Personal genomes: The case of the missing heritability

When scientists opened up the human genome, they expected to find the genetic components of common traits and diseases. But they were nowhere to be seen. Brendan Maher shines a light on six places where the missing loot could be stashed away.
Brendan Maher
If you want to predict how tall your children might one day be, a good bet would be to look in the mirror, and at your mate. Studies going back almost a century have estimated that height is 80–90% heritable. So if 29 centimetres separate the tallest 5% of a population from the shortest, then genetics would account for as many as 27 of them.
This year, three groups of researchers scoured the genomes of huge populations (the largest study looked at more than 30,000 people) for genetic variants associated with the height differences. More than 40 turned up.
But there was a problem: the variants had tiny effects. Altogether, they accounted for little more than 5% of height's heritability — just 6 centimetres by the calculations above. Even though these genome-wide association studies (GWAS) turned up dozens of variants, they did "very little of the prediction that you would do just by asking people how tall their parents are", says Joel Hirschhorn at the Broad Institute in Cambridge, Massachusetts, who led one of the studies.

Height isn't the only trait in which genes have gone missing, nor is it the most important. Studies looking at similarities between identical and fraternal twins estimate heritability at more than 90% for autism5 and more than 80% for schizophrenia. And genetics makes a major contribution to disorders such as obesity, diabetes and heart disease. GWAS, one of the most celebrated techniques of the past five years, promised to deliver many of the genes involved. And to some extent they have, identifying more than 400 genetic variants that contribute to a variety of traits and common diseases. But even when dozens of genes have been linked to a trait, both the individual and cumulative effects are disappointingly small and nowhere near enough to explain earlier estimates of heritability. "It is the big topic in the genetics of common disease right now," says Francis Collins, former head of the National Human Genome Research Institute (NHGRI) in Bethesda, Maryland. The unexpected results left researchers at a point "where we all had to scratch our heads and say, 'Huh?'", he says.


Although flummoxed by this missing heritability, geneticists remain optimistic that they can find more of it. "These are very early days, and there are things that are doable in the next year or two that may well explain another sizeable chunk of heritability," says Hirschhorn. So where might it be hiding?


Right under everyone's noses
The inability to find some genes could be explained by the limitations of GWAS. These studies have identified numerous one-letter variations in DNA called single nucleotide polymorphisms (SNPs) that co-occur with a disease or other trait in thousands of people. But a given SNP represents a much bigger block of genetic material. So, for example, if two people share one of these variants at a key location, both may be scored as having the same version of any height-related gene in that area, even though one person actually has a relatively rare mutation that has a huge effect on height. The association study might identify a variant responsible for the height difference, says Teri Manolio, director of the Office of Population Genomics at the NHGRI, but averaging across hundreds of people could give the appearance that its effects are pretty wimpy. "It's going to be diluted," she says.
Finding this type of missing heritability is conceptually easy, because it involves closer scrutiny of the genes already in hand. "Just exploring, in a very dense way, genetic variation at the loci that have been discovered is probably going to [explain] another increment of missing heritability," Hirschhorn says. Researchers will need to sequence candidate genes and their surrounding regions in thousands of people if they are to unearth more associations with the disease.
Helen Hobbs and Jonathan Cohen of the University of Texas Southwestern Medical Center in Dallas did this in an attempt to capture all the variation in ANGPTL4, a gene their studies had linked to cholesterol and triglyceride concentrations. They sequenced the gene in around 3,500 individuals from the Dallas Heart Study and found that some previously unknown variants had dramatic effects on the concentration of these lipids in the blood. Mark McCarthy of Britain's Oxford Centre for Diabetes, Endocrinology and Metabolism says that such studies could reveal much of the missing heritability, but not a lot of people have had the enthusiasm to do them. This could change as the cost of sequencing falls.

