Thirty years of Alzheimer's disease genetics: the implications ... - Nature

9 downloads 0 Views 472KB Size Report
autosomal-dominant forms of the disease are caused by mutations in ... Abstract | The genetic underpinnings of Alzheimer's disease (AD) remain largely elusive.
REVIEWS

Thirty years of Alzheimer’s disease genetics: the implications of systematic meta-analyses Lars Bertram and Rudolph E. Tanzi

Abstract | The genetic underpinnings of Alzheimer’s disease (AD) remain largely elusive despite early successes in identifying three genes that cause early-onset familial AD (those that encode amyloid precursor protein (APP) and the presenilins (PSEN1 and PSEN2)), and one genetic risk factor for late-onset AD (the gene that encodes apolipoprotein E (APOE)). A large number of studies that aimed to help uncover the remaining disease-related loci have been published in recent decades, collectively proposing or refuting the involvement of over 500 different gene candidates. Systematic meta-analyses of these studies currently highlight more than 20 loci that have modest but significant effects on AD risk. This Review discusses the putative pathogenetic roles and common biochemical pathways of some of the most genetically and biologically compelling of these potential AD risk factors.

Autosomal-dominant inheritance A type of inheritance in which the phenotype of a trait is determined completely by one of two alleles on the non-sexchromosomes. There can be either one (heterozygous state) or two (homozygous state) copies of the dominant allele.

Genetics and Aging Research Unit, MassGeneral Institute for Neurodegenerative Diseases (MIND), Department of Neurology, Massachusetts General Hospital, 114 16th Street, Charlestown, Massachusetts 02129, USA. Correspondence to L.B. e‑mail: [email protected]. harvard.edu doi:10.1038/nrn2494

Alzheimer’s disease (AD) is the most common neuro­ degenerative disease, and one of the most common of all diseases in the industrialized world. Genetically, AD is heterogeneous and complex, displaying no single or simple mode of inheritance. Rare, early­onset autosomal-dominant forms of the disease are caused by mutations in three genes, all of which alter production of the amyloid­b (Ab) peptide, the principal compo­ nent of senile plaques1. Ab is generated from the amyloid precursor protein (APP) by sequential cleavage by two enzymes: b­secretase and g­secretase. All mutations that are currently known to cause AD in early­onset (that is, onset before 60 to 65 years of age) autosomal­dominant families are located either in the APP gene itself or in the genes that encode the proteins that lie at the catalytic centre of the g­secretase complex: presenilin 1 (PSEN1) and presenilin 2 (PSEN2) (TABLE 1). By contrast, suscep­ tibility for late­onset AD shows less­obvious or no appar­ ent familial aggregation (hence it is sometimes called ‘sporadic’ AD) and is likely to be governed by an array of common risk alleles across a number of different genes. These genes affect various pathways, many of which are likely to be involved in the production, aggregation and removal of Ab (FIG. 1). Although the total number of AD risk genes and their precise identity remain elusive, there is good evidence suggesting that in combination they have a substantial impact on disease predisposition and age of onset2–4 (BOX 1).

768 | O cTOBER 2008 | vOlumE 9

In the quest to uncover the elusive AD genes, a vast body of data has been accrued and more than 500 genes have been genetically assessed as potential risk factors using primarily a candidate­gene approach (BOX 2). However, with the exception of one genetic variant (polymorphism), the e4 allele of the apolipoprotein E gene (APOE)5,6, none of these candidates has been proved to consistently influence disease risk or onset age in more than a handful of samples7. Instead, most reports of ‘novel AD genes’ have been followed by a large number of con­ flicting reports that challenge the genes’ contribution to disease risk. Owing to the exceedingly large number of studies, it has become virtually impossible to systematically follow, evaluate or interpret these findings. To date, information garnered from studies of the established AD genes has guided the development of some promising therapeutics that are currently in clini­ cal trials. Thus, it can be expected that the identification of new risk factors and/or onset­age modifiers will improve the diagnosis, treatment and prevention of AD. It has been estimated that delaying the onset of AD by just a few years could substantially reduce the number of patients with AD over the next 50 years8,9. In this Review, we discuss the potential functional implications of some of the most interesting current candidate AD risk fac­ tors, selected on the basis of systematic meta­analyses of the entire AD genetics literature. For each gene we summarize the genetic epidemiological evidence, review www.nature.com/reviews/neuro

REVIEWS Senile plaque An extracellular pathological lesion in the brains of patients with AD. Senile plaques have a core of aggregated amyloid-b protein and a periphery that consists primarily of dystrophic neurites.

Polymorphism Genetic variation (for example, a single base change or the insertion or deletion of a piece of DNA) that occurs with at least 1% frequency in a population.

Linkage disequilibrium (LD). The non-random association of alleles at two or more loci. In other words, when a combination of alleles at different loci occurs more (or less) frequently in a population than would be expected on the basis of the alleles’ frequencies. The degree of LD can be quantified by different measures, for example r2 or D′. Both of these measures can take values ranging from 0 to 1, where 0 indicates no LD and 1 indicates strong LD.

Genetic-association study A study that aims to determine whether a certain allele or set of alleles at polymorphic sites shows differences in distribution between samples of disease-affected and -unaffected individuals. In the simplest setting, single base changes (single-nucleotide polymorphisms (SNPs)) are assayed across unrelated cases and controls.

the potential functional­genomic consequences of the implicated sequence variants and discuss these variants’ possible role in the build­up to neurodegeneration and dementia.

Approaches to identify new AD genes Genome-wide association analysis. An alternative to the traditional candidate­gene approach is offered by recent advances in large­scale genotyping technolo­ gies that enable researchers to perform comprehensive, unbiased genome­wide association (GWA) analyses (BOX 2). Three groups have reported the results of AD GWA analyses10–12. The first10 tested roughly 17,000 polymorphisms in or near to genetic coding regions (coding single­nucleotide polymorphisms (cSNPs)). The only cSNPs that were found to consistently associ­ ate with AD risk across the different samples are located in close proximity to APOE and probably reflect linkage disequilibrium (lD) with the APOE­e4 allele. Although the authors proposed a number of additional genes to be potential AD genes, none of these genes showed the same consistency of effect or the same level of statistical significance as the e4­related variants. The second study11 tested ~500,000 SNPs in ~1,100 unrelated AD cases and controls. Again, with the exception of a single SNP that had strong lD with APOE­e4, no significant signals were observed. In a follow­up paper13, the same group reported evidence of an AD association with variants in the gene that encodes GRB2­associated binding protein 2 (GAB2, which is located on chromosome 11q14), which only became evident when AD cases and controls were divided into carriers and non­carriers of the e4 allele of APOE. Finally, a third group12 tested nearly 470,000 markers and reported an AD association with variants in golgi membrane protein 1 (GOLM1, which is located on chromosome 9q22) and two currently uncharacterized loci on chromosomes 9p and 15q. Although the findings from these GWA studies hold the promise of pinpointing new pathways and mechanisms that are important in AD pathogenesis, it will be important to ensure that they can be consistently replicated by independent investigators. Systematic meta-analysis. GWA studies actually sub­ stantially compound the problem that plagues human genetic studies today — how to determine which of the many reported risk factors are real, as opposed to

statistical artefacts resulting from the large number of tests performed. In an attempt to alleviate this situation, we have created a publicly available database — called AlzGene — that systematically collects, summarizes and meta­analyses all AD genetic-association studies (including GWA studies)7. After thorough (and ongoing) searches of the available scientific literature, studies published in peer­reviewed journals that are available in English are included in the database. Key variables (such as ances­ try, type of AD diagnosis, sample size, onset age, and genotype distributions) are extracted from the original publications. Furthermore, published genotype data from independent case–control samples are systemati­ cally subjected to random effects meta­analyses on the basis of allele contrasts (BOX 3). Because this approach quantitatively synthesizes all of the published genotype data for each polymorphism, it facilitates the overall interpretation of association findings: rather than rely­ ing on the either ‘positive’ or ‘negative’ outcomes of individual studies, it produces a summary risk estimate (called the odds ratio (OR)) that takes into consideration within­ and between­study variation. currently AlzGene includes nearly 1,100 individual studies (the first of which was published 30 years ago14) and showcases the results of over 200 individual meta­ analyses. In the meta­analyses, more than 20 genes that are not related to the well­established APOE­e4 allele show nominally significant risk effects (see Supplementary information S1 (table)). Interestingly, approximately one­ third of these genes were originally implicated by GWA analysis. The average allelic summary ORs for non­APOErelated effects are modest (~1.25 for ‘risk’ alleles and ~0.8 for ‘protective’ alleles) compared with that of a single copy of the APOE­e4 allele (~3–4). This has important implica­ tions for the design of future genetic­association studies of AD and of other complex diseases: sample sizes need to be very large in order to be able to detect or exclude ORs of ~1.25 with sufficient confidence. For instance, to detect an OR of 1.25 with 80% power at a P­value of 0.05, sample sizes between ~1,250 and ~6,150 are needed for minor disease­allele frequencies ranging from 0.50 to 0.05 (for more details see rEF. 7).

