Genome-Wide Association Studies: Back to the Future
Each cell expresses hundreds or thousands of genes, and most genes are expressed in many different types of cells or in varying cellular contexts. Complex traits are spatio-temporal collectives of multiple interacting gene products, changing over time (from conception onward) in any target organ or tissue, and sculpted by increasingly well-documented phenomena including largely unknown environmental influences.
—Anne Buchanan, Samuel Sholtis, Joan Richtsmeier, and Kenneth Weiss, “What Are Genes ‘for’ or Where Are Traits ‘From’?”1
Polymorphisms—variations at a single DNA site or locus—are common in nature, and much more frequent in the human genome than had been anticipated prior to its mapping. They constitute the genetic basis for human variation and diversity.2 The APOE polymorphic variation, the alleles ε2, ε3, and ε4, are clinally distributed, that is, they occur universally, but they do not everywhere have the same prevalence. And, as we have seen, it is well established that individuals with an 4 allele are at increased risk for AD under circumstances that are as yet poorly understood. Given that APOEε4 is neither necessary nor sufficient to cause AD, it has long been assumed that other genes must be involved in the incidence of this condition. It is also widely accepted that gene/gene and gene/environment interactions without doubt contribute to the initiation of AD pathology. Furthermore, given that intensive research over the years has not revealed any other gene that, similar to APOEε4, clearly has a moderately large effect on AD risk, current understanding has it that it is very unlikely that more low-hanging fruit will be found, although other genes that contribute in one way or another to AD causation are probably numerous. These genes are presumed to have small effects that may be cumulative in circumstances yet to be spelled out. Until relatively recently it has not been possible to carry out research to investigate this knowledge lacuna effectively.
The development of high-speed technologies over the past eight years or so (and hence not available when the human genome was first mapped) permits so-called deep sequencing of genomes, a technology that has enabled the investigation of AD genetics to advance in new directions. One such sequencing technique is known as GWAS (genome-wide association studies); this is an approach in which a great number of DNA sequences are randomly assayed (examined) for comparative purposes. The usual objective is to detect genetic variants known as SNPs (single nucleotide polymorphisms) in individual subjects.3 SNPs are understood as biological markers that identify a region of the human genome believed to increase risk for a disease under investigation, but SNPs do not specify genetic causation. However, when SNPs occur within a gene, or in a regulatory region near a gene, they may have a significant role in disease occurrence by affecting the function of the gene in question.
Over 1,200 GWA studies had been carried out by the end of 2010, designed to uncover differences among SNPs that may be implicated in over 200 diseases. In contrast to other methods of DNA investigation that specifically test one or a few genetic regions, GWA studies examine the entire genome. GWAS has been put to use most frequently to determine the genetic contribution to several common multicausal diseases including heart disease, cancer, and Alzheimer disease. If one type of variant is more frequent in people who have the condition under consideration, that SNP is said to be “associated” with the disease. The purpose when investigating AD, for example, is to systematically detect SNPs that appear to be associated with AD traits, such as memory loss or amyloid deposition, across the population under examination.
Today, SNP assays allow researchers to sample and identify 500,000 or more SNP sites in the genome of individual human subjects, that can then readily be compared with the SNPs present in the genomes of very many, often thousands, of other subjects. This process facilitates an examination of all or most of the genes of any given organism on a massive scale. The cost of such sequencing has been greatly reduced over the years, and the time to conduct a SNP assay has also been shortened, making use of GWAS increasingly attractive. Research subjects donate a sample of cells, usually procured using mouth swabs, or at times from hair; DNA is then extracted from the cells and spread on SNP chips, from which millions of DNA sequences can then be read. It must be reiterated that GWA studies identify chromosomal sites of interest and not genes per se, from which deductions are then made about which genes, their protein products, and noncoding parts of the DNA are likely to be implicated. Such research was originally based on the assumption that the genetics of complex, “common” diseases such as heart disease, cancer, and Alzheimer’s, among others, would be best explained using the “common disease-common variant hypothesis,” in which it is assumed that one or else several common genetic variants account for disease causation.4 In practice, once Alzheimer GWA studies commenced it quickly became evident that this hypothesis did not appear to apply, as a good number of involved experts had already suspected, and that many genes with small effects appeared to be of more significance. However, further investigation has shown that the matter is yet more complex, as we will see below.
Most GWAS are described as hypothesis-free (a fishing expedition as some biologists describe it), although there is nevertheless an underlying assumption that the trait under examination has an identifiable genetic component. GWA studies usually compare the DNA of two groups of participants: identified cases of the disease or trait under study and unaffected controls. A problem immediately arises with such an approach in connection with AD research because one cannot be certain that healthy controls are not harboring prodromal AD. As one researcher noted when talking with me, “[T]here is always the fear that you have people in your so-called healthy control samples who are already harboring AD but are not symptomatic. I read somewhere that something like 10% of the controls turn out not to be controls at a later date, or when autopsied. Of course, this effect can be diluted a bit if you use really large samples.” A further problem raised in previous chapters compounds this matter, because heated debate persists as to whether or not biomarkers do indeed reliably detect so-called prodromal AD, and, furthermore, the replicability of the clinical AD phenotype is questionable. The situation is one in which both an evasive genotype and an elusive phenotype are under investigation, making it exceptionally difficult, perhaps impossible, to establish robust associations between the two.
