Schizophrenia is one of those important human traits that has eluded understanding despite heavy research investment. It is elusively variable and hence challenging to diagnose as a single entity or to decide how to split it up into causally distinct subsets. It seems highly familial in terms of its increased risk among family members, and hence seems clearly to have a genetic component. But the specific genes have been elusive--they must be there in the genome, but where are they?
A recent paper in Nature ("Common polygenic variation contributes to risk of schizophrenia and bipolar disorder", The International Schizophrenia Consortium, published online 1 July 2009) looked at large amounts of data on schizophrenia from several study populations. The authors did an extensive amount of genotyping and then various kinds of analysis (they looked, for example, at about a million variable sites (SNPs) in the genome, to identify regions where a particular variant marker was found more often in some 3322 cases than 3587 controls--pretty large studies for this kind of trait.
No really strong signal, that is that explained a high fraction of the disorder, was found. But through a series of analytic approaches, including computer simulations to test a range of possible genetic causal models to see which fit best, the authors (and this is one of those papers with a huge list of authors) concluded that many thousands of genes (classically they'd be known as 'polygenes') contribute to the trait. Most of the contributing variants are rare, but more importantly, they have individually very small effects.
Regardless of the details of the study, which could include all sorts of artifacts or be affected by the methods and assumptions of the authors, the study seems convincing that schizophrenia is like many other traits of a polygenic nature. The authors confirmed current ideas that bipolar disorder may involve many of the same genes, as well.
There are good evolutionary and biological reasons why this makes sense. In a nutshell, it's because so many processes are involved in brain development and function, each of them subject to mutational variation, that there are many ways to end up with the same trait. Natural selection only prunes those who can't reproduce as successfully, but the effect is distributed across these many parts of the genome, and hence acts only very weakly against any one of them. The result is an accumulation of variation that, at each individual region is essentially undetectably abnormal. The frequency of the individual variants changes over generations (and over geographic space in our species) mainly by chance (genetic drift).
The individual components have to work together--the 'cooperation' that is at the core of life as we outline in our book The Mermaid's Tale, but there is plenty of tolerance for variation, what we refer to as functional 'slippage'. It all makes sense biologically, evolutionarily, and causally.
In addition to its consistency with evolutionary expectations, this flies in the face of current predominant thinking about the prospects for what is being called 'personalized medicine', that is, medicine based on each individual person's genotype. If genotypes are poorly predictive, as in this case they seem to be, then they are of no real use to a clinician. In fact, as with so many similar studies, the total identified effect was small: based on various assumptions, the polygenic component identified by this geomewide search accounted for only 3 to 20% of the total disease risk, which itself is only 1%! Schizophrenia is an important problem (1% of the population is a lot of people), but clearly the predictive power of these gene-sets is modest, and this assumes that environmental effects will retain their current overall nature and impact (many of the genes probably have effects that vary depending on environment).
Many researchers will try to develop synthesizing methods to make individual sense of polygenotypes, so that treatment might be varied accordingly. How well they succeed only time will tell. But this is another case in which extensive study of a trait based on modern high-intensity technology has documented the nature of complex traits.
Showing posts with label association studies. Show all posts
Showing posts with label association studies. Show all posts
Tuesday, July 14, 2009
Monday, June 15, 2009
The simple facts of life
Here we report on some reflections after our participation in a meeting on the 'New Genomics in Medicine and Public Health' held at the University of Bristol, UK. The talks were varied and interesting, including a talk about Mendel, reports of successful and unsuccessful genomewide association studies, plaudits for the UK Biobank, and discussion of clinical applications of genomic findings.
An important question these days, related to various methods in genetics and its role in medicine and public health, is how causally complex life really is--a question at the heart of most of the work reported at the meeting. Some normal traits as well as diseases clearly are genetic, in that their variation is clearly caused by variation in a single gene (or a small number of genes, in a way that's well understood). But others are less clear cut, as we've discussed here a number of times.
Vested interests of all sorts, including venal and careerist interests, but also strongly held scientific conviction affect this area these days. One way to put the question is: "How causally complex is life?" Here the interest is mainly in genes, environment getting some but usually rather casual or minimal attention, and the question boils down to how well phenotypes can be predicted from known or knowable genotypes. Sometimes this means using individual variation to predict individual disease risk--this is the major original purpose of GWAS (genome-wide association studies). Sometimes it means using natural variation as a tactic to identify genetic pathways that are responsible for some normal trait; the idea here is either that, when mutant, the pathway (or 'systems' or 'network') genes could lead to disease and/or that these genes, when known, can be used as general preventive or therapeutic targets.