Potential implications of AD candidate genes Below we review ten genes that we have selected from a current list (current as of 31 march 2008; see

Table 1 | Summary of genetic findings for early-onset autosomal-dominant forms of AD Gene (and protein)

Chromosomal location

Total number of pathogenic mutations (affected families)

Relevance to AD pathogenesis

Initial study references

APP (amyloid precursor protein)

21q21.3

29 (78)

Increase in Ab production or Ab42/Ab40 ratio; mutations in the Ab sequence or close to the band g-secretase site of APP; locus duplications

98

PSEN1 (presenilin 1)

14q24.3

166 (362)

Increase in Ab42/Ab40 ratio; mutations throughout molecule; enzymatic role in g-secretase complex

99

PSEN2 (presenilin 2)

1q31–42

10 (18)

Increase in Ab42/Ab40 ratio; mutations throughout molecule; enzymatic role in g-secretase complex

100,101

The data in the ‘Total number of pathogenic mutations’ column is from the AD & FTD Mutation Database. The ‘Initial study references’ column lists the first publication(s) to describe Alzheimer’s disease (AD)-causing mutations in the respective genes. Ab, amyloid-b; APP, amyloid precursor protein. Table modified, with permission, from rEF. 102  (2001) Current Science Inc.

NATuRE REvIEWS | neuRosCIenCe

vOlumE 9 | O cTOBER 2008 | 769

REVIEWS APP

Cu, Zn, Fe

CHRNB2

β-secretase α-secretase γ-secretase PSEN1 and 2 Aβ SORL1 GAB2 CH25H Oligomerization APOE

Cerebrovascular events APOE ACE CH25H



MAPT GAB2

Clearance from brain APOE

ACE

Neurofibrillary tangles (NFTs) Impairment of LTP APOE Fibrillization CST3 CHRNB2 MAPT PRNP TF GAB2 Inflammation Oxidative stress Senile (neuritic) plaques

Degradation

Neuron death

Figure 1 | summary of possible pathogenetic roles for candidate early-onset Nature Reviews | Neuroscience familial AD genes covered in this Review. The early-onset familial Alzheimer’s disease (AD) gene APP encodes the amyloid precursor protein, which gives rise to Ab through serial cleavage by b-secretase and g-secretase. g-Secretase is a complex of four proteins in which the enzymatic components are encoded by the early-onset familial AD genes presenilin 1 (PSEN1) and presenilin 2 (PSEN2). In contrast to cleavage by b-secretase and g-secretase, cleavage by a-secretase precludes Ab production. SORL1 (rEFS 75,76) and CH25H (rEF. 27) have both been reported to affect APP processing and Ab generation, SORL1 by affecting APP trafficking, and CH25H through production of 25-hydroxycholesterol. GRB2-associated binding protein 2 (GAB2) binds GRB2, which has been reported to bind APP and both presenilins53. The established late-onset AD gene apolipoprotein E (APOE) and CH25H affect cholesterol metabolism, and angiotensin I converting enzyme (ACE) affects blood pressure — these genes might modulate the effects of cerebrovascular events on Ab production. Once it has been generated, Ab can be transported out of the brain with the assistance of ApoE. Alternatively, Ab can undergo proteolytic degradation (for example, by ACE1)17. As Ab accumulates it can aggregate, and this is influenced by ApoE. Aggregation into oligomers (for example, dimers and trimers) can lead to impairment of long-term potentiation (LTP). The function of the product of CHRNB2, a component of some pentameric nicotinic receptor complexes, might be affected by Ab97. Ab oligomers can further aggregate into fibrils, which might ultimately be deposited in senile plaques. Ab fibrillization is affected by ApoE, cystatin 3 (CST3)45,46 and prion protein (PRNP)70. Senile plaques and Ab aggregates can induce inflammatory responses and oxidative stress. Transferrin (TF) regulates the metabolism of iron, a reactive metal that is involved in free-radical generation and, thus, in oxidative stress. Oxidative stress and iron84 have been associated with abnormal tau phosphorylation and aggregation, and with the formation of neurofibrillary tangles (NFTs). The principal component of NFTs is tau, which is encoded by MAPT. GAB2 has been reported to affect tau phosphorylation and NFT formation13. Ab oligomers have been reported to induce NFT formation32. NFTs induce neuron death, which can result in further inflammation and oxidative stress. In turn, inflammation and oxidative stress can enhance further Ab deposition, resulting in a vicious cycle. CHRNB2 has been proposed to have a neuroprotective role34, and mutations in lamin A/C (LMNA) cause a broad array of laminopathies, which involve cell death. Thus, these genes could influence neuron survival. CHRNB2 also influences cholinergic activity and might thus have pathogenic effects on cognition. Established AD genes are depicted in green; AD candidate genes are depicted in blue.

Supplementary information S1 (table)) of AlzGene top results, and discuss their potential role in the molecu­ lar mechanisms that lead to AD. Although we consider these genes to be particularly interesting for a number of reasons, including the strength and consistency of the 770 | O cTOBER 2008 | vOlumE 9

genetic­association findings, their presumed or proven role in AD pathogenesis or the fact that they were first identified in a GWA study, this collection is by necessity subjective. We strongly encourage the reader to consult the AlzGene website for up­to­date summaries and rank­ ings of these and other findings. The studies that contrib­ uted to the meta­analysis of each of these ten genes are listed in Supplementary information S2 (box). ACE. ACE encodes angiotensin I converting enzyme 1 (AcE1), a ubiquitously expressed zinc metalloprotease that is involved in regulating blood pressure. Three different polymorphisms currently show significant summary ORs in AlzGene. The most­significant effects are found for two variants (rs1800764 and rs4291), and these effects are particularly strong when using recessive models. The variants are located ~0.2 kb and ~3.8 kb, respectively, from the transcription start site in the pro­ moter of the gene. The most widely studied ACE variant in AD, however, is the alu repeat insertion (I)/deletion (D) polymorphism (rs1799752) located in intron 16 of the gene. carriers of the minor I allele show ~20% increased risk for AD (based on a combined sample size of over 18,000 cases and controls originating from over 30 independent samples). This variant is one of the best­studied polymorphisms in humans, and has been associated with a number of different phenotypes. Several studies have demonstrated that decreased AcE1 serum levels are found in carri­ ers of the I allele15. When the large number of different polymorphisms at this locus were condensed into dif­ ferent haplotype clades, AcE1 levels were consistently found to be lowest in carriers of clade A, highest in car­ riers of clade B and intermediate in carriers of clade c16. Although it is probably not the functional variant itself, the I allele at the intron 16 site is in perfect lD with clade A in most studies, whereas clades B and c are aggregated in carriers of the D allele. Of potential relevance to AD is the observation that AcE1 can degrade naturally secreted Ab in vitro17,18 (FIG. 1): this could explain the increased risk for AD in carriers of the I allele. It remains to be shown, however, whether the Ab­degrading activity is also relevant in vivo: this was not supported in two recent reports19,20. Another involvement of AcE1 in AD pathogenesis could be due to its role in blood pressure regulation, as some studies have shown an increased risk for AD in indviduals with high mid­life blood pressure21. This observation has led to the testing of pharmacological inhibitors of AcE1 (a commonly prescribed class of anti­hypertensive medica­ tions) for efficacy in treating AD, with some success22. The interpretation of AcE1’s role in AD pathogenesis is further complicated by the observation that its levels might actually be increased in some brain regions, in particular in perivascular areas23. more recently, AcE1 activity has been reported to be increased in AD brains in proportion to the parenchymal Ab load24. Finally, in addition to AD, ACE variants (in particular the D allele) have been associated with risk for a number of diseases, including atherosclerosis, diabetic nephropathy, coronary heart disease and stroke15. www.nature.com/reviews/neuro

REVIEWS Box 1 | Background on the effect of genes and environment on AD risk A common feature of many neurodegenerative diseases — including Alzheimer’s disease (AD) — is a dichotomy of familial (rare, early-onset) and seemingly non-familial (common, late-onset) forms. Although the latter are frequently described as ‘sporadic’ or ‘idiopathic’, there is a growing body of evidence that suggests that a large proportion of these cases are actually also significantly influenced by genetic factors. Much of this evidence stems from twin studies that suggest there is a higher concordance of AD in monozygotic twins than in dizygotic twins2,4,89,90. Early estimates suggested a heritability (a measure of the proportion by which phenotypic variation is determined by genetic variation) of AD of ~40–80% (rEFS. 2,90). More recently this range was narrowed to 60–80% in larger samples4,91. Regardless of the precise value of the estimate, these studies collectively indicate that genetic variation plays a crucial (probably the most crucial) role in determining the risk of late-onset AD. However, as yet we have no clues as to which or how many risk genes are implicated in disease susceptibility and onset-age variation. One study3 estimated the existence of up to seven AD susceptibility genes, four of which possibly have effect sizes similar to that of the apolipoprotein E (APOE) gene. Assuming that the above heritability estimates are correct, they imply that non-genetic factors also make a substantial, albeit smaller, contribution to AD risk. Indeed, there is a vast body of literature that implies that there are a number of potential environmental risk factors92, including metal concentrations in drinking water, pesticides and other environment pollutions, dietary factors, head trauma, acute or chronic medical conditions (such as viral or bacterial infections), nicotine and alcohol abuse. The problem with many of these factors is that they are difficult to reproducibly measure — and therefore difficult to compare across samples — and almost impossible to adequately assess in a cross-sectional, post-hoc study. Sufficiently sized, longitudinal population- or community-based samples are needed to more reliably assess the influence of such variables on disease susceptibility. The situation is further complicated by the fact that physiological or pathogenetic responses to many of the ‘environmental exposures’ are also genetically controlled. Thus, although it is almost certain that non-genetic factors make a sizeable contribution to AD risk as a whole, their effects are currently even more difficult to establish than those of genetic factors.

