Genetic and epigenetic studies of amyotrophic lateral sclerosis

Abstract

The identification of genetic and epigenetic factors that are associated with an increased risk of developing amyotrophic lateral sclerosis (ALS), or that modify the age of onset or rate of progression, requires a multimodal research strategy, facilitated through international collaboration. The discovery of several ALS genes strongly linked to RNA biology, the proteasome pathway, and axonal transport suggest they have an important role in pathogenesis, but the immense complexity of these processes is also apparent. The increasing rate of genetic discoveries brings the hope of designing more targeted and efficacious therapies.

Key words::

• ALS
• genetic
• epigenetic
• familial ALS
• modifier genes

Introduction

Amyotrophic lateral sclerosis (ALS) is a devastating neurological disease in which neurodegeneration, predominantly of motor neurons, leads to progressive paralysis of voluntary muscles until death follows, usually from respiratory failure, typically within about five years of symptom onset. Research into the causes of ALS is essential for the rational design of therapies. A simple view is that there can only be two causes of ALS: genetic variation and environmental exposures. A third possibility is also important to consider and is known as epigenetic variation: environmental exposures, random processes or genetic variation result in changes to DNA, principally, but not exclusively, by methylation. The study of epigenetic factors is in its infancy but making rapid progress. Environmental factors are important and the subject of epidemiological studies, but have the difficulty that an environmental factor may have acted well before the onset of symptoms and could therefore be difficult to capture accurately. Genetic studies are feasible and, with the rapid advances in genetic technology, becoming easier.

It is known that ALS has a genetic component because a family history of ALS is reported by about 5% of patients (1); twin studies show that the heritability is about 60% (2), and disease-causing mutations have been identified in up to 60% of those with a family history, and between 5% and 15% of those without to date (3).

One problem is that genes described as ALS genes may produce atypical phenotypes including very young age of onset, very slow progression, or purely upper or lower motor neuron signs, while other genes regarded as typical ALS genes may give rise to ALS with frontotemporal dementia (ALS-FTD). Another issue is the distinction made by many between familial ALS (FALS) and apparently sporadic ALS (SALS), because this is useful when gene-hunting. However, FALS is not well defined by the research community (4); the identification of another affected individual depends on the recall and knowledge of the patient, and there is a major statistical bias in whether a large effect gene will affect only one person in a pedigree, dependent on the family size and how penetrant the disease gene is (5). Thus, the distinction is artificial and it is therefore expected that genes mutated in FALS should also be found to cause a proportion of what is called SALS.

ALS genes

How to find ALS genes

The frequency of a genetic variation is a measure of how large an effect it has on a disease. This is because when a mutation first arises, if it has a very deleterious effect it will be strongly selected against and is unlikely to rise to a high frequency. If it has a neutral or small effect it can become more frequent by chance. A direct consequence of this is that strongly deleterious genes, which may be sufficient to cause ALS alone, are more likely to be found in a few families only, while small effect genes, which only increase the risk of ALS a little, need to act in concert and may therefore lead to familial clustering without necessarily resulting in a strong family history.

There are two broad strategies for identifying disease genes − linkage studies and association studies. In linkage analysis, a statistical test is performed in a family to determine if a genetic variant occurs in affected individuals more often than would be expected by chance. If it does, this is called linkage (6). In association analysis, the frequency of a genetic variant is compared between a group of unrelated affected individuals and a group of unrelated controls. If the frequency differs between the groups then the variant shows association with disease (7). Thus, linkage studies are useful in pedigrees and therefore in finding rare, highly deleterious genes, while association studies are useful in large groups of unrelated individuals.

Linkage studies use an indirect approach, employing known genetic variants as markers for the true disease-causing gene variant, and produce a robust estimate of the likely location of a disease gene (8). The resolution of linkage is low, so the estimate of the location may be very many megabases long. On the other hand, association studies are prone to false positive findings but produce an estimate of the true disease gene location that is precise and often just a few kilobases in length.

Association studies historically used direct sequencing of genes thought to be good candidates for disease causation, and the general burden of variants in cases compared with controls. With advances in genetic technology, an indirect method that interrogates the entire genome became possible in which hundreds of thousands of genetic markers could be tested in thousands of individuals. This is known as a genome-wide association study (GWAS). It is now becoming possible to directly assay all genes by sequencing, so-called whole genome or whole exome sequencing.

