Effect of the Epstein-Barr Virus on Multiple Sclerosis

Effect of the Epstein-Barr
Virus on Multiple Sclerosis
A Mendelian Randomisation Study
[email protected]
Epstein-Barr Virus (EBV)
Risk of developing Infectious Mononucleosis
(IM)
Hesla, H.M. et al. Herpesvirus infections and allergic sensitization in children of families with anthroposophic and non-anthroposophic lifestyle - the ALADDIN birth cohort. Pediatr Allergy Immunol
24, 61-65 (2013).
Morris, M.C. et al. Sero-epidemiological patterns of Epstein-Barr and herpes simplex (HSV-1 and HSV-2) viruses in England and Wales. J Med Virol 67, 522-7 (2002).
Crowcroft, N.S., Vyse, A., Brown, D.W. & Strachan, D.P. Epidemiology of Epstein-Barr virus infection in pre-adolescent children: application of a new salivary method in Edinburgh, Scotland. J
Epidemiol Community Health 52, 101-4 (1998).
Epstein-Barr Virus - Genetic Component
University students tested at the beginning and the end of their undergraduate studies.
237 Seropositive Students
49 Seronegative Students
Two SNPs located to the left of HLA-A associated with an increased risk of EBV and IM after EBV
Allele Frequencies
Seropositive
Seronegative
rs6457110
AA
AT
TT
46.9%
41.3%
11.8%
34.7%
44.9%
20.4%
(p=0.01)
rs356966
AA
AT
TT
38%
49.3%
12.7%
51%
38.7%
10.2%
(p=0.03)
McAulay, K.A. et al. HLA class I polymorphisms are associated with development of infectious mononucleosis upon primary EBV infection. J Clin Invest 117, 3042-8 (2007).
Multiple Sclerosis
MS - Genetic Component
9949 Controls
5667 Cases
Immunochip
Controls
Cases
rs6457110
AA
AT
TT
47.7%
42.2%
10%
38.3%
47%
14.5%
rs356966
(proxy)
AA
AT
TT
42.6%
44.06%
13.2%
44.7%
44.04%
11%
Nat Genet. 2013 Nov;45(11):1353-60. doi: 10.1038/ng.2770. Epub 2013 Sep 29.
MS and EBV
Environmental Risk Factors for Multiple Sclerosis. Part I: The Role of Infection. Alberto Ascherio, MD, DrPH,1–3 and Kassandra L. Munger, MSc1
MS and EBV
EBV +
99.5%
EBV -
95%
Environmental Risk Factors for Multiple Sclerosis. Part I: The Role of Infection. Alberto Ascherio, MD, DrPH,1–3 and Kassandra L. Munger, MSc1
Goal
What is the relationship between the Epstein-Barr
Virus and Multiple Sclerosis ?
Mendelian Randomisation - Instrumental Variables
Unobserved
variable
EBV
MS
Unobserved
variable
EBV
MS
Mendelian Randomisation - Instrumental Variables
In an ideal world...
Unobserved
variable
MS
do EBV = x
Intervention to set this variable to
some chosen values
Unobserved
variable
do EBV = x
MS
Instrumental Variable
Use a variable that has an effect on MS only
through its effect on EBV.
Unobserved
variable
Z
EBV
Instrumental Variable
MS
Mendelian Randomisation
Unobserved
variable
Genetic
Variant
EBV
MS
Random allocation of genotypes at meiosis
Mendelian Randomisation
Unobserved
variable
Genetic
Variant
EBV
But there might be pleiotropy
MS
HLA Class I Single Nucleotide Polymorphism
Unobserved
variable
rs6457110
{0,1,2}
EBV
MS
{0,1}
{0,1}
Discrete Outcome (i.e. MS) :
Likelihood not identified
Usual Regression Based Approaches Give Biased Estimates of Causal Odds Ratio
Counterfactuals
Every individual is perfectly described by the data
of :
Universe 1
- What happens if they get infected?
and
Function : {Universe 1,Universe 2}
Universe 2
- What happens if they do not get infected?
{MS= 0 ,MS = 1}
Bayesian Approach - Parametrisation
Unobserved
variable
SNP
EBV
{0,1,2}
{0,1}
Possible Compliance
Types
MS
{0,1}
Possible Response Types
Confounder acts as a selector on the set of possible mappings
Fraction of the population in a given
category
A Clinician's Tool for Analyzing Non-compliance. David Maxwell Chickering and Judea Pearl.
Average Causal Effect - Counterfactual Queries
Prior on the fraction of the population falling in each one of the categories :
Average Causal Effect : What is the fraction of the population whose lives were changed by EBV?
ACE ranges from -1 to 1. If ACE < 0, then EBV protects from MS ; if ACE > 0, then EBV causes MS.
Counterfactual Query : Regardless of my genotype, if I am an individual infected by EBV and affected by MS,
what is the probability that I would not have been affected, had I not had EBV?
A Clinician's Tool for Analyzing Non-compliance. David Maxwell Chickering and Judea Pearl.
Posterior Distribution
Prior. ~ Dirichlet(1,1,...)
Posterior
Sensitivity to Prior Distribution
Prior
Posterior
~ Dir(1,0.8,1.2,...)
~ Dir(1,1,1,...)
A Counterfactual Query
Conclusions
●
Based on available data, there is a strong statistical evidence for a causal effect of
the Epstein-Barr virus, when infected as a young adult, on multiple sclerosis.
●
Model with transparent parameterisation and interpretation of the results (e.g.
counterfactual queries).
●
Deal with discrete graphical models => when the likelihood is not identified.
Further Work
●
Collect more data. Hard to find a seronegative cohort.
●
Use joint distribution : P(EBV, MS | SNP1, SNP2) to get a better handle on pleiotropy.
Acknowledgements
Supervision
Helpful Discussions
Luke Jostins (WTCHG, Oxford)
Gil McVean (Statistics, WTCHG, Oxford)
Robin Evans (Statistics, Oxford)
Loukas Moutsianas (WTCHG, Oxford)
Charlotte Houldcroft (Sanger, Cambridge)
Main References
A clinician's tool for analyzing non-compliance. D. M. Chickering, Judea Pearl
Transparent parametrizations of models for potential outcomes. Richardson, Evans and Robins
Code for the Gibbs Sampler
http://nbviewer.ipython.org/url/www.stats.ox.ac.uk/~frot/ACliniciansTool-LipidData.ipynb
Model
~ Dirichlet(1,1,...)
\nu
CR1
CR2
...
CRn
Obs1
Obs2
...
Obsn
P(cr_i | Data, \nu) : Draw from a multinomial
P(\nu | Data,cr_i) = P(\nu,cr_i) : Draw from a
Dirichlet (conjugate prior)
Pleiotropy