Odds ratio (Or). A measure of effect size (for example, of risk effects). The Or measures the ratio of the odds of an event occurring in one group (for example, disease cases) to the odds of that same event occurring in another group (for example, healthy controls). An Or of 2 indicates that carriers of a certain risk factor are at twice the risk of developing the disease as non-carriers; an Or of 0.5 indicates that the risk in carriers is only half that in non-carriers.

CH25H. cholesterol 25­hydroxylase (c25H) is respon­ sible for the synthesis of 25­hydroxycholesterol, a potent regulator of gene transcription (in particular the transcription of genes that are involved in cholesterol and lipid metabolism). The gene that encodes c25H, CH25H, maps within a broad AD linkage region on the long arm of chromosome 10 and, thus, represents a func­ tional and positional AD candidate gene. The first report of a potential association between variants in CH25H and AD risk was published in 2002 (rEF. 25) and was followed­up in seven independent studies, five of which also showed significant results. It is important to note that the five samples that produced significant results were all reported by the same group that published the first CH25H AD­association paper. This potential bias aside, the genetic effect sizes estimated for CH25H are among the strongest of all current AlzGene top results (see Supplementary information S1 (table)). In a recent study26, a functional basis for the observed genetic association was suggested by the observation that the risk­associated CH25H haplotype (contain­ ing the rs13500 risk allele) not only leads to increased cerebrospinal fluid (cSF) concentrations of lathosterol, a metabolic precursor of cholesterol, but also is associ­ ated with a higher brain Ab load and with lower levels of one form of Ab, Ab1–42, in the cSF of non­demented elderly subjects (FIG. 1). more recently it was shown that 25­hydroxycholesterol, which is produced by c25H, can lead to changes in Ab levels by altering the processing and trafficking of APP in vitro27.

NATuRE REvIEWS | neuRosCIenCe

Despite these promising results, interpretation of the CH25H findings is complicated by the fact that the SNP that is associated with AD risk, rs13500, maps ~6.4 kb upstream of the CH25H start codon, and actually lies much closer to another nearby locus, namely that of lipase A (LIPA). Therefore, future studies not only need to independently replicate the putative association between disease risk and rs13500 and other CH25H variants, but also need to address whether this associa­ tion is due to functional changes in CH25H, LIPA or both. Both of these genes, as well as APOE, are involved in cholesterol metabolism. Ab production and clear­ ance have been shown to be regulated by cholesterol, and drugs that inhibit cholesterol synthesis lower Ab in cellular and animal models28. Thus, it will be inter­ esting to monitor by meta­analysis whether additional cholesterol­related genes will yield significant effects on AD risk. CHRNB2. Nicotinic acetylcholine receptors (nAchRs) are ligand­gated ion channels that respond to the bind­ ing of acetylcholine (Ach) by increasing the intracellular ca2+ concentration in pre­ and postsynaptic neurons. They are widely expressed in the cNS, where the b2 subunit is particularly abundant and forms heteropenta­ meric a4b2 receptors with a4 subunits29. The CHRNB2 gene, which encodes the nAchR b2 subunit, maps to chromosome 1q21, close to a previously described whole­genome linkage region. To date, only two inde­ pendent genetic­association studies have investigated the potential involvement of CHRNB2 in AD risk, in a total of four independent case–control samples30,31. These studies suggest that the minor (T) allele at an intronic SNP (rs4845378) reduces the risk of develop­ ing AD by ~50%. Although no published studies have directly assessed the functional­genomic consequences of the associated variant, it is located only 14 bp in the 3′ direction from exon 5 of CHRNB2, suggesting that it might affect alternative splicing rather than induce changes in gene or protein expression. The reduction in the levels of nAchRs and the loss of cholinergic neurons in disease­relevant brain regions is one of the major neurochemical hallmarks of AD32, and several studies have suggested that an age­dependent decrease in protein and/or mRNA levels of a4 and, in particular, b2 subunits occurs in the cortex and hip­ pocampus of healthy individuals33. Furthermore, aged CHRNB2­knockout mice display significant impair­ ments in visual–spatial memory tasks, suggesting that b2­containing nAchRs contribute to both neuronal survival and the maintenance of cognitive performance during aging34. Finally, pharmacologically increasing cerebral Ach by inhibiting acetylcholinesterase activity is currently one of the most widely applied and success­ ful treatments for the symptoms of AD — it results in a temporary slowing of cognitive decline. It has been suggested that Ab might directly inhibit the a4b2 complex35, and that nicotine, possibly by acti­ vating nAchRs, can mediate the toxic effects of Ab and might also be involved in phosphorylating tau in vitro and in vivo36. This in turn suggests a multitude of potential vOlumE 9 | O cTOBER 2008 | 771

REVIEWS Box 2 | Background and examples of genome-wide association studies in AD GWA study Population references

Platform

number of snPs

Featured genes

10

United States and UK

Celera (cSNPs)

17,343

APOE*, ACAN, BCR, CTSS, EBF3, GALP, GWA_14q32.13, GWA_7p15.2, LMNA, LOC651924, MYH13, PCK1, PGBD1, THEM5, TNK1, TRAK2, UBD

11,13

United States and the Netherlands

Affymetrix (500K)

502,627

APOE*, GAB2

12

Canada

Affymetrix (500K)

469,438

APOE*, GOLM1, GWA_15q21.2, GWA_9p24.3

Until recently, usually only a handful of genetic markers were chosen per genetic-association study. Choice of markers was typically based on a prior hypothesis that implicated certain genes in Alzheimer’s disease (AD) pathogenesis (such genes were termed ‘candidate genes’). By nature of their design, candidate-gene studies usually do not allow conclusions that lie beyond the scope of the initial hypothesis, such as the elucidation of pathogenetic pathways beyond those that initially drove the selection of the tested candidate genes, to be reached. Genome-wide association (GWA) analyses simultaneously test a large number of genetic markers, usually several hundreds of thousands, in a largely hypothesis-free (or ‘unbiased’) fashion. The markers on a GWA array typically consist of single-nucleotide polymorphisms (SNPs) that are chosen on the basis of their ability to cover common variation in the human genome. More recent arrays also include probes that allow a systematic assessment of copy-number variants (CNVs: deletions or multiplications of certain chromosomal segments). Other GWA arrays only assay SNPs that are located in known or predicted coding regions (cSNPs). This leads to an enrichment of potentially functionally relevant variants at the expense of overall genome-wide coverage. Besides technical issues related to the testing of several hundreds of thousands of SNPs in a single experiment, one of the main challenges posed by the GWA approach is that of multiple testing. For instance, in a GWA study using 500,000 independent markers, 25,000 can be expected to show nominally (that is, P‑value ≤5×10–2) significant association by chance alone, 5 of which will be significant at a P-value of 1×10–5. Therefore, there is a need for strategies that effectively correct for the large number of comparisons in the GWA setting while maintaining the power to uncover ‘real’ genetic effects. Several approaches have been proposed; one popular approach is to use a threshold of P ≤5x10–8 (rEFS 93,94). Regardless of their statistical significance, the first and essential step in differentiating real from false-positive findings is to provide independent replication of the association. In AD it remains to be shown whether any of the currently proclaimed GWA signals will prove to replicate more consistently than results from conventional candidate-gene analyses. The table summarizes all AD GWA studies (all of which used case–control designs) published prior to 31 March 2008. The ‘Featured genes’ column lists genes that were emphasized in the original publications, usually because they showed some degree of genetic association after completion of all analyses, such as correction for multiple comparisons and/or replication in multiple independent data sets. Asterisks indicate surrogate markers for the apolipoprotein E (APOE) e4 allele.

Autosomal-recessive inheritance A type of inheritance in which a certain phenotype of a trait arises only if two copies of a particular allele are present on the non sex-chromosomes (that is, the homozygous state).

pathogenetic effects resulting from nAchR dysfunction or loss. Another human condition that is known to result from nAchR dysfunction is autosomal­dominant noc­ turnal frontal lobe epilepsy, which is caused by gain­of­ function mutations in CHRNB2, as well as in CHRNA4 and CHRNA2, which encode the a4 and a2 nAchR subunits, respectively37,38. The latter two genes have also been investigated in AD, although neither currently has sufficient data to warrant meta­analysis.

