Spatial data of PPTC

Pediatric papillary thyroid carcinoma (PPTC) exhibits heterogeneous clinical behavior, yet the epithelial and microenvironmental features distinguishing nodular, primary, and metastatic tissues remain incompletely defined. Here, we generated single-cell resolution spatial transcriptomic data using the 10x Genomics Xenium platform and integrated it with scRNA-seq to construct a multimodal atlas spanning pediatric thyroid nodules, peritumoral tissues, primary tumors, and lymph node metastases. Spatial analyses revealed immunosuppressive niches enriched for regulatory T cells and malignant epithelial components, with NECTIN3-TIGIT signaling preferentially linking tumor cells and Tregs in PPTC and metastatic lesions. In parallel, VEGFA-KDR signaling highlighted enhanced angiogenic activity in PPTCs. Across modalities, we identified conserved malignant-associated epithelial states characterized by stem-like, partially dedifferentiated, and immune-regulatory transcriptional programs, whereas benign lesions retained differentiated follicular features. Building on the reproducibility of these epithelial signatures, we derived a machine-learning classifier based on 30 epithelial-intrinsic genes, including both known and previously unreported markers. This classifier outperformed existing biomarker sets and generalized across independent scRNA-seq and proteomic cohorts. Together, these findings define a coordinated epithelial–microenvironmental architecture in pediatric thyroid lesions and provide an externally validated molecular panel for distinguishing malignant from nodular epithelial states.