Genes causing ALS

A database of ALS genes (ALSoD) (9) with their associated phenotypes and a measure of their credibility is available at the World Federation of Neurology and European Network for the Cure of ALS (ENCALS) project website http://alsod.iop.kcl.ac.uk, generously supported by ALSA, ALS Therapy Alliance, MNDA and ALS Canada.

Familial ALS genes, focusing on UBQLN2

The genes most widely accepted as causing typical ALS when mutated are SOD1 (10), TARDBP (coding for TDP-43) (11–13), and FUS (14,15). Genes that may have an effect confined to specific populations are ANG (16) and OPTN (17). Genes for ALS and ALS-FTD include UBQLN2 (18) and a gene identified within a few days of this meeting, C9orf72 (19,20). The C9orf72 gene causes ALS through a massive expansion of a hexanucleotide repeat motif between non-coding exons 1a and 1b (GGGCC)_n. In unaffected individuals, the sequence is usually repeated just once or twice, although repeat sizes up to the 20s and 30s have been recorded. In ALS or ALS-FTD the expansion is hundreds of repeats in length. The mechanism by which this results in disease is unknown, and it is not obvious if the gene itself is relevant or the expansion would be pathological if inserted into other genes. The C9orf72 gene expansion probably comes from a single founder, but it is also possible that the haplotype on which it occurs is inherently unstable and results in spontaneous expansion. Because it is a relatively common cause of ALS, and is found in apparently sporadic ALS as well, mutation in C9orf72 is a landmark discovery of great clinical importance.

The UBQLN2 gene is of particular interest because it is involved in the ubiquitin pathway, and aberrant ubiquitination is a common feature of all pathological protein inclusions in ALS. A family with ALS showing X-linked inheritance was studied by linkage, and a locus identified in the pericentromeric region (18). A mutation was found in the UBQLN2 gene that segregated with disease. Testing other families, including families with ALS-FTD, identified mutations in four other pedigrees; these mutations all occurred in the same part of the gene.

The existing major ALS genes are associated with distinct pathology, unrelated to TDP-43 or FUS (SOD1), related to FUS (FUS) or related to TDP-43 (TARDBP, other genes and ALS without an identified genetic basis). Interestingly, antibodies to UBQLN2 show that it colocalizes in ALS in inclusions containing TDP-43, FUS and OPTN. The pathology of individuals with ALS caused by UBQLN2 mutation shows a novel aggregate in radial neurites with a membrane bound inclusion in the hippocampus. In patients with dementia or ALS not caused by UBQLN2 mutation, UBQLN2 aggregates are still found, but TDP-43 aggregates are not always present, suggesting that the UBQLN2 pathway represents a more unifying process than that of TDP-43.

Thus, mutations in UBQLN2 cause both ALS and dementia, in adults and juvenile patients, with a novel pathology extending to sporadic ALS and ALS dementia and other forms of non-ubiquilin familial diseases, even when TDP-43 pathology is not present.

Genes for apparently sporadic ALS

Several non-overlapping GWAS have been conducted for SALS, the first reporting in 2007 (21–32), including genome-wide association studies of copy number variation (33–36). The early studies failed to find replicable associations, even with quite large sample sizes. Furthermore, as sample sizes increase, the effect of any genetic variant identified tends to be smaller, which some would argue makes it of less relevance. Thus, it is perhaps not unreasonable to conclude that GWAS has failed in ALS. However, a large scale GWAS identified two susceptibility loci for SALS; one on chromosome 19p13.3 that maps to a haplotype block within the boundaries of the UNC13A (unc-13 homolog A) gene and one on chromosome 9p21.2 (28). The second locus was of particular interest since it coincided with a known linkage region for ALS-FTD and was therefore a more robust finding. It was then replicated in two further GWAS studies, one in UK samples and one in Finnish samples (29,30). Thus, the weakness of association studies, that they are prone to false positive results, was not at play here, and there was only the benefit of association, that the disease gene location is more precisely known. The disease gene was identified within a few months of the publications and a few days of this conference, as C9orf72 (19,20), now known to be a moderate effect gene, involved in up to 60% of FALS and up to 15% of SALS (37,38), varying considerably by population.