772 | O cTOBER 2008 | vOlumE 9

CST3. cystatin c (cysc), which is encoded by CST3, is the most abundant extracellular inhibitor of cysteine proteases. It is ubiquitously expressed, is found in high levels in most body fluids — in particular the cSF — and is upregulated following acute injury39. In the brain it is produced mostly by astrocytes and activated microglia, but also by neurons40. Both of the CST3 SNPs that have been analysed in AlzGene showed significant association with AD, with rs1064039 (Ala25Thr) showing the most pronounced effects in homozygous carriers of the Thr25 allele, and both SNPs are in complete lD41. In vitro experiments have suggested that there is ~50% less secretion of cysc in cells that express the Thr25 allele than in cells that express the Ala25 allele42,43. This difference is probably due not to changes in cysc expression, but to defective intracellular processing and/ or protein maturation and subsequent impaired secre­ tion. Of potential relevance to AD, cysc was found to bind Ab44 and to inhibit Ab­fibril formation in a concentration­dependent manner in vitro45 and in vivo46. These findings, together with the observation that there is a general reduction in neuroprotection when levels of cysc are decreased in vitro, might be the functional correlates of the observed epidemiological association between Thr25 and AD42 (FIG. 1). However, injection of cysc into the rat hippocampus was shown to induce apoptotic cell death, suggesting that hyperactivity of this protease inhibitor might be neurotoxic47. cysc is also involved in the development of one form of cer­ ebral amyloid angiopathy (cAA): hereditary cystatin c amyloid angiopathy (HccAA) is a rare, early­onset hereditary form of cAA that is caused by a mutation at codon 68 (leu68Gln). It is important to add that cAA lesions are also observed in AD brains, where they are composed of Ab bound to cysc48. GAB2. GAB2 is a member of a family of evolutionarily highly conserved scaffolding/adaptor proteins that are involved in multiple signalling pathways and, in particular, in the transduction of cytokines and in growth­receptor signalling49,50. GAB2 is ubiquitously expressed, but is found at particularly high levels in white blood cells, the prefrontal cortex and the hypothalamus. All ten of the GAB2 SNPs that have been meta­analysed in AlzGene show significant association with AD. Specifically, a pro­ tective role is indicated for all of the minor alleles. All ten of the SNPs display a high degree of lD, and therefore probably point to a single underlying signal. Of the ten associated markers only one (rs1385600) is predicted to map within the coding region of GAB2, where it does not invoke a change in the amino­acid sequence. The association between AD and GAB2 was first identified in one of the recently published GWA analy­ ses13 (BOX 2), and was nominally — that is, at a P­value ≤0.05 — confirmed in another GWA study using the same marker panel12. Although the association between GAB2 and AD risk was not found to be significant in a more recent study that investigated case–control sam­ ples from France and the uK51, it is noteworthy that the distribution of genotypes showed the same direction of effect that was seen in the two GWA studies, indicating www.nature.com/reviews/neuro

REVIEWS Box 3 | Background, interpretation and limitations of meta-analyses in genetic epidemiology research

Alu repeat A dispersed, repetitive DNA sequence that is present in the human genome in ~300,000 copies. It is named after the restriction endonuclease (AluI) that cleaves it.

Haplotype Alleles located in close proximity on the same chromosome that, as a result, are inherited together. In a population, the genome is partitioned into haplotype blocks of varying length depending on the strength of the LD between the alleles, and the different combinations of alleles that are located in these blocks are called haplotype clades or haplotype alleles.

Alternative splicing A process whereby different mrNAs can be produced from a single gene through the differential incorporation of exons into the mature transcript during splicing. Frequently, various mature proteins are generated from a single gene.

Adaptor protein An accessory to the main signalling proteins in a signal-transduction pathway. Adaptor proteins tend to lack any intrinsic enzymatic activity themselves, but instead mediate specific protein–protein interactions that drive the formation of protein complexes.

Hardy–Weinberg equilibrium (HWE). A principle that postulates that the genotype frequencies of a population remain constant over time. In the case of a bi-allelic polymorphism with frequencies p and q, genotype frequencies can be calculated as 1 = p^2 + 2pq + q^2. HWE can be disturbed by effects such as selection, mutations and non-random mating. In genetic-association studies, large deviations from HWE can be an indicator of genotyping error.

Meta-analysis provides a means of quantitatively synthesizing Search literature databases for geneticdata or results across independent studies. If study-level association studies in AD samples raw data are available, they can be pooled and analysed jointly. This allows the inclusion of certain co-variates, such as Include if: onset age or gender. If study-level raw data are not available • peer reviewed • published in English (which is the case for essentially all AD genetic-association studies), summary data, such as genotype or allele distributions in cases versus controls, can be used as Enter study details into internal database surrogates. For meta-analyses performed in the AlzGene and double-check all entries database, summary odds ratios (ORs) and 95% confidence intervals (CIs) are calculated from study-level ORs on the basis Calculate allelic random-effects of the genotype or allele distributions of the individual meta-analysis if genotype data are samples. These ORs and CIs are based on allelic contrasts available from at least four (comparisons of the number of minor alleles and the number independent case–control samples of major alleles at each polymorphism — for example, C 95 versus T) that use random-effect models , which allow withinMake data and results available online and between-study heterogeneity to be accounted for. Alternative approaches include calculating genotype-based ORs (for example, assuming autosomal dominant (CC + CT versus TT) or recessive (CC versus CT + TT) inheritance) or Nature Reviews | Neuroscience fixed-effects models when no evidence of heterogeneity is found. Summary ORs greater than 1 indicate ‘risk’ effects, whereas those below 1 indicate ‘protection’ conferred by the allele or genotype under study. Nominal statistical significance is reached when both 95% CIs remain on the same side of 1 as the summary OR. Naturally, the overall quality of any meta-analysis is only as high as the quality of the study-level data that is used in its calculations. Furthermore, the outcome of any meta-analysis is susceptible to various different biases, which can only partly be assessed or controlled. For AlzGene a number of such potential biases are addressed (including publication bias, bias that is due to an outlying result in the initial study, and bias that is due to deviations from Hardy–Weinberg equilibrium)7. Methods are currently being implemented to systematically grade the ‘epidemiological credibility’ of any significant meta-analytic finding using recently proposed criteria. This will further facilitate the interpretation of positive results96. Regardless of how many precautions are taken, false-positive results can never be excluded with 100% certainty, and every significant result should be interpreted cautiously until a functional basis for the association has been established. The flow chart summarizes the data-management and -analysis procedures that are used in the AlzGene database (for more information, see rEF. 7). As of 31 March 2008, the AlzGene database included 1,086 studies, 531 loci and 203 meta-analyses. Of the loci, 27 contained at least 1 variant that showed nominally significant association by meta-analysis (see Supplementary information S1 (table)).

that the minor alleles confer protection against AD. In the AlzGene meta­analyses, combining the geno­ type data from all three studies results in a robust and relatively significant association that is strongest in homozygous carriers of the putative protective allele. Because this locus has only been considered an AD candidate gene for a short time, little data has accrued regarding the potential functional mechanisms that underlie the genetic findings. However, preliminary evi­ dence reported in the original GWA paper13 suggested that changes in GAB2 expression could potentially affect glycogen synthase kinase 3 (GSK3)­dependent phospho­ rylation of tau and the formation of neurofibrillary tan­ gles (NFTs), a hallmark of AD (FIG. 1). moreover, growth factor receptor­bound protein 2, which binds GAB2, has been reported to bind tau52, APP and presenilin 1 and 2 (rEF. 53). These interactions have been proposed to regulate signal transduction (for example, through the extracellular­signal­regulated kinase 1 (ERK1) and ERK2 pathway). Although more independent genetic and functional­genomic data are needed before the potential role of GAB2 in AD risk can be evaluated more definitively, this association nurtures the hope that the unbiased GWA approach might indeed provide new insights into the genetic causes and pathogenetic mechanisms of AD.

NATuRE REvIEWS | neuRosCIenCe

LMNA. lamins are intermediate­filament proteins that are a component of the nuclear lamina. Alternative splicing of LMNA produces two isoforms (lamin A and lamin c), which are ubiquitously expressed, and mutations in LMNA have been found to cause a large number of dif­ ferent disorders54. LMNA maps ~1.5 mb distal (that is, towards the long arm) of CHRNB2 on chromosome 1, close to the presumed AD linkage region at 1q22–q25. To date, only one group has reported an association between variants in LMNA and AD10 in a GWA of cSNPs. In that study, the LMNA association ranked second in terms of genetic effect size of all non­APOE­e4­related genes, with an OR of 1.35, which is identical to the current AlzGene estimates. The associated SNP, rs505058, produces a synony­ mous change at codon 446. It could thus affect mRNA splicing or stability, although no functional study has yet assessed that possibility. The same allele that was associated with increased risk for AD was also reported to increase risk for type­2 diabetes mellitus in a recent meta­analysis 55. The same codon was found to be mutated in families with Emery Dreifuss muscular dystrophy (EDmD)54, an early­onset disorder that is characterized by muscular contractures and weakness as well as cardiac defects. Other heritable diseases that are caused by mutations in LMNA include (but are not vOlumE 9 | O cTOBER 2008 | 773

REVIEWS limited to) familial partial lipodystrophy, dilated cardio­ myopathy, progeria syndrome and a particular subtype of charcot­marie Tooth disease, all of which are termed ‘laminopathies’ (rEFS 54,56). The potential association between LMNA and AD adds late­onset neurodegenera­ tion to the broad spectrum of phenotypes that are caused by a dysfunction of lamin A/c. Despite the overall consistency of the genetic effects that have been reported across five independent case– control cohorts10, the current lack of truly independent replication data makes LMNA the least probable AD candidate gene of this Review. Interestingly, however, the gene maps within ~15 kb of semaphorin 4A (SEMA4A), a homologue of SEMA5A, which has been associ­ ated with risk for Parkinson’s disease (PD) in a recent GWA study57. It also maps within ~60 kb of ubiquilin 4 (UBQLN4), a homologue of UBQLN1, which has been associated with risk for AD in family samples58. Thus, future studies that investigate LMNA should take into account the possibility of lD with these nearby loci.