GWAS follow-up studies often limit the assay to SNPs located in known or predicted regions of the genome, as established by a first run. This practice results in an “enriched” sample of potentially functionally relevant areas of the genome, at the expense of aiming for genome-wide coverage—a practice that raises yet more problems when interpreting findings.
Prior to 2011, findings from eight Alzheimer GWAS had been published or were in press, all of which confirmed that the APOEε4 gene puts certain individuals at risk for AD, especially in its homozygous form. The neurogeneticists Bertram and Tanzi carried out one of these studies and published a review article in 2009 summarizing what they believe had resulted to date from GWAS research. They commenced their review by reiterating the cautionary note they first made in 2004, cited in the previous chapter, about the conflicting results that have emerged over the past 30 years in connection with AD genetics; then they set out GWAS findings of the day:
The most common perception is that late-onset AD is likely to be governed by an array of low-penetrance common risk alleles across a number of different, currently only ill-defined loci. These genes likely affect a variety of pathways, many of which are believed to be involved in the production, aggregation and removal of Aβ.
Their summary position is as follows:
As additional GWAS are carried out on larger datasets and higher-resolution arrays, we can expect the list of novel AD gene candidates to keep growing over the coming years. For all of these putative associations, replication attempts and meta-analyses across multiple independent samples will be essential to determine the identity of bona fide AD susceptibility genes. Despite the rapid progress being made in these still early days of the GWAS era, it should be emphasized that for none of the novel AD candidate genes that have thus far emerged from genome-wide screening, do we have conclusive functional genetic evidence that would allow us to unequivocally establish any of these loci as genuine AD risk genes.5
Bertram and Tanzi note that because of the exceedingly large number of studies taking place, it has become virtually impossible to systematically follow and evaluate them all. With this in mind they set up an “AlzGene” database, easily accessible on the Internet. As of 2009, drawing on the findings of various meta-analyses, AlzGene was listing 32 loci that contain at least one genetic variant that shows a “nominally significant association” with AD causation. However, an assumption has to be made that, statistically, all the studies used in the meta-analysis are comparable, which is often not the case.6 The findings of 10 of the completed GWA studies are assessed at length by Bertram and Tanzi in their review article noted above, and they conclude that over two dozen novel potential AD-associated loci have been uncovered that demand careful, repeated replication.
One of the first efforts to carry out a “cross platform comparison” on four of the Alzheimer GWA studies reached the following conclusion: “The number of replicating association signals we observed is no higher than would be expected due to chance.” The researchers suggest that increasing the power by using additional data from larger studies may result in more encouraging findings.7
On September 7, 2009, AlzGene issued a “Paper Alert” in which they reported on findings made by the two largest GWA studies to date, findings that had first appeared online. This research was published as a pair of articles in Nature Genetics, a company-run journal devoted almost entirely to the distribution of GWAS findings. AlzGene comments, “Netting three new risk genes between them, these two studies stand out not only for their collaboration and data pooling—together they engaged several dozen institutions from 10 countries—but also for fingering three new genes robustly, without the lingering doubt of small sample size and missing replication that has accompanied most prior gene association results except APOE.”8
The chief scientific director of Alzheimer’s Research UK, Julie Williams, was lead author of one of the articles. The research conducted by Williams’s group had been funded by the Wellcome Trust, the Medical Research Council, the Alzheimer’s Research Trust, and the Welsh Assembly Government, among other funders. The project involved scientists at universities in Cardiff, London, Cambridge, Nottingham, Southampton, Manchester, Oxford, Bristol, and Belfast who collaborated with researchers at Irish, German, Greek, Belgian, and American institutions. Julie Williams notes that in the study “over half a million differences in the DNA of each of 4,000 people with AD were compared with 8,000 people without the disease.” In addition to APOE, two other genes, CLU (clusterin, also known as apolipoprotein J) and PICALM, showed “overwhelming evidence for a relationship with AD.” These findings were then replicated using another sample of over 2,000 diagnosed AD cases and 2,000 controls. Williams claims, “The findings are significant and conclusive.”9 The chief executive of the Alzheimer’s Research Trust provided a nationalistic footnote: “These findings are a leap forward for dementia research. … The work of Professor Williams and colleagues shows how British researchers lead the world in the struggle to understand and defeat dementia.”10
The second GWA study was led by Philippe Amouyel, the director general of the Institut Pasteur in Lille, an MD with a PhD in cellular and molecular biology. This study made use of 2,032 cases and 5,328 controls, and the findings were replicated in a second study involving 3,978 AD cases and 3,297 controls—samples that originated from centers located in Belgium, Finland, Italy, and Spain. The findings of the Williams and Amouyel teams together allowed researchers to conclude that three genes in addition to APOE had now been shown to be definitively associated with risk for late-onset Alzheimer disease. Clusterin, produced by the CLU gene, is associated with the clearance of cellular debris and with apoptosis (cell death); it had already been found prior to this study in high levels in the blood of patients with Alzheimer’s, and is associated with cognitive decline. PICALM is involved in the transport of molecules into and inside nerve cells, and is associated with memory formation in addition to other brain functions. CRI (complement receptor 1 gene) is the third gene that proved to be significant, initially only in the French study, but after combining the findings from the French and British projects, the finding was regarded as solid. CRI appears to play a key role in connection with the functioning of the immune system, is associated with inflammation, and, if overproduced, can be very damaging to tissue, including that of the central nervous system. Prior to the GWA studies, CLU and CRI had already been shown to be involved with clearance of Aβ. In the following years a small number of other GWA studies replicated these findings.
