By some measures, ApoE found its first companions in 2009 when scientists reported three additional genes robustly linked to late-onset Alzheimer disease. Now, in this week’s JAMA, a genomewide association study (GWAS) of more than 35,000 people—the largest to date for AD—has revealed two more risk genes and confirmed two of the 2009 loci. None confer nearly as much risk as ApoE, but all have reached genomewide significance in multiple independent GWASs—something the vast majority of genes emerging from smaller GWASs fail to do. One of the novel genes in the current analysis is near BIN1, which encodes a protein involved in clathrin-mediated trafficking. The second lies close to two genes (BLOC1S3 and MARK4) in pathways linked to AD pathology. The previously identified CLU and PICALM also came up as major hits, yet did not improve AD risk prediction when added to a model that included age, gender, and ApoE status. Nevertheless, these and other small-effect genes may help guide researchers to biological pathways that contribute to AD pathogenesis.

For AD genetics research, and GWASs in particular, 2009 was a banner year. At the International Conference on Alzheimer’s Disease in Vienna (see ARF related conference story) and a few months later in two Nature Genetics papers (Lambert et al., 2009; Harold et al., 2009), scientists announced what looked to be the first three AD risk genes besides ApoE to reach genomewide significance in several independent GWASs. Those genes were CLU (clusterin/apolipoprotein J), PICALM (phosphatidylinositol-binding clathrin assembly protein), and CR1 (complement receptor 1).

In the current study, an international research team led by Monique Breteler of Erasmus University, Rotterdam, The Netherlands, and Sudha Seshadri of Boston University, Massachusetts, started by analyzing GWAS data from six cohorts in the U.S. and Europe. Two cohorts came from larger case-control studies (Translational Genomics Research Institute public release database and the Mayo AD GWAS); the others came from CHARGE (Cohorts for Heart and Aging Research in Genomic Epidemiology). More than 2,000 of the nearly 17,000 participants had AD. Because the cohorts had their DNA sequenced on different platforms and may have had other slight differences due to geography, the investigators figured it may be inappropriate to simply pool all datasets right from the start, Seshadri told ARF.

Rather than doing a more conventional GWAS—which pulls top hits from a large pool of samples and replicates them in independent cohorts—Seshadri and colleagues used a three-stage sequential approach. In the first stage, they identified 2,708 single nucleotide polymorphisms (SNPs) with moderate AD association in the six initial cohorts. All but 38 (on 10 loci) were left in the dust after the second round, which subjected the first-stage SNPs to stricter criteria in an expanded sample set containing the original six plus an additional European cohort. The strongest hits proceeded to the final stage—replication in a meta-analysis of all second-round samples plus a different non-overlapping cohort—bringing the total number analyzed to 8,371 AD cases among some 35,000 participants. Besides ApoE, just four SNPs emerged from the final gauntlet with genomewide significance—PICALM, CLU, and two novel loci on chromosomes 2 (near BIN1) and 19 (near BLOC1S3 and MARK4). These four associations were confirmed in an independent Spanish sample set.

It is interesting that the winners’ circle lacked CR1, the third gene identified in the 2009 GWAS. However, 13 SNPs within CR1 fell just below the predefined cutoff, and did reach genomewide significance in a post-hoc analysis. “We believe it’s real,” Seshadri said.

Using data on incident AD from the two largest CHARGE cohorts that had the information available, the researchers determined that PICALM and CLU did little to improve existing models for predicting AD risk. This is not surprising. In existing models, age and gender already explain a large part of the risk, suggested coauthor Philippe Amouyel of Institute Pasteur, Lille, France, who led one of the 2009 GWASs (Lambert et al., 2009). “Thus, adding one or two susceptibility genes or any other risk factor will not, in a general population sample, largely improve this prediction,” he wrote in an e-mail to ARF.

However, those genes may still prove useful in a clinical setting. Amouyel made this point by citing an example from heart disease. Genetic variability in HMG (3-hydroxy-3-methyl-glutaryl)-CoA reductase explains less than 10 percent of the variance of low-density-lipoprotein blood cholesterol levels. Yet this enzyme is currently the major target of statins, one of the most widely used drugs for staving off heart disease.

