Mendelian randomisation
Nic Timpson
Epigenetic Epidemiology 2012
Mendelian randomisation (“MR”)
Objectives:
Build on “epigenetic epidemiology” and “strengthening causal inference” sessions
Introduce MR specifically as a means to strengthening causal inference
Illustrate the process/underpinnings of MR give worked examples
Should should be able to:
Describe what MR is
Describe why MR allows causal inference
Describe an example of MR
Formulate a crude MR experiment
Epigenetic Epidemiology 2012
D.A.G
A
B
D
C
Epigenetic Epidemiology 2012
What is Mendelian randomisation?
Genetic association study or observational epidemiology?
- Conceptual approach to Mendelian randomisation should be
taken from the point of view of conventional epidemiology.
- Observational epidemiology is limited, however problems can be
avoided given the existence of “better measures”.
- In an observational framework what could we use as a
“better measure”?
- Mendelian randomisation aims to use genetic variation related to risk
factors of interest in efforts to re-assess observational estimates.
Epigenetic Epidemiology 2012
Assumptions of MR analysis
U
Z
X
• Z associated with X
• Z is independent of U
• Z is independent of Y given U and X
Epigenetic Epidemiology 2012
Y
Why are genotypes good instruments?
U
An instrumental variable is a variable that is:
(i)
Z
X
Y
Associated with an exposure of interest
(ii) NOT associated with the outcome of interest – except through its
association with the exposure of interest.
HDL & Ischaemic Heart Disease (IHD)
???
CETP variation & HDL
http://csg.sph.umich.edu/locuszoom/
Epigenetic Epidemiology 2012
Number of
comparisons
Expected
number
Observed
number
P
(Ho: OBS=EXP)
96 non-genetic
variables
4560
228
2447
<p*10-6
23 candidate
loci
253
12.7
14
0.66
Correlations here taken at p≤0.05
Epigenetic Epidemiology 2012
SNP/NONSNP
23 SNPs and
96 variables
110/120
P=0.35
Total number of phenotypes: 121
Total number of pairs: 7260
p value
Expected
Observed
0.05
363
1843
0.01
72
1281
0.0001
0
684
6.88 x 10-6 *
0
568
1.37 x 10-6 *
0
504
1.37 x 10-8 *
0
421
*bonferroni corrected results of significance values 0.05, 0.01 and 0.0001, respectively
• Phenotypes are highly over
correlated, as expected.
Number of SNPs with r2<0.25 = 112
Number of phenotypes = 121
Number of pairs at r2<0.25 = 13552
p value
Expected
Observed
0.05
678
709
0.01
136
167
0.0001
1
4
3.69 x 10-6 *
0
1
7.40 x 10-7 *
0
1
7.40 x 10-9 *
0
0
*bonferroni corrected results of significance values 0.05, 0.01 and 0.0001, respectively
Michelle Taylor
Epigenetic Epidemiology 2012
Conventional observational epidemiology
Confounders
(Factors associated with both exposure and
outcome, including unmeasured confounders and
confounders measured with error, e.g. smoking,
socioeconomic position, diet, physical activity)
Modifiable exposure
Outcome
(Phenotype, affected by
genotype and by
environment,
e.g. CRP)
(Usually a disease or health
status, e.g. coronary heart
disease)
It is often impossible to exclude unmeasured or residual
confounding as an explanation for observed
exposure/outcome associations
Epigenetic Epidemiology 2012
Mendelian randomisation approach
Confounders
(Factors associated with both exposure and
outcome, including unmeasured confounders and
confounders measured with error, e.g. smoking,
socioeconomic position, diet, physical activity)
Instrumental
variable
(Genotype, randomly
determined,
e.g. CRP)
Modifiable exposure
Outcome
(Phenotype, affected by
genotype and by
environment,
e.g. CRP)
(Usually a disease or health
status, e.g. coronary heart
disease)
Epigenetic Epidemiology 2012
Mendelian randomisation approach
Confounders
(Factors associated with both exposure and
outcome, including unmeasured confounders and
