24 July 2009. On the last morning of the International Conference on Alzheimer’s Disease (ICAD), held 11-16 July in Vienna, Austria, investigators from different research groups marked a milestone in AD genetics. They reported results from different genomewide association studies (GWAS) that, thanks to pooling samples by many different laboratories, reached a size where new risk genes of small effect not only showed up with genomewide significance but also, importantly, surfaced repeatedly in several independent GWAS. Not until a gene finding stands up across the board, as ApoE does time and again, does it persuade scientists that it is incontrovertibly true. Last Thursday’s ICAD session breached a barrier of sorts as two groups not only presented data from GWAS that included up to 20,000 samples, but also confirmed each other’s signals. More broadly, the session reflected a shared optimism that growing the sample size for GWAS—both by recruiting more patients and by pooling datasets—has brought them within reach of a list of genes that together account for much of the population’s attributable risk for AD. By upping sample sizes even further past 20,000, some geneticists expect to have a replicated set of genes within the next few years. “Today we saw the best data in a long time. This is an exciting time for complex genetics, which had difficulty this past decade replicating findings. Things are changing and we are beginning to find robust results,” said Julie Williams of Cardiff University in the U.K.
Genetics has been driving AD research ever since it entered the scene. The first wave uncovered rare mutations in APP and presenilins 1 and 2, which account for about 1 percent of cases, plus one common susceptibility gene, ApoE4, which accounts for roughly 15 percent of the population risk, said Philippe Amouyel of Institute Pasteur in Lille, France. Since ApoE’s discovery by Allen Roses at Duke University in 1993, more than 1,200 linkage and association studies have unearthed some 580 additional genetic locations that might contain susceptibility genes. However, initial fanfare about most new proposed genes quickly died down when researchers were unable to replicate a new finding in their own samples. The sample sizes of most everyone’s studies were too small to pick up genes of small effect. To organize and analyze this increasingly unwieldy literature, Lars Bertram, who recently moved from Massachusetts General Hospital to the Max-Planck Institute for Molecular Genetics in Berlin, Germany, developed the science behind, and curates, the publicly available AlzGene database. At ICAD, Bertram noted in his talk that nearly 30 proposed genes show at least one variant with a modestly significant odds ratio when all published studies on them are taken together. AlzGene was the entry point to many genetics talks in Vienna, as groups around the world—besides looking for new genes—are now routinely testing which of the AlzGene Top Results their own studies confirm.
The big push in genetics these days is to conduct GWAS. In these studies, researchers use arrays containing some 20,000 to one million sequence variations called single nucleotide polymorphisms (SNPs). With those they scan a person’s entire DNA for association of any SNPs with disease risk in an unbiased way. In other diseases, for example, diabetes, a trove of new genes were found in large GWAS projects by the Wellcome Trust Case Control Consortium, the Diabetes Genetics Initiative of the Broad Institute/Novartis, and a group coordinated by the National Human Genome Research Institute (Saxena et al., 2007; Zeggini et al., 2007; Scott et al., 2007). In the two years since then, 18 more genes have been published and 31 more are unpublished, according to Williams. These most recent advances came out of meta-analyses and further data merging, which boosted sample size to 55,000 cases. “We need to be thinking about this number of cases to find genes in AD, too,” said Williams.
And the field may be ready to get there. Alison Goate of Washington University, St. Louis, who contributed 800 samples to the British/U.S./German GWAS, put it this way: “We all worked for 10 painful years on our own samples and did not get anywhere. So everyone saw that there was something to be gained from pooling samples.” Earlier this spring, the National Institute on Aging announced that it has awarded nearly $20 million to fund an AD Genetics Consortium. Headed by Jerry Schellenberg at the University of Pennsylvania, Philadelphia, the ADGC is charged with finding the remaining LOAD risk genes in a set of 20,000 case/control samples to which many genetics groups in the U.S. will contribute. “It is important to applaud people who are willing to share their data with other groups to move the overall field forward,” said Williams.
