The quest to understand the genetic causes of Alzheimer’s disease has followed an ancient mythic ratio—scads of thorny challenges and obstacles for every triumph. This was true of the four best-known Alzheimer’s genes, APP, PS1, PS2, and APOE. Mutant versions were uncovered only after years of effort. Over the past decade, hundreds of genetic variants have been offered up as possible risk factors, mostly for the late-onset form of the disease, only to be knocked down by disconfirming studies.
“People would say ‘Hey, we’ve found a great gene,’ and another paper would come out saying, ‘We don’t see it, it’s not an Alzheimer’s gene,’” said Lars Bertram, HMS assistant professor of neurology at Massachusetts General Hospital. “It was like a ping-pong game, back and forth.” Three years ago, Bertram had an idea: why not collect every study of every putative Alzheimer’s susceptibility gene and collate them in such a way that viewers could evaluate the merits of a particular candidate? Working with Rudolph Tanzi, Matthew McQueen, and colleagues, he launched the database, AlzGene, in 2004.
Now the researchers have gone a step further. They have systematically analyzed 789 papers covering 802 polymorphisms in 277 genes—all the papers collected in AlzGene up to December 2005. From these, they have identified the most viable susceptibility gene candidates. The 13 genes are published in the January Nature Genetics.
“One of the milestones achieved by this paper is to assess where we have arrived after the Alzheimer’s disease gene–association craze of the last 10 years,” said Tanzi, HMS professor of neurology at MGH. “Out of hundreds of genes, here is this list of 13, and the effects are small.” In fact, one of the genes has already been pushed off the list, a casualty of recent studies. Rather than serve to demoralize researchers, Tanzi and Bertram hope the list will help scientists overcome two opposing tendencies that have held back the field. Many have been racing to test every new single-nucleotide polymorphism (SNP) that comes along. Others have been playing it safe, devoting their efforts to the tried-and-true APP, APOE, and the two presenilins, which together account for about a third of the genetic basis of the disease.
“Here we are as a field putting all our biological research, all our drug research, into understanding four known genes, and they probably represent only 30 percent of the genetic puzzle. That is why it is so important to find the rest of the disease genes,” said Tanzi, who is director of the genetics and aging research unit at MGH. “For the first time there’s a paper that objectively says, ‘Pay attention to these for now.’”
Intriguingly, two genes identified in Tanzi’s lab are not on the list, yet one of them, ubiquilin 1, is the subject of three new publications coming out of Tanzi’s lab—which points to a paradox. Some genes may play a crucial role in the biology of the disease and yet may not appear altered in Alzheimer’s patients. “Beta secretase is the number one target of drug companies for Alzheimer’s. It is not genetically associated with disease, but it is clearly a biological player,” said Tanzi. “So your gene doesn’t have to be on this list or be genetically associated with disease to matter.”
A Database for Disease Risk
Figuring out which genes to study had been a frustrating exercise for Bertram. Trained as a physician in Germany, where he saw hundreds of Alzheimer’s patients, he wanted to work on the most compelling genes, but found it difficult to pick them out from the growing mountain of association studies. Part of the problem is that unlike APP, PS1, PS2, and APOE, which exert relatively strong genetic effects, most of the new candidates are quite weak—they increase a person’s risk for developing Alzheimer’s only slightly and so may show up only in larger populations. He hatched the AlzGene idea, which pools samples from several studies. With Tanzi, he approached the Alzheimer Research Forum, which agreed to host it on the forum website. They started work in 2003, collecting and reading every peer-reviewed, English-language study of a putative Alzheimer’s susceptibility gene.
The next step was to extract and display the essential features of each study. Association studies typically measure how often a particular polymorphism occurs in Alzheimer’s patients and controls, and then calculate the odds that a person carrying the gene variant will develop disease. These genotype distributions and odds ratios—along with information about patient and control populations, such as size, age, gender, and ethnic makeup—are displayed in a user-friendly form in AlzGene. The database was launched at the 2004 International Conference on Alzheimer’s Disease and Related Dementias and immediately created a buzz. “AlzGene has become the authority you have to answer to,” Tanzi said.
Bertram, who conceived of the database as a means of finding the most promising susceptibility genes to study, was ready to carry out his search. He and McQueen, then a doctoral student in the Department of Epidemiology at HSPH, decided they would pool the genotype distributions for each polymorphism and calculate a summary odds ratio. They tried it out on APOE, the subject of a previous, more labor-intensive meta-analysis and compared the results. “They were almost identical,” said Bertram.
Having validated the method, they performed meta-analyses for the other polymorphisms in the database. Though the database contained 802 variants, only 127, occurring in 69 genes, had enough studies to carry out a meaningful analysis. Of these, 20 variants in 13 genes had summary odds ratios that were statistically significant, though the effects appear weak.
“The genes in the final list lend themselves to pretty easy biological stories,” said Tanzi, which, he added, is not surprising since they were chosen as markers because of their purported role in the disease. Two of the genes on the list give rise to the proteins nicastrin and presenilin1, members of a key Alzheimer’s agent, the gamma-secretase complex. Another pair, angiotensin-converting enzyme (ACE) and insulin-degrading enzyme (IDE), play a role in degrading Abeta, the precursor to amyloid plaques.
Tanzi proposed IDE as a susceptibility gene several years ago, though two other picks are not on the list, A2M and ubiquilin 1. “Ubiquilin 1 didn’t make the list, but that has not in any way dissuaded me from working on ubiquilin biology,” said Tanzi. In fact, Mikko Hiltunen, a former HMS research fellow, and Alice Lu, an HMS graduate student, along with Tanzi and colleagues, report in the Oct. 22 Journal of Biological Chemistry, that ubiquilin 1 may act as a gatekeeper inside neurons, controlling how much of the Abeta precursor, APP, gets to the cell surface.
The paradox, that genes with compelling biological stories may not be good predictors of disease, has a flip side: the most promising association genes may be those that do not yet have a neat biological story. “Genetics is best done with no biological bias,” said Tanzi. In fact, he, Bertram, and colleagues are currently screening the entire genome of 1,500 patients and controls. With funding from the Cure Alzheimer’s Fund, a private foundation, they are looking at 500,000 SNPs. Preliminary results are intriguing. “I would not have picked one of them, which is exactly what we want,” he said. “Genetics should surprise you.”
And yet, even the whole-genome approach may not be without bias. “It’s a technological advance but, in essence, it will produce the same type of result as the many previous single locus studies,” said Bertram, who is currently establishing an AlzGene-like database for Parkinson’s disease and schizophrenia. “You are going to have a resulting set of genes that look very promising, but still you are going to have to throw that set out there. Is it going to be replicable in other studies? Will it hold up in meta-analysis once enough studies are published?”