The researchers involved with this research made it clear at the time that, in their opinion, yet larger samples were needed if the “promise” of GWAS was to be fulfilled. Accordingly, a massive intercontinental project was undertaken, the results of which were published in May 2011 as a pair of articles, once again in Nature Genetics. This time the input from the United States was substantially increased. One article lists 155 authors and the other lists 172 authors whose specific contributions to the article are set out under the following headings, among yet others: “Study Management and Coordination,” “Statistical Methods and Analysis,” “Study Design,” “Manuscript Writing Group.” Several authors contributed to both of these articles. The following excerpt gives readers a glimpse into the complexity of these undertakings:
To identify genetic variants associated with risk for Alzheimer disease, the [Alzheimer’s Disease Genetics Consortium] ADGC assembled a discovery data set [over 8,000 cases and over 7,000 controls] using data from eight cohorts and a ninth newly assembled cohort from the 29 National Institute on Aging (NIA)-funded Alzheimer’s Disease Centers (ADC’s) with data coordinated by the National Alzheimer’s Coordinating Center (NACC) and samples coordinated by the National Cell Depository for Alzheimer’s Disease (NCRAD). For the stage 2 replication we used four additional datasets and additional samples. … The stage 3 replication used the results of association analyses provided by three other consortia [the largely European-based samples discussed above] … because the cohorts were genotyped using different platforms, we used imputation to generate a common set of 2,324,889 SNPs. We applied uniform stringent quality control measures to all data sets.11
Findings reported on the basis of this research state that, in addition to APOE, nine other genes had now been shown to be associated with risk for AD. Although each of these genes individually is attributed with a very small risk, the “cumulative population attributable fraction” of these new loci is estimated to be 35%, although it is acknowledged that these estimates will probably vary widely among studies as further research findings accrue. The authors conclude that although their study was “well powered” in order to reveal risk alleles of “small effect sizes,” finding additional loci, that they are certain exist, will require yet larger studies with “increased depth of genotyping to test for the effects of both common and rare variants.”12
Clearly, numerous questions are raised by these findings, above all, under what circumstances are these genes of low effect expressed or inhibited in vivo? Are their respective effects cumulative, and what exactly are the biological pathways that they instigate, overactivate, or block, and under what circumstances? It is understandable that the involved scientists are excited about what they are doing; the technology alone and the speed with which an enormous number of samples can be screened are mind-boggling, and no doubt highly stimulating to manipulate and manage. But I am reminded of the much greater hype that surrounded the mapping of the human genome, and the cautionary note sounded over 10 years ago in connection with the limits of what such mapping could tell us, a caution that was disregarded for many years by a large number of experts.13 The majority of involved GWAS researchers are well aware that, if the AD world is to evolve toward prevention as its primary approach to management of this condition, then it is essential to move beyond association studies in order to understand how these genes and their products function throughout the life course of individuals—in other words, to document in detail the pathways that are involved through time. This is a daunting task, one that GWAS cannot begin to address; on the contrary, the findings make it abundantly clear that decontextualized information about genes is merely a first, hesitant step toward confronting the complexity associated with AD genetics—a situation fully appreciated by the GWAS scientists to whom I talked. Gina Kolata published a predictably upbeat article about GWAS findings in the New York Times,14 but the comments she cites by knowledgeable scientists are carefully measured: “This is a big, solid step,” states Dr. Nelson Freimer, who directs the Center for Neurobehavioral Genetics at the University of California, Los Angeles.
