Abstract
Current staging is inadequate for predicting clinical outcome of esophageal squamous cell carcinoma (ESCC). Aberrant expression of LOXL2 and actin-related proteins plays important roles in ESCC. Here, we aimed to develop a novel molecular signature that exceeds the power of the current staging system in predicting ESCC prognosis. We found that LOXL2 colocalized with filamentous actin in ESCC cells, and gene set enrichment analysis (GSEA) showed that LOXL2 is related to the actin cytoskeleton. An ESCC-specific protein-protein interaction (PPI) network involving LOXL2 and actin-related proteins was generated based on genome-wide RNA-seq in 15 paired ESCC samples, and the prognostic significance of 14 core genes was analyzed. Using risk score calculation, a three-gene signature comprising LOXL2, CDH1, and FN1 was derived from transcriptome data of patients with ESCC. The high-risk three-gene signature strongly correlated with poor prognosis in a training cohort of 60 patients (P = 0.003). In mRNA and protein levels, the prognostic values of this signature were further validated in 243 patients from a testing cohort (P = 0.001) and two validation cohorts (P = 0.021, P = 0.007). Furthermore, Cox regression analysis revealed that the signature was an independent prognostic factor. Compared with using the signature or TNM stage alone, the combined model significantly enhanced the accuracy in evaluating ESCC prognosis. In conclusion, our data reveal that the tumor-promoting role of LOXL2 in ESCC is mediated by perturbing the architecture of actin cytoskeleton through its PPIs. We generated a novel three-gene signature (PPI interfaces) that robustly predicts poor clinical outcome in ESCC patients.