Conclusion
A model combining CDH4 and STAT4 and LMP1 was proposed to be a feasible model for diagnosing NPC and predicting late stage of NPC.
Methods
The transcriptomic data of primary NPC cell strains were used as a discovery tool. Linear regression approach was used to define signatures distinctive between early and late stage of NPC. Expressions of candidates were validated with an independent set of biopsies (n = 39). Leave-one-out cross-validation technique was employed to estimate the prediction accuracy on stage classification. The clinical relevance of marker genes was verified using NPC bulk RNA sequencing data and IHC analysis.
Purpose
Nasopharyngeal carcinoma is highly metastatic but difficult to detect in its early stages. It is critical to develop a simple and highly efficient molecular diagnostic method for early detection of NPC in clinical biopsies.
Results
Three genes comprising CDH4, STAT4, and CYLD were found to have a significant differentiating power to separate NPC from normal nasopharyngeal samples and predicting disease malignancy. IHC analyses showed stronger CDH4, STAT4, and CYLD immunoreactivity in adjacent basal epithelium compared with that in tumor cells (p < 0.001). EBV-encoded LMP1 was exclusively expressed in NPC tumors. Using an independent set of biopsies, we showed that a model combining CDH4, STAT4, and LMP1 had a 92.86% of diagnostic accuracy, whereas a combination of STAT4 and LMP1 had a 70.59% accuracy for predicting advanced disease. Mechanistic studies suggested that promoter methylation, loss of DNA allele, and LMP1 contributed to the suppressive expression of CDH4, CYLD, and STAT4, respectively.