Compendia Bioscience is preparing to release in the fall a new version of its Oncomine database of cancer gene-expression profiles that will feature enhanced analysis tools, including a new “meta-analysis” tool that the company and its collaborators recently used to identify a novel biomarker for prostate cancer.
The new tool, called Meta-COPA, is an extension of an approach that the company previously developed called COPA, or Cancer Outlier Profile Analysis, which predicts candidate oncogenes from gene-expression data by identifying genes with high expression levels in only a subset of cases. Meta-COPA applies this principle across multiple studies to rank genes by the number of data sets in which they demonstrate strong outlier profiles.
In a study published in the June 10 issue of Cancer Cell, a research team led by the University of Michigan used Meta-COPA to identify a specific gene expressed in a subset of prostate cancers that holds potential both as a diagnostic biomarker and a therapeutic target. The authors demonstrate in the paper that the gene, serine peptidase inhibitor, Kazal type 1, or SPINK1, is “specific to a subset of aggressive ETS-negative prostate cancers” that comprise around 10 percent of all prostate cancer cases.
“Traditionally, pharma has looked for targets that are very prevalent across an entire cancer population: ‘What’s the best target for colon cancer? What the best target for prostate cancer?’” said Daniel Rhodes, Compendia’s CEO and co-founder and a co-author of the paper. But as research reveals the heterogeneity of cancer, the “best targets are probably specific to particular subsets of cancer,” he explained.
Asked by BioInform via e-mail if he deemed the method and the study’s finding interesting, William Isaacs, professor of urology and oncology at Johns Hopkins School of Medicine, responded: “Quite!” This study represents “another excellent example of the use of bioinformatics to make discoveries by this productive group,” said Isaacs, who did not participate in the Cancer Cell paper.
Rhodes believes that when the meta-analysis capability is added to the Oncomine suite it will be able to nominate a whole new field of cancer targets such as SPINK1.
Rhodes led the development of Oncomine and is also a part-time faculty member at the University of Michigan. Rhodes and several of his colleagues founded Compendia as a UM spinout in 2006 [BioInform 02-16-07].
Oncomine is a curated set of more than 25,000 transcriptome profiles that is updated quarterly. It includes an analysis engine with web-based functionality for data mining and visualization. Oncomine is free for academic and non-profit users and the company offers a “professional” version for biotech and pharmaceutical customers with additional features, such as the ability for users to export data to Ingenuity’s Pathway Analysis software.
“This study really shows the power of this meta-analysis combined with outlier analysis and now our development team is building that into the product.”
Rhodes said that Compendia sees this study as validation for the meta-analysis feature of Oncomine, which will be included in the product’s fall release. The company is rolling out “almost a complete overhaul of the Oncomine system, really focusing on improved search and browse functionality, meta-analysis functionality, being able to easily ask questions across the collection of data, and export functionality,” he said.
“This study really shows the power of this meta-analysis combined with outlier analysis, and now our development team is building that into the product,” said Rhodes.
Meta-COPA, he said, can empower scientists to use Oncomine in different ways, not only to look over data that has been published over the last several years, but to “look at it with this novel analysis method to nominate novel cancer targets.”
Scott Tomlins, a researcher in UM’s department of pathology and a co-author on the study, explained to Bioinform that different tumors may develop via different pathways and so it is important to find the molecular drivers of a specific cancer.
“A biotech or pharmaceutical company might find that 20 percent of the tumors respond to their compound and then focus on [the question], ‘What is it that is making those 20 percent, but not the other 80 percent, [respond]?’” said Tomlins.
COPA reveals candidate genes that have “very high expression in only a fraction of samples,” he said. “That gives you a prioritized list of genes that are interesting and then it is up to understanding the biology and figuring out of any of those have a functional role.”
Pinpointing an Evil-doer
The Cancer Cell study indicates that SPINK1 plays a role in an aggressive subtype of prostate cancer with a less favorable prognosis for patients. SPINK1 is not a novel gene. It has “been extensively characterized in pancreatitis,” Isaacson told BioInform. However, its role in aggressive prostate cancer was previously unknown.
The UM researchers previously used Oncomine and COPA to identify fusions in two genes of the ETS family of transcription factor genes in prostate cancer, work that was published in Science in 2005. Gene fusions in oncogenes play a role in 50 percent to 70 percent of prostate cancers, leading to the hypothesis that these molecular changes may play a role in other solid tumors as well, Rhodes explained.
Following up on that research, the team applied the Meta-COPA method to seven prostate cancer profiling studies that had been integrated in the Oncomine database to calculate which genes were outliers in multiple data sets and also had high expression in the fusion-negative cancers only.
That analysis delivered SPINK1 as the most likely candidate oncogene for ETS-negative prostate cancers, and the possible driver gene for this aggressive subset of prostate cancers.
As Rhodes explained, in traditional microarray analysis scientists look for genes that are upregulated in all prostate cancers relative to normal prostate. “The outlier analysis is very different in that it looks for remarkable changes in expression, even if in only small subsets of patients,” he said.
“When you think about it, it doesn’t make a whole lot of sense to group all prostate cancers together and expect to find the key driver genes, knowing that it is a heterogeneous collection of diseases,” he said.
What made this study different is that the team did not just apply the COPA method to one microarray study. “We applied it to seven different studies, and prioritized our candidates by which genes were high-ranking outliers in multiple studies: The analysis nominated SPINK1,” Tomlins explained.
“This idea of meta-analysis across multiple independent studies provides lots of power in discovery — if something shows up multiple independent times then it is almost certain to be true,” Rhodes said.
A further analysis step revealed that SPINK1 cases and ETS gene-fusion cases were mutually exclusive. “So what it looks like is that prostate cancer can be driven by the ETS gene fusions or by SPINK1,” Rhodes said.
The team validated their in silico results with immunohistochemistry and fluorescent in situ hybridization in other prostate cancer patient cohorts.
“We used Oncomine for the in silico part of the study to nominate SPINK1 as a candidate and then we did a lot of experimental validation on different cohorts using different techniques and we found that the results were the same,” said Tomlins.
“I do not know of any other software that is similar, but this is not my area [of expertise],” explained Johns Hopkins’ Isaacson. However, he added, “It seems to be very useful, given the result.”