Questions Triggered by GWAS
I traveled to Cardiff, Wales, to talk with Julie Williams and two of her colleagues at the Center for Neuropsychiatric Genetics and Genomics in November 2009, shortly after the publication of the findings of the first two GWA studies. Williams made it clear at once that she does not believe that “we’re anywhere near the stage of being able to predict from the genes we’ve found if someone is going to get Alzheimer’s or not. What we have are at least four susceptibility genes in addition to APOE [details about one were unpublished when we talked], and each one possibly accounts for around 2% of actual risk.” When I asked if one can assume that these effects are additive, Williams agreed that this is probably the case, but even so such knowledge does not improve individual risk prediction. She went on to stress that simply asking whether or not these genes have an effect on the clearance of β-amyloid is not adequate:
[A] number of things contribute—genetic risk factors and probably environmental risk factors, and once you are over a certain burden you cascade into the disease … but there are going to be several processes involved, and some of the genes will code for molecules that have a number of roles, I suspect, so it’s not going to be straightforward. But what finding these genes will do is that it will help to focus research in the right place—we’ll be dealing with the right causal pathways.
Williams is particularly excited by findings that strongly suggest that the immune system and inflammation are implicated in AD causation. She reminded me that clusterin has a role in dampening down inflammation in the brain and went on to emphasize that usually inflammation, long noted as associated with risk for AD, has been viewed as a secondary effect of the disease, whereas amyloid as primary. She states that the GWAS results suggest that researchers may have been thinking about things “the wrong way round,” and that inflammation may precede amyloid deposition. Williams adds,
I think the real motivator for us, rather than just identifying a list of genes that can be used to predict individual risk or possibly develop drugs, is to understand the complexity of the pathways involved so as we can understand causation much better. My view is that plaques and tangles may be correlational but I think there are new mechanisms that we are now beginning to see that we haven’t been measuring well. What we haven’t looked at much is absolute cell loss, but this actually correlates better with cognitive changes and this fits with the complement story we are now beginning to unravel—recent work suggests that complement proteins have functions we hadn’t paid attention to, like synaptic pruning, and this is something I’m excited about. GWAS really makes it clear that we have to think about much more than the production of APP [amyloid precursor protein]—we are beginning to see new patterns, and that’s exciting. And because so many “normals” with plaques don’t have AD, it seems clear that plaques are a risk factor but not the whole story. But at the moment I think some neuropathologists feel threatened by this.
At one point I asked if any attempt was being made in GWA studies to control for ethnicity in the sampling procedures. The reply from Williams’s colleague sitting in on the discussion was that in effect the research team had deliberately controlled their sample to be as Caucasian as possible. This researcher added that the team wants to investigate different populations based on ethnicity, but that it is virtually impossible to obtain the large number of cases that are needed to carry out such a study. Williams added, “[R]ace and ethnicity are confounding factors—they are a genetic conundrum, so we have to try and simplify the question for the time being.” In effect, the demands of the technology and the urgent need for continued funding constrain the research methods.
In May 2010 I met with Gerard Schellenberg, who heads up the major Alzheimer GWAS project taking place in the United States. I had previously met Schellenberg in Montréal right after he had agreed to become the director of this project, at which time he had said that he was having trouble sleeping at nights, wondering what he had let himself in for. Sitting in his office at the University of Pennsylvania School of Medicine at 8:30 in the morning a year later, he seemed more relaxed, but even so emphasized the incredible pressures being experienced by his team, several of whom regularly sleep in the lab. Schellenberg has had years of experience working on the genetics of AD and, in 1995, along with Rudolph Tanzi and other colleagues, identified presenilin-2, which is causal of familial AD present in the families of Volga Germans.
Like virtually every other researcher in molecular genetics I have talked to, Schellenberg stressed that carrying out genetic tests on individuals to predict their risk for late-onset AD makes no sense at all. We talked at some length about sampling for GWAS, and he emphasized that despite a great deal of care with sample collection “there will never be a perfect AD diagnosis and so samples cannot be perfect.” Schellenberg added that difficulties exist with “harmonizing” the huge samples that his team works with because they have been procured from different “genotyping platforms.”
When asked about ethnic representation, Schellenberg replied that about 1,000 African Americans are in their GWA sample, and lower numbers of Hispanics and Asians. He added, somewhat ruefully, that in order to say anything significant about a minority population, one would need at least 10,000 cases.
Schellenberg is skeptical of claims beginning to be made about AD pathways based directly on GWAS findings:
Pathway analysis is in its infancy. Researchers claim that here’s a pathway that shows up statistically significant, but this pathway has not been assembled or vetted by an expert in the field. And then you look at this pathway and you say this is … not real. If you are somewhat familiar with the pathway you know that it just doesn’t make sense what is being claimed directly, just from GWAS findings.
You know, the experience with diabetes was when they got up to 30,000 cases [using GWAS] they’d got about 30 genes, or something like that. It’s not clear when there would be an end to finding more genes of small effects, so working out pathways is quite a problem. They’re probably additive; so the more of them you know, the more risk you’ll be able to predict, but it won’t be useful to give this kind of information to patients.
