Single-cell transcriptomics and machine learning unveil ferroptosis features in tumor-associated macrophages: Prognostic model and therapeutic strategies for lung adenocarcinoma

Background Lung adenocarcinoma (LUAD) is a major cause of cancer-related mortality worldwide. Tumor-associated macrophages (TAMs) play a crucial role in the tumor microenvironment (TME), influencing tumor progression and immune response. Ferroptosis, an iron-dependent form of regulated cell death, has been implicated in tumor biology, but its role within TAMs in LUAD remains unclear. Aim This study aimed to screen key genes associated with ferroptosis in macrophages and construct a prognostic risk model for LUAD based on these genes. Methods Integrating the TCGA-LUAD, GSE131907, and GSE13213 datasets, macrophage heterogeneity was analyzed through single-cell dimensionality reduction clustering, pseudotime analysis, and cell-cell communication. Using the GeneCards ferroptosis gene set (1515 genes), ferroptosis-related differentially expressed genes in macrophages were screened. Eight machine learning algorithms (LASSO, SVM, XGBoost, etc.) were leveraged to identify prognostic genes and build a Cox regression risk model. The functional roles of key genes were validated through immune infiltration analysis, drug sensitivity prediction, and Western blot analysis. Results Single-cell analysis revealed that macrophages in LUAD lead intercellular communication through the MIF (CD74+CXCR4) ligand-receptor interaction, with ferroptosis-related genes (FRGs) highly expressed in macrophages. 73 macrophage FRGs were identified, and through multi-algorithm cross-validation, HLF, HPCAL1, and NUPR1 were determined as core genes. The risk model (Risk Score = HLF × (−0.153) + HPCAL1 × 0.261 + NUPR1 × (−0.21)) demonstrated robust predictive performance in both the TCGA and GSE13213 cohorts, with 1-, 3-, and 5-year AUC values of 0.756, 0.753, and 0.705. The high-risk group was enriched in tumor progression pathways (like epithelial-mesenchymal transition, cell cycle checkpoints), exhibited low expression of immune checkpoint genes (BTLA, CD47), and showed increased sensitivity to cyclophosphamide and crizotinib. Western blotting confirmed the expression levels of HLF, HPCAL1, and NUPR1 were remarkably lower in LUAD cell lines compared to normal bronchial epithelial cells (P < 0.05). Conclusion The research is the first to build a LUAD prognostic model based on macrophage ferroptosis-related genes (HLF, HPCAL1, NUPR1), revealing the immune microenvironment characteristics and drug sensitivity differences in the high-risk group. These findings provide new strategies for precision therapy targeting ferroptosis in tumor-associated macrophages (TAMs).