Mendelian randomization

1
Mendelian randomization

• George Davey Smith
• Department of Social Medicine
• University of Bristol

2
Why new approaches to the study of modifiable
causes of disease?
3
(No Transcript)
4
CHD risk according to duration of current Vitamin
E supplement use compared to no
use
RR
Rimm et al NEJM 1993 328 1450-6
5
Vitamin E supplement use and risk of Coronary
Heart Disease
Stampfer et al NEJM 1993 328 144-9 Rimm et al
NEJM 1993 328 1450-6 Eidelman et al Arch
Intern Med 2004 1641552-6
6
Vitamin C and coronary heart disease
observational study and RCT
7
Vitamin C levels and risk factors Womens Heart
Health Study
Childhood SES Manual social class No car
access State pension only Smoker Daily
alcohol Exercise Low fat diet Obese Height
Leg length
Lawlor et al, Lancet 2004
8
(No Transcript)
9
(No Transcript)
10
Vitamin C and incident CHD
Lawlor et al, Heart 2005
11
(No Transcript)
12
Mendel on Mendelian randomization
the behaviour of each pair of differentiating
characteristics in hybrid union is independent of
the other differences between the two original
plants, and, further, the hybrid produces just so
many kinds of egg and pollen cells as there are
possible constant combination forms   (Sometimes
called Mendels second law the law of
independent assortment)   Gregor Mendel, 1865.
Mendel in 1862
13
Mendelian randomization
  Genotypes can proxy for some modifiable
environmental factors, and there should be no
confounding of genotype by behavioural,
socioeconomic or physiological factors (excepting
those influenced by alleles at closely proximate
loci or due to population stratification), no
bias due to reverse causation and lifetime
exposure patterns can be captured
14
(No Transcript)
15
Mendelian randomisation and RCTs
16
Clustered environments and randomised genes (93
phenotypes, 23 SNPs)
Davey Smith et al. PLoS Medicine 2007 in revision
17
Exposure propensity how does alcohol intake
influence the risk of disease?
18
My Doctor said Only 1 glass of alcohol a day.
I can live with that.
19
Metabolism of alcohol
Ethanol
Acetaldehyde
Acetic acid
ADH
ALDH
CYP2E1
Mainly occurs in the liver, but some activity
is also present in the oral cavity and digestive
tract
20
Relationship between alcohol intake and ALDH2
genotype in men
Alcohol intake ml/day
ALDH2 genotype

Takagi et al, Hypertens Res 200225677-81
21
Relationship between characteristics and ALDH2
genotype
Age
Smoker
Percent
Years
BMI
Cholesterol
kg/m2
mg/dl
Takagi et al, Hypertens Res 200225677-81
22
ALDH2 genotype and systolic blood pressure
Chen et al, submitted
23
ALDH2 genotype and systolic blood pressure
Chen et al, submitted
24
Relationship between HDL cholesterol and ALDH2
genotype
HDL mg/dl

P lt0.0001
Takagi et al, Hypertens Res 200225677-81
25
Intermediate phenotypes cholesterol and CHD
26
APOB mutation, cholesterol and CHD

• Arg 3500 Gln carriers have cholesterol levels
  2.6 mmol/l higher than non-carriers
• Genotype unrelated to triglycerides, fibrinogen,
  glucose, BMI, W/H ratio etc etc

Tybjæg-Hansen et al NEJM 19983381577-1584
27
APOB mutation, cholesterol and CHD

• Arg 3500 Gln carriers CHD risk
• OR 7.0 (95 CI 2.2-22)
• Central estimate higher than predicted by trials
  BUT not attenuated by measurement error and
  reflects life-time differences in cholesterol
  levels (Same finding with PCSK9 variants, Cohen
  et al, NEJM 2006)

Tybjæg-Hansen et al NEJM 19983381577-1584
28
Implications of triangulation of cholesterol,
APOB genotype and CHD risk

• Lowering cholesterol will reduce CHD risk in the
  whole population
• NOT
• Screening for APOB mutations to detect high CHD
  risk
• Genotype effect size or PAR is NOT the issue!

29
Conventional observational epidemiology
Confounders (Factors associated with both
exposure and outcome, including unmeasured
confounders) and/or Reverse causation (Disease
alters the modifiable exposure of interest,
rather than vice versa)
Outcome (Disease or health status measure)
Modifiable exposure (Phenotype, e.g. alcohol
intake, cholesterol level)
30
Conventional observational epidemiology
Confounders (Factors associated with both
exposure and outcome, including unmeasured
confounders) and/or Reverse causation (Disease
alters the modifiable exposure of interest,
rather than vice versa)
Outcome (Disease or health status measure)
Modifiable exposure (Phenotype, e.g. alcohol
intake, cholesterol level)
It is often impossible to exclude confounding and
/or reverse causation as an explanation for
observed exposure/outcome associations
31
Mendelian randomisation approach
Confounders (Factors associated with both
exposure and outcome, including unmeasured
confounders) and/or Reverse causation (Disease
alters the modifiable exposure of interest,
rather than vice versa)
Instrumental variable (Genotype, randomly
determined at conception,e.g. CRP)
Modifiable exposure (Phenotype, e.g. alcohol
intake, cholesterol level)
Outcome (Disease or health status measure)
32
Mendelian randomisation approach
Confounders (Factors associated with both
exposure and outcome, including unmeasured
confounders) and/or Reverse causation (Disease
alters the modifiable exposure of interest,
rather than vice versa)
Instrumental variable (Genotype, randomly
determined at conception,e.g. CRP)
Modifiable exposure (Phenotype, e.g. alcohol
intake, cholesterol level)
Outcome (Disease or health status measure)
33
Mendelian randomisation approach
Confounders (Factors associated with both
exposure and outcome, including unmeasured
confounders) and/or Reverse causation (Disease
alters the modifiable exposure of interest,
rather than vice versa)
Instrumental variable (Genotype, randomly
determined at conception,e.g. CRP)
Modifiable exposure (Phenotype, e.g. alcohol
intake, cholesterol level)
Outcome (Disease or health status measure)
We use associations that are not affected by
confounding or reverse causation to estimate the
causal exposure-outcome association
34
Medelian randomisation and RCTs
35
MR and RCTs

