Identification of papillary thyroid carcinoma-associated epithelial cell subpopulations and diagnostic biomarkers: integrating machine learning with single-cell analysis.

Papillary thyroid carcinoma (PTC) represents the most prevalent sort of thyroid malignancy, and its incidence has been consistently increasing worldwide. Although most PTC cases exhibit indolent behavior, a subset demonstrates aggressive characteristics, leading to recurrence, metastasis, and poor clinical outcomes. This study aims to unveil the molecular mechanism of PTC and identify potential biomarkers for its early diagnosis and individualized treatment. Single-cell RNA sequencing (scRNA-seq) data were utilized for the analysis of alterations in cellular composition within PTC tissues, with a particular focus on differentially expressed genes (DEGs) in every cell population. Co-expression gene module analysis of tumor-associated epithelial cells (Tumor-epi) was carried out via the high-dimensional weighted gene co-expression network analysis (hdWGCNA). Subsequently, multiple machine learning (ML) algorithms, including Random Forest (RF), Extreme Gradient Boosting (XGBoost), Boruta feature selection algorithm (Boruta), and support vector machine-recursive feature elimination (SVM-RFE), were leveraged to refine and select three pivotal genes, XPR1, SH3RF1, and TLE1, for the construction of a diagnostic model. Finally, quantitative real-time polymerase chain reaction (qRT-PCR) was performed in the PTC cell line (BCPAP) and normal thyroid epithelial cells (NC) to validate the expression of these hub genes. A notable elevation in epithelial cell percentage was detected in tumor tissues. Further analyses revealed that Tumor-epi cells exhibited distinct molecular characteristics and biological functions potentially linked to PTC invasiveness and immune evasion. Through hdWGCNA, co-expression gene modules were identified within Tumor-epi cells, which reflect their putative role in PTC pathogenesis. The diagnostic model, based on the three selected genes (XPR1, SH3RF1, TLE1), demonstrated high discriminatory power in distinguishing PTC from normal thyroid tissue [area under the curve (AUC) >0.90]. qRT-PCR validation in the BCPAP and NC confirmed that XPR1 was significantly upregulated, whereas SH3RF1 and TLE1 were downregulated in PTC (all P<0.001), consistent with computational predictions. Moreover, these genes were found to be linked to the immune cell distribution in the tumor immune microenvironment, suggesting their involvement in mechanisms of immune escape. This study presents new perspectives on the molecular pathogenesis of PTC and highlights the crucial role of epithelial cells in tumor progression. The three identified genes, namely XPR1, SH3RF1, and TLE1, serve as prospective diagnostic biomarkers and may facilitate diagnostic refinement and personalized treatment strategies for PTC. Additionally, their correlation with the immune microenvironment offers a new avenue for exploring the mechanisms of tumor immune evasion.