Williams claimed that experience with large GWAS in other diseases has already done away with the myth that these expensive undertakings merely produce a jumble of random genes without meaning. To the contrary, the genes that come up tend to fall into clusters that highlight new biological pathways at play in a given disease. An example from diabetes would be pancreatic β cell regeneration, which came out of the large GWAS. Moreover, this work has also shown that proteins whose genes have merely a small genetic effect can be biologically important, Williams said. An example there is PPARγ. Its odds ratio in diabetes genetics is but 1.14, but its role as the target for some of the insulin-sensitizing drugs prescribed for diabetes leaves no doubt about its biological importance.
In AD research to date, nine GWAS have been published (see AlzGene GWAS page). They are smaller than the new ones reported at ICAD; each group finds ApoE but beyond that mostly weak signals that fail to reach genomewide significance and vary from study to study. This should not be seen as a problem, Williams emphasized. “A smaller GWAS is in no way wasted if it does not find results right away. If we combine these to grow the dataset, we may still nail genes with compelling evidence,” said Williams. Her team’s large GWAS, as well, brought up a smattering of candidate genes below genomewide significance that even larger sample sizes may be able to separate into true hits and false-positives.
On this study, Williams worked with investigators at 11 institutions in the U.K., Ireland, Germany, and the U.S. to pull together the largest GWAS in AD to date (see ARF related news story). In toto, the study included some 20,000 samples. The team screened nearly 5,000 cases and 10,000 aged controls. After quality control, they included some 4,000 cases and 9,000 controls in the first discovery set, which was typed with chips containing some 500,000 SNPs. The top hits from this round were then checked in a separate replication set of 3,000 AD cases and 4,000 controls. Besides ApoE, this study fingered two genes that reached genomewide significance. One is ApoJ/clusterin on chromosome 8. An intron in this gene accounted for 2.3 percent of AD risk in the sample, Williams reported. The other is PICALM, on chromosome 11, and it showed up via a SNP in its 5’ untranslated region. Just below the cutoff for genomewide significance, the researchers spotted 13 variants, many more than the four they had predicted to see by chance. “These probably are false-negatives worth exploring,” said Williams.
One of these runners-up is CR1 on chromosome 1, a gene encoding a complement receptor. CR1 happens to be one of the two major hits of the second-largest GWAS in AD to date, also first presented in Vienna. (The other major hit of that second study was—ApoJ/clusterin.) Presented by Amouyel, this is a European study that included some 2,300 AD cases from the French cities of Lille, Bordeaux, Montpellier, Rouen, Paris, and Dijon, and compared them to 7,000 well-characterized controls from the population-based Three Cities (3C) study. After quality control, about 2,000 cases and 5,300 controls remained for analysis in the discovery sample. The stage 2 replication set included samples from 15 centers in other European countries, including Belgium, Finland, Italy, Spain, and the U.K., and amounted to some 4,000 cases and slightly fewer controls. This study used a chip containing some 600,000 SNPs. “In our studies, no other gene [besides ApoE] has shown an association with AD as strong as CR1 and ApoJ,” Amouyel said in his talk.
Regarding PICALM, Amouyel said that he had heard about it only the night before and that his colleagues were checking for it at present. This gene did, however, pop up in a third, smaller GWAS presented the same morning by Matthias Riemenschneider of the Technical University, Munich. His study used some 1,000 cases and 1.500 controls from Germany for discovery, and 1,300 cases and 1,700 controls from several groups in Australia, Italy, and Sweden for replication. In his talk, Riemenschneider focused on his group’s discovery of TMEM23, a gene encoding a sphingomyelin synthase, and on secondary analyses linking genes to endophenotypes; however, in response to a question from the session chairs about ApoJ and PICALM, Riemenschneider said that his team saw significant associations for those two genes as well.
