Where is the Missing Heritability •  We will assume that the heritability

calculations are correct, and that GWAS is somehow not finding all the heritability"

•  Note that this assumption is not necessarily true: some of the missing heritability might be because the heritability estimates are wrong, but many seem to be robust"

•  Given this assumption, what might account for the missing heritability?"

Currently Four Major Theories •  Infinitesimal model: many variants of small

effect"•  Rare allele model: many rare alleles of large

effect"•  Epistasis and genotype x environment

effects: interactions with variants that differ between individuals"

•  Epigenetic variation: inherited variation underlies disease, but its epigenetic rather than genetic variation"

In!nitesimal model •  Disease is caused

by a large number of common variants, each of small effect"

•  Disease comes from cumulative effects of these"

•  Effect sizes are so small that you need very large sample sizes to detect them"

•  Evidence that suggests this: if variants each have a small effect, they will not be purged by natural selection, and so are expected to be present"

•  Variance is typically found to occur across the genome, and animal breeders successfully use markers across the genome to continuously improve traits. Furthermore, model organism research usually finds multiple variants with a range of effect sizes"

•  GWAS is not powered to detect variants with small effect sizes, so that from a distribution of effect sizes, it can only detect the variants with the largest effects – i.e. missing heritability would be expected"

In!nitesimal model

•  However, lots of variants of small effects would be expected to produce a blending (continuous distribution) of phenotypes, rather than the patchiness (like traits running in families) we observe"

•  This model cannot explain some population disease incidence, such as the differences in disease rates for the same ethnic group living in different countries"

•  What are these variants actually doing? Is it realistic to think that so many genes could be having an effect on a single trait?"

In!nitesimal model

In!nitesimal model •  The smaller

the effect size, the greater the sample size needed to detect it"

•  Current studies can only detect risks greater than 20% above average (odds ratio above 1.2)"

•  Disease is caused by a number of rare variants, each of large effect"

•  Disease comes from having just one or a few"

•  Effect sizes are large, but most individuals with the disease have different variants"

Rare alleles model

•  Disease-causing alleles are expected to be rare because of selection against them, and this is known to be true"

•  Large effect rare alleles have been found (e.g. BRCA1 mutations) "

•  However, simulations show that most rare alleles would be detected by GWAS (BRCA1), and the rates of diseases recurrence found in families is not consistent with only one or two alleles being involved"

•  Changes in disease frequencies over time (e.g. increased incidence of heart disease) suggests more than just one or two large-effect alleles"

Rare alleles model

Genotype x environment •  These come where the effect of the genotype

is modulated by the environmental conditions"•  The idea is that a certain genotype will result

in a certain phenotype only under some, but not other, environmental conditions"

•  Therefore, unlike fully environmental or fully genetic effects, the phenotype is the result of an interaction between the genotype and environment!

Genotype x environment •  These are obviously known to occur, and

heritability measures try to exclude these"•  What we mean by “environment” is very

broad – includes the abiotic environment, the biotic environment, and the cultural/behavioural environment"

•  However, often we don’t know what the environmental effects are that interact with a given genotype to influence a given trait"

Epistasis (genetic interactions) •  “The interaction between nonallelic genes at two

or more loci resulting in one gene masking the phenotypic expression of another gene”"

•  Therefore, the phenotypic effect of one gene variant is dependent on the state of one or more other gene variant"

•  In this case, the phenotype that arises is dependent on the combination of variants that an individual has – the effects of the variants are not additive"

Epistatic effects

•  The assumption is that these alleles contribute additively to risk"

•  However, maybe one determines whether some of the others do result in increased risk or not"

Epistatic effects

•  The assumption is that these alleles contribute additively to risk"

•  However, maybe one determines whether some of the others do result in increased risk or not"

X X X X

Epistasis (genetic interactions) •  Because variants work in combinations under

epistasis, we need to assess the effect of combinations of variants on the risk of disease"

•  “very large samples are required to find sufficient individuals of each genotype combination to measure small effects accurately, and the number of comparisons scales exponentially with the number of interactions, so the testing burden is enormous.”"

•  “some will soon ask, as seismologists have, whether we are trying to predict the unpredictable.” [Jason Moore, Dartmouth Medical School]"

Epistasis (genetic interactions) •  But is epistasis really a major force in the

phenotypic outcome of genotype?"•  Controversial – some say there is little

evidence for epistasis having large effects in human studies"

•  Others note that significant contributions of epistasis have been found in model organisms, and that it is known that usually a number of genes interact in each biological pathway"

Epigenetic variants •  Gene activity is not only determined by the primary

DNA sequence, but also from epigenetic factors: primarily DNA methylation and histone modifications"

•  Therefore changes in these epigenetic factors without any change in the DNA can affect phenotype"

•  The major question is whether these epigenetic changes are heritable (or sufficiently heritable) to contribute to missing heritability"

•  DNA methylation can be heritable, but is controversial whether it is in practice. Histone modification is less obviously heritable"

•  More evidence is accumulating for the heritability of these epigenetic marks, but its still controversial"

The concept of “continuous liability” of disease

•  Diseases are typically considered in a binary sense (you have them or you don’t)"

•  However, many human traits are quantitative, not qualitative like this (e.g. height; http://www.nature.com/scitable/nated/topicpage/multifactorial-inheritance-and-genetic-disease-919)"

•  Are we therefore mistaken in thinking that diseases are qualitative traits – are they more likely to be quantitative?"

•  We know that many/all diseases have different levels of severity for the same disease"

The concept of “continuous liability” of disease

•  This quantitative view of disease is sometimes called the continuous liability of disease or the threshold model of disease"

•  If we accept this, then it requires a different approach to doing GWA studies"

•  In the traditional approach, there are likely to be controls at nearly the same risk of disease as the cases"

•  Using a quantitative framework can avoid this"