Multiple Sclerosis Journal - October 2017 - 1463

J Devorak, LE Mokry
MS.9,24 If summary statistics were not available for
an index SNP in either study, a highly correlated
proxy (r2 > 0.8) was selected first from the
Immunochip study and then from the IMSGC/
WTCCC2 study, if the former was unavailable.
Linkage disequilibrium (LD) for proxies was measured using UK10K samples (n = 3781).25
SNP validation
LD assessment. MR studies require that the SNPs not
be in LD, since strong correlations between selected
SNPs may bias results.20 To ensure that the adiponectin-associated SNPs met this requirement, LD was
measured between all selected SNPs using European
samples from the UK10K project using PLINK software version 1.90.26 SNPs were excluded from analyses if their measured LD was r2 > 0.05.
Pleiotropy assessment. MR analyses assume that the
SNPs influence the outcome (MS) solely through the
exposure of interest (adiponectin). To assess for the
presence of pleiotropy, MR-Egger regression was
performed as previously described.27 This approach
is based on Egger regression, which has been used to
examine publication bias in the meta-analysis
literature.28 In brief, the SNP's effect upon the exposure variable is plotted against its effect upon the outcome, where an intercept distinct from the origin
provides evidence for pleiotropic effects. Funnel plots
can also be used for visual inspection of symmetry. In
addition, a systematic PubMed literature search was
conducted to investigate possible pleiotropic mechanisms of the selected SNPs on MS, using a previously described method24 (S1 Methods). Finally,
pleiotropy was assessed by examining only the SNP
at ADIPOQ, which encodes adiponectin. Pleiotropy
is less likely to influence results at this locus, since it
is likely that genetic variation at ADIPOQ influences
adiponectin levels directly.29
Population stratification. To reduce this potential
source of bias, selected SNPs and summary statistics
for both adiponectin and MS were obtained from
analyses involving individuals of European descent
only. In addition, a literature search was conducted to
investigate potential residual population stratification
that may exist among European subgroups with
respect to adiponectin levels.30 To the best of our
knowledge, no epidemiological studies have investigated adiponectin levels across European subgroups;
therefore, mean adiponectin serum concentrations
from the ADIPOGen European cohorts were compared to investigate potential differences in population adiponectin levels across Europe. Shapiro-Wilk's
journals.sagepub.com/home/msj

test was used to assess normality of mean adiponectin
concentration for the following countries: United
Kingdom, United States, the Netherlands, Germany,
Italy, and Finland. Analysis of variance (ANOVA)
was then performed to investigate potential differences in adiponectin concentrations across these
countries. Shapiro-Wilk's test and ANOVA were performed using GraphPad Prism 6 software (GraphPad
Software Inc., La Jolla, CA, USA).
MR estimates
In this previously described two-sample MR study
design24,31 where independent SNPs evaluate the
association of exposure to genetically altered adiponectin levels with MS risk, MR estimates were
obtained by weighting each of the adiponectin-associated SNPs by the magnitude of its effect upon
natural-log-transformed adiponectin level. The
individual estimates were then meta-analysed using
a fixed-effects model to obtain a summary measure
for the effect of genetically increased adiponectin
on risk of MS.
Sensitivity analyses
If a given SNP violated any of the underlying assumptions of MR, MR estimates were re-calculated excluding that SNP. Further sensitivity analyses were
undertaken using (1) only the lead SNP from
ADIPOGen, located near the adiponectin-encoding
gene ADIPOQ, to reduce potential bias from pleiotropy;29 and (2) only the SNPs genotyped in both
ADIPOGen and either of the MS studies to reduce
potential bias from random error introduced by use of
proxy SNPs
All statistical analyses were performed using R version 3.2.2 software32 unless otherwise noted.
Results
SNP selection
ADIPOGen identified 12 SNPs as genome-wide
significant (p < 5 × 10−8) for adiponectin level in
European populations.21 Of these, none were genotyped directly in the Immunochip study; however,
four were found in the IMSGC/WTCCC2 GWAS:
rs1108842 (within GNL3), rs12922394 (within
CDH13), rs1597466 (near TSC22D2), and
rs2925979 (within CMIP) (Table 1). Proxies
(r2 > 0.80) were identified for six of the eight
remaining SNPs: one from the IMSGC Immunochip
study (rs6810075, near the adiponectin-encoding
1463
https://journals.sagepub.com/home/msj
Table of Contents for the Digital Edition of Multiple Sclerosis Journal - October 2017
