Shared Genetic Risk Factors of Intracranial, Abdominal, and Thoracic Aneurysms

20 Pages·2016·3.44 MB·English

ORIGINAL RESEARCH
Shared Genetic Risk Factors of Intracranial, Abdominal, and Thoracic
Aneurysms
Femke N. G. van ’t Hof, MD; Ynte M. Ruigrok, MD, PhD; Cue Hyunkyu Lee, MSc; Stephan Ripke, MD, PhD; Graig Anderson, MD;
Mariza de Andrade, PhD; Annette F. Baas, MD, PhD; Jan D. Blankensteijn, MD, PhD; Erwin P. B€ottinger, MD; Matthew J. Bown, MD;
Joseph Broderick, MD; Philippe Bijlenga, MD, PhD; David S. Carrell, PhD; Dana C. Crawford, PhD; David R. Crosslin, PhD;
Christian Ebeling, PhD; Johan G. Eriksson, MD; Myriam Fornage, PhD; Tatiana Foroud, PhD; Mikael von und zu Fraunberg, MD, PhD;
Christoph M. Friedrich, PhD; Em(cid:2)ılia I. Ga(cid:2)al, MD, PhD; Omri Gottesman, MD; Dong-Chuan Guo, PhD; Seamus C. Harrison, PhD, FRCS;
Juha Hernesniemi, MD, PhD; Albert Hofman, MD, PhD; Ituro Inoue, MD, PhD; Juha E. J€a€askel€ainen, MD, PhD; Gregory T. Jones, PhD;
Lambertus A. L. M. Kiemeney, PhD; Riku Kivisaari, MD, PhD; Nerissa Ko, MD; Seppo Koskinen, MD, PhD; Michiaki Kubo, MD, PhD;
Iftikhar J. Kullo, MD; Helena Kuivaniemi, MD, PhD; Mitja I. Kurki, PhD; Aki Laakso, MD, PhD; Dongbing Lai, MSc; Suzanne M. Leal, PhD;
Hanna Lehto, MD; Scott A. LeMaire, MD; Siew-Kee Low, PhD; Jennifer Malinowski, PhD, MS; Catherine A. McCarty, PhD;
D
o Dianna M. Milewicz, MD, PhD; Thomas H. Mosley, PhD; Yusuke Nakamura, MD, PhD; Hirofumi Nakaoka, PhD; Mika Niemel€a, MD, PhD;
w
nlo Jennifer Pacheco, BA; Peggy L. Peissig, PhD; Joanna Pera, MD, PhD; Laura Rasmussen-Torvik, PhD; Marylyn D. Ritchie, PhD;
ad Fernando Rivadeneira, MD, PhD; Andre M. van Rij, MD; Regie Lyn P. Santos-Cortez, MD, PhD; Athanasios Saratzis, PhD;
e
d from ASigtnaieHs.zkVaerSmloewuliekn,,MPDh,DP;hGDa;oATt.suWsahnigT,aPkahhDa;sAhni,eMurDys,mPhCDo;nGseorratirudmT;roVmaspc,uPlahrDR;eAsnedarr(cid:2)echG.CUonitstoerrltiinudmeno,fPNheDw; SZheeaflaalnidS;*. VBeurhmma,HManS,;PhD;
h Gabri€el J. E. Rinkel, MD, FRCP(E); Paul I. W. de Bakker, PhD
ttp
://ja
h
a.a Background-—Intracranial aneurysms (IAs), abdominal aortic aneurysms (AAAs), and thoracic aortic aneurysms (TAAs) all have a
h
ajo familial predisposition.Giventhataneurysmtypesareknowntoco-occur,wehypothesizedthattheremaybesharedgeneticrisk
u
rn factors for IAs, AAAs, and TAAs.
a
ls
.o MethodsandResults-—Weperformedamega-analysisof1000GenomesProject-imputedgenome-wideassociationstudy(GWAS)
rg
a/ dataof4previouslypublishedaneurysmcohorts:2IAcohorts(intotal1516cases,4305controls),1AAAcohort(818cases,3004
t U controls),and1TAAcohort(760cases,2212controls),andobservedassociationsof4knownIA,AAA,and/orTAAriskloci(9p21,
n
ive 18q11,15q21,and2q33)withconsistenteffectdirectionsinall4cohorts.WecalculatedpolygenicscoresbasedonIA-,AAA-,and
rs
ite TAA-associated SNPs and tested thesescores for associationtocase-controlstatus intheotheraneurysmcohorts; this revealed
its no shared polygenic effects. Similarly, linkage disequilibrium–score regression analyses did not show significant correlations
b
iblio between any pair of aneurysm subtypes. Last, we evaluated the evidence for 14 previously published aneurysm risk single-
th nucleotide polymorphisms through collaboration in extended aneurysm cohorts, with a total of 6548 cases and 16 843 controls
e
ek (IA)and4391casesand37 904controls(AAA),andfoundnominallysignificantassociationsforIArisklocus18q11nearRBBP8to
U
tre AAA (odds ratio [OR]=1.11; P=4.1910(cid:1)5) and for TAA risk locus 15q21 near FBN1 to AAA (OR=1.07; P=1.1910(cid:1)3).
c
h
t o Conclusions-—AlthoughtherewasnoevidenceforpolygenicoverlapbetweenIAs,AAAs,andTAAs,wefoundnominallysignificant
n
S effects of two established risk loci for IAs and TAAs in AAAs. These two loci will require further replication. (J Am Heart Assoc.
e
p
te 2016;5:e002603 doi: 10.1161/JAHA.115.002603)
m
b
er 1 Key Words: abdominal aortic aneurysm(cid:129)genome wide association study(cid:129)intracranial aneurysm(cid:129)thoracic aortic aneurysm
6
, 2
0
1
6
I
ntracranial aneurysms (IAs), abdominal aortic aneurysms all lead to rupture with a