Zeitschrift für klinische und medizinische Genomik

Gene Variability between Perineural-Positive and Perineural-Negative Squamous Cell Skin Cancers

Abstract

Ashley C Mays, Jeff Chou, Ann L Craddock, Lance Miller and J Dale Browne

Background: Recurrent cutaneous squamous cell cancer (CSCC) is associated with poor outcomes with perineural invasion reported as a frequent finding in such lesions. Given the morbidity associated with late recurrence, identifying aggressive subtypes of CSCC at the time of primary excision is all the more necessary. This project sought to identify differentially expressed genes (DEGs) between perineural invasion-positive (PP) and perineural invasion-negative (PN) CSCC. Gene-based classification models for diagnosis of perineural invasion in CSCC were also developed.

Method: Forty fresh-frozen surgical specimens of CSCC with presence or absence of histopathological perineural invasion were processed for RNA isolation and hybridization to Affymetrix-U219 DNA microarrays. Raw gene expression data were normalized by Robust Multi-array Averaging (RMA) and log2 transformed. DEGs were identified by empirical Bayes statistics using the Bioconductor limma package. BRB-ArrayTools software was used to develop gene expression-based sample classification models. Using leave-one-out cross-validation, the resulting accuracies of eight different classification algorithms were evaluated.

Results: Twenty-one PP and 19 PN samples were analyzed. At a stringent limma p-value (p<0.001), 24 genes were differentially expressed between specimens. The cross-validated performance of the eight classification models exhibited a mean accuracy of 85-95%. Diagonal linear discriminant was most accurate at 95%, followed by Bayesian compound covariate at 94%. The poorest accuracy (85%) was observed for 1-Nearest neighbor and Support vector machines. For all eight methods, the sensitivities and specificities ranged from 79%-95%.

Conclusion: Gene expression distinguishes between PP and PN CSCC. Classification models based on these gene patterns distinguish PP and PN cancers with strong statistical accuracy and may potentiate more timely and objective diagnosis of perineural invasion that could guide more comprehensive adjuvant therapies.