Fine­mapping and functional­genomics studies recently converged in identifying an H1 subclade (termed ‘H1c’ or ‘H1B’) that is associated with increased risk for AD64 and PSP65. This clade is tagged by the A allele of the SNP rs242557 (present in ~40% of caucasian popu­ lations, but currently not showing significant effects in AlzGene). Although both studies were in agreement that the A allele at rs242557 increased risk for both AD and PSP, their in vitro functional results deviated in show­ ing both increased65 and decreased64 transcriptional activity after cloning of the rs242557 risk allele into the MAPT promoter. The observation that neither this nor any other SNP that is in lD with the H1c/H1B subclade currently shows an association that is as pronounced as the H2 versus H1 comparison for both PD and PSP does not preclude the interpretation that the real pathogeneti­ cally active MAPT variant might actually be found on the H2 background and might confer protection, pos­ sibly by reducing the overall expression of tau mRNA65,66 in response to the degree of brain amyloid deposition67.

MAPT. Accumulation of the microtubule­associated protein tau as NFTs is one of the major neuropathologi­ cal hallmarks of AD. The gene that encodes tau, MAPT, has long been suspected to harbour disease­causing mutations, but the search for such mutations in AD has thus far been unsuccessful. However, MAPT mutations were found to cause another form of dementia, fronto­ temporal lobar degeneration with parkinsonism linked to chromosome 17 (FTDP­17)59. Despite the lack of MAPT mutations in AD, it is likely that tau dysfunction still contributes to AD risk60. MAPT is located in a region of chromosome 17q21 that represents one of the longest known lD regions in the human genome61. The extensive lD is explained by an inversion of a ~900 kb interval that gives rise to two heterologous haplotype clades — denoted H1 and H2 — between which recombination is prevented62. As a result, H2, which is particularly frequent (~20%) in caucasians and nearly absent in subjects of Southeast Asian decent, shows very little variation and can be precisely tagged by a number of different SNPs. This has been carried out in over 20 different AD case– control association studies, a meta­analysis of which suggests that the H2 clade has a modest but significant protective effect compared with the H1 clade (see Supplementary information S1 (table)). As expected, this effect was restricted to samples of caucasian ances­ try. It is interesting that a similar, albeit much more significant, protective effect of H2 versus H1 is also observed in meta­analyses of PD association studies (see the PDGene database) and in at least two related disorders (progressive supranuclear palsy (PSP) and corticobasal degeneration63). This supports the notion of a more general role for tau in modulating risk for neurodegeneration. For AD, it is possible that at least part of the apparent association between MAPT and disease risk could be due to phenocopy effects in clini­ cally diagnosed (non­autopsy­confirmed) patients who really suffer from tau­related diseases, such as FTD, or PD­related dementia.

PRNP. Prion protein (PRP) is a membrane­tethered glycoprotein that is highly expressed in all regions of the brain, particularly in neurons. Although its physiological function remains elusive, mutations and polymorphisms in PRNP, the gene that encodes PRP, were found to be major determinants of familial and sporadic prion dis­ eases (such as creutzfeldt–Jakob disease (cJD)). most forms of prion diseases are characterized by rapidly pro­ gressing neurodegeneration with spongiosis and amyloid plaques that consist of misfolded PRP aggregates. more than two dozen different amino­acid­changing muta­ tions in PRNP that are transmitted in an autosomal­ dominant fashion with nearly 100% penetrance have been identified as causes of familial prion diseases68. clinical presentation and disease progression are fur­ ther modified by a common non­synonymous polymor­ phism (rs1799990; met129val), and it was reported that the homozygous state for either allele (met/met or val/ val) was disproportionally more frequent in sporadic cJD than the heterozygous genotype (met/val). This was attributed to a decreased tendency of heterozygous PRP to form amyloid69. Ab­positive plaques in AD brains often contain PRP deposits, and recent evidence from studies that used APP–PRNP double­transgenic mice suggests that PRP might actually promote plaque formation in AD. This is likely to be mediated by increased Ab aggregation70 (FIG. 1) . AlzGene meta­analyses show a significantly decreased risk of AD in carriers of the minor val allele (based on a combined sample size of over 4,000 cases and controls) (TABLE 2). conversely, this result could also indicate a recessive risk effect of the met allele. In con­ trast to sporadic cJD, the comparison of homozygous and heterozygous carriers did not reveal any significant effects. Although the precise functional implications of this association for AD remain obscure, it is tempting to speculate that the met allele of PRP facilitates the forma­ tion of Ab fibrils. This agrees with recent observations of PRP­amyloid formation71, although a similar finding was earlier reported for the val allele72.

774 | O cTOBER 2008 | vOlumE 9

www.nature.com/reviews/neuro

REVIEWS SORL1. Sortilin­related receptor (SORlA, also known as lR11) belongs to a family of sorting receptors that contain a vPS10 (vacuolar­protein­sorting protein 10) domain, through which they mediate various intracel­ lular sorting and trafficking functions73. SORlA also binds to ApoE74 and APP (possibly leading to reduced Ab generation75,76), and is highly expressed in the brain. As a result of these findings, the gene that encodes SORlA, SORL1, was examined in an association study that tested 29 different genetic variants in over 5,000 cases and controls drawn from population­based cohorts and families with AD77. Although multiple different association signals throughout SORL1 were detected, none of the implied SNPs showed any clear and consist­ ent evidence of association across all data sets, possibly indicating the presence of allelic heterogeneity. Since the original publication, the potential association between SORL1 and AD has been assessed in six other studies, with mixed results. AlzGene meta­analyses currently show 8 of the 29 SNPs to be associated with summary ORs ranging from 1.09 to 1.21 (see Supplementary information S1 (table)). It is noteworthy that three of the positive SNPs (rs1699102, rs2070045 and rs3824968) result in syn­ onymous coding changes that are located in a region of strong lD in the 3′ half of the gene, possibly indicating that the postulated functional effects on AD pathogen­ esis originate from the 3′ part of the protein. This would be in line with biochemical evidence which suggests that SORlA directly influences the production of Ab by affecting the processing and/or trafficking of APP75,78, which binds to a complement­type repeat domain located near the region that contains the associated SNPs76 (FIG. 1). SORlA might mediate re­internalization of APP from the cell surface to endocytic compart­ ments, where it is processed into Ab by b­secretase and

g­secretase cleavage. In agreement with these in vitro results are studies which suggest that SORlA expression is reduced in the brains of patients with AD79 or preclini­ cal AD80, and the fact that SORL1­knockout mice show increased brain Ab levels75. However, despite the seem­ ingly converging genetic and biochemical evidence that implicates SORL1 as a new AD risk factor, it should be noted that more independent data are needed to enable a better evaluation of the epidemiologic relevance of the potential association. TF. Transferrin (TF) is the major circulating glyco­ protein involved in iron metabolism, and is highly expressed in the brain. It carries iron into cells by receptor­mediated endocytosis. There is a vast body of literature which suggests that iron misregulation promotes neurodegeneration, possibly through the generation of reactive oxygen species81. Iron has been found to be increased in the brains of patients with AD82, where it is associated with plaques and NFTs83. more recently it was suggested that iron might also have a role in aggregating hyperphosphorylated tau into insoluble paired helical filaments, one of the core components of NFTs84 (FIG. 1). In addition, the binding of metals to Ab has been shown to modulate several physiochemical properties of the peptide, along with its aggregation rate and pathogenicity, and iron has also been demonstrated to regulate the translation of APP85. Thus, iron levels can modulate the generation of both tangles and plaques in AD. Although a number of polymorphisms in TF are known, only rs1049296 (Pro570Ser) has been examined for association with AD to date. In AlzGene, carriers of the minor Ser allele show an OR of 1.21 (based on over 6,000 cases and controls) (TABLE 2). Because this variant constitutes an amino­acid substitution, it was

Table 2 | Meta-analysis details of genes that show significant summary ORs in AlzGene Gene

Variant

Location Linkage Allelic summary oR region (95% CI)

Genotypic summary oR (95% CI)

stratum n (# samples)

ACE*

rs1800764

17q23

CC vs CT + TT: 0.70 (0.54–0.91)

Cau

1,565 (4)

103

CH25H

rs13500

10q23

Yes

T vs C: 1.44 (1.08–1.93)

TT + TC vs CC: 1.47 (1.08–2.00)

All

3,413 (7)

25,26

1q23

Near

T vs G: 0.67 (0.50–0.90)

TT + TG vs GG: 0.69 (0.51–0.94)

All

1,363 (4)

31

CHRNB2 rs4845378

No

C vs T: 0.79 (0.68–0.92)

Initial study references

CST3*

rs1064039

20q11

No

C vs G: 1.17 (1.04–1.32)

CC vs GC + GG: 1.72 (1.17–2.54)

All

4,057 (10)

GAB2*

rs2373115

11q14

No

T vs G: 0.79 (0.67–0.94)

TT vs TG + GG: 0.56 (0.36–0.88)

All/Cau

5,961 (7)

104,105

LMNA

rs505058

1q23

Near

C vs T: 1.35 (1.12–1.63)

CC + TC vs TT: 1.36 (1.11–1.65)

All/Cau

3,663 (5)

10

MAPT*

rs2471738

17q21

No

T vs C: 1.24 (1.01–1.53)

TT + CT vs CC: 1.25 (1.00–1.56)

All/Cau

3,145 (6)

106,107

PRNP

rs1799990

20p13

No

G vs A: 0.88 (0.81–0.96)

GG + GA vs AA: 0.81 (0.67–0.96)

Cau

7,258 (13)

108,109

SORL1*

rs2070045

11q24

No

G vs T: 1.21 (1.10–1.34)

GG + GT vs TT: 1.25 (1.12–1.39)

Cau

5,953 (8)

