NUSAP1 Is a Potential Diagnostic Biomarker for High-Grade Endometrial Stromal Sarcoma: An Integrated Machine Learning and Clinical Validation Study

Abstract

High-grade endometrial stromal sarcoma (HGESS) and low-grade endometrial stromal sarcoma (LGESS) show overlapping histological morphologies, and the low specificity of existing markers makes identification difficult. To address the need for better diagnostic tools, this study aimed to discover and validate novel diagnostic biomarkers for HGESS by integrating transcriptome sequencing with machine learning. A total of 24 HGESS and 11 LGESS samples were selected as the training cohort. Transcriptome RNA-seq was conducted to identify differentially expressed genes. Key genes were screened by combining the following 3 distinct machine learning techniques: LASSO regression, XGBoost, and the Boruta algorithm. These candidate biomarkers were then validated by immunohistochemistry in an independent cohort of 24 HGESS, 34 LGESS cases, and 33 other uterine mesenchymal tumors. Machine learning shortlisted Centromere protein E (CENPE) and nucleolar and spindle-associated protein 1 (NUSAP1). NUSAP1 immunostaining was positive in 17 of 24 HGESS (70.8%), and negative in 34 LGESS (0%) and 33 other uterine mesenchymal tumors (0%). For HGESS versus LGESS, NUSAP1 yielded 70.8% sensitivity, 87.9% accuracy, and 82.9% negative predictive value, comparable to Cyclin D1 and superior to BCOR, while maintaining 100% specificity and 100% positive predictive value, thus outperforming CD10. CENPE expression was higher in HGESS than in LGESS (58.3% versus 32.4%), but its specificity was compromised by positivity in several mesenchymal tumors. These findings indicate that NUSAP1 is a novel immunohistochemical marker with high specificity that can serve as an important complement to existing markers such as Cyclin D1 and BCOR, providing a new practical tool for the precise pathological diagnosis of HGESS.