Similarly, targeting AD risk genes with small effects could help keep disease at bay. If interventions could wipe out the disease-related effects of the handful of genes identified thus far, “you could remove 20 to 30 percent of all AD cases,” said coauthor Julie Williams of Cardiff University, U.K. For common diseases like AD, “we all have some risk factors,” she said. “But it’s the accumulation of risk factors that is important. The accumulation takes us to a point that tips us into a disease state. By simply removing some effects of these risk factors, you can actually take people out of the risk zone.” It might also postpone age of onset. “Disease that starts at age 75 may not develop until age 90, for example,” she said.

Aside from clinical relevance, one of the most exciting things coming out of GWASs may be the identification of potential biological pathways that influence AD pathology. “A pattern is developing in the genes we identify,” Williams said. “These are not random genes. They are telling us a story.” One story from the current paper relates to BIN1, which was just underneath the threshold for genomewide significance in her 2009 GWAS (Harold et al., 2009). “Now it is a definite hit,” she said. “Both PICALM and BIN1 affect the same process in the brain—clathrin-mediated endocytosis and trafficking. You’ve got two genes that are telling us something new about components or factors that trigger disease.”

Beyond the potential biological insight, the study represents somewhat of a methodological milestone. “I think it's encouraging that researchers within the field are willing to pool data and are more enthusiastic to do that,” Williams said. In her mind, the gold standard would be a huge GWAS where multiple datasets are treated uniformly with the same quality control measures and pooled in a structured way, “so we can really extract all the information we can, and follow up what comes out of that in independent samples,” she said, noting this has succeeded for other complex diseases such as diabetes and hypertension. “That would be a really powerful way to get at more genes undoubtedly out there that are related to AD.”

At present, she is heading up a GWAS that will involve around 50,000 people. “I’m optimistic that we will find several more genes within the next few months,” Williams told ARF. Seshadri predicts GWASs could net 20 to 25 new loci over the next five years.

Seshadri expects future GWASs to have deeper computation and more complex analytical approaches—sequencing areas of interest, for instance, or looking at methylation and gene-environment interactions. “Maybe a gene only has an effect if you’re obese in midlife,” Seshadri said.

In the meantime, the current data reinforce the idea that “at the individual level, it makes far more sense to focus on things we know affect risk prediction and development of disease,” Seshadri said. “I would focus more on midlife vascular risk factors than on trying to determine whether you have the minor or major allele of the CLU locus.”

Potential impact of environmental interactions was also highlighted by Nancy Pedersen of Karolinska Institutet, Stockholm, Sweden, in her JAMA commentary on the present GWAS. “Very large sample sizes, on the order of those in GWAS consortia such as those Seshadri et al. report, are necessary for detecting significant gene-environment interactions,” she wrote. “Possibly more could be gained by focusing efforts in these consortia on incorporating information on environmental risk and protective factors in further collaborative efforts than in further pursuit of gene identification or replication.”—Esther Landhuis


  1. The methodology in this study is slightly different from the one we are used to: a GWAS developed on a large population sample with a selection of SNPs that reach statistical significance (p -8) and are then replicated in large, independent samples. In this study, the first stage involved the selection of 2,708 SNPs with genetic association (p -3) from a meta-analysis of six different studies (four constituting the CHARGE consortium). Then these 2,708 SNPs were evaluated in our GWAS EADI consortium. A meta-analysis of the six studies together with the EADI consortium allowed the selection of 38 SNPs in 10 loci that have a p -5. In the third stage, the most significant SNPs from these 10 loci were meta-analyzed with the GWAS GERAD consortium, allowing the discovery of two SNPs reaching the threshold for genomewide significance. Finally, the two new loci, one on chromosome 2 in the vicinity of BIN1, and one on chromosome 19 within BLOC1S3/EXOC3L2/MARK4, were confirmed in an independent Spanish case-control study. Thus, the major conclusion of this paper is that two new potential genetic susceptibility factors are identified that deserve in-depth analyses and other independent replications.