77

TF

rs1049296

3q22

No

C2 vs C1: 1.18 (1.04–1.33) C2 vs non-C2: 1.21 (1.03–1.42)

All

7,375 (14)

110

13

Based on data from Alzgene and current as of 31 March 2008. Note that genotype-based results are not displayed in the database and that new studies might have been published for some of these polymorphisms since the above date, therefore summary odds ratios (ORs), 95% confidence intervals (CIs) and sample sizes online might be different from those displayed here. Genes marked with an asterisk had more than one polymorphism that showed significant summary ORs (allelic contrast). The ‘Initial studies’ column lists the first publication(s) to indicate a significant association between Alzheimer’s disease (AD) and the respective gene. ‘Linkage region’ is as listed on AlzGene: ‘Yes’ = within implied linkage region; ‘near’ = within ~10 Mb of implied linkage region. ‘Stratum’: ‘All’ = samples of all ethnic backgrounds; ‘Cau’ = Caucasian-only samples; ‘All/Cau’ = meta-analysis results equivalent in both strata. ‘N’: sum of AD cases and controls included in meta-analysis. ‘Samples’: number of independent case–control samples. For more details on all underlying test statistics, see rEF. 7. ACE, angiotensin I converting enzyme; CH25H, cholesterol 25-hydroxylase; CHRNB2, cholinergic receptor, nicotinic, b2; CST3, cystatin C; GAB2, GRB2-associated binding protein 2; LMNA, lamin A/C; MAPT, microtubule-associated protein tau; PRNP, prion protein; SORL1, sortilin-related receptor, L(DLR class) A repeats-containing; TF, transferrin; vs, versus.

NATuRE REvIEWS | neuRosCIenCe

vOlumE 9 | O cTOBER 2008 | 775

REVIEWS speculated to affect the iron­binding properties of TF. This hypothesis was not confirmed in at least two stud­ ies86,87. As is often the case in complex­disease genetics, this could indicate either another functional correlate of this SNP or the presence of lD with a still­elusive AD­predisposing variant. It is interesting to note that a recent AD association study concluded that the TF Ser allele might act epistatically with the Tyr allele at another locus that is involved in iron metabolism (rs1800562 (cys282Tyr) in the haemochromatosis (HFE) gene on chromosome 6p22 (rEF. 88)), although these findings have yet to be independently replicated.

Conclusions much remains to be learned about the genetics of late­ onset AD, a highly heritable but genetically poorly defined neurodegenerative disease. To date, more than 20 non­APOE­related loci exhibit nominally significant association with disease risk in systematic meta­analyses of the available AD genetic literature7. These findings implicate many of the potential culprits that have long been believed to be involved in the development of neurodeneration and dementia (such as APP metabo­ lism, Ab degradation and clearance, signal transduc­ tion, tau dysfunction, protein trafficking, cholinergic deficits, cholesterol metabolism and the homeostasis of heavy metals). This is not surprising, as approximately two­thirds of the currently implicated genetic variants are located in candidate genes that were originally tested on the basis of these functional hypotheses. Elevating

Tanzi, R. E. & Bertram, L. Twenty years of the Alzheimer’s disease amyloid hypothesis: a genetic perspective. Cell 120, 545–555 (2005). 2. Bergem, A. L., Engedal, K. & Kringlen, E. The role of heredity in late-onset Alzheimer disease and vascular dementia. A twin study. Arch. Gen. Psychiatry 54, 264–270 (1997). 3. Daw, E. W. et al. The number of trait loci in late-onset Alzheimer disease. Am. J. Hum. Genet. 66, 196–204 (2000). 4. Gatz, M. et al. Role of genes and environments for explaining Alzheimer disease. Arch. Gen. Psychiatry 63, 168–174 (2006). This is the most recent and the most extensive AD twin study, and provides probably the most realistic heritability estimates. 5. Strittmatter, W. J. et al. Apolipoprotein E: highavidity binding to beta-amyloid and increased frequency of type 4 allele in late-onset familial Alzheimer disease. Proc. Natl Acad. Sci. USA 90, 1977–1981 (1993). 6. Saunders, A. M. et al. Association of apolipoprotein E allele epsilon 4 with late-onset familial and sporadic Alzheimer’s disease. Neurology 43, 1467–1472 (1993). 7. Bertram, L., McQueen, M. B., Mullin, K., Blacker, D. & Tanzi, R. E. Systematic meta-analyses of Alzheimer disease genetic association studies: the AlzGene database. Nature Genet. 39, 17–23 (2007). This paper describes the first complex-disease meta-analysis database and the methodology behind the AlzGene approach. 8. Sloane, P. D. et al. The public health impact of Alzheimer’s disease, 2000–2050: potential implication of treatment advances. Annu. Rev. Public Health 23, 213–231 (2002). 9. Brookmeyer, R., Gray, S. & Kawas, C. Projections of Alzheimer’s disease in the United States and the public health impact of delaying disease onset. Am. J. Public Health 88, 1337–1342 (1998). 10. Grupe, A. et al. Evidence for novel susceptibility genes for late-onset Alzheimer’s disease from a genome-wide 1.

776 | O cTOBER 2008 | vOlumE 9

11.

12.

13.

14. 15. 16. 17.

18.

19.

AD genetics research to the next level will require the development of new bioinformatics tools and systems­ biology approaches to facilitate the elucidation of intersecting functions and biochemical pathways for a growing set of potential AD genes derived from candi­ date­gene association studies, sytematic meta­analyses and high­density GWA screens. The integration of these approaches should offer new hypotheses and molecular mechanisms that should eventually sharpen the overall picture (FIG. 1) of the pathogenetic forces that lead to AD and to other neurodegenerative diseases. It is likely that many of the currently most promis­ ing AD genes — including those that were discussed in more detail above — will eventually prove not to be rel­ evant modifiers of AD risk or related phenotypic traits. conversely, as the field develops more powerful asso­ ciation testing strategies, additional loci with significant risk effects will emerge. The coming and going of pro­ claimed ‘novel disease genes’ has been characteristic of all genetic epidemiology research into complex diseases. AlzGene and related field synopses provide powerful tools for continuously and quantitatively monitoring the progress within and across diseases, making results easily accessible to investigators of various different dis­ ciplines. Eventually, however, only combining the forces of genetics, genomics, proteomics and other disciplines will give rise to new diagnostic and therapeutic targets and thus herald a new era of AD research that will enable the early prediction and therapeutic management of this devastating disease.

association study of putative functional variants. Hum. Mol. Genet. 16, 865–873 (2007). This was the first genome-wide AD association study. It focused on polymorphisms in or near coding regions. Coon, K. D. et al. A high-density whole-genome association study reveals that APOE is the major susceptibility gene for sporadic late-onset Alzheimer’s disease. J. Clin. Psychiatry 68, 613–618 (2007). Li, H. et al. Candidate single-nucleotide polymorphisms from a genomewide association study of Alzheimer disease. Arch. Neurol. 65, 45–53 (2008). This high-density genome-wide association study identified several potential new AD loci, including several that had previously been implicated by meta-analyses. Reiman, E. M. et al. GAB2 alleles modify Alzheimer’s risk in APOE e4 carriers. Neuron 54, 713–720 (2007). This was the first high-density genome-wide AD association study. It implicated GAB2 as a potential new AD locus. Henschke, P. J., Bell, D. A. & Cape, R. D. Alzheimer’s disease and HLA. Tissue Antigens 12, 132–135 (1978). Sayed-Tabatabaei, F. A., Oostra, B. A., Isaacs, A., van Duijn, C. M. & Witteman, J. C. ACE polymorphisms. Circ. Res. 98, 1123–1133 (2006). Keavney, B. et al. Measured haplotype analysis of the angiotensin-I converting enzyme gene. Hum. Mol. Genet. 7, 1745–1751 (1998). Hu, J., Igarashi, A., Kamata, M. & Nakagawa, H. Angiotensin-converting enzyme degrades Alzheimer amyloidb-peptide (Ab); retards Ab aggregation, deposition, fibril formation; and inhibits cytotoxicity. J. Biol. Chem. 276, 47863–47868 (2001). Hemming, M. L. & Selkoe, D. J. Amyloidb-protein is degraded by cellular angiotensin-converting enzyme (ACE) and elevated by an ACE inhibitor. J. Biol. Chem. 280, 37644–37650 (2005). Eckman, E. A. et al. Regulation of steady-state b-amyloid levels in the brain by neprilysin and endothelin-converting enzyme but not angiotensin-

20.

21.

22. 23.

24.

25.

26.

27.