When I asked Schellenberg whether he thinks it is possible to make a clear distinction between normal and pathological aging, he responded immediately by saying that such an argument seemed to make sense in the 1960s but today things are not so clear:
What’s interesting is that when we started to understand something about APP and presenilin-1 and -2, you could argue that the onset is earlier for this kind of AD, and the course is a little bit faster (although not always). But, if you hand a neuropathologist two brains, they cannot distinguish between a brain from an early-onset family and a late-onset brain. And when my group showed, years ago, that APOE interacts with presenilin-1 to affect the age of onset, and APOE also acts in the same way with late onset populations, then this all strongly suggested that we are looking at one single disease. But recently there’s been a lot of research with tau and ubiquitin and other molecules that suggest we should be able to separate this condition into two or more diseases, based on pathways—but we haven’t got there yet, and in the end it may turn out that they are just variations that influence the eventual expression of Alzheimer’s.
Schellenberg concluded our hour-long discussion by stating firmly that he is “highly skeptical” that we will ever be able to set up trials for preventive medications: “You’d have to give somebody who doesn’t have Alzheimer’s, or even any symptoms, trial drugs for ten years at least, and then at the next stage you’d have to give these drugs to 100,000 people for ten years, and you’re going to have to follow them all this time—I’m not optimistic about this!”15 I met with Schellenberg before the trial using Basque Colombians as subjects was being discussed in public. From what Schellenberg told me in 2010, it appears that he would be ill at ease with this trial in terms of both sample size (too small) and duration (too short).
Born in Kenya, Peter St George-Hyslop, like Gerard Schellenberg, Rudolph Tanzi, and other neurogeneticists I talked to, is someone who has been deeply involved in Alzheimer research for many years, and whose team identified several of the key genes and a key protein implicated in nerve cell degeneration. Hyslop’s primary interest for many years has been in dominantly inherited Alzheimer disease. I first talked to him at the University of Toronto where he has worked since 1991, becoming director of the Center for Research in Neurodegenerative Diseases in 1995. Then in 2009, I talked with him again, in Cambridge, where he also holds a position as the Wellcome Trust Principal Research Fellow at the Cambridge Institute for Medical Research. At our second meeting, immediately following the initial publication in Nature Genetics about the GWAS findings of three new AD genes in addition to APOE, Hyslop expressed considerable concern about GWAS as a method:
A major problem with GWAS is that when you get a statistical result and then someone else gets the same result, particularly when they use an overlapping population, you always have to be very cautious about what this really means … does it mean that there is a genetic signal there, or is it a statistical explanation that is not due to genetics? Second, I think you have to be very cautious about the magnitude of the results. People [researchers] are very attracted to the p value and less attracted to the biological impact the genes may have.16 So, you know, if you test 1000 people and you get a p value of 0.003, for example, and then you test 10,000 people and you get a higher p value, you believe the higher p value, but the biological information is exactly the same. The real information is in the odds ratio,17 and the odds ratios they are getting with GWAS are not high. So what are you getting out of these GWA studies? What exactly have you got? Not a lot. Especially because at least two of the genes being reported are ones that we already know about—there are already literally hundreds of studies out about complement and APOEJ.
Hyslop then went on to stress that no doubt many different pathways are likely to be involved, but added that he believes the most “robust” way to proceed at this point is to work out several of the pathways in detail, prior to making claims about risk associated with specific genes. One cannot infer pathways from genes alone, he insisted, and although his name is attached to one of the GWAS projects, his primary interest is now in proteomics, with the hope of obtaining a deeper understanding of involved pathways. Hyslop is a strong supporter of the need to better recognize gene/environment diversity among populations, and in this respect he also finds GWA studies to be wanting. But proteomic findings will not radically change the dominant, reductionistic direction of AD research, although it moves to the cellular level, rather than focusing exclusively on genes.
Two years after he talked with me, Hyslop did an interview with CBC radio in April 2011, in which he discussed the contribution made to current understanding of AD genetics by the Canadian team he leads that contributes to the NIA-funded Alzheimer’s Disease Genetics Consortium (ADGC) that conducts GWAS research. Hyslop had the following to say about the 5 genes recently added to the list of genes believed to increase risk for AD, making 10 in all:
This represents new information about the pathway that causes Alzheimer’s. … Several of the genes were genes we didn’t know about and they are going to quite richly tell us more about the disease. I think they are going to be very valuable in the next few years and might even lead us to diagnostic or treatment markers with potential to slow down the disease. … Now we will need to go back and look at a whole new range of possibilities. … Things that we thought were end stage of life events will have to be looked at again. We have to do a full scale re-think.18
It seems that Peter Hyslop was taken by surprise with the latest round of GWAS findings, and that he now believes it is quite possible that this technology will result in significant genetic findings that he had not previously anticipated.
