Towards a transcriptomic biomarker for the classification of melanocytic neoplasms.

Abstract

Histopathologic diagnosis of thin, invasive cutaneous melanoma (CM) is only 34-62% accurate. Therefore, we sought to develop a transcriptomic biomarker to distinguish benign from malignant melanocytic neoplasms. We generated a targeted RNA-Sequencing dataset (TempO-Seq) of benign nevi (BN; n = 50) and CM (Breslow depth ≤ 1.0 mm; n = 51) and demonstrated enrichment of immune-related pathways among the 450 differentially expressed genes. Next, we trained a putative transcriptomic biomarker in two datasets, including BN and CM, and one dataset with CM in association with a nevus, macrodissected into CM and nevus regions. We refer to the nevus portion of CM in association with a nevus as progressing nevi (PN), since these nevi progressed to CM. Principal component analysis showed that PN samples clustered in a component intermediate to BN and CM. Ordinal regularized regression selected PYGL, AP000845.1, PHYHIP, WSCD1, FBXO7, TRPM1, SLC4A4, NALCN, FRMD4B, HHATL, COL1A1, CRYM, EPOP, RGS1, KRT6C, IGHG1, CNTN1, MMP11, GZMM, AP001880.1, TTYH3, TMEM132A, and PRAME; these genes were consistently selected in 1000 models using data from bootstrap resamples and had a single model predictive accuracy of at least 0.90 (area under the receiver operator characteristics curve). Linear regression models fit with these 23 genes in the TempO-Seq data, and publicly available microarray datasets from BN, dysplastic nevi, and CM, showed high consistency in the magnitude and directionality of gene expression differences between nevi and CM. Furthermore, immunohistochemical staining showed consistent protein-level changes in MMP11 and PYGL. These results illuminate the potential for a transcriptomic biomarker to differentiate benign from malignant melanocytic neoplasms and improve the accuracy of melanoma diagnosis.