Expression changes of four genes in seemingly normal tissues in the mouth could predict whether oral cancer will recur locally and may inform treatment decisions, an early-stage study by researchers at the Ontario Cancer Institute has found.
The team, led by Suzanne Kamel-Reid and Igor Jurisca, analyzed gene expression in 96 samples located near the tumors of 24 patients with oral squamous cell carcinoma. They combined this data with a meta-analysis of five additional public microarray datasets to identify a signature of four genes overexpressed in otherwise histologically normal surgical margins, which they then tested against a 136-sample validation set from 30 additional patients.
The four-gene signature was able to predict recurrence in the independent validation cohort with performance, measured by an area under the receiver operating characteristic curve, of 0.73, the researchers reported in BMC Cancer this month. According to the authors, the group has applied for relevant patents and has received feedback from a few companies interested in the possibility of developing a commercial prognostic test based on their research.
"We were happy it worked that well, and with only four genes," Jurisca told PGx Reporter in an e-mail. "It is quite exciting since it has a true potential for clinical use, [though] further validation is [necessary and] underway."
Currently, a complete surgical resection — with histologically clear tumor margins — is one of the only prognostically significant factors in oral carcinoma, Kamel-Reid and research associate Patricia Reis, the paper's first author, told PGx Reporter this week.
"But we know that genetic changes will occur before histological changes, so we thought if we look at the genetic profile of margins, can we identify any biomarkers that will help us predict which margins may have genetic changes that could predispose them to recurrence," Kamel-Reid said.
This is important because even among patients with normal surgical margins, "about 20 percent to 30 percent will still have local recurrence around the region excised," she explained. "And of course, recurrence is associated with mortality — with treatment failure and with death."
Oral squamous cell carcinoma accounts for about 24 percent of all head and neck cancers, and is a "major cause of cancer death worldwide," largely due to recurrence, the study authors wrote.
"So obviously histology isn't giving us all the information we require," Kamel-Reid said.
To establish their predictive signature, the group evaluated a total of 232 tissue samples from 54 patients divided into independent training and validation sets.
Using Affymetrix Human Genome HG-U133A Plus arrays, the researchers measured gene expression in the 96-sample training set and combined this profiling with a meta-analysis of five additional published microarray studies to yield an additional 199 samples, the authors reported.
Choosing to focus only on over-expressed genes based on the hypothesis that these would be more reliable markers, the group took an intersection of genes identified in both the meta-analysis and their in-house microarray experiment, which produced a list of 138 potential genes upregulated at least two-fold in tumor versus normal tissue. Following statistical analysis, the researchers then drilled down to four genes with the largest predictive power: MMP1, COL4A1, P4HA2, and THBS2.
The team performed quantitative PCR validation of the four-gene signature in another cohort, which showed that the markers of interest were up-regulated in margins from patients who experienced disease recurrence compared to margins from patients who did not recur.
Using a survival receiver operating characteristic analysis, the group found that using a high threshold to divide high-risk and low-risk patients gave the best predictive power with a low false-positive rate, with an AUC of 0.73.
"The [signature we found] predicted the majority of recurrences at a low false-positive rate," Reis said. "And this is a real advantage because you're not just looking at histology, but you have anther parameter you can look at," Kamel-Reid added.
"In fact, in our study we were able to predict two patients that actually did recur [but were not known to have recurred in the study dataset]," she said.
A larger study may be able to even further fine-tune the cutoff threshold to increase sensitivity and specificity, the authors wrote in their discussion.
In terms of treatment, the potential for the group's signature to act as an additional prognostic indicator is exciting, Kamel-Reid said.
Patients at a higher recurrence risk might benefit from closer monitoring, or from post-operative radiation treatment, the authors reported. This would be particularly beneficial for those with histologically clear margins who may still be at risk.
Though the research is early stage, Kamel-Reid said that the team was able to achieve much more than previous studies seeking useful biomarkers in oral carcinoma. Though many genes are known to be deregulated in the disease compared to normal tissue, none had previously been shown to be at all predictive of recurrence.
"We were fortunate in that we were able to get enough patient material to analyze," she said. "It's quite difficult to get because surgical margins themselves are very small. But … we feel our sample set was actually pretty good," she said. "At the end of the day we tested hundreds of margins: all together 232 samples from 54 patients."
Kamel-Reid said the group is working on further validating the signature, and Jurisca added in an e-mail that if this goes well, a clinical trial would be a possibility.
According to Kamel-Reid, the team is also working on simplifying its methods for clinical use, with an eye toward possibly commercializing the test. "We could do it as a protein immunohistochemistry test," she suggested. "That would be great because any pathology lab could do that. If not, then a simple molecular test would also be a possibility."
Kamel-Reid said the group has had interest from a few companies to "talk further" about how they could potentially use their work.
"We are not in this for profit. But I think it would be very useful for this to be more widely available," she said.
Have topics you'd like to see covered in Pharmacogenomics Reporter? Contact the editor at mashford [at] genomeweb [.] com.