Three other genes are of note here: UNC13A, ELP3, and ATXN2 (28,32,39). The first two were found by GWAS with corroborating evidence, and implicate disease pathways already thought to be involved in ALS. The ALS-associated SNP within UNC13A is, by meta-analysis, the most strongly associated ALS genetic variant known. It is of interest because the ubiquitin pathway is implicated in ALS by pathology, mutation in the SQSTM1 (p62) gene, and the findings of the UBQLN2 mutations mentioned above. The genetic variants associated with ALS in the ELP3 gene are of interest because they are microsatellite alleles rather than SNPs. ELP3 is a component of RNA polymerase II and RNA processing is emerging as a common theme in ALS pathogenesis; mutagenesis experiments in flies have shown that ELP3 is essential for correct neuronal development.

Recently, following a genetic screen in yeast, ataxin 2 protein (coded for by ATXN2) was found to modulate TDP-43 toxicity. Because of the central role played by TDP-43 in several neurodegenerative diseases including ALS, it then became the subject of a genetic study. ATXN2 and TDP-43 interact in an RNA-dependent complex, and both become mislocalized in spinal cord neurons of patients with ALS.

The ATXN2 (12q24.1) polyQ tract shows a normal size range that extends between 14 and 31 repeats, 22 and 23 repeats being the most frequent (40). Expansions over 34 repeats are associated with the development of spinocerebellar ataxia type 2 (SCA2), an autosomal dominant disorder characterized by progressive cerebellar gait and limb ataxia, slow saccadic eye movements, supranuclear ophthalmoplegia, and hyporeflexia (41), and there have been a few case reports of motor neuron degeneration associated with SCA2. Alleles with 32 and 33 repeats are also known to cause SCA2 but these are associated with exceptionally late onset disease (42). The genetic study found that intermediate repeat lengths predispose to ALS, a finding now replicated in several other studies.

Thus, there are now two repeat sequences definitively associated with ALS: the trinucleotide repeat in ATXN2, and the hexanucleotide repeat in C9orf72, suggesting that repeat variation may be a common theme with more to be found in the future.

Genes that modify the phenotype of ALS

Although genes for susceptibility to ALS will reveal the pathway leading to neurodegeneration, other factors such as those that modify the age of onset, the site of first symptoms, the burden of upper and lower motor neuron disease, and the rate of disease progression are also very important. Modifying the rate of disease progression is the goal of all therapeutic trials in ALS, and finding gene variants that have the same effect would have profound implications for the design of rational therapies. Similarly, factors that raise the age of onset would be useful in delaying disease in gene carriers. It is likely that there is considerable overlap between the effects of different modifying factors, as well as between modifier genes and susceptibility genes. For example, genes that act to reduce the age of onset will have a similar effect to susceptibility genes, and some susceptibility genes are known to be associated with specific patterns of upper or lower motor neuron disease, or rates of disease progression.

Modifiers of survival

Some mutations confer a survival effect in a predictable manner. This is best characterized for SOD1 mutations. The A4V, G41S, G93A and R115G variants of SOD1 have consistently poor survival while D90A homozygosity, G41D and H46R have a much slower course than average (3). Homozygosity or compound heterozygosity for ALS-causing mutations is also associated with a more rapid disease course.

The UNC13A gene, in which a SNP confers susceptibility to ALS, also appears to modulate survival, with the risk allele leading to shorter survival (43).

A GWAS that used a prevalent clinic based sample identified variation in the KIFAP3 gene associated with survival, confirmed in multiple populations within the study. The effect was similar to that of riluzole (27). A subsequent GWAS that used an incident population based sample did not replicate the original finding, but this may be a result of the younger, generally longer lived population that is seen in prevalence rather than incidence cohorts (44). Primary lateral sclerosis (PLS) has also been associated with the same variant in KIFAP3 and it is conceivable that since PLS has a generally longer survival than ALS, different inclusion criteria between the studies account for the different results (45).

Modifiers of age of onset

A study of heritability for age of ALS onset that examined the age at which members of pedigrees with two different SOD1 mutations (A4V or D90A) were affected, found heritability to be 0.29 (0–0.42), suggesting that genetic factors may play a role (46). In other words, the implication is that there are genes that increase the risk of developing ALS (susceptibility genes) and other genes that only modify phenotype without apparently increasing risk.

Homozygosity for SOD1 mutations or compound heterozygosity for ALS-causing mutations is associated with younger onset age. SOD1 mutations specifically associated with a young onset age include I104F, G37R, L38V and G114A. APOE genotype is reported to modify age of onset in ALS (47).

