Identification and functional validation of lipid droplet-associated prognostic biomarkers in breast cancer via integrative multi-omics and machine learning approaches.

Lipid droplet (LD)-associated metabolic reprogramming plays a critical role in breast cancer progression and immune modulation, yet robust prognostic biomarkers and their functional mechanisms remain incompletely understood. This study aimed to identify LD-associated biomarkers with prognostic and therapeutic relevance through multi-omics integration and functional validation. Bulk transcriptomic, single-cell RNA sequencing, and spatial transcriptomic data were integrated using machine learning to construct a prognostic model in the TCGA-BRCA cohort, validated in merged GEO datasets (GSE24450 and GSE42568). Functional enrichment, immune infiltration analyses, and in vitro/in vivo experiments-including 3T3-L1 adipogenesis, co-culture, orthotopic tumor models, and clinical adipose tissue validation were performed to characterize candidate genes. The StepCox[both]+plsRcox algorithm generated an optimal prognostic model that independently stratified patients across molecular subtypes, outperforming ER/PR/HER2 status. High-risk patients exhibited reduced immune infiltration and T-cell dysfunction. SQLE and SOCS3 emerged as key LD-associated genes with opposing expression patterns: SQLE enriched in tumor-associated adipocytes and SOCS3 in immune cells. Functional assays confirmed SQLE promoted while SOCS3 inhibited adipogenesis. Modulating these genes in adipocytes suppressed tumor growth and EMT and polarized macrophages toward an M1-dominant phenotype. SQLE and SOCS3 serve as functionally significant LD-associated prognostic biomarkers and represent promising therapeutic targets in breast cancer, revealing novel mechanisms in tumor-adipocyte crosstalk.