What do these genes do? Each points to different functions that could fit within the amyloid hypothesis, but could also contribute to LOAD more broadly via lipid metabolism and innate immunity, said Williams. Apolipoprotein J/clusterin is an abundant glycoprotein that occurs in all body fluids; in the brain, astrocytes secrete it. Sometimes called a chaperone, clusterin has been linked to protein clearance in general and Aβ clearance in particular. Genetically, ApoJ/clusterin has been a bit player until now (see ApoJ on AlzGene), but its history in AD research goes back 15 years to its initial description in Blas Frangione’s group (Koudinov et al., 1994). Its role was seen as being intimately tied up with ApoE and with Aβ transport and clearance (e.g., Oda et al., 1995; Zlokovic, 1996; Demattos et al., 2004). Also at ICAD, Madhav Thambisetty from the National Institute on Aging in Baltimore reported on behalf of the London-based AddNeuroMed consortium that ApoJ/clusterin was the only protein that came up as an antecedent plasma biomarker in studies combining plasma proteomics with PIB amyloid imaging in participants of the Baltimore Longitudinal Study of Aging. (This finding still needs replication in independent datasets.) More generally in biomedicine, though, ApoJ is being studied in the context of aging, injury response, and modulation of the innate immune system.
The innate immune system is where CR1 fits in, too. This gene encodes the receptor for proteins of the classical complement pathway, such as C3b/C4b, which mediates clearance of antigen-antibody complexes and other substances through the complement cascade. Complement activation around amyloid plaques has been spotted even during the early wave of interest in inflammation in AD (Eikelenboom and Stam, 1982). It’s not clear yet whether complement activation in AD fuels or slows the disease process. At ICAD, Amouyel said that data gathered to date suggest that CR1 tends to play a protective role via Aβ phagocytosis (e.g., Rogers et al., 2006; Webster et al., 2000). Mouse data, as well, connect complement activation with amyloid deposition (e.g., Maier et al., 2008 and ARF related news story; Wyss-Coray et al., 2002). Beyond that, basic research raises other possibilities for a role of complement in the brain as well. For example, one study has implicated the classical complement cascade in the physiological, activity-dependent pruning of synapses in development, raising the question of whether the loss of synapses in AD could have something to do with complement activation as well (Stevens et al., 2007). For her part, Williams encouraged the audience to take a wider perspective in future efforts at understanding the functional importance of CR1 and the other new genes.
The third new gene, PICALM, is short for “phosphatidylinositol-binding clathrin assembly protein,” and also goes by the alias CALM. It participates in clathrin-mediated endocytosis and intracellular trafficking. This gene and its product are poorly understood, but some previous data hint at a role at synapses. That is because clathrin-mediated endocytosis of APP from the cell membrane affects not only APP’s subsequent cleavage, but also the activity-dependent release of Aβ and Aβ levels in the interstitial fluid (e.g., Cirrito et al., 2008 and ARF related news story; also Rudinskiy et al., 2009). An ongoing collaboration with David Sattelle’s group at the MRC Functional Genomics Unit in Oxford, U.K., at ICAD reported initial findings indicating that PICALM influences plaque deposition in a C. elegans amyloid model, Williams noted.
Readers may wonder where some of the usual suspects fell in the new GWAS data. In the British-international GWAS, APP and the presenilin genes did show up but fell far below genomewide significance. Tau was equally weak, even though it plays a role in the complex genetics of PD (see tau in PDGene; ARF related Prague story). “This is telling us that overproduction of Aβ is not the primary route into LOAD, nor is tau dysregulation,” said Williams. TDP-43 did not show up either, even though it is present upon postmortem pathology in a fraction of AD cases. Ubiquitin was absent as well.