No modifiers of limb or bulbar onset have yet been found. Similarly, in ALS-FTD, no modifier that determines ALS rather than FTD has yet been found, although international efforts are underway to explore this.

Despite the increasing success in identifying genes that underlie familial ALS (FALS), which results from penetrant monogenic mutations, the etiology of most cases of ALS (sporadic ALS or SALS) remains unknown, and more complex genetic models may need to be considered. Such models could, for example, take into account mutations in several genes, possibly with variable penetrance, with additive effects in the presence or absence of environmental factors. Therefore, variations in a modifier gene could affect disease parameters such as age of onset, duration and site of onset. Several modifier genes have been proposed for ALS, including SMN1, the paraoxonase cluster (PON1-PON3), VEGF, ANG and others (48–51).

RNA processing, epigenetics and ALS

Non-coding RNA, RNA networks and epigenetics

Epigenetics is a new genomic science that involves DNA methylation, histone, nucleosome and chromatin remodeling, non-coding RNAs, RNA editing and DNA recoding. These processes work together and regulate expression of the genome. Epigenetics is a mechanism through which gene-environment interactions are mediated.

RNA is the ‘sister’ molecule of DNA. Although about 2% of the genome is transcribed into mRNA that encodes proteins, the rest is not. Much of it is transcribed into RNA that is thought to have sophisticated regulatory functions; up to 98% of the genome is non-coding RNAs, which are frequently transcribed, functional, and often biologically context-specific. Our knowledge of this area is still limited, but the discovery of many different functional subclasses of small and long RNAs and the ongoing studies of their function mean this is a very fruitful direction of research.

There is a rapidly evolving body of evidence that ALS is associated with complex and nuanced alterations in RNA regulatory networks (52,53). These include processes associated with transcriptional regulation, post-transcriptional processing, deployment and function of different subclasses of non-coding RNAs, local control of translation and additional epigenetic processes. Post-transcriptional processing include pre-RNA splicing; RNA editing; RNA stabilization, degradation, quality control, nuclear export, intracellular transport, and intercellular trafficking (54). Epigenetic processes include DNA methylation, histone, nucleosome and higher-order chromatin remodeling (55). The epigenome differs from the genome in that every cell has its own unique epigenome and this changes continuously as the environment interacts with it. Three common methods to examine the epigenome are: direct RNA sequencing to examine different RNA subclasses, species and splice variants; chip sequencing in which antibodies are used to immunoprecipitate a complex of interest and the DNA it contains is sequenced to examine histone, nucleosome and higher order chromatin remodeling factors and associated DNA methylation cofactors; and the third method is to examine DNA methylation using bisulfite sequencing or methylation- sensitive endonucleases.

The non-coding RNAs, which include microRNAs and so-called long non-coding RNAs, exert their dynamic, versatile and environmentally-responsive local as well as long-distance genomic effects, in large part, through genomic site-specific targeting of more generic DNA methylation, histone and chromatin modifying proteins and associated epigenetic regulatory complexes and epigenetic remodeling and reprogramming modules (56,57). Associated non-coding RNA regulatory networks have preferential roles in nervous system development, homeostasis, plasticity, connectivity and aging (58).

RNA is increasingly being recognized as a universal biosensor of environmental, interoceptive and homeostatic signals, as a transducer of biological processes utilizing a fraction of the bioenergetic costs of equivalent protein species and as a unique signaling molecule. It is able to function as an integral component of both digital, sequence-specific DNA/RNA codes and analogue, conformational protein-based codes as well as a reservoir and interpreter of embedded genetic information. This is because RNA has a digital code in the sequence of nucleotides, so it can interact with DNA in a digital manner, and also has an analogue code in having a tertiary structure and therefore can interact with proteins in an analogue manner.

The ability of RNA to participate in seminal biological functions at a molecular, genetic and cellular systems level, endows it with a uniquely broad range of complex and nuanced functional properties, including complex epigenetic memory states.

Alterations in RNA processing are implicated in ALS/MND pathogenesis through many pathways, including defects in axonal transport and cytoskeletal stabilization, DNA replication and repair, mitochondrial dysfunction, connectivity, protein degradation and stress responses (59–61). The field of RNA biology, in general, and RNA regulatory networks, in particular, represents an important, challenging and rapidly evolving scientific discipline for helping to decipher the molecular pathogenesis of ALS (62).

