America's National Cancer Institute (NCI) has confirmed that its much-heralded cancer genome atlas is off to a slow start. And scientists say that the obstacles facing the project are just a taste of those that will soon confront all researchers who use large-scale methods to analyse material from tissue banks.

The techniques used to store tissue samples might distort the genetic information. Credit: H. MORGAN/SPL

Nine months ago, US National Institutes of Health director Elias Zerhouni announced the “scientific start” of the Cancer Genome Atlas, which aims to catalogue the genetic characteristics of various cancers at a previously impossible level of detail. The project was controversial even before it started, with some scientists afraid that it would divert resources from investigator-initiated research in tight budget times (see Nature 438, 894; 2005). But it prevailed, thanks in part to strong support from Francis Collins, director of the National Human Genome Research Institute (NHGRI) in Bethesda, Maryland.

Now, the NCI has confirmed that the project stumbled as soon as it left the starting blocks. The problem lies with brain-cancer samples, which are removed surgically from patients and stored at a tumour bank at the MD Anderson Cancer Center in Houston, Texas. The centre was chosen to provide the Cancer Genome Atlas with samples of glioblastoma multiforme — a type of brain cancer that is almost always fatal. Glioblastoma is one of three cancers that will be analysed in the atlas's US$100-million three-year pilot programme, which is funded jointly by the NCI and the NHGRI. The other two diseases being analysed are lung and ovarian cancers.

The pilot aims to analyse at least 500 samples of each cancer. But this February, when scientists started to retrieve tissue samples from the bank, “we started realizing the scene wasn't as rosy as we had hoped”, says Carolyn Compton, director of biorepositories and biospecimen research at the NCI.

Deadly diseases: the fatal brain cancer glioblastoma can be hard to sample. Credit: PHOTOTAKE INC./ALAMY

So far, the project has analysed fewer than half the number expected because about half of the samples at the MD Anderson bank haven't met the project's stringent quality tests, according to cancer geneticist Ronald DePinho from Harvard Medical School in Boston, Massachusetts, who is co-chair of the atlas's external scientific committee.

Compton and other scientists say this problem isn't unique to the MD Anderson tumour bank. Variable sample quality is a concern with all tissue banks, scientists say, because every bank has its own standards, and samples aren't usually checked for quality until someone needs them for a project.

“We haven't had the large number of samples we expected,” says Lynda Chin, a cancer geneticist at the Dana-Farber Cancer Institute in Boston, Massachusetts, and co-head of one of the atlas's seven Cancer Genome Characterization Centers. “But this is the reality of the limitations of the existing tissue banks.”

The Cancer Genome Atlas set high standards for its samples to ensure high-quality data. All the samples included in the project must have at least 80% viable cells, because the project is analysing DNA and RNA only from live cells. The samples also have to be quite large — weighing at least 200 milligrams — because the tissue has to be shared between the many analysis centres involved in the project.

The high standards are part of the reason that the project stumbled with the glioblastoma samples. Glioblastoma tumours tend to be spread throughout the brain, and must by definition contain necrotic, or dead, tissue. So scientists expected that some tumours from the MD Anderson bank would contain too much necrotic tissue to meet their criteria — but not this many.

The failure of so many samples has prompted the Cancer Genome Atlas to expand its criteria to admit brain-cancer samples with more necrotic tissue. It has also started to take samples from brain tumour banks at the Henry Ford Health System in Detroit, Michigan, and the University of California, San Francisco.

Compton says that the lung and ovarian cancer projects shouldn't have the same problems as the one for glioblastoma, because those tumours tend to be larger. She still hopes that the pilot project will complete its work in the scheduled three years.

But changing the criteria for glioblastoma samples won't solve the bigger problems with tissue banking. Foremost among them is how the procedures used in tissue banking affect the nucleic acids and proteins in the stored samples. For instance, it's not clear how anaesthesia affects gene expression during surgery to remove tumours, Compton says. That could mean that signatures gleaned from banked tissues might just be by-products of procedures used in surgery and storage — not markers of disease processes. “The state of the science right now is pathetic,” Compton says.

Because the science is so poor, the NCI has decided, for the first time, to fund studies on biospecimen banking. On 28 June, Compton is scheduled to present the agency's biospecimen research plan to its board of scientific advisers. If it is approved, the NCI will award scientists money to study how collection and banking processes affect the nucleic acids and proteins embedded within tissue samples.

The NCI is also working with research agencies around the world to try to encourage tissue banks everywhere to adopt uniform standards. Researchers in many countries are grappling with the same problem. The International Agency for Research on Cancer, a World Health Organization body based in Lyon, France, with 20 member states, is also trying to address it.

In the meantime, DePinho insists that the Cancer Genome Atlas is doing well for a new venture of its scope. “This is an extraordinarily difficult project, so these hiccoughs at the embryonic stage are to be expected,” he says.