Background: Due to recent advances in colonoscopic techniques, submucosal colorectal cancers (T1 CRCs) can now be removed endoscopically. Among these, 70% of T1 CRCs are considered as “high risk” because they demonstrate presence of lymphovascular invasion, poor differentiation, and the depth of tumor is >1000um. However, post-surgical pathology results suggest that only ~10-15% of all T1 CRCs are truly lymph node positive, while majority of high risk patients undergo unnecessary surgical treatments with current criteria. Since current pathological criteria have limitations, availability of molecular biomarkers that can identify ‘genuine high risk patients with lymph node (LN) metastasis’ will reduce the burden of surgical overtreatment. Since gene expression-based classification of CRC could identify patients with poor prognosis, we sought to identify a gene expression signature which can detect T1 CRCs with LN metastasis.
Methods: Two independent publicly available genome-wide mRNA expression datasets were used for mRNA biomarker discovery (n=125) and in-silico validation (n=56). Genome-wide unbiased gene expression signature was developed from The Cancer Genome Atlas (TCGA) RNA-Seq data by comparing the expression profiles between 16 LN-positive and 109 LN-negative T1/2 CRC patients. In addition to the selection of most differentially expressed genes between the two groups, we used (ROC) based back-step elimination methodology to identify a robust mRNA panel. The gene panel was validated in an independent publicly available dataset (n=56), followed by analytical validation in two independent T1 CRC patient cohorts (n=134 and n=67) using RT-PCR assays.
Results: The in silico genome-wide comprehensive discovery led to the identification of an eight gene mRNA classifier that significantly predicted LN-metastasis with an AUC of 1.0, and the subsequent validation in an independent public data set resulted in an AUC of 0.93. The RT-PCR based training and validation of this eight gene classifier in two independent clinical cohorts robustly identified LN metastasis-positive T1 CRC patients with an AUC of 0.86 (95% CI 0.74-0.97, P=0.001) and 0.79 (95% CI 0.69-0.92, P=0.001) respectively.
Conclusions: In conclusion, we have identified a novel mRNA-based classifier that can detect high risk T1 CRCs with Lymph node metastasis. Further validation of these biomarkers in endoscopically collected biopsies will aid in clinical decision making and improving the clinical management of such patients.