A research team, led by the University of Texas MD Anderson Cancer Center, has identified a protein signature that predicts ovarian cancer patients’ risk of recurrence, which they hope can be developed into a clinical prognostic test.
Led by Roel Verhaak and other colleagues involved in The Cancer Genome Atlas' analysis of ovarian cancer, the study — published this month in the Journal of Clinical Investigation — followed on previous work using TCGA gene expression data to develop a signature to predict prognosis.
Verhaak told PGx Reporter that in the team's earlier gene expression study, the resulting signature had a significant association with overall survival, but its value for predicting treatment response — based on time to recurrence — was limited. So the group decided to turn to more recently released TCGA protein data to see if they could develop a predictor for time to recurrence.
"Our earlier paper was the most robust gene expression based method published to that point for ovarian cancer prognosis," Verhaak said. "However, a limitation was that we predicted outcome, but not so much response to therapy."
"The protein component of the TCGA ovarian cancer project … only became available about a year ago, but [the data was generated] at MD Anderson, so we had direct access. We started playing with the protein profiles, and saw that they were, actually, robustly able to predict time to recurrence," he said.
In the recent study Verhaak and his colleagues used reverse-phase protein arrays to generate ovarian carcinoma expression profiles for 222 cases from the TCGA ovarian cancer study set with data available on progression-free survival times.
Based on protein levels differentially expressed in cases with better and worse outcomes, the researchers identified a nine-protein signature they called the protein-driven index of ovarian cancer, or PROVAR, which significantly discriminated between high- and low-risk cancers as well as long- and short-term survivors.
To validate the finding, the group then tested PROVAR in an independent sample set of 226 tumors supplied by Japanese collaborators, showing that the signature maintained its predictive power.
In the validation set median progression-free survival times were 15 months for those predicted to be high risk by PROVAR, and 24 months for those predicted to be low risk. Median overall survival was 39 months in the predicted high-risk group and more than 60 months in the predicted low-risk group.
Compared to gene expression-based outcome classifiers, the authors wrote, the protein panel showed "significantly improved capacity" to predict tumor progression.
While Verhaak said there is no direct evidence of why the protein-based approach was more effective than gene expression predictors, he said the team's assumption is that the fact that protein expression is more closely related to tumor biology than gene expression plays some role.
According to the study authors, a prospective validation of the signature will be required before PROVAR could be adopted for clinical use.
Though there are not targeted therapies currently approved to treat ovarian cancer, a prognostic tool like PROVAR could help physicians choose current therapies with better effectiveness but increased rates of side-effects, or combination therapy strategies for those patients identified as being at higher risk of early recurrence, the authors wrote.
Verhaak said the group is planning to first translate the signature to an immunohistochemistry-based assay, a necessary step for adapting PROVAR for clinical use.
"The signature only contains eight or nine proteins— a small enough number that you could more easily turn this into an actual test than the more complicated gene expression sets that have been developed," Verhaak said. "We think that, given the methodology, we could turn it into a relatively straightforward immunohistochemistry-based test."
The team is also in discussions with its clinical colleagues at MD Anderson about designing a prospective validation trial and raising funds to support that study, he said.
According to Verhaak, the group has also done some work to test whether a combination of protein and gene-expression might be an even stronger prognostic predictor, but he said that it looks, so far, like gene expression "didn't really add much," to the PROVAR signature.
However, he said, the team also has plans to look at integrating clinical variables, like tumor residue after surgery and tumor stage and grade, to improve the predictor.