By Deborah Janssen
August 18, 2004 | Scientists at the University of Michigan’s Comprehensive Cancer Center have identified a common “meta-signature” of 67 genes that appears to be required to change normal human cells into cancerous ones. The report, published in the June 7 online edition of the Proceedings of the National Academy of Sciences, is based on statistical analysis of a cancer microarray database called Oncomine (www.oncomine.org), which the Center has established during the past year.
More than 200 studies of global gene expression in cancer have been published, including millions of data points from thousands of experiments using different microarray technology platforms. But sifting through those gene-expression patterns to identify those changes directly relevant to cancer development is a formidable task.
Oncomine is a bioinformatics infrastructure that allows cancer biologists to mine published oncology microarray data. The Michigan scientists, led by Oncomine co-developer Daniel Rhodes, surveyed the database, which includes some 90 data sets from more than 7,000 microarray experiments, to identify common core groups of genes that are activated in cancer, regardless of the cancer type. They developed a technique called metaprofiling, which performs a common statistical student’s t-test on each data set, and compares statistical measures across data sets to examine if common genes have changed.
“We performed the t-test across all data sets that had cancer tissues of a particular origin and normal tissues of that same origin, identified the genes activated in each respective cancer type, and used the metaprofiling approach to assess if a core set of genes existed that are continually activated in cancer,” explains Rhodes, the paper’s first author. The analysis revealed 67 statistically significant genes, and was enriched across multiple data sets in most major cancer types.
Rhodes notes that microarray data sets are typically statistically challenging because each data set collected uses different platforms and experimental methods, hindering direct comparison. The uniqueness of the meta-profiling approach is that instead of directly comparing data points from data sets, the method performs the statistical test on each data set individually and subsequently compares the statistical measures. The group validated the meta-signature by applying the 67 genes to 10 to 12 additional independent data sets, including several new cancer types.
The Michigan team also looked for a common signature of genes in undifferentiated cancer samples. (Undifferentiated tumors, which don’t resemble the tissue of origin, typically behave poorly and lead to poor patient outcomes.) “Across multiple different cancer types, there is a meta-signature of 69 genes that is always activated in the undifferentiated cancers but not the well-differentiated cancers, which points to a common signature of genes that leads cancer cells to undifferentiate and behave aggressively,” Rhodes says.
Interestingly, there is an overlap between the 67- and 69-gene signatures. The majority of genes in common are connected to cellular proliferation, while a subset of genes unique to the undifferentiated meta-signature is associated with early embryonic development. Rhodes suspects a link between the genes that are activated in early development and those that become activated in undifferentiated cancer. “We speculate that the genes are programmed to remain ‘off’ in normal human tissues, but when they get turned ‘on’ these genes lead cancer down the path to become undifferentiated,” he says.
Some of the genes in the universal cancer meta-signatures have the potential to become new therapeutic targets. For example, TOP2A, present in the 67-gene meta-signature, encodes a protein, topoisomerase, which is the target of several chemotherapeutic agents. “We believe that there are other important therapeutic targets in this gene set, and by attacking a number of these targets simultaneously, we will be able to effectively treat cancer,” Rhodes says.