confounders measured with error, e.g. smoking,
socioeconomic position, diet, physical activity)
Instrumental
variable
(Genotype, randomly
determined,
e.g. CRP)
Modifiable exposure
Outcome
(Phenotype, affected by
genotype and by
environment,
e.g. CRP)
(Usually a disease or health
status, e.g. coronary heart
disease)
Epigenetic Epidemiology 2012
Example – using instruments for alcohol consumption
U
ADH2
Z
Alcohol
X
Epigenetic Epidemiology 2012
Y
Metabolism of alcohol
Ethanol
Acetaldehyde
ADH
Acetic acid
ALDH
CYP2E1
* Mainly occurs in the liver, but some activity is also present in the oral cavity and
digestive tract
Epigenetic Epidemiology 2012
ALDH2 genotype by alcohol consumption,
g/day: 5 studies, n=6815
60
Men
Alcohol g/day
50
40
Women
30
20
10
0
*1*1
*1*2
*2*2
Chen, Davey Smith et al, PLoS Med 2008
Epigenetic Epidemiology 2012
Relationship between characteristics and ALDH2 genotype
Smoker
Years
Percent
Age
Cholesterol
mg/dl
kg/m2
BMI
Epigenetic Epidemiology 2012
ALDH2 genotype and systolic blood pressure
(heterozygote vs heterozygote)
(homozygote vs homozygote)
Epigenetic Epidemiology 2012
Example – using instruments for adiposity
U
BMI loci
Z
Adiposity
X
Epigenetic
Y
GIANT_3 (n~250,000) – Speliotes et al 2010
Epigenetic Epidemiology 2012
BUT, are these “genes for obesity”?
FTO effect ~0.1SD
across the distribution
Effect mirrored in a simple
Cumulative frequency plot
Copenhagen data – n=~38000 cross sectional
Epigenetic Epidemiology 2012
Example – using instruments for adiposity
U
FTO
Adiposity
Traits of
metabolism
CRP/BMI
IHD
Epigenetic Epidemiology 2012
Diabetes, 2008
Epigenetic Epidemiology 2012
Timpson et al Hypertension 2009
Epigenetic Epidemiology 2012
Components of analysis
Observational effects/trend
Linear regression result
Adjusted linear regression result
Instrumental variable regression result
A test for difference (p(DWH))
Epigenetic Epidemiology 2012
aa
Aa
AA/Aa/aa
AA
AA Aa aa
Epigenetic Epidemiology 2012
MR not just with continuous outcomes…
Confounders
BMI
genotype
BMI
IHD
~25%
~50%
Nordestgaard et al, PLoS Med, 2012
Epigenetic Epidemiology 2012
MR not just with simple instruments…
(1)
(2)
(1) Basic score and BMI gradient in ALSPAC samples with GWAS
Red lines indicate the cut off points for the top and bottom 5%
Score from Speliotes et al 2010 explains ~2.4% variance
(2) Histograms of BMI by the two groups generated from the two groups from figure 1.
Red lines indicate the mean BMI for each – these are ~20 and 23kg/m2 respectively
** A change of threshold to a score using the top 50, ooo markers from the GIANT
GWAS for BMI takes advantage of extra variance explained (>3.5% - continuous)
Epigenetic Epidemiology 2012
Limitations to Mendelian randomisation
1- Heterogeneity
- Genetic heterogenetity through linkage disequilibrium
or by pleiotrophy.
- Phenotypic heteogeneity & complexity (e.g. EC-SOD).
- Population genetic heterogeneity in instrument choice.
Epigenetic Epidemiology 2012
LMS allowing adjustment for age-specific
heteroscedasticity and skewness.
Median (or M) curves were modelled with 9
degrees of freedom, with corresponding
curves for the coefficient of variation (S) with
5 d.f. and skewness (L) with 3.
Epigenetic Epidemiology 2012
Limitations to Mendelian randomisation
1- Heterogeneity
- Genetic heterogenetity through linkage disequilibrium
or by pleiotrophy.
- Phenotypic heteogeneity & complexity (e.g. EC-SOD).
- Population genetic heterogeneity in instrument choice.
2- Population stratification
Epigenetic Epidemiology 2012
Limitations to Mendelian randomisation
1- Heterogeneity
- Genetic heterogenetity through linkage disequilibrium
or by pleiotrophy.
- Phenotypic heteogeneity & complexity (e.g. EC-SOD).
