Monday 23 October 2023
Lecture 18 – Adaptation Genomics: Complex Traits
Simple traits – where one gene controls the presence or absence of a trait
Pitx1 controls the loss of pelvic spines in stickleback
Cortex controls the peppered moth pigmentation
MCR1 loss of function mutation controls the pigmentation of pale desert lizards
Large body size of dogs
Due to strong selective pressures, where mutations with large effects have swept through
the population
Genome-Wide Association Studies (GWAS) – for traits controlled by many genes
For a GWA or “Case Control” study, get a population of >10k individuals to identify “cases”—
subjects who show the phenotype (breast cancer – BRCA1 cancer gene, schizophrenia,
Alzheimer's, diabetes, cardiovascular disease etc.) and “controls”—those who don’t. Then
genotype all subjects with a high-density panel of SNPs (assaying for the minor alleles of
polymorphic sites) or sequence entire genomes, and determine the association of SNPs with
phenotype – the probability of SNP for a particular polymorphism appearing in the case and
control populations
If statistically significant higher probability of SNP in case population than control
population, infer the SNP marks a genetic location that is associated with an
increased risk of the disease – the SNP is not causally responsible, only linkage
disequibirum
Odds Ratio
The metric of association between a SNP and a phenotype is the odds ratio (OR).
Calculate the odds of SNP appearing in the case and control population and then the ratio:
For each SNP, the odds, O, that it will appear in each group.
For example, if pcase is the probability of SNP occurring in the cases, the odds of that
happening are:
pcase
O case =
1−p case
If pcont is the probability of SNP occurring in the controls, the odds of that happening
are:
p
Ocont = cont
1−p cont
The odds ratio, OR, is:
p case
1− pcase
¿=
p cont
1− p cont
OR = 1 – the SNP is equally likely to occur in cases than in controls
OR > 1 – the SNP is more likely to occur in cases than in controls = increased risk
, Monday 23 October 2023
OR < 1 – the SNP is less likely to occur in cases than in controls = protective
The chi-squared distribution can be used for significance testing, and given large sample
sizes, many significant odds ratios are significant SNPs
Human Height
Human height is a trait that is influenced by the environment but is also highly heritable with
a narrow-sense heritability of 80%
V
h2 = A = Proportion of total phenotypic variation attributable to the additive
VP
phenotypic effects of alleles
In 2008 several major studies identified ≈ 44 QTLs associated with human height and the
chromosomal locations of height loci
The 2010 meta-analysis of European GWA height studies by the GIANT consortium –
180,000 subjects, 180 height QTLs. Calculated 12% of the heritability & 9.6% of phenotypic
variance accounted for. Implies the existence of many more height genes
The 2014 meta-analysis of European GWA height studies by the GIANT consortium –
250,000 subjects, 697 SNPs–423 height QTLs. Calculated 20% of the heritability & 16% of
phenotypic variance accounted for
What accounts for the “the missing variance”? Increasing the sample size alone does not
solve this issue
One explanation is that the genome-wide significance values (P < 5x10 −8) used in these
studies are too stringent. If you use SNPs associated with height with a less stringent cut-off.
(P < 5x10−3 → 9500 SNPs) then you can explain 36% of the heritability and 29% of the
phenotypic variance. You’re less sure which SNPs truly affect height, but collectively you can
predict height from just looking at genotype alone really rather well.
i.e. attempting to show that individual loci have a statistically significant affect upon
height by setting highly significant p values for the association of every SNP and
height (the high p values decreases the chance of false positives but also increases
chance of false negatives – excluding genes that may have an association but not as
significantly high) so using a less stringent cut-off can explain more of the variance
But a more fundamental explanation for the missing variance is that GWAS studies are all
based on common SNPs — with minor allele frequencies (MAF) > 5% — that’s how the chips
are made. But what if a lot of the variants are rare (MAF < 1%)? You might be missing many
height variants. So, use an ExomeChip: ≈ 241k rare variants (MAF > 0.004%) on 711k
individuals. 83 variants now can explain 27.4% of the heritability of height. Variants with big
effects are rarer than those with small effects. Ultimately, deep sequencing will surely
identify more.