Raising the GWAS Stakes
In November 2009 Rudolph Tanzi, the Harvard neuroscientist, participated in an event at Capitol Hill together with Francis Collins, the director of the National Institutes of Health. These scientists, both billed on the Internet as rock stars of science, joined up with Aerosmith’s Joe Perry and together they played a rendition of “The Times They Are a-Changin’ ” in which the second verse was liberally adapted to direct the attention of Congress to the urgent need for more funding for Alzheimer disease. Joe Perry performed “unplugged,” Collins was the lead guitar, and Tanzi played the harmonica, an ensemble strikingly portrayed in a poster to commemorate the event. A line appears on the video informing viewers that the sponsors of the event were the pharmaceutical companies Wyeth and Elan. In an accompanying interview conducted on the steps of Capitol Hill, Tanzi, who is an excellent guitar player, informs everyone that he has been an Aerosmith fan since he was 16 years old and adds that Joe Perry and he are in agreement that what is needed in both science and music is “novel thinking; thinking outside the black box.” Tanzi then puts in a serious plug for Alzheimer funding (which is why he takes time out from his lab to participate in a performance such as this) noting that at present only $400 million a year is spent on AD—a drop in the bucket, as he describes it.19 A 2012 request sent to the U.S. government is for funding to a level of $2 billion per year.20
I met Rudy Tanzi in his Harvard office attached to his laboratory facilities in early 2010. Tanzi, grinning, started out by stating,
[W]e are sadomasochists here, we don’t believe in using core facilities and outsourcing, we do everything ourselves because then you can really trust it. Our GWAS is family-based and we used four different samples of over 1,300 families.
GWAS is telling us we have—if you take it at face value—just looking at the SNPs, many AD candidate genes that are common variants exerting tiny effects on risk. Is that what AD is? Well, it fits very nicely with a model I helped propagate in ’99: common variant-common disease, rare mutation-rare disease. But that’s in question now. Here’s the thing: penetrance and prevalence affect what determines the effect of a gene—right? Penetrance means the chance of getting the disease when you inherit a mutation or a variant. Prevalence is the percentage of the population or patients carrying it [the mutation or variant]. And for early onset AD it looks like this: heavy weighting on penetrance and low on prevalence—rare mutations, full penetrance, in effect. APOE looks like heavy prevalence with moderate penetrance, and with GWAS what we are getting is huge prevalence, tiny penetrance—if you take them at face value.
When I asked if the GWAS genes found thus far work in consort, Tanzi said,
They probably all work in consort with each other, but it’s difficult to do gene, to gene, to gene, to gene, to gene, to gene interaction studies. You get Bonferronied [a technique that makes use of multiple testing to avoid false positives] to death, as I say, we can’t get a reliable signal to noise ratio with all these interactions. You can imagine, we might have a risk factor in the same biological pathway as a protective factor sometimes, and at other times other risk and protective factors are in different pathways. So, depending on pathways they may be simply additive, or synergistic, or antagonistic.
So dozens of novel AD candidate genes contain common variants of SNPs that appear to exert tiny effects on the risk for AD. But are they actually pathogenic? Or, and this is the most important thing I want to say to you, do these common variants tag ancestral haplotypes [a set of alleles at closely linked loci on a chromosome that tend to be inherited together over generations] that harbor and cluster rare mutations with large effects on AD risk in a small portion of the population?
At this juncture, Tanzi referred to the recent work of several molecular geneticists who have produced widely cited articles in which attempts have been made to account for what is known as the “missing” heritability of complex diseases.21 One such article published in Nature in 2009 sets out with this statement:
Genome-wide association studies have identified hundreds of genetic variants associated with complex human diseases and traits, and have provided valuable insights into their genetic architecture … most variants identified so far confer relatively small increments in risk, and explain only a small proportion of familial clustering, leading many to question how the remaining, “missing” heritability can be explained.22
These authors then make an interesting observation, namely that the vast majority of disease-associated variants (greater than 80%) fall outside coding regions on the genome. They then list the usual suggestions made to account for “missing” heritability, including the following: that there are many more variants of small effect to be found, and this demands further GWA studies with larger and larger sample sizes. Second, there may be variants that are very rare, possibly with larger effects that are unlikely to be detected using the present GWAS approach, because it focuses only on those variants that occur in at least 5% or more of the population under study. Third, detecting gene-gene interactions is not well “powered” in GWAS, and, last, there is inadequate accounting of shared environment among the families that are studied. It is noted, “Consensus is lacking … on approaches and priorities for research to examine what has been termed ‘dark matter’ of genome-wide association—dark matter in the sense that one is sure it exists, can detect its influence, but simply cannot ‘see’ it yet.”
Among the proposed reasons for “missing heritability,” the one that most attracts Rudolph Tanzi and certain other AD geneticists is the theory about rare variants with large effect size. The article in Nature elaborates:
A probable contributor to small genetic effect sizes observed so far [by GWAS] is that current investigations have incompletely surveyed the potential causal variants within each gene. … Notably, 11 out of 30 genes implicated as carrying common variants associated with lipid levels also carry known rare alleles of large effect … suggesting that genes containing common variants with modest effects on complex traits may also contain rare variants with larger effects.