28. 29.

converting enzyme. J. Biol. Chem. 281, 30471–30478 (2006). Hemming, M. L., Selkoe, D. J. & Farris, W. Effects of prolonged angiotensin-converting enzyme inhibitor treatment on amyloid b-protein metabolism in mouse models of Alzheimer disease. Neurobiol. Dis. 26, 273–281 (2007). Takeda, S., Sato, N., Ogihara, T. & Morishita, R. The renin-angiotensin system, hypertension and cognitive dysfunction in Alzheimer’s disease: new therapeutic potential. Front. Biosci. 13, 2253–2265 (2008). Ohrui, T. et al. Effects of brain-penetrating ACE inhibitors on Alzheimer disease progression. Neurology 63, 1324–1325 (2004). Savaskan, E. et al. Cortical alterations of angiotensin converting enzyme, angiotensin II and AT1 receptor in Alzheimer’s dementia. Neurobiol. Aging 22, 541–546 (2001). Miners, J. S. et al. Angiotensin-converting enzyme (ACE) levels and activity in Alzheimer’s disease, and relationship of perivascular ACE-1 to cerebral amyloid angiopathy. Neuropathol. Appl. Neurobiol. 34, 181–193 (2008). Papassotiropoulos, A. et al. Genes involved in brain cholesterol metabolism are associated with the risk for Alzheimer’s disease and with disease related traits. Neurobiol. Aging 23, S268 (2002). Papassotiropoulos, A. et al. Cholesterol 25-hydroxylase on chromosome 10q is a susceptibility gene for sporadic Alzheimer’s disease. Neurodegener. Dis. 2, 233–241 (2005). Zerbinatti, C. V. et al. Oxysterol-binding protein-1 (OSBP1) modulates processing and trafficking of the amyloid precursor protein. Mol. Neurodegener 3, 5 (2008). Puglielli, L., Tanzi, R. E. & Kovacs, D. M. Alzheimer’s disease: the cholesterol connection. Nature Neurosci. 6, 345–351 (2003). Kalamida, D. et al. Muscle and neuronal nicotinic acetylcholine receptors. Structure, function and pathogenicity. FEBS J. 274, 3799–3845 (2007).

www.nature.com/reviews/neuro

REVIEWS 30. Kawamata, J. & Shimohama, S. Association of novel and established polymorphisms in neuronal nicotinic acetylcholine receptors with sporadic Alzheimer’s disease. J. Alzheimers Dis. 4, 71–76 (2002). 31. Cook, L. J. et al. Candidate gene association studies of the a4 (CHRNA4) and b2 (CHRNB2) neuronal nicotinic acetylcholine receptor subunit genes in Alzheimer’s disease. Neurosci. Lett. 358, 142–146 (2004). 32. Oddo, S. & LaFerla, F. M. The role of nicotinic acetylcholine receptors in Alzheimer’s disease. J. Physiol. (Paris) 99, 172–179 (2006). 33. Tohgi, H., Utsugisawa, K., Yoshimura, M., Nagane, Y. & Mihara, M. Age-related changes in nicotinic acetylcholine receptor subunits a4 and b2 messenger RNA expression in postmortem human frontal cortex and hippocampus. Neurosci. Lett. 245, 139–142 (1998). 34. Zoli, M., Picciotto, M. R., Ferrari, R., Cocchi, D. & Changeux, J. P. Increased neurodegeneration during ageing in mice lacking high-affinity nicotine receptors. EMBO J. 18, 1235–1244 (1999). 35. Wu, J. et al. b-Amyloid directly inhibits human a4b2nicotinic acetylcholine receptors heterologously expressed in human SH-EP1 cells. J. Biol. Chem. 279, 37842–37851 (2004). 36. Oddo, S. et al. Chronic nicotine administration exacerbates tau pathology in a transgenic model of Alzheimer’s disease. Proc. Natl Acad. Sci. USA 102, 3046–3051 (2005). 37. De Fusco, M. et al. The nicotinic receptor b2 subunit is mutant in nocturnal frontal lobe epilepsy. Nature Genet. 26, 275–276 (2000). 38. Marini, C. & Guerrini, R. The role of the nicotinic acetylcholine receptors in sleep-related epilepsy. Biochem. Pharmacol. 74, 1308–1314 (2007). 39. Palm, D. E., Knuckey, N. W., Primiano, M. J., Spangenberger, A. G. & Johanson, C. E. Cystatin C, a protease inhibitor, in degenerating rat hippocampal neurons following transient forebrain ischemia. Brain Res. 691, 1–8 (1995). 40. Yasuhara, O. et al. Expression of cystatin C in rat, monkey and human brains. Brain Res. 628, 85–92 (1993). 41. Balbin, M. & Abrahamson, M. SstII polymorphic sites in the promoter region of the human cystatin C gene. Hum. Genet. 87, 751–752 (1991). 42. Benussi, L. et al. Alzheimer disease-associated cystatin C variant undergoes impaired secretion. Neurobiol. Dis. 13, 15–21 (2003). 43. Paraoan, L. et al. Unexpected intracellular localization of the AMD-associated cystatin C variant. Traffic 5, 884–895 (2004). 44. Vinters, H. V., Nishimura, G. S., Secor, D. L. & Pardridge, W. M. Immunoreactive A4 and gammatrace peptide colocalization in amyloidotic arteriolar lesions in brains of patients with Alzheimer’s disease. Am. J. Pathol. 137, 233–240 (1990). 45. Sastre, M. et al. Binding of cystatin C to Alzheimer’s amyloid b inhibits in vitro amyloid fibril formation. Neurobiol. Aging 25, 1033–1043 (2004). 46. Kaeser, S. A. et al. Cystatin C modulates cerebral b-amyloidosis. Nature Genet. 39, 1437–1439 (2007). This study applied a range of in vitro and in vivo experiments to elucidate the potential functional role of CST3 in animal models. We highlight it not for the specific results, but for the experimental approach. 47. Nagai, A. et al. Neuronal cell death induced by cystatin C in vivo and in cultured human CNS neurons is inhibited with cathepsin B. Brain Res. 1066, 120–128 (2005). 48. Levy, E., Jaskolski, M. & Grubb, A. The role of cystatin C in cerebral amyloid angiopathy and stroke: cell biology and animal models. Brain Pathol. 16, 60–70 (2006). 49. Liu, Y. & Rohrschneider, L. R. The gift of Gab. FEBS Lett. 515, 1–7 (2002). 50. Sarmay, G., Angyal, A., Kertesz, A., Maus, M. & Medgyesi, D. The multiple function of Grb2 associated binder (Gab) adaptor/scaffolding protein in immune cell signaling. Immunol. Lett. 104, 76–82 (2006). 51. Chapuis, J. et al. Association study of the GAB2 gene with the risk of developing Alzheimer’s disease. Neurobiol. Dis. 30, 103–106 (2008). 52. Reynolds, C. H. et al. Phosphorylation regulates tau interactions with SH3 domains of phosphatidylinositol-3-kinase, phospholipase cg1, GRB2 and SRC-family kinases. J. Biol. Chem. 8 May 2008 (doi:10.1074/jbc.M709715200).

NATuRE REvIEWS | neuRosCIenCe

53. Nizzari, M. et al. Amyloid precursor protein and presenilin1 interact with the adaptor GRB2 and modulate ERK 1,2 signaling. J. Biol. Chem. 282, 13833–13844 (2007). 54. Rankin, J. & Ellard, S. The laminopathies: a clinical review. Clin. Genet. 70, 261–274 (2006). 55. Duesing, K. et al. Evaluating the association of common LMNA variants with type 2 diabetes and quantitative metabolic phenotypes in French Europids. Diabetologia 51, 76–81 (2008). 56. Capell, B. C. & Collins, F. S. Human laminopathies: nuclei gone genetically awry. Nature Rev. Genet. 7, 940–952 (2006). 57. Maraganore, D. M. et al. High-resolution wholegenome association study of Parkinson disease. Am. J. Hum. Genet. 77, 685–693 (2005). 58. Bertram, L. et al. Family-based association between Alzheimer’s disease and variants in UBQLN1. N. Engl. J. Med. 352, 884–894 (2005). 59. Mackenzie, I. R. & Rademakers, R. The molecular genetics and neuropathology of frontotemporal lobar degeneration: recent developments. Neurogenetics 8, 237–248 (2007). 60. Ballatore, C., Lee, V. M. & Trojanowski, J. Q. Taumediated neurodegeneration in Alzheimer’s disease and related disorders. Nature Rev. Neurosci. 8, 663–672 (2007). This is an excellent review on tau function and dysfunction in AD and in other neurodegenerative diseases. 61. Hinds, D. A. et al. Whole-genome patterns of common DNA variation in three human populations. Science 307, 1072–1079 (2005). 62. Stefansson, H. et al. A common inversion under selection in Europeans. Nature Genet. 37, 129–137 (2005). 63. Pittman, A. M., Fung, H. C. & de Silva, R. Untangling the tau gene association with neurodegenerative disorders. Hum. Mol. Genet. 15 Spec. No. 2, R188–R195 (2006). 64. Myers, A. J. et al. The MAPT H1c risk haplotype is associated with increased expression of tau and especially of 4 repeat containing transcripts. Neurobiol. Dis. 25, 561–570 (2007). 65. Rademakers, R. et al. High-density SNP haplotyping suggests altered regulation of tau gene expression in progressive supranuclear palsy. Hum. Mol. Genet. 14, 3281–3292 (2005). 66. Kwok, J. B. et al. Tau haplotypes regulate transcription and are associated with Parkinson’s disease. Ann. Neurol. 55, 329–334 (2004). 67. Kauwe, J. S. et al. Variation in MAPT is associated with cerebrospinal fluid tau levels in the presence of amyloid-b deposition. Proc. Natl Acad. Sci. USA 105, 8050–8054 (2008). 68. Gambetti, P., Kong, Q., Zou, W., Parchi, P. & Chen, S. G. Sporadic and familial CJD: classification and characterisation. Br. Med. Bull. 66, 213–239 (2003). 69. Palmer, M. S., Dryden, A. J., Hughes, J. T. & Collinge, J. Homozygous prion protein genotype predisposes to sporadic Creutzfeldt–Jakob disease. Nature 352, 340–342 (1991). 70. Schwarze-Eicker, K. et al. Prion protein (PrPc) promotes b-amyloid plaque formation. Neurobiol. Aging 26, 1177–1182 (2005). 71. Lewis, P. A. et al. Codon 129 polymorphism of the human prion protein influences the kinetics of amyloid formation. J. Gen. Virol. 87, 2443–2449 (2006). 72. Baskakov, I. et al. The presence of valine at residue 129 in human prion protein accelerates amyloid formation. FEBS Lett. 579, 2589–2596 (2005). 73. Yamazaki, H. et al. Elements of neural adhesion molecules and a yeast vacuolar protein sorting receptor are present in a novel mammalian low density lipoprotein receptor family member. J. Biol. Chem. 271, 24761–24768 (1996). 74. Taira, K. et al. LR11, a mosaic LDL receptor family member, mediates the uptake of ApoE-rich lipoproteins in vitro. Arterioscler. Thromb. Vasc. Biol. 21, 1501–1506 (2001). 75. Andersen, O. M. et al. Neuronal sorting proteinrelated receptor sorLA/LR11 regulates processing of the amyloid precursor protein. Proc. Natl Acad. Sci. USA 102, 13461–13466 (2005). 76. Andersen, O. M. et al. Molecular dissection of the interaction between amyloid precursor protein and its neuronal trafficking receptor SorLA/LR11. Biochemistry 45, 2618–2628 (2006). 77. Rogaeva, E. et al. The neuronal sortilin-related receptor SORL1 is genetically associated with

