Researchers identify gene signature for improved oral cavity cancer predictions

Oral squamous cell carcinoma (OSCC) impacts over 30,000 people in the U.S. each year. While tobacco and heavy alcohol consumption are the major risk factors for OSCC, an increasing number of cases are diagnosed among individuals without these conventional risk factors.

Patients diagnosed with early stage OSCC often undergo an extensive neck surgery for doctors to accurately assess whether their cancer has metastasized, or spread, to lymph nodes in their neck. In many cases, patients turn out to have negative pathological results after undergoing this highly invasive procedure. Recently, researchers at the University of Chicago Medicine Comprehensive Cancer Center, in collaboration with Johns Hopkins University and Cornell University, developed a cancer prediction tool that may potentially reverse this trend in treating this devastating disease.

“Clinical decisions are currently based on criteria that often don’t properly predict the patient’s lymph node (LN) metastatic status,” said Evgeny Izumchenko, PhD, Assistant Professor of Medicine at UChicago. “The combination of imaging methods, tumor biopsies and observations by the clinician is sometimes not strong enough to reliably predict the metastatic status of the patient’s LN. As a result, many patients undergo unnecessary procedures.”

To address this issue, researchers led by Izumchenko identified a gene signature in primary oral cavity malignancy that may accurately predict whether patients with early stage disease have cancer that has already spread to their lymph nodes. These results may have an important impact on reducing the number of patients who undergo unnecessary invasive procedures.

In a study, researchers used a k-Top scoring pair algorithm that identified gene pairs included in the reported gene signature. This classifier-based algorithm provides a relative ranking for the genes in a lymph node’s positive and negative pathological state and scores each of the gene pairs based on the magnitude and consistency of the ranking. Six gene pairs with the highest average scores were chosen to predict the status of a sample.

After training the classifier in a large number of gene expression profiles obtained from public sequencing repositories, researchers then validated the gene signature for its ability to predict lymph node status with PCR using an independent cohort of patients with known pathological LN status. The k-Top classifier was able to accurately distinguish a positive pathological state in 19 metastatic primary tissue samples that were examined. Researchers also saw that the genes that promoted tumor growth were highly expressed in the tumors that were positive for lymph node metastasis, further supporting a rationale for their assay.

In the clinic, this predictive tool can be combined with other standard and approved methods to help define the nodal metastatic status of a patient. “The tool is a supportive predictor. While additional evaluation is required to further assess its true potential and limitations, this tool can step in to support the confidence of surgeons and oncologists when making clinical decisions,” said Izumchenko.

A major advantage of this k-Top classifier is that the predictions are independent of the platform used to evaluate the expression level of the 12 genes included in the signature. This would allow researchers and clinicians to obtain consistent results across various assays, in contrast to previously suggested predictive methods, increasing the potential translation of this method to clinical settings.

In fact, in their statistical model, the researchers focused on gene pairs that take part in biological processes related to cell growth and cancer progression. This effectively trained a purely statistical model to look for a biologically relevant signature for oral cell carcinoma.

“It was really a collaborative work,” Izumchenko said. He described the project as an interdisciplinary collaboration involving investigators with complementary and integrated expertise, including oral surgeons, oncologists, geneticist and a biostatistician.

“To move this work further, we need to create a large database of samples with pathological data from tumors that have been evaluated by the pathologist,” he said. His group is now collecting tumor samples from patients with known pathological LN status. Confirmation of the high performance of the signature in this and other independent datasets, together with its simple decision rules, makes it a potential candidate for developing a clinically approved method to guide the neck treatment plan in patients with early-stage disease.

Additional study authors included: Yasmin Ghantous, Esther Channah Broner and David Sidransky from Johns Hopkins University in Baltimore; Imad Abu El-Naaj of Bar Ilan University in Israel; Mohamed Omar and Luigi Marchionni of Weill Cornell Medicine, New York; Nishant Agrawal, Alexander Pearson, and Ari Rosenberg of the University of Chicago. Support for the study was provided by grants from the National Institutes of Health and the National Cancer Institute.