A number of issues remain unresolved. What are the roles of long non-coding RNAs in ALS pathogenesis? The main roles appear to be regulating adjacent protein coding genes as well as site-specific targeting of multiple genes throughout the genome that form functional gene networks. What are the roles of newly identified non-coding RNAs transcribed from the mitochondrial genome? What are the roles of alterations in developmental processes in the selective vulnerability of the motor unit? What are the implications of abnormal RNA biology in the identification of novel molecular targets and the design of innovative therapeutic strategies?

RNA editing

A feature of some RNA transcripts is that they undergo RNA editing. This is a process in which the nucleotide sequence is changed enzymatically resulting in a different sequence from that encoded by the original DNA. RNA editing occurs most efficiently in the central nervous system and the commonest change in mammals is deamination from adenosine to inosine. When editing occurs in the coding regions of mRNA, this may change the amino acid sequence from that encoded by the gene.

Glutamate receptors come in a variety of types. The AMPA receptor is a non-NMDA ionotropic receptor mediating fast synaptic neurotransmission. In vertebrate AMPA receptor proteins a section of the pore lining domain has two alternative structures with either a glutamine (Q) or arginine (R) amino acid. This variation, known as a Q/R site, has a major influence on the receptor properties. The R containing variant of the mammalian AMPA receptor subunit is not coded by the DNA sequence, but is created by post-transcriptional RNA editing activities.

Motor neurons express abnormal Q/R site-unedited GluA2 or GluR2 mRNA in sporadic ALS patients in a disease-specific and motor neuron- selective manner (63–65). GluA2 plays a regulatory role in Ca²⁺ permeability of the AMPA receptor after conversion of adenosine for the Q/R site to inosine (A-to-I conversion) by the RNA editing enzyme called adenosine deaminase acting on RNA 2 (ADAR2).

In AR2 mice in which the ADAR2 gene is conditionally knocked out in motor neurons, the motor neurons undergo slow death by failure to edit the GluA2 Q/R site (66). Notably, the ADAR2-lacking motor neurons are rescued when they are genetically engineered to express only edited GluA2 in AR2 mice. These lines of evidence suggest that complete editing of GluA2 Q/R site is necessary for neuronal survival and that expression of unedited GluA2 resulting from the reduction of ADAR2 activity may be a cause of death of motor neurons in sporadic ALS. Together with the finding that ADAR2 expression is always reduced in ALS motor neurons that exhibit TDP-43 pathology (67), a hallmark of sporadic ALS, one hypothesis is that there is a progressive down-regulation of ADAR2 activity in ALS motor neurons, in which the pathological process commences once unedited GluA2 is expressed. Normalization of GluA2 RNA editing in motor neurons may therefore be a therapeutic strategy for sporadic ALS (68).

Next steps

Every method for gene discovery in ALS has limitations. The problem comes down to the relationship between frequency and effect size. A large effect gene is rare and relevant to a few people, while a small effect gene is common and therefore of importance for ALS in general, but of such small impact that the therapeutic potential is more limited. Moderate effect genes with a penetrance in the range of 20–40% would be ideal targets in being relevant to a significant proportion of people and having a relevant effect size, and such variants could be detected with either linkage or association studies.

For GWAS, the problem is that several thousand well phenotyped cases are needed to find genetic factors of modest effect. Any genetic heterogeneity in SALS might limit the identification of causative alleles, since different disease-causing alleles may exist within the same gene in different people. If this is the case, the strength of association signals for that gene will be diluted. Although rare highly penetrant mutations are expected to cause a proportion of SALS, the identification of such variations by GWAS will inherently be more difficult since the variant needs to be frequent enough for detection.

For linkage, the problem is that large well phenotyped pedigrees are needed to find genetic causes of limited relevance since in most cases the variant will be of high penetrance and therefore rare. Since ALS is generally a disease of older age groups, atypical ALS affecting younger people or progressing more slowly than normal is more commonly studied.

For whole exome or whole genome sequencing, the problem is that huge datasets are generated, consisting of petabytes of data, and very many genetic variants are identified in each person without necessarily a simple way to tell which ones are relevant to ALS.