A commonly invoked motivation for human genetics work these days is that we will be able to implement 'personalized medicine', to predict disease or treatment, or to suggest preventive measures, based on each person's genotype. Many companies are promoting this, and the molecular genetics community is hyping it very heavily (here, there is no doubt of strong material vested interests, even if some actually believe it will work as advertized).
There are hundreds of diseases for which a, or often the causative gene is known. Sickle cell anemia, Huntington disease, Phenylketonuria (PKU), Cystic fibrosis (CF), and Muscular dystrophy (MD) are just a few examples. For these, predictive power already exists, though clinical application is not necessarily based on genotype. There are other examples where the latter is true, but these are generally rare in the population. Promising gene-based molecular therapy is in the works for CF, MD, and maybe even for some forms of inherited breast cancer (due to BRCA1/2 mutations). For these diseases, causation is clear even if there are substantial variation in risk, age of onset, or severity. Causation here is usually thought of as simple.
But for most common and/or chronic diseases, the story is far from clear as we've mentioned in various earlier posts (and as is widely discussed in the literature). These traits usually have substantial heritability (i.e., familial risk--if a close family member is affected, your chance of getting the same disease is greater than that of a random member of the population to which you belong). That means that, unless we are somehow badly understanding things, genetic variation plays a major role in risk (at least in current environments). Yet after many sophisticated, large studies, identified genes account for only a small fraction of the familial risk. The data suggest that many genes, say 'countless' genes, contribute substantial risk in aggregate, but individually their contribution is so small as to be unidentifiable by feasible (or cost-justifiable) studies. That would suggest that the disorder is caused by numerous combinations of huge numbers of individually weak, and rare, genetic variants. This is known classically as 'polygenic' inheritance, and if it's what's going on, things are very complex indeed.
Others, focused on the many clearly 'Mendelian' (single-gene) traits, simply don't believe life is that complicated. They suggest at least two other possibilities. One is that only a modest number of genes contribute, but most of the culpable alleles (sequence variants) are so rare and weak that genomewide association studies cannot pick them up. At such genes, there may be one or two strong, common alleles and these have high penetrance (when present, the disease usually occurs) and so they can be identified in family or GWAS. Those variants only account for a small amount of overall genetic contributions. But once the gene is known, we can sequence it in many patients and, lo and behold!, we find many other alleles that, some argue, contribute the rest of the observed family risk.
There is some truth to this: we have done simulations to show that there can be high heritability but only a few contributing genes, for just such reason (heterogeneity of the frequency and effects of existing alleles).
Another possibility is that a modest number of genes have variants with rare, but not very rare frequency. These will be identified by the panoply of existing methods, and once that's done it will be possible to genotype everyone at these genes, identify each person's individual set of variants, and determine risk. These are called 'oligogenic' effects, because the number of genes involved is small rather than huge. This view acknowledges the current problem, but assumes it will go away with enough data--and, importantly, that business as usual is a right approach.
Presentations at the Bristol meeting, including Ken's, show clearly that causation is a spectrum of aggregate vary rare genetic effects, a larger but still small fraction of oligogenic effects, major gene effects, and polygenic effects.
The question is: what do we do if this is true? Where is the practical limit below which attempts to identify all the genes are futile or not worth the investment, and is it likely that current attempts will at least identify the bulk of genetic effects and the networks involved so that the disease can be eliminated in whole or at least major part?
There is no single consensus in this area. Some are more skeptical than others. Some argue that knowing the genetic contributions to disease may make diagnosis more specific (the doctor can test for which gene is contributing to a given case), even if genotype-based prediction will remain a dream in the eyes of venture capitalists. Other computophiles believe that if enough computers are used on enough DNA sequence, the problem will, like infectious diseases, be solved. We won't be sick any more.
Time will tell where in the causal spectrum most traits lie. One thing we can be sure of, though: in this contentious area in which huge career, institutional, and commercial investments are at stake, in years to come, retrospective evaluation will always claim victory! Few will look back and say that we knew better than to make the level of investment in genetic causation that we are currently making.
An important question these days, related to various methods in genetics and its role in medicine and public health, is how causally complex life really is--a question at the heart of most of the work reported at the meeting. Some normal traits as well as diseases clearly are genetic, in that their variation is clearly caused by variation in a single gene (or a small number of genes, in a way that's well understood). But others are less clear cut, as we've discussed here a number of times.