78.

79. 80. 81.

82. 83.

84.

85.

86. 87.

88.

89. 90.

91.

92.

93.

94.

95. 96. 97. 98. 99. 100.

101.

Alzheimer disease. Nature Genet. 39, 168–177 (2007). Offe, K. et al. The lipoprotein receptor LR11 regulates amyloid b production and amyloid precursor protein traffic in endosomal compartments. J. Neurosci. 26, 1596–1603 (2006). Scherzer, C. R. et al. Loss of apolipoprotein E receptor LR11 in Alzheimer disease. Arch. Neurol. 61, 1200–1205 (2004). Sager, K. L. et al. Neuronal LR11/sorLA expression is reduced in mild cognitive impairment. Ann. Neurol. 62, 640–647 (2007). Brewer, G. J. Iron and copper toxicity in diseases of aging, particularly atherosclerosis and Alzheimer’s disease. Exp. Biol. Med. (Maywood) 232, 323–335 (2007). Loeffler, D. A. et al. Transferrin and iron in normal, Alzheimer’s disease, and Parkinson’s disease brain regions. J. Neurochem. 65, 710–724 (1995). Smith, M. A., Harris, P. L., Sayre, L. M. & Perry, G. Iron accumulation in Alzheimer disease is a source of redox-generated free radicals. Proc. Natl Acad. Sci. USA 94, 9866–9868 (1997). Yamamoto, A. et al. Iron (III) induces aggregation of hyperphosphorylated tau and its reduction to iron (II) reverses the aggregation: implications in the formation of neurofibrillary tangles of Alzheimer’s disease. J. Neurochem. 82, 1137–1147 (2002). Huang, X., Moir, R. D., Tanzi, R. E., Bush, A. I. & Rogers, J. T. Redox-active metals, oxidative stress, and Alzheimer’s disease pathology. Ann. NY Acad. Sci. 1012, 153–163 (2004). Lee, P. L., Ho, N. J., Olson, R. & Beutler, E. The effect of transferrin polymorphisms on iron metabolism. Blood Cells Mol. Dis. 25, 374–379 (1999). Zatta, P. et al. The C2 variant of human serum transferrin retains the iron binding properties of the native protein. Biochim. Biophys. Acta 1741, 264–270 (2005). Robson, K. J. et al. Synergy between the C2 allele of transferrin and the C282Y allele of the haemochromatosis gene (HFE) as risk factors for developing Alzheimer’s disease. J. Med. Genet. 41, 261–265 (2004). Raiha, I., Kaprio, J., Koskenvuo, M., Rajala, T. & Sourander, L. Alzheimer’s disease in Finnish twins. Lancet 347, 573–578 (1996). Meyer, J. M. & Breitner, J. C. Multiple threshold model for the onset of Alzheimer’s disease in the NASNRC twin panel. Am. J. Med. Genet. 81, 92–97 (1998). Pedersen, N. L., Posner, S. F. & Gatz, M. Multiplethreshold models for genetic influences on age of onset for Alzheimer disease: findings from Swedish twins. Am. J. Med. Genet. 105, 724–728 (2001). Lahiri, D. K., Maloney, B., Basha, M. R., Ge, Y. W. & Zawia, N. H. How and when environmental agents and dietary factors affect the course of Alzheimer’s disease. Curr. Alzheimer Res. 4, 219–228 (2007). McCarthy, M. I. et al. Genome-wide association studies for complex traits: consensus, uncertainty and challenges. Nature Rev. Genet. 9, 356–369 (2008). This is a state-of-the-art review of the current status of genome-wide association studies and their implications for complex diseases. Hoggart, C. J., Clark, T. G., De Iorio, M., Whittaker, J. C. & Balding, D. J. Genome-wide significance for dense SNP and resequencing data. Genet. Epidemiol. 32, 179–185 (2008). DerSimonian, R. & Laird, N. Meta-analysis in clinical trials. Control. Clin. Trials 7, 177–188 (1986). Ioannidis, J. P. et al. Assessment of cumulative evidence on genetic associations: interim guidelines. Int. J. Epidemiol. 37, 120–132 (2008). Xu, H. et al. Estrogen reduces neuronal generation of Alzheimer b-amyloid peptides. Nature Med. 4, 447–451 (1998). Goate, A. et al. Segregation of a missense mutation in the amyloid precursor protein gene with familial Alzheimer’s disease. Nature 349, 704–706 (1991). Sherrington, R. et al. Cloning of a gene bearing missense mutations in early-onset familial Alzheimer’s disease. Nature 375, 754–760 (1995). Rogaev, E. I. et al. Familial Alzheimer’s disease in kindreds with missense mutations in a gene on chromosome 1 related to the Alzheimer’s disease type 3 gene. Nature 376, 775–778 (1995). Levy-Lahad, E. et al. Candidate gene for the chromosome 1 familial Alzheimer’s disease locus. Science 269, 973–977 (1995).

vOlumE 9 | O cTOBER 2008 | 777

REVIEWS 102. Bertram, L. & Tanzi, R. E. Of replications and refutations: the status of Alzheimer’s disease genetic research. Curr. Neurol. Neurosci. Rep. 1, 442–450 (2001). 103. Kehoe, P. G. et al. Variation in DCP1, encoding ACE, is associated with susceptibility to Alzheimer disease. Nature Genet. 21, 71–72 (1999). 104. Crawford, F. C. et al. A polymorphism in the cystatin C gene is a novel risk factor for late-onset Alzheimer’s disease. Neurology 55, 763–768 (2000). 105. Finckh, U. et al. Genetic association of a cystatin C gene polymorphism with late-onset Alzheimer disease. Arch. Neurol. 57, 1579–1583 (2000). 106. Lilius, L. et al. Tau gene polymorphisms and apolipoprotein E e4 may interact to increase risk for Alzheimer’s disease. Neurosci. Lett. 277, 29–32 (1999). 107. Bullido, M. J. et al. A polymorphism in the tau gene associated with risk for Alzheimer’s disease. Neurosci. Lett. 278, 49–52 (2000). 108. Casadei, V. M. et al. Prion protein gene polymorphism and Alzheimer’s disease: one modulatory trait of

778 | O cTOBER 2008 | vOlumE 9

cognitive decline? J. Neurol. Neurosurg. Psychiatry 71, 279–280 (2001). 109. Dermaut, B. et al. PRNP Val129 homozygosity increases risk for early-onset Alzheimer’s disease. Ann. Neurol. 53, 409–412 (2003). 110. van Rensburg, S. J., Carstens, M. E., Potocnik, F. C., Aucamp, A. K. & Taljaard, J. J. Increased frequency of the transferrin C2 subtype in Alzheimer’s disease. Neuroreport 4, 1269–1271 (1993).

Acknowledgements

This work was sponsored by grants from the National Institute on Aging (5R01AG23667 to L.B.) and the National Institute of Mental Health (5R37MH60009 to R.E.T.). The AlzGene database was developed in collaboration with the Alzheimer Research Forum and is funded by the Cure Alzheimer’s Fund.

Competing interests statement

The authors declare competing financial interests: see web version for details.

DATABASES Alzgene: www.alzgene.org 1q23 | 1q31–42 | 3q22 | 10q23 | 11q14 | 11q24 | 14q24.3 | 17q21 | 17q23 | 20p13 | 20q11 | 21q21.3 | ACE | CH25H | CHRNB2 | CST3 | GAB2 | LMNA | MAPT | PRNP | rs13500 | rs505058 | rs1049296 | rs1064039 | rs1799990 | rs1800764 | rs2070045 | rs2373115 | rs2471738 | rs4845378 | SORL1 | TF

FURTHER INFORMATION AD and FTD mutation database: http://www.molgen.ua. ac.be/ADMutations/ Alzheimer Research Forum: http://www.alzforum.org PDGene: http://www.pdgene.org/ SzGene: http://www.szgene.org

SUPPLEMENTARY INFORMATION See online article: S1 (table) | S2 (box) ALL LInks ARe ACTIVe In The onLIne PDF

www.nature.com/reviews/neuro