Cervical lymph node metastasis is the strongest prognostic factor in oral tongue carcinoma, yet current clinical guidelines rely primarily on depth of invasion to guide elective neck dissection. This approach results in unnecessary surgery in up to 70% of patients. Cancer-testis antigens (CTAs) are a family of genes associated with tumor aggressiveness and may serve as predictive biomarkers for nodal spread. A multi-step analysis integrating large-scale public datasets, including microarray (GSE78060), bulk RNA-seq emerging from the cancer genome atlas (TCGA), and single-cell RNA-seq (GSE103322), was employed to identify CTA genes active in oral tongue cancer. Selected genes were validated using NanoString nCounter RNA profiling of 16 patients undergoing curative glossectomy with elective neck dissection. Machine learning algorithms, including decision trees, t-distributed stochastic neighbor embedding (t-SNE), and convolutional neural networks (CNN), were applied to assess predictive power for nodal metastasis. Computational analysis initially identified 40 cancer-active CTA genes, of which four genes (LY6K, MAGEA3, CEP55, and ATAD2) were most indicative of nodal spread. In our patient cohort, NanoString nCounter profiling combined with machine learning confirmed these four genes as highly predictive. We present a proof-of-concept CTA-based genetic diagnostic tool capable of discriminating nodal involvement in oral tongue cancer. This approach may reduce unnecessary neck dissections, minimizing surgical morbidity.
Smith et al. (Mon,) studied this question.