Vested interests of all sorts, including venal and careerist interests, but also strongly held scientific conviction affect this area these days. One way to put the question is: "How causally complex is life?" Here the interest is mainly in genes, environment getting some but usually rather casual or minimal attention, and the question boils down to how well phenotypes can be predicted from known or knowable genotypes. Sometimes this means using individual variation to predict individual disease risk--this is the major original purpose of GWAS (genome-wide association studies). Sometimes it means using natural variation as a tactic to identify genetic pathways that are responsible for some normal trait; the idea here is either that, when mutant, the pathway (or 'systems' or 'network') genes could lead to disease and/or that these genes, when known, can be used as general preventive or therapeutic targets.
A commonly invoked motivation for human genetics work these days is that we will be able to implement 'personalized medicine', to predict disease or treatment, or to suggest preventive measures, based on each person's genotype. Many companies are promoting this, and the molecular genetics community is hyping it very heavily (here, there is no doubt of strong material vested interests, even if some actually believe it will work as advertized).
There are hundreds of diseases for which a, or often the causative gene is known. Sickle cell anemia, Huntington disease, Phenylketonuria (PKU), Cystic fibrosis (CF), and Muscular dystrophy (MD) are just a few examples. For these, predictive power already exists, though clinical application is not necessarily based on genotype. There are other examples where the latter is true, but these are generally rare in the population. Promising gene-based molecular therapy is in the works for CF, MD, and maybe even for some forms of inherited breast cancer (due to BRCA1/2 mutations). For these diseases, causation is clear even if there are substantial variation in risk, age of onset, or severity. Causation here is usually thought of as simple.
But for most common and/or chronic diseases, the story is far from clear as we've mentioned in various earlier posts (and as is widely discussed in the literature). These traits usually have substantial heritability (i.e., familial risk--if a close family member is affected, your chance of getting the same disease is greater than that of a random member of the population to which you belong). That means that, unless we are somehow badly understanding things, genetic variation plays a major role in risk (at least in current environments). Yet after many sophisticated, large studies, identified genes account for only a small fraction of the familial risk. The data suggest that many genes, say 'countless' genes, contribute substantial risk in aggregate, but individually their contribution is so small as to be unidentifiable by feasible (or cost-justifiable) studies. That would suggest that the disorder is caused by numerous combinations of huge numbers of individually weak, and rare, genetic variants. This is known classically as 'polygenic' inheritance, and if it's what's going on, things are very complex indeed.
Others, focused on the many clearly 'Mendelian' (single-gene) traits, simply don't believe life is that complicated. They suggest at least two other possibilities. One is that only a modest number of genes contribute, but most of the culpable alleles (sequence variants) are so rare and weak that genomewide association studies cannot pick them up. At such genes, there may be one or two strong, common alleles and these have high penetrance (when present, the disease usually occurs) and so they can be identified in family or GWAS. Those variants only account for a small amount of overall genetic contributions. But once the gene is known, we can sequence it in many patients and, lo and behold!, we find many other alleles that, some argue, contribute the rest of the observed family risk.
There is some truth to this: we have done simulations to show that there can be high heritability but only a few contributing genes, for just such reason (heterogeneity of the frequency and effects of existing alleles).
Another possibility is that a modest number of genes have variants with rare, but not very rare frequency. These will be identified by the panoply of existing methods, and once that's done it will be possible to genotype everyone at these genes, identify each person's individual set of variants, and determine risk. These are called 'oligogenic' effects, because the number of genes involved is small rather than huge. This view acknowledges the current problem, but assumes it will go away with enough data--and, importantly, that business as usual is a right approach.
Presentations at the Bristol meeting, including Ken's, show clearly that causation is a spectrum of aggregate vary rare genetic effects, a larger but still small fraction of oligogenic effects, major gene effects, and polygenic effects.
The question is: what do we do if this is true? Where is the practical limit below which attempts to identify all the genes are futile or not worth the investment, and is it likely that current attempts will at least identify the bulk of genetic effects and the networks involved so that the disease can be eliminated in whole or at least major part?
There is no single consensus in this area. Some are more skeptical than others. Some argue that knowing the genetic contributions to disease may make diagnosis more specific (the doctor can test for which gene is contributing to a given case), even if genotype-based prediction will remain a dream in the eyes of venture capitalists. Other computophiles believe that if enough computers are used on enough DNA sequence, the problem will, like infectious diseases, be solved. We won't be sick any more.
Time will tell where in the causal spectrum most traits lie. One thing we can be sure of, though: in this contentious area in which huge career, institutional, and commercial investments are at stake, in years to come, retrospective evaluation will always claim victory! Few will look back and say that we knew better than to make the level of investment in genetic causation that we are currently making.
Subscribe to:
Posts (Atom)