Nevertheless, each of these approaches has advantages too. GWAS with larger sample sizes is likely to uncover many more genes and therefore increase our understanding of the network of problems causing ALS. Linkage techniques are improving and generate robust results. Whole genome and whole exome sequencing will become easier as computing and statistical methods catch up with genetic technology. Each genetic discovery provides another view of the pathway to neurodegeneration that we hope will at some point come into sharp focus, allowing the design of targeted treatments and possibly personalized medicine.

Controversies and discussion points

There remains a strong argument in favour of simple candidate gene studies, focused on a single gene. The model is one in which biology is used to identify the gene of interest and genetic variants are then sought. This is the technique that led to the identification of ATXN2, and combined with linkage, led to the identification of TARDBP.

Similarly, combining high throughput, whole exome sequencing with association studies examining the burden of genetic variants in each gene in cases and controls allows a hybrid study in which a non-biologically based approach can still identify genes in which multiple alleles in a single gene contribute to SALS. For the reasons described above, such a scenario is likely to be less common than rare variants in families and common variants in SALS, but it is still likely to occur and will not be easily detected by other methods.

A key problem faced by researchers is that innovative ideas may be seen as high-risk projects in terms of funding, even by agencies that express a wish to consider innovative ideas. Most funders require a precedent, or evidence to directly support a hypothesis. For example, there are about 500 kinases in the genome and, even though studies show that all are important, only a few have ever been studied, possibly because it is not easy to obtain funding for research into kinases that have no existing data. There may be other methods that would find other types of genetic variation, but currently these are hidden.

A major issue is in establishing the pathogenic mechanism of ALS in pathways revealed by genetic studies. For example, is TDP-43 protein toxic when accumulated, or is there a loss of function caused by mislocalization away from the nucleus? These ideas are critical for the design of rational therapies, since it may or may not be necessary to remove the inclusions seen pathologically in ALS. One view is that these are not mutually exclusive since it is quite possible that a protein develops a novel toxic function for example by accumulating, and in being removed by the aggregate also results in a loss of function. The aggregates themselves may be toxic, may be simple gravestone-type passive markers, may be protective and part of cell defence, or may be aggregating some other protein.

Another key issue is whether future investment should be targeted towards genetic or environmental studies of ALS. The view of the geneticists in the room was that the genetic question represents a finite space and therefore a tractable question, whereas the environment is not only an infinite research space, but also impossible to address in the same detail and with the same reliability as a genome. Nevertheless, it is meaningless to understand genes without the context of the environment in which they act. Another issue is that it is possible to examine the genome without a prior hypothesis using linkage, GWAS or whole genome sequencing, but it is difficult to examine all environmental factors in the same way, so any environmental study is potentially prone to bias in the selection of factors to examine. If clusters of cases could be mapped geographically with a common factor identified, or some other such environmental clue found without a prior hypothesis, then such bias would disappear, but this is not straightforward. On the other hand, genetic research has been funded for about 20 years and can only explain about 15% of ALS, and so perhaps the time has come for environmental studies to be given priority. A solution is to study gene-environment interactions (although these require huge sample sizes), or genetic epidemiology, or to devote some resources to developing new disease models, e.g. examining the initial regional CNS nidus and subsequent spread of ALS, or cell models that might highlight the multiple environmental impacts that could trigger degeneration.

The identification of genetic factors runs into an issue to do with frequency and effect size, as discussed above. Rare genes have small impact on the disease numerically because they only affect a few people, but have a large effect biologically. On the other hand, small effect common genes may impact many people but only contribute a little individually. A gene of moderate effect might have the ‘best’ of both worlds. The main reason to identify genes is therefore not to design a specific gene therapy but to understand the biology of ALS, and so to design a rational therapy.

Acknowledgements

We thank the Motor Neurone Disease Association of Great Britain and Northern Ireland, the ALS Association, the Les Turner ALS Foundation, the ALS Society of Canada, and the Canadian Institutes of Health Research (CIHR) for support.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.

The research leading to these results has received funding from the European Community ‘s Health Seventh Framework Programme. AAC receives salary support from the National Institute for Health Research (NIHR) Dementia Biomedical Research Unit at South London and Maudsley NHS Foundation Trust and King’s College London. The views expressed are those of the author and not necessarily those of the NHS, the NIHR or the Department of Health. GAR holds the Canada Research Chair and a Jeanne-et-J-Louis-Levesque Chair for the Genetics of Brain Diseases.