i.e. SNP chips are based on genes with a minor allele frequency of >5% (to ensure it is
found in a large number of individuals) so these are common SNPs, this means the
Lecture 18 – Adaptation Genomics: Complex Traits
Simple traits – where one gene controls the presence or absence of a trait
Pitx1 controls the loss of pelvic spines in stickleback
Cortex controls the peppered moth pigmentation
MCR1 loss of function mutation controls the pigmentation of pale desert lizards
Large body size of dogs
Due to strong selective pressures, where mutations with large effects have swept through
the population
Genome-Wide Association Studies (GWAS) – for traits controlled by many genes
For a GWA or “Case Control” study, get a population of >10k individuals to identify “cases”—
subjects who show the phenotype (breast cancer – BRCA1 cancer gene, schizophrenia,
Alzheimer's, diabetes, cardiovascular disease etc.) and “controls”—those who don’t. Then
genotype all subjects with a high-density panel of SNPs (assaying for the minor alleles of
polymorphic sites) or sequence entire genomes, and determine the association of SNPs with
phenotype – the probability of SNP for a particular polymorphism appearing in the case and
control populations
If statistically significant higher probability of SNP in case population than control
population, infer the SNP marks a genetic location that is associated with an
increased risk of the disease – the SNP is not causally responsible, only linkage
disequibirum
Odds Ratio
The metric of association between a SNP and a phenotype is the odds ratio (OR).
Calculate the odds of SNP appearing in the case and control population and then the ratio:
For each SNP, the odds, O, that it will appear in each group.
For example, if pcase is the probability of SNP occurring in the cases, the odds of that
happening are:
pcase
O case =
1−p case
If pcont is the probability of SNP occurring in the controls, the odds of that happening
are:
p
Ocont = cont
1−p cont
The odds ratio, OR, is:
p case
1− pcase
¿=
p cont
1− p cont
OR = 1 – the SNP is equally likely to occur in cases than in controls
OR > 1 – the SNP is more likely to occur in cases than in controls = increased risk
, Monday 23 October 2023
OR < 1 – the SNP is less likely to occur in cases than in controls = protective
The chi-squared distribution can be used for significance testing, and given large sample
sizes, many significant odds ratios are significant SNPs
Human Height
Human height is a trait that is influenced by the environment but is also highly heritable with
a narrow-sense heritability of 80%
V
h2 = A = Proportion of total phenotypic variation attributable to the additive
VP
phenotypic effects of alleles
In 2008 several major studies identified ≈ 44 QTLs associated with human height and the
chromosomal locations of height loci
The 2010 meta-analysis of European GWA height studies by the GIANT consortium –
180,000 subjects, 180 height QTLs. Calculated 12% of the heritability & 9.6% of phenotypic
variance accounted for. Implies the existence of many more height genes
The 2014 meta-analysis of European GWA height studies by the GIANT consortium –
250,000 subjects, 697 SNPs–423 height QTLs. Calculated 20% of the heritability & 16% of
phenotypic variance accounted for
What accounts for the “the missing variance”? Increasing the sample size alone does not
solve this issue
One explanation is that the genome-wide significance values (P < 5x10 −8) used in these
studies are too stringent. If you use SNPs associated with height with a less stringent cut-off.
(P < 5x10−3 → 9500 SNPs) then you can explain 36% of the heritability and 29% of the
phenotypic variance. You’re less sure which SNPs truly affect height, but collectively you can
predict height from just looking at genotype alone really rather well.
i.e. attempting to show that individual loci have a statistically significant affect upon
height by setting highly significant p values for the association of every SNP and
height (the high p values decreases the chance of false positives but also increases
chance of false negatives – excluding genes that may have an association but not as
significantly high) so using a less stringent cut-off can explain more of the variance
But a more fundamental explanation for the missing variance is that GWAS studies are all
based on common SNPs — with minor allele frequencies (MAF) > 5% — that’s how the chips
are made. But what if a lot of the variants are rare (MAF < 1%)? You might be missing many
height variants. So, use an ExomeChip: ≈ 241k rare variants (MAF > 0.004%) on 711k
individuals. 83 variants now can explain 27.4% of the heritability of height. Variants with big
effects are rarer than those with small effects. Ultimately, deep sequencing will surely
identify more.
i.e. SNP chips are based on genes with a minor allele frequency of >5% (to ensure it is
found in a large number of individuals) so these are common SNPs, this means the
