Transcript Microsoft PowerPoint - NCRM EPrints Repository

What is
Mendelian Randomisation?
Frank Dudbridge
Short answer
• Instrumental Variable Analysis
with a gene as the instrument
• A method for making causal statements from
observational data
Association is not causation
• Two major problems in observational data are:
• Confounding
Socio-economic
status
(Endogeneity)
Alcohol
consumption
• Reverse causation
?
•Associated with both
exposure and outcome
•Not on causal pathway
from exposure to outcome
Heart disease
?
Bodyweight
Well-being
?
Observation vs Randomisation
• In randomised intervention studies, confounding and
reverse causation are reduced or even eliminated
– Groups differ only in the intervention received, and by no
other characteristics
– Events that follow treatment (in time) are more likely to be
caused by the treatment
• But randomisation is not always possible
– Effects of smoking or alcohol
– Socio-economic effects
– Observational studies are the only ethical option
Mendelian Randomisation principle
• Genes are randomly allocated, independent of
confounders
• Genes cause phenotypes, but phenotypes do not
cause genes
• Therefore, a gene that causes the exposure of
interest can be a proxy for that exposure
Socio-economic
status
X
Gene that
influences
alcohol
consumption
?
X
Heart disease
Mendel’s first law
• Law of segregation: Two copies of a gene have an
equal probability of being transmitted to an offspring
– “Random allocation at conception”
50%
50%
Mendel’s second law
• Law of independent assortment: each gene is
distributed randomly and independently from genes
for other characteristics
– “Random allocation at conception” independent of
confounders
25%
25%
25%
25%
Randomisation to judge causality
Genetics
Intervention trial
RCT
Mendelian randomisation
Sample
Population
Randomisation
Random allocation of alleles
Intervention
Control
Event rate
lower
Event rate
higher
Genotype aa
Genotype AA
Event rate
lower
Event rate
higher
Mendelian Randomisation technique
Instrumental variable analysis
MR relies upon three key assumptions
1. The gene Z is associated with exposure of interest X
2. The gene Z is independent of confounding factors U (that
confound X-Y association)
3. The gene Z is related to the outcome Y only via its
association with the modifiable exposure X
Confounders
U
Z
Genotype
X
Modifiable Exposure
Y
Outcome
Alcohol and Cardiovascular disease
Corrao et al., 2000
• 1 UK alcohol unit = 8g ethanol
• UK recommended limits = 3-4 units per day for men; 2-3 units per day women
Slide by Caroline Dale
Alcohol metabolising genes
• Heritability of alcoholism estimated at 40-60% (Goldman et al., 2005)
• Many genes influence propensity to drink
– Alcohol metabolising enzymes (ALDH2, ADH1B, ADH1C)
– Neurotransmitter receptors (GABRA2, DRD4, DRD2)
acetate
ALCOHOL
DEHYDROGENASE (ADH)
alcohol
acetaldehyde
ALDEHYDE
DEHYDROGENASE-2 (ALDH2)
“Flushing” response
water
• More active ADH variants / less active ALDH2 variants associated with higher
acetaldehyde concentration  unpleasant side effects  lower alcohol intake
Caroline Dale
ADH1B – CVD collaboration
Large collaborative effort to describe causal effects of alcohol
intake on cardiovasular markers
• >16 studies with >56,000 subjects and >45 measurements
SES, diet etc
U
Z
ADH1B Genotype
X
Alcohol intake
Y
Markers for CVD
Objective 1:
Association with alcohol phenotypes
SES, diet etc
U
Z
ADH1B
X
Alcohol
Y
Markers for CVD
Objective 2:
No association with potential confounders
SES, diet etc
X
Z
ADH1B
X
Alcohol
U
Y
Markers for CVD
Objective 3:
Association with CVD biomarkers
SES, diet etc
U
Z
ADH1B
Y
Markers for CVD
X
Alcohol
X
Causal effect of alcohol on blood pressure
• Effect of gene on systolic BP = -1.19 mmHg
• Effect of gene on log weekly units of alcohol = -0.16 units
• Therefore, effect on SBP of increasing alcohol by one log unit
is -1.19/-0.16 = 7.4375 mmHG
– Equivalent to multiplying alcohol intake by 2.72 units
SES, diet etc
-1.19
-0.16
Z
ADH1B Genotype
X
Alcohol intake
U
1.19/0.16
Y
Systolic BP
Issues in Mendelian randomisation
Among many other issues, two important ones are
• Pleiotropy
– Gene affects traits other than the phenotype in question
– This can invalidate the MR assumptions
• Weak instrument bias
– If the gene predicts the phenotype poorly, then the MR estimate is
biased and unstable
– Most genes have very weak effects on modifiable exposures
• Both problems can be reduced by combining multiple genes
into a composite instrument
Multiple genes used for Mendelian Randomisation
Among many other issues, two important ones are
• Pleiotropy
– Gene affects traits other than the exposure in question
– This can invalidate the MR assumptions
• Weak instrument bias
– If the gene predicts the exposure poorly, the MR estimate
becomes biased and unstable
– Most genes have very weak effects on complex traits
• Both problems can be reduced by combining
multiple genes into a composite instrument
Juan-Pablo Casas
Thanks
LSHTM
•
•
•
•
•
Caroline Dale
Michael Holmes
Richard Silverwood
Juan Pablo Casas
Dave Leon
Bristol
• Tom Palmer
• Debbie Lawlor