- Population genetic heterogeneity in instrument choice.
2- Population stratification
C. Waddington, 1957
3- Canalisation
‘the cell can take specific permitted
trajectories, leading to different cell fates’
Epigenetic Epidemiology 2012
Limitations to Mendelian randomisation
1- Heterogeneity
- Genetic heterogenetity through linkage disequilibrium
or by pleiotrophy.
- Phenotypic heteogeneity & complexity (e.g. EC-SOD).
- Population genetic heterogeneity in instrument choice.
2- Population stratification
3- Canalisation
4- Power (instrument “strength” and basic power)
Epigenetic Epidemiology 2012
CEBP(2010)19:1341-48
Epigenetic Epidemiology 2012
“Importantly, despite its size… study had low power…”
- relatively weak relationship between genotype and milk consumption
(an often ignored characteristic in the examination of lactase
persistence genotypes & with weak instruments comes instability and
bias - Burgess)
- modest observational association between milk consumption and
RCC
A study with ∼37,000 cases and 37,000 controls would be needed to
achieve 80% power under the same framework.
- Not to mention the possibly complicating nature of variation at MCM6
Epigenetic Epidemiology 2012
Objectives:
Build on “epigenetic epidemiology” and “strengthening causal inference” sessions
Introduce MR specifically as a means to strengthening causal inference
Illustrate the process/underpinnings of MR give worked examples
Should should be able to:
Describe what MR is
Describe why MR allows causal inference
Describe an example of MR
Formulate a crude MR experiment
Epigenetic Epidemiology 2012
References
- Davey Smith, G. Reflections on the limitations to epidemiology. Journal of Clinical Epidemiology 54, 325-31 (2001).
- Davey Smith, G. & Ebrahim, S. Mendelian Randomisation: prospects, potentials and limitations. International Jounal of Epidemiology 33, 30-42 (2004).
- Davey Smith, G. et al. Genetic epidemiology and public health: hope, hype, and future prospects. Lancet 366, 1484-98 (2005).
- Lawlor, D.A., Harbord, R., Sterne, J., Timpson, N.J. & Davey Smith, G. Mendelian randomisation: using genes as instruments for making causal inferences in
epidemiology. Statistics in Medicine 27, 1133-1163 (2008).
- Davey Smith, G. et al. Clustered environments and randomized genes: a fundamental distinction between conventional and genetic epidemiology. PLoS Medicine 4,
e352 (2007).
- Greenland, S. An introduction to instrumental variables for epidemiologists. International Jounal of Epidemiology 29, 722-729 (2000).
- Burgess, S., Thompson, S.G. & Collaboration, C.C.G. Avoiding bias from weak instruments in Mendelian randomization studies. International Jounal of Epidemiology
40, 755/764 (2011).
- Chen, L., Davey Smith, G., Harbord, R.M. & Lewis, S.J. Alcohol intake and blood pressure: a systematic review implementing a Mendelian randomization approach.
PLoS Medicine 5, e52 (2008).
- Speliotes, E.K. et al. Association analyses of 249,796 individuals reveal 18 new loci associated with body mass index. Nat Genet 42, 937-948 (2010).
- Freathy, R.M. et al. Common variation in the FTO gene has a metabolic impact consistent with its effect on BMI. Diabetes 57, 1419-1426 (2008).
- Timpson, N.J. et al. Does greater adiposity increase blood pressure and hypertension risk? Mendelian randomisation using FTO/MC4R genotype. Hypertension 54,
84-90 (2009).
- Nordestgaard, B.r.G. et al. The Effect of Elevated Body Mass Index on Ischemic Heart Disease Risk: Causal Estimates from a Mendelian Randomisation Approach. PLoS
Medicine 9, e1001212 (2012).
- Sovio, U. et al. Association between Common Variation at the FTO Locus and Changes in Body Mass Index from Infancy to Late Childhood: The Complex Nature of
Genetic Association through Growth and Development. PLoS Genet 7, e1001307 (2011).
- Waddington, C.H. Canalisation of development and the inheritance of acquireed characters. Nature 150, 563-565 (1942).
- Waddington, C.H. The Strategy of the Genes, (George Allen & Unwin., London, 1957).
Epigenetic Epidemiology 2012