Transcriptomic sequencing analysis of the tumor microenvironment atlas and potential mechanisms of metastasis in papillary thyroid carcinoma.

Papillary thyroid carcinoma (PTC) is the most prevalent thyroid malignancy and its incidence continues to rise. Although prognosis is generally favorable, overdiagnosis, overtreatment, and occasional lethal complications persist. To clarify the micro-environmental basis of PTC metastasis, we integrated single-cell RNA-seq from tumor (T), peritumoral thyroid (PT) and lymph-node (LN) tissues with bulk transcriptomes. Metastatic (M) and non-metastatic (NM) cases were compared, and machine-learning models were built to predict lymph-node status and outcome. Our atlas delineates highly heterogeneous ecosystems across T, PT and LN samples. Within T/NK lymphocytes, both immune-activating and immune-suppressive subsets coexist; NK and CD4+TNFRSF4 cells were selectively enriched in M tumors, implying heterogeneous anti-tumor responses. Among myeloid cells, a Mac-APOC1 subset displayed tumor-associated macrophage/M2 features, consistent with immune escape and dissemination. Dendritic cells differed markedly in antigen presentation and chemotaxis, while RGS5+ myCAFs, markedly expanded in M samples from T and LN, up-regulated metabolic and ribosomal pathways. Trajectory analysis uncovered dynamic CAF state transitions. Tumor epithelial cells followed multiple developmental branches, with those linked to LN invasion predicting poorer survival. A random-forest classifier achieved 0.98 accuracy for N-stage, and a lasso-Cox signature (C-index > 0.9) robustly stratified survival, underscoring the clinical relevance of the identified genes. Collectively, this work reveals the cellular and functional heterogeneity that drives PTC progression, identifies metastasis-associated subpopulations, and demonstrates the utility of machine learning for precise prediction of lymph-node metastasis and patient prognosis, offering a foundation for tailored therapeutic strategies.