Integrating multi-omics and machine learning to unravel mechanisms of lymph node metastasis in papillary carcinoma with and without thyroiditis.

Background: Papillary thyroid carcinoma (PTC) is the most common thyroid malignancy, with lymph node metastasis (LNM) significantly influencing prognosis. PTC cases were categorized based on the presence or absence of coexisting thyroiditis: those with thyroiditis were designated as PTC-thyroiditis, and those without as PTC-blank. Although both subtypes exhibit distinct clinical and genomic profiles, the mechanisms underlying LNM-particularly the role of extracellular matrix (ECM) remodeling and cancer-associated fibroblasts (CAFs)-remain poorly understood. This study aimed to compare the molecular and cellular heterogeneity of PTC-blank and PTC-thyroiditis, identify key drivers of LNM, and develop a predictive model for clinical use. Methods: We integrated bulk RNA-seq data from the TCGA database and single-cell RNA-seq (scRNA-seq) data from GSE184362 to analyze clinical, genomic, and tumor microenvironment (TME) differences between PTC-blank and PTC-thyroiditis. Differential expression analysis, Gene Set Enrichment Analysis (GSEA), and immune cell infiltration analysis were performed. Fibroblast subpopulations were characterized at single-cell resolution. Machine learning algorithms (LASSO, random forest, KNN) were applied to identify LNM-related genes and construct a predictive model. Mendelian randomization (MR) and molecular docking were used to validate causal genes and potential drug interactions. Results: PTC-blank exhibited higher T, N, and M stages and increased mutations in BRAF and MUC16 compared to PTC-thyroiditis. LNM in PTC-blank was associated with ECM remodeling and collagen fiber accumulation, associated with a distinct PI16+ fibroblast subcluster with active ECM organization functions. In contrast, LNM in PTC-thyroiditis involved immune-related pathways without significant fibroblast infiltration or ECM changes. A 17-gene predictive model for LNM was developed, with the KNN classifier demonstrating high accuracy. MR analysis identified SHISA5 as a causal risk gene for thyroid cancer, and molecular docking revealed strong binding affinity with acetaminophen, suggesting therapeutic potential. Conclusions: PTC-blank and PTC-thyroiditis exhibit distinct LNM mechanisms: ECM remodeling and fibroblast infiltration are associated with metastasis in PTC-blank, while immune dysregulation appears to be more prominent in PTC-thyroiditis. The identified 17-gene model offers robust predictive value for LNM risk, and SHISA5 represents a novel causal gene and potential therapeutic target. These findings provide insights into subtype-specific management strategies for PTC patients.