glutamine uptake and phagocytic activity were assessed to evaluate functional enhancement.

TAMs in breast cancer exhibited substantial metabolic dysregulation, particularly impaired glutamine metabolism accompanied by decreased expression of the glutamine transporter SLC38A2. Overexpression of SLC38A2 in anti-HER2 CAR-Ms, compared with conventional anti-HER2 CAR-Ms, enhanced glutamine uptake and markedly augmented phagocytosis of HER2⁺ breast cancer cells.

Metabolic engineering via SLC38A2 restored glutamine fitness and enhanced the antitumor activity of HER2-targeted CAR-Ms, thus providing a promising strategy to boost CAR-M–mediated tumor suppression in solid tumors.

In a prospective dual-center cohort of 155 patients with breast cancer, 747 freshly bisected lymph node slides were obtained via D-FFOCT. Surgeons interpreted each slide with histopathology as the gold standard. A deep learning model was trained on 28,911 patches (corresponding to 590 slides) and tested on 7,736 patches (corresponding to 157 slides). The results were mapped to the slide level for potential intraoperative evaluation.

D-FFOCT strongly correlated with hematoxylin and eosin (H&E)-stained histological images. Surgeons achieved 97.10% specificity in nodal diagnosis with D-FFOCT. The performance of the artificial intelligence (AI) model was not inferior to that of human experts and had a sensitivity/specificity of 87.88%/91.94% and an area under the receiver operating characteristic curve of 0.899 at the slide level. The human–AI collaborative system reduced labor requirements by 75% and increased the specificity by 6.5%, to 98.39%.

D-FFOCT has excellent potential as a tool for assessing lymph node metastatic status without tissue preparation or consumption. The integration of D-FFOCT with deep learning decreases labor demands and maintains high accuracy, thereby enabling streamlined nodal prediction independent of routine pathology procedures.

A total of 1,047 patients from TCGA and 166 patients from 3 external centers were included. The histopathology model used a transformer-based pretrained encoder (H-optimus-0) and a clustering-constrained attention multiple instance learning (CLAM-SB MIL) classifier to generate WSI-level representations. The clinical model incorporated engineered clinical variables and an extreme gradient boosting (XGBoost) model. A decision-level late fusion strategy (Multimodal PIK3CA Model, MPM) combined probabilistic outputs from both branches. Performance was evaluated with the area under the curve (AUC) and secondary metrics. Interpretability was assessed via attention heatmaps and shapley additive explanations (SHAP) analysis.

MPM outperformed single-modality models. It achieved an AUC of 0.745 on TCGA and maintained stable performance across external cohorts (0.695, 0.690, and 0.680). SHAP analysis identified molecular subtype as the most influential clinical feature, whereas attention maps highlighted mutation-associated morphological regions.

The developed multimodal framework effectively integrates complementary morphological and clinical information, and provides a robust and generalizable method for predicting PIK3CA mutation status. Strong multicenter adaptability and biological interpretability support its potential use as a clinical decision-support tool and an accessible alternative to molecular testing.