An important consideration is that the overwhelming majority of GWAS and other genetic studies have been limited to European ancestry populations, whereas genetic variation is greatest in populations of recent African ancestry and studies in non-Europeans have yielded intriguing new variants.23
These authors add that even when numerous rare variants can be detected in a gene or region, they may have disparate effects on the phenotype, thus greatly complicating matters. Nevertheless, their conclusion is that GWAS continues to be an efficient means of examining unexplained heritability, and that sampling should be extended to non-European populations.
A 2010 article published by the group headed up by David Goldstein at Duke University extends the above discussion along similar lines. The concept of “synthetic association” is introduced in which it is suggested that many of the signals being detected in GWA studies could very well result from rare variants with large effect sizes capable of acting over large genomic distances. These variants, because they are rare, are not themselves detected by GWAS as yet but form crucial associations with the much more common variants being detected in GWAS. Furthermore, if this is indeed the situation, examination of single haplotypes is unlikely to be sufficient to account for the associations being observed; years of further empirical work is called for.24 Related research introduces a concept of “phantom inheritability” in which it is argued that insufficient attention has been paid to “epistasis”—the interaction of genes one with another—leading to overestimations of heritability.25
The recent review on the genetics of Alzheimer disease published by Lars Bertram, Christina Lill, and Rudolph Tanzi in Neuron in 2010 points out that virtually all of the recent AD GWA studies have detected loci that are linked to Aβ metabolism in one or more ways, notably the aggregation or clearance of Aβ in the brain. They argue that with improved high-sequencing technologies “missing heritability” will increasingly be assessed in “unprecedented” detail, and that rare sequence variants that are no doubt implicated will come to light.26 They imply in this article that the amyloid cascade hypothesis stands firm as a result of genetic research, but, as Tanzi emphasized when I talked with him, he believes that researchers must now move upstream:
[W]hat people are saying about AD is that lots of SNPs with tiny little effects all work together somehow to finally give you the disease—they hit a threshold. No! What is really happening is that all these little effects of these SNPs is just an echo, the wake of a large boat that they are following. GWAS is the very first step in the journey, a necessary first step. GWAS tells you which haplotypes to look at to find rare mutations in clusters. We have rare mutations associated with early onset that we know about and we have late onset mutations also with strong effects, but they don’t guarantee the disease until 100 or 120 years old. And most people die before that! So, it only looks like partial penetrance because people don’t live long enough.
Well, guess what, when people were dying around 40 years old APP and presenilin mutations [associated with dominantly inherited Alzheimer disease] looked as though they had partial penetrance too.
It is of note that Evelyn Fox Keller, in her book The Mirage of a Space between Nature and Nurture, argues that because interactions of genes and environment are inseparably entangled throughout the lives of individuals, their relative contribution to phenotypes cannot be assessed independently of each other, thus making the very concept of “missing heritability” questionable, see chapter 9.
Tanzi and his colleagues have recently been doing research on a gene called ADAM10 using samples taken from their “enriched” GWA study composed of families they have been following for years. Out of 1,004 families screened, they found 7 families that all appeared to be at an increased risk for AD on the basis of common SNPs clustered in each of these families. The GWA research with much larger cohorts had shown that these particular SNPs increase risk for AD by a small amount. Tanzi’s group found that these 7 families all had very rare mutations of the well-conserved gene ADAM10. These mutations were present in 70% of the affected family members and were found in only one of about 1,000 subjects unaffected by AD. What was striking about this finding was that these mutations apparently did not bring about dementia until the age of 70 or thereabouts. “We got lucky with this one,” said Tanzi, “imagine how we are going to document the ones that don’t strike until 80 or 90 years old.” He added, “[T]he odds ratio we’re finding for ADAM10 for increased risk of AD is actually a little stronger than APOE. But it’s rare, and its effect is huge, and it is associated with a whole lot of common variants with tiny effects. Here you have an example of what Goldstein’s group calls ‘synthetic associations.’ I was really happy to see that Goldstein who was trashing GWAS has turned right around and now he’s telling us how we can do GWAS.”
