Gene Signature to Predict Nodal Status in Oral Cavity Squamous Cell Carcinoma

Purpose: For patients with early-stage oral cancer who do not have clinically evident neck metastasis, determination of risk for metastasis is critical for survival and quality of life. As none of the current imaging modalities could reliably detect occult metastasis, surgeons usually opt for a prophylactic neck dissection because failure to surgically resect occult neck metastasis at the time of oral cancer diagnosis reduces survival. In this study we use cancer tissue and matched contralateral control tissue to analyze the methylation and gene expression signature of metastasizing and non-metastasizing oral cancer.

Methods: We enrolled 41 oral cavity squamous cell carcinoma patients. We collected and snap froze cancer tissue and matched contralateral normal tissue at the time of surgical resection. The advantage of using matched contralateral normal tissue was that the patients served as their own control, and gene expression was normalized according to expression of the normal tissue. Nodal metastasis was determined based on pathological staging. We extracted DNA and RNA from the tissues. We performed bisulfite treatment of DNA to prepare the samples for methylation array analysis. We used the Illumina Infinium Human Methylation 450 BeadChip Array to perform a genome-wide methylation analysis. We processed RNA transcript into cRNA and performed a gene expression array using the Illumina HumanHT-12 v4 Expression BeadChip. We analyzed the methylation and gene expression data in parallel.

Results: Firstly, we constructed a classifier of 51 genes based on differential expression patterns to distinguish metastatic (node-positive) from non-metastatic (node-negative) patients.   Principal component analysis (PCA) using the 51-gene classifier separated node-positive from node-negative patients. Leave-p-out cross-validation (p = 2) demonstrated no false positives or false negatives with a receiver operating characteristic (ROC) value for the nodal status classifier of 1.0. By using the 51-gene classifier we were able to differentiate metastasizing (node positive) from non-metastasizing (node negative) oral SCC with a false discovery rate (FDR) of 2.85x10^-5 and a q-value of 1.06x10^-3.  We then analyzed the methylation results and established 18 methylation sites that could distinguish between node-positive and node-negative patients. Using leave-p-out cross-validation we were able to predict metastasis with a negative predictive value of 94% and positive predictive value of 100%. Lastly, we combined the gene expression and methylation data and showed concordance between hypermethylation and silenced gene expression, with p=1.03x10^­-10, demonstrating that the significant methylation changes led to biologically relevant changes.

Conclusions: This study is the first demonstration of using combined methylation and gene expression array analysis to develop a gene signature that predicts nodal metastasis. Also, in contrast to previous studies that used healthy subjects as controls, we used matched cancer and contralateral normal tissues, so that patients serve as their own controls. We show robust methylation and expression signatures that could predict nodal metastasis.

Viet CT et al. Methylation array analysis of preoperative and postoperative saliva DNA in oral cancer patients. Cancer Epidemiology Biomarkers and Prevention. 15(12): 3603-11, 2008.

Viet CT et al. Understanding oral cancer in the genome era. Head Neck. 32(9): 1246-1268, 2010.