Sometimes a project doesn't quite produce expected results. In their search for copy-number variants associated with Parkinson's disease, Nathan Pankratz and his colleagues thought at first that they'd uncovered new CNVs, but those findings did not hold up under further scrutiny. "We thought that we had a couple of good leads in the current study but when we tried to validate them molecularly, it failed," says Pankratz, who was at Indiana University School of Medicine at the time. "We learned a lot more about methodology and what needs to be done to validate things than we did anything else."
Copy-number variants have long been associated with Parkinson's disease. Two main genes are associated with the disease: park 1, or α-synuclein, and park2, or parkin. Parkin, Pankratz notes, has always been known to contain all sorts of mutations, including point mutations, deletions, and duplications. In addition, it is linked to early disease onset. More recently, Pankratz adds, researchers discovered that α-synuclein could be tripled so that there are extra copies of the normal, wild-type gene, which also leads to Parkinson's. Researchers have also found duplications in α-synuclein. "The interesting thing was that triplication had an early onset and more severe phenotype than those with duplications," Pankratz says. "We were hoping to find some other genes that were related to Parkinson's disease that might be deleted or duplicated more in cases than in controls."
In a study published in Neurology in 2009, Pankratz and his colleagues reported that a single dosage mutation in parkin could be a risk factor for Parkinson's disease — usually, two mutations in parkin are needed. "There's been a couple of papers supporting this theory," he says. "I know of one large paper that did not find this association so the jury's still out on whether single dosage mutation can lead to increased risk or not. "
More recently in an August PLoS One paper, Pankratz and his colleagues searched for novel Parkinson's-related CNVs in 816 cases and 856 controls from whole blood samples and cell lines, respectively, using genotyping chips and two copy-number-calling algorithms. From this, they homed in on two prospective CNVs, DOCK5 and USP32. The researchers then followed up on those prospective CNVs, and found that they were in fact artifacts.
The artifacts, Pankratz says, were likely due to characteristics of both the study samples and the calling algorithms used. The biggest problem, he says, is that the cases came from whole blood while the controls came from cell lines. "That introduced its own set of artifacts," he says. A Wellcome Trust Case Control Consortium study saw similar artifacts when using samples from both whole blood and cell lines, he adds. That the calling algorithms gave different results is not unheard of. "Anybody who has looked at CNVs in detail has noticed that if you use a different algorithm, you get breakpoints and you get different CNVs that are called," Pankratz says.
However, when that occurs, larger CNVs — like those that had previously been detected in parkin — are easier to validate, given their larger size.
Some researchers then turn to consensus calls from a number of calling algorithms, but Pankratz says that, too, has its problems. In this case, the two calling algorithms the researchers used, PennCNV and QuantiSNP, would not have reached significance on parkin, which has been validated and is a true positive, he says.
There are newer, dense arrays that are better suited to uncovering CNVs, but Pankratz says his lab won't be using them for discovery work. Instead, his lab aims to find CNVs associated with cancer and cardiovascular disease, which are major foci of research at University of Minnesota School of Medicine, where he now is. "I am using this knowledge, but I am not following up on PD," he says.