Galton and human height
Human height is the archetypal quantitative trait (anything caused by many genes and the environment- most talked about - human height)
- Normal distribution
- Highly heritable (caused by many genes, each of small effect- can use to test if this is a typical quantitative trait)
Can use it to test if many or a few genes explain the heritability
What did Francis Galton do
Francis Galton (1822-1911)- came up with the first method of determining height from parents height - he recognised that offspring tended to resemble parents (left - height of mum and dad and right- height of children), slope of children is slightly flatter than the parents height- parents with extreme values aren’t as extreme as they are
Wrote Hereditary Genius (1869)
Fisher’s 1918 paper- resolved the biggest debate in genetics
- Although published in 1918, Fisher had done the work by 1911, when he was 21
- It ended the argument between the Biometricians and the Mendelians (argued that blending inheritence)
- Mendelialns : single genes, biometricians: many genes and the environment (Quantitative)
- It paved the way for the Modern Synthesis in Evolutionary Biology
- Invented quantitative genetics in one paper
What do we actually mean by ‘heritability’?
The broad-sense (less useful) heritability is the proportion of variation explained by all genetic effects. (proportion of VP / VG)
We can break phenotypic variation (VP) down into genetic variation (VG) and environmental variation (VE)
VP = VG + VE
Which means the broad sense heritability is VG/VP
However, when people say the heritability, it is more often the narrow-sense heritability (written as h2) that matters- more relevent
The narrow sense heritability is the proportion of phenotypic variation explained by additive genetic variation (VA)
what do we mean by additive genetic variation?
- Three SNP genotypes, AA, AB and BB (measured lots of individuals with genotypes- each dot represents an individual, on average AA - lowest value etc.
- Substitution from an A to a B results in an increase of ~0.5- addititive
It is additive
- In this example, AB has nearly the same mean as BB.
- Allele B is dominant to allele A.- shifted heterozygotes up
- The difference between the mean of AB ~1.9) and the midpoint of AA and BB (~1.5) is called the dominance component (also additive)
- Dominance is not heritable to offspring in the same way as additive variants
Narrow-sense heritability
It is the additive genetic variation that is inherited from parent to offspring, and so the narrow-sense heritability is most relevant if we want to know how complex traits are inherited.
We measure heritability on a zero to 1 scale.
h2 = 0 The trait is not heritable at all
h2 = 1 The trait is completely heritable
Why is narrow sense heritability useful
- We can use it to predict how similar relatives are
- We can use it to predict a response to selection (only if a traits heritable)
- We can use it to predict the risk of a disease
How do we measure heritability?
To measure heritability, we classically compare how similar relatives are.
If a trait is heritable, then the more related two individuals are, the more similar they will be. A famous approach is twin studies, but these can have problems.
We first need to calculate relatedness. (further apart two individuals are in a pedigree (12:20)
For any given pedigree, we can estimate the relatedness (or kinship coefficient) of relatives, by looking at the number of family links required to connect them.
If the number of links is l, then the relatedness is 0.5l
- B and C (paternal half-sibs) are joined by two links – relatedness is 0.52 = 0.25
- A and B (father and son) are joined by one link – relatedness is 0.51 = 0.50
- A and C (father and daughter) are joined by one link - relatedness is 0.51 = 0.50
- A and D (uncle and niece) are joined via E and via F. Each join involves 3 steps, so the relatedness is 0.53 + 0.53 = 0.125 + 0.125 = 0.25
- D is cousins with both B and C, and they share two ancestors (E and F). From D to B via E, involves 4 steps, and from D to B via F involves 4 steps. Therefore, the relatedness is 0.54 + 0.54 = 0.0625 + 0.0625 = 0.125
Using the relatedness estimates
- Galton used regression and the assumption that tall parents had tall children
- Sib-design approaches make the assumption that sibs are more similar than non-relatives
- Twin studies use a similar idea. In addition, monozygous twins (identical; relatedness is 1) should be more alike than dizygous twins (relatedness is 0.5)
why could all of these approaches give upwardly biased estimates of heritability?
Environment also plays a role
Using all kinds of relatives at once
We can use complicated pedigrees, simultaneously analysing many types of relative to measure heritabilities. These are typically done using a statistical approach called The Animal Model. The animal model can be applied in any organism, but was first popularised in animal breeding, hence the name.