    Another interest of this study is that two cohorts from the CHARGE consortium—the Rotterdam Study and the CHS—allowed us to estimate risk of incident Alzheimer disease in the general population. The risk estimation improvement associated with these genetic susceptibility factors is weak compared to the information from classical risk factors (age and gender). This conclusion is also observed with other diseases like type 2 diabetes (Talmud et al., 2010). Should these results suggest that GWASs identify loci that have little influence on disease risk and thus do not add significant information to our knowledge of the disease?

    If we examine the estimation obtained in the Rotterdam Study of the AUC for a prediction model that includes only age and gender, it already reaches 0.826, which means age and gender explain a very large part of the risk. Thus, adding one or two susceptibility genes or any other risk factor will not, in a general population sample, largely improve this prediction. Therefore, in that situation, the interest of these risk factors as a screening tool for risk prediction is not obvious in the general population. However, that does not mean that genes coming out from GWASs are not helpful. Let us take the example of HMG CoA reductase enzyme, whose genetic variability explains less than 10 percent of the variance of the LDL-cholesterol blood level. This enzyme is today the major target for one of the most active classes of therapeutics for cardiovascular risk reduction—the statins. Thus, it is very difficult to estimate the importance of a new hit or a new pathway from the predictive value of its genetic variability in the general population. The genes identified in AD GWASs—i.e., CLU, PICALM, CR1, and now BIN1 and BLOC1S3/EXOC3L2/MARK4—need further in-depth investigation to understand how they interfere with Alzheimer disease.

    A last remark about the individual clinical interest of these genetic susceptibility risk factors comes from a recent article (Ashley et al., 2010). It describes the potential use of an individual whole-genome sequence in the estimation of the risk of a 40-year-old man with a family history of coronary artery disease and sudden death, and with clinical characteristics within normal limits. In this patient's genome, several common variants were found associated with increased coronary and type 2 diabetes risks, together with rare variants in three genes associated with sudden cardiac death, offering potentially useful information for the possible care of this patient. This personal genome case, which will probably not be isolated in the near future due to the dramatically decreasing costs of whole-genome sequencing, gives an example of clinical utility of the genes emerging from GWASs.

    Even if the results of GWASs do not seem to add significant information to the risk prediction in the general population, they pave the way for personalized medicine and tailored care of the diseases.


    . Utility of genetic and non-genetic risk factors in prediction of type 2 diabetes: Whitehall II prospective cohort study. BMJ. 2010;340:b4838. PubMed.

    . Clinical assessment incorporating a personal genome. Lancet. 2010 May 1;375(9725):1525-35. PubMed.

    View all comments by Philippe Amouyel

Make a Comment

To make a comment you must login or register.


News Citations

  1. Vienna: In Genetics, Bigger Is Better—Data Sharing Nets Three New Hits

Paper Citations

  1. . Genome-wide association study identifies variants at CLU and CR1 associated with Alzheimer's disease. Nat Genet. 2009 Oct;41(10):1094-9. PubMed.
  2. . Genome-wide association study identifies variants at CLU and PICALM associated with Alzheimer's disease. Nat Genet. 2009 Oct;41(10):1088-93. PubMed.

External Citations

  1. BIN1
  2. CLU
  4. CR1
  5. Translational Genomics Research Institute public release database

Further Reading


  1. . Genome-wide association study identifies variants at CLU and CR1 associated with Alzheimer's disease. Nat Genet. 2009 Oct;41(10):1094-9. PubMed.
  2. . Genome-wide association study identifies variants at CLU and PICALM associated with Alzheimer's disease. Nat Genet. 2009 Oct;41(10):1088-93. PubMed.

Primary Papers

  1. . Genome-wide analysis of genetic loci associated with Alzheimer disease. JAMA. 2010 May 12;303(18):1832-40. PubMed.
  2. . Reaching the limits of genome-wide significance in Alzheimer disease: back to the environment. JAMA. 2010 May 12;303(18):1864-5. PubMed.