• The fact that many, many environmental and
  genetic factors influence the target phenotype in
  RCTs clearly does not vitiate causal
  interpretation regarding treatment
• The fact that the treatment explains a low
  proportion of the variance in the target
  phenotype in RCTs also is of no concern to causal
  interpretation
• The same applies to MR

36
A few examples where MR approaches have been
applied

• Cholesterol and coronary heart disease
• Maternal folate and offspring NTDs
• Milk intake and bone mineral density / fractures
• CRP and insulin resistance
• IGF-1 and breast cancer
• Maternal caffeine intake and stillbirth
• Passive smoking and lung cancer / COPD
• NSAIDS and colon cancer
• Fibrinogen and atherothrombosis
• Inflammation and atherothrombosis
• Maternal glucose levels and offspring birthweight
  

37
Mendelian randomisation reflects the phenocopy /
genocopy dialectic

• e.g. Hartnups syndrome and pellagra
• no doubt all environmental effects can be
  mimicked by one or several mutations
• (Zuckerkandl and Villet, PNAS 1988)
• Gene / environment equivalence gene
  environment interchangability (West Eberhard
  2003)

38
Gene-environment interaction and evidence of
causal effects of modifiable exposures
39
NAT2 (slow versus fast acetylator), smoking and
bladder cancer
Garcia-Closas et al. Lancet 2005366649-659
40
NAT2 (slow versus fast acetylator), smoking and
bladder cancer
Garcia-Closas et al. Lancet 2005366649-659
41
Gene by environment interactions may provide
important information regarding the causal
effects of environmental exposures of interest
(while in their own right being rather
uninteresting journal padding)
42
Mendelian randomisation

• Potential limitations

43
The art of invigorating and prolonging life

• W Kitchiner, 1822

44
The art of invigorating and prolonging life, to
which is added the pleasure of making a will

• W Kitchiner, 1822

45
Failure to establish reliable genotype-phenotype
associations

• Not particular to Mendelian randomisation

46
Given a true genotype-phenotype association
when are interpretations misleading?
47
Mendelian randomisation why may interpretations
be misleading?

• Pleiotropy (multiple phenotypic effects)
• This could include alternative splicing
• E.g. ApoE genotype, lipids and CHD

48
Mendelian randomisation why may interpretations
be misleading?

• Confounding
• Allele under study in LD with allele that
  influences disease risk or confounding by
  population substructure

49
Multiple instruments in Mendelian randomization
approaches
Gene 1
Gene 2
Exposure
Outcomes
Confounders reverse causation bias
50
Mendelian randomisation why may interpretations
be misleading?

• Canalization
• Morphogenic stability / developmental adaptation
  / canalization create important caveats to the
  interpretation of such studies.
• . Mice without myoglobin

51
In RCTs randomisation is generally in middle-age
.

• in Mendelian randomisation, this occurs
  during gamete formation and conception.

52
Lack of suitable polymorphisms for studying
modifiable exposure of interest

• E.g. in the case of vitamin C, haptoglobin
  polymorphism related to substantial differences
  in plasma vitamin C, but also to several other
  physiological parameters
• Despite extensive search no functional
  polymorphsim found for human vitamin C
  transporter SVCT1

53
Lack of suitable polymorphisms for studying
modifiable exposure of interest

• E.g. in the case of vitamin C, haptoglobin
  polymorphism related to substantial differences
  in plasma vitamin C, but also to several other
  physiological parameters
• Despite extensive search no functional
  polymorphsim found for human vitamin C
  transporter SVCT1

54
Mendelian randomisation conclusions
Polymorphism association studies can provide
powerful evidence on mechanisms of disease and
potential interventions Such studies are
considerably less prone to confounding and
reverse causation than conventional risk-factor
epidemiology   Morphogenic stability /
developmental adaptation / canalization create
important caveats to the interpretation of such
studies this has been under-appreciated in some
presentations of Mendelian randomisation The
findings apply to whole population interventions
and such studies are NOT aimed at motivating
genetic screening strategies Effect sizes
likely to be small sample sizes need to be very
large   Provides greater public health relevance
to genetic epidemiology
55
Thanks

• Lina Chen
• Sarah Lewis
• Shah Ebrahim
• Jonathan Sterne