Out of sight
Other variants, for which GWAS haven't even begun to provide clues, will prove even harder to find. In the past, conventional genetic studies for inherited diseases such as cystic fibrosis identified rare, mutated genes that have a high penetrance, meaning that the gene has an effect in almost everyone who carries it. But it quickly became apparent that high-penetrance variants would not underlie most common diseases because evolution largely keeps them in check.
What powered the push into genome-wide association was a hypothesis that common diseases would be caused by common, low-penetrance variants when enough of them showed up in the same unlucky person. Now that hypothesis is being questioned. "A lot of people are recognizing that screening for common variation has delivered less than we had hoped," says David Goldstein, professor of genetics at Duke University in Durham, North Carolina.
But between those variants that stick out like a sore thumb, and those common enough to be dredged up by the wide net of GWAS, there is a potential middle ground of variants that are moderately penetrant but are rare enough that they are missed by the net. There's also the possibility that there are many more-frequent variants that have such a low penetrance that GWAS can't statistically link them to a disease.
These very-low-penetrance variants pose some problems, says Leonid Kruglyak professor of ecology and evolutionary biology at Princeton University in New Jersey. "You're talking about thousands of variants that you would have to invoke to get near 80% or 90% heritability." Taken to the extreme, practically every gene in the genome could have a variant that affects height, for example. "You don't like to think about models like that," Kruglyak says.
If rare, moderately penetrant or common, weakly penetrant variants are the culprits, then bumping up the number of people in existing association studies could help find previously missed genetic associations. Peter Visscher of the Queensland Institute of Medical Research in Brisbane, Australia, says that a meta-analysis of height studies covering roughly 100,000 people is in the works. Lowering the stringency with which an association is made could drag up more, but confidence in the hits would drop.
At some point it might make sense to stop using SNPs, and start sequencing whole genomes. Collins suggests that the NHGRI's 1,000 genomes project, which aims to sequence the genomes of at least 1,000 people from all over the world, could go a long way towards finding hidden heritability, and many more genomes may become possible as the price of sequencing falls.
Not everyone supports an all-out sequencing onslaught. Goldstein warns against continuing to "turn the crank" without devising a more rational approach, such as sequencing the genomes of people who exhibit extreme manifestations of diseases. "I'm not really sold on doing the sequencing version of what we did with [GWAS]," he says. "It's a big enough, costly enough job, that I think we want to think a little bit harder about exactly who gets re-sequenced."

In the architecture
Some researchers are now homing in on copy-number variations (CNVs), stretches of DNA tens or hundreds of base pairs long that are deleted or duplicated between individuals. Variations in these features could begin to explain missing heritability in disorders such as schizophrenia and autism, for which GWAS have turned up almost nothing. Two recent studies looked at hundreds of CNVs in normal people and in those with schizophrenia, and found strong associations between the disease and several CNVs. They commonly arise de novo — in an individual without any family history of the mutation.
These structural variants might account for a lot of the genetic variability from person to person and could account for some of those rare 'out-of-sight' mutations with moderate penetrance that GWAS can't pick up. Many CNVs go undetected because they don't alter SNP sequences. Duplicated regions can also be difficult to sequence.
A standard technology for uncovering CNVs is array comparative genomic hybridization, in which scientists examine how genetic material from different individuals hybridizes to a microarray. If certain spots on an array pick up more or less DNA, it could indicate that there's a CNV. This and several other techniques are being tested by a consortium called the Copy Number Variation Project, run out of the Wellcome Trust Sanger Institute in Cambridge, UK. The consortium is dedicated to characterizing as many CNVs as possible so that associations can be made between them and diseases. McCarthy says that the role hidden CNVs have in heritability "should play out in the next six months to a year". But Goldstein argues that current technologies will miss many of the smaller CNVs, from 50 base pairs down to repeats of just two bases. "All we'll have verification of is the big whopping CNVs that are identifiable, and they clearly do not account for much of the missing heritability."

In underground networks
Most genes work together with close partners, and it is possible that the effects of one on heritability cannot be found without knowing the effects of the others. This is an example of epistasis, in which one gene masks the effect of another, or where several genes work together. Two genes may each add a centimetre to height on their own, for example, but together they could add five. GWAS don't cope with epistasis very well, and efforts to find these interactions usually require good up-front guesses about the interacting partners.
Joseph Nadeau, a geneticist at Case Western Reserve University in Cleveland, Ohio, says that 'modifier' genes act even in some straightforward single-gene diseases. "That's a simple kind of epistasis," he says. Cystic fibrosis, for example, is usually caused by mutations in one gene, CFTR, yet can vary greatly in symptoms and severity. The suspicion has been that modifier genes are one cause of this variability.
But despite the years of study, researchers still struggle to pin down these genes. "People haven't modelled truly the effect of epistasis," says population geneticist Sarah Tishkoff at the University of Pennsylvania in Philadelphia.
It's no surprise that genetics is more complicated than one gene, one phenotype, or even several genes, one phenotype, but it's humbling to realize how much more complex things are starting to look. In a now classic study, Kruglyak and his colleagues found that expression of most yeast genes is controlled by several variants, often more than five. To fill in all the heritability blanks, researchers may need better and more varied models of the entire network of genes and regulatory sequences, and of how they act together to produce a phenotype. At some point this process starts to look more like systems biology, and researchers are already applying systems methods to humans and other organisms. "What we're learning from these studies is that we need to think about the more complex of the complex models rather than the more simple of the complex models," Kruglyak says.Chronic Schizophrenia and Adult Autism: Issues in Diagnosis, Assessment, and Psychological TreatmentAutism Spectrum Disorders: The Complete Guide to Understanding Autism, Asperger's Syndrome, Pervasive Developmental Disorder, and Other ASDs