Has also been applied to multigenerational human pedigrees; e.g. Pre-Industrial Finns
Do molecular methods make classic quantitative genetics redundant?
Now that we can find causal loci, e.g. by GWAS, doesn’t it make old-fashioned pedigree-based methods a bit pointless?
Instead of assuming that there are lots of unknown genes that do this
Genome wide association studies of human height
An allelic substitution typically adds/decreases 2-5mm (i.e. small effects)
Percentage variation explained is nowhere near 80%!
If you add the effects of those genes together- doesn’t tell us about variation
Can GWAS results predict phenotype?
The 54 significant SNPs from the three previous studies only explained 4-6% of variance, and could not predict whether somebody was tall or short.
Midparent values (Galton’s approach) explained 40% of variance, and could reliably predict tall / short stature
A bigger GWAS of human height
Lango Allen et al. (2010) combined data from 46 studies
133,653 individuals; 2,834,000 SNPs
180 separate loci detected, but only 10.5% of variance is explained
Probably the biggest GWAS ever…..
The ‘missing heritability’ problem
- Why do the significant SNPs explain nowhere near the full heritability?
- Does this mean GWAS studies are doomed to failure?
- Crohn’s Disease Example
- Black dots – pre GWAS
- Green dots – GWAS
- blue dots – pooling GWAS studies
- 71 loci explain 20% of the heritability
- Might need 20,000 loci to explain 50% of the heritability
The ‘missing heritability’ problem solutions
- Measure the effect of all SNPs, not just the significant ones (sometimes known as genomic prediction or genomic selection)
- Analyse the effects of different genomic regions, e.g. different chromosomes, one at a time
Could look at the effect size- described as personalised medicine - more reliable than looking at the statistically significant ones - amount of variation starts to go up - can explain 60% of the heritability, infinite number of SNPs that can contribute
Missing heritability and polygenicity
One chromosome at a time, markers used to estimate identity-by-descent (relatedness) between all (distantly unrelated) individuals.
Clockwise (from top left)
Height, BMI, Qti (electrocardiographic measure) and von Willebrand Factor (Mendelian Trait)
Approach known as chromosome partitioning
If traits are polygenic, then additive effects of each chromosome should scale with chromosome size ………………. they usually do.
Molecular quantitative genetics
Using very large numbers of markers, instead of pedigrees can be useful for three reasons
For some traits, we simply don’t have pedigree data
Markers can give more accurate estimates of relatedness
Markers can be used to distinguish between very distantly related individuals. This gets around the ‘shared environment’ problem.
e.g. Relatives with r of 0.03 are three times more genetically similar than relatives with an r of 0.01, but there is no expectation that either pair will have a common environment, because they are only distant relatives
Why markers can be more accurate than pedigrees
- Ground-breaking paper that showed that full sibs will always have a pedigree-based expected relatedness of 0.5, but the actual or realised relatedness can range from <0.4 to >0.6. The higher the relatedness, the more similar the sibs.
- Also, all pairs should have same amount of shared environment, regardless of their realised relatedness
- Therefore, markers can give greater precision in estimating heritability.
Some applications of molecular quantitative genetics
- Looked at 1940 unrelated people in Scotland who took IQ tests aged 11 and again in old age. Typed at >500,000 SNPs to estimate relatedness
- Markers gave an estimated heritability of IQ (called it cognition) of at least 0.24
- Cognition at age 11 and old age were genetically correlated (0.62), but some environmental component to how it changes over time as well.
3322 cases, 3587 controls, typed at ~1 million SNPs - Relatively few genes found in GWAS (although immune-related MHC was significant)
- Showed that there were many more common SNPs of small effect, which explained 1/3rd of the variance in liability of schizophrenia
- Showed that the same genes also increased risk of bipolar disorder
- Showed that immune-related genes could be involved (trade-off?)
Why do so many disease alleles persist?
They were ancestrally neutral
Fitness costs are too recent for purifying selection to have removed them
- Balancing selection.
Alleles that cause disease may also have a selective advantage e.g. sickle-cell anaemia/malaria at HBB locus. Predicts disease alleles could be quite common. - Mutation-selection balance
Purifying selection removes alleles, but mutation creates new ones. If correct, predicts SNPs with low frequency/quite high penetrance.
Models 3 (and to some extent 2) best seem to fit the data followed by two followed by one