Tanzi believes that similar findings will be made once researchers start to “dig deeper” in connection with GWAS research. He thinks, for example, that this kind of approach will show how rare variants are associated with the innate immune system, inflammation, and in turn AD. However, when I checked in July 2012 on the AlzGene list of genes, ADAM10 was ranked as only 39. The fortunes of genes come and go on this ranking list; two others, TOMM40 and SORL1, have both been reported in recent years as very promising candidate genes, but they are currently ranked at 22 and 24, respectively, and neither came up in the large GWA studies. Just about every GWA investigation reveals new candidate genes, simultaneously heightening excitement and uncertainties, but these findings are rarely repeated by other studies, and how to interpret them remains unanswered. Clearly some deep digging is in order, or even “some other kind of digging altogether”!27
It seems likely that this recent insight into the possible contribution made by rare variants to AD causation may account for Peter Hyslop’s apparent change of heart when he commented in an interview that recent GWAS findings indicate that a “rethink” is called for. It appears that at times rare variants can bring about complex effects as a result of “synthetic associations” in their genetic milieu and also that common variants with small, gradually accumulating effects are indeed implicated as part of this extraordinary story. The biologist Ken Weiss insists that greater understanding will come about only once it is better recognized by researchers that “a mix of allelic variation in all likelihood contributes to various levels of risk that differ according to context. No one hypothesis about risk will be appropriate for all situations.” Weiss insists that scientists have to learn to deal with the fact of the mix and the epistemological challenge it presents (personal communication).
In February 2012 an article appeared with findings about the genetics of 439 individuals selected from families where at least four relatives had been diagnosed with late-onset Alzheimer disease (see chapter 5). The study showed for the first time how rare variants and pathogenic mutations of the APP, PSEN1, and PSEN2—the genes associated with dominantly inherited Alzheimer’s—are at times implicated in cases of late-onset AD. The researchers conclude, “Clearly factors other than the mutation can impact the age of onset and penetrance of at least some variants causing AD.” And, furthermore, “familial aggregation is more important than age at onset in determining the likelihood of an individual carrying a disease-causing variant.”28 It is noted by the authors of this article, as other researchers in the above discussion have noted, that GWA studies are thus far able to identify only common variants associated with disease, and that rare variants go undetected in such studies. In this recent study it is concluded, in common with Tanzi, that rare variants could indeed explain a significant proportion of genetic heritability of AD. But the epigraph at the head of this chapter makes clear that the contribution of variants, common or rare, to phenotypic traits cannot be assessed decontextualized from the environments, ranging from cellular to social, in which they function.
Clearly much remains to be uncovered in connection with AD genetics, and it is striking that research has barely begun to touch on gene/gene interactions, life course exposures to toxins and other types of environmental effects, and the social milieus in which people are raised, all of which can modify the expression of genes and influence risk for AD.
Thanks largely to technological transformations that make it increasingly possible to manage ever larger data sets at a lower cost than previously, and at an increasingly rapid rate, many geneticists, “hungry” for more data, are strongly advocating whole genome sequencing. I have been told informally that several researchers, changing their former positions, now firmly embrace GWAS, and that this “crossing of the floor” is almost certainly in part due to political expediency as far as funding goes, in addition to the lure of manipulating such huge data sets. No doubt the image of scientists united in a worldwide humanitarian effort to defeat “the robber of the very thing that makes human beings unique,”29 also captivates a good number of researchers.
The senior researchers cited above have known each other for more than 20 years and meet regularly at conferences and workshops. My impression is that, on the whole, their relationships are of reasonably friendly competition. John Hardy told me in 2010 that he and Peter Hyslop have breakfast about once a month; this is time well spent, no doubt, when future directions are taking shape for research involving massive amounts of money and hundreds of labs worldwide. But it is important to mark what is so often forgotten in the world of AD research. In a 2011 issue of the journal Alzheimer’s & Dementia a long detailed article prepared by the Alzheimer’s Association sets out the “facts and figures” of Alzheimer disease. The closing sentences run as follows:
Although available data do not permit definitive estimates of how many individuals have undiagnosed dementia, the convergence of evidence from numerous sources indicates that as many as half of [the] people satisfying diagnostic criteria for dementia have never received a diagnosis. Some lines of evidence suggest that as many as 80% or more of affected individuals are never diagnosed.30
This article is focused entirely on the American situation, and a large portion of it falls under the heading of “Caregiving,” in which the plight of the nearly 15 million Americans who provide unpaid care for demented relatives is discussed at length. Such care amounted to 17 million hours of work in 2010, a contribution to the nation of approximately $202 billion. The political and economic pressures to defeat AD in the United States, and virtually everywhere else these days, are enormous, but because a cure, or better still prevention, is always just around the corner, or so it seems, the incredible burden that falls on caregivers, especially those who are not well-off, continually drops out of sight. In 2011 the National Alzheimer’s Project Act was signed into law by the Obama administration, and in 2012 the war on Alzheimer’s was upgraded when a plan was announced that targets 2025 as the goal for preventing or treating Alzheimer’s disease.
The following chapter is devoted largely to a consideration of the responses of healthy individuals who come from families where AD has been diagnosed and who volunteered to have their APOE genotype tested as part of a randomized controlled trial. What was striking while interviewing these people was that very many of them, above all else, wanted to tell my research team about the demands that they have to shoulder daily in connection with caregiving in their families. Pondering the implications of the results of their own genotyping, as we asked them to do, was very often relegated to a distant second place in our discussions with these individuals, many of whom were exhausted when we met them.