As an aside to the GWAS coverage here, attendees also asked about Tomm40. This gene could not be studied individually in these GWAS, because the Illumina chips lack ApoE itself and scientists actually use Tomm40 as their ApoE marker. Questions about this gene came up repeatedly because earlier in the conference, Allen Roses of Duke University in Durham, North Carolina, had presented data from three separate series of AD patients and controls. The discovery set included 74 patients and 31 controls (with 210 alleles between them); the conformation set included 72 patients and 60 controls (with 264 alleles). For the demonstration of the age of onset of Tomm40 variants connected to ApoE3, Roses used 40 autopsy-confirmed AD patients with the ApoE3/4 genotype whose age of onset had been part of their Duke ADRC record. This work was very different from a GWAS. It used deep sequencing and a phylogenetic analysis of the DNA in the vicinity of ApoE. It set out to explore whether other genes near ApoE could help account for this gene’s extraordinarily large contribution to Alzheimer disease risk and age at onset. Roses reported considerable variation of poly-T variants (long and short forms) of the Tomm40 gene. In people who inherited a long form of Tomm40 along with ApoE3, the age at clinical onset was seven years lower than in people who carried ApoE3 along with less risky short variants of Tomm40. In this way, Tomm40 could help explain why many ApoE3 carriers develop AD, Roses suggested. All Tomm40 variants connected to ApoE4 were long. If confirmed, this finding would mean that combined genotyping of ApoE and Tomm40 might predict a person’s risk and likely age at onset with more precision than do current ApoE tests alone. Roses plans to test this prediction within a diagnostic population-based validation study coupled to a prevention trial for people who are at high risk within the next five to seven years. This project is designed and managed by a startup company called Zinfandel Pharma (see also WSJ story).
Tomm40 is biologically interesting because it encodes a translocase enzyme that mediates protein transport across the outer membrane of mitochondria, which are drawing intense interest in AD research these days. For example, a study by scientists including Eric Reiman at the Banner Alzheimer Institute in Phoenix, Arizona, who donated samples to Roses study, showed that most genes encoding subunits of the mitochondrial electron transport chain were downregulated in neurons taken from regions of the brain whose glucose metabolism declines early on in AD (Liang et al., 2008).
At ICAD, other geneticists were variously intrigued and skeptical about Roses’ study. Some dismissed the idea of two major AD genes occurring so close together as “intelligent design,” and claimed that sequence-independent variation in ApoE expression explains the finding more parsimoniously. Others noted that the small size of the study made it prone to error, and cautioned that the complexity of this genetic locus would make replication challenging. Yet others, however, said that they would follow up the finding and try to tease apart effects of the ApoE alleles and Tomm40 variants.
Wrapping up the last genetics plenary, Williams encouraged a renewed focus on studying the function of all new genes above and just below genomewide significance in GWAS and Tomm40. Based on an initial pathway analysis of these genes, Williams concluded that cholesterol and sterol factors, as well as innate immunity, are primary to disease development. “That is my crucial take-home message. Now we need to know why cholesterol and inflammation are important. We should make them center stage in AD research,” she said.
A lot of work remains. Scientists are already starting to use these new datasets to probe them for links to the presence of psychosis, depression, or vascular risk factors in AD, or to try to link them to biomarker data. Importantly, scientists agreed they need to study expression, epigenetic modification, and function of the genes in the requisite tissues and understand how the AD-linked SNPs change things away from normal. This is easier said than done. After a talk at ICAD describing his efforts to unravel the details of this process for tau, Richard Wade-Martins of the University of Oxford, U.K., said, “We still don’t understand exactly how the tau H1 haplotype leads to increased risk, even though that association was first described in 1997. I hope we won’t say that about ApoJ, CR1, and PICALM in 12 years’ time.”
When all was said and done, John Hardy, a neurogeneticist at University College London, who chaired the session with Christine van Broeckhoven from VIB/University of Antwerp in Belgium, called into the auditorium, “Can we now declare that we have three new susceptibility genes for Alzheimer disease? Lars Bertram, are you here? You be the referee.” Stepping out of the dark, the AlzGene guru said, “I appreciate the authority. I’ll need to see more data, but based on what I saw today, I’d say we are now as close to declaring three new genes for AD as we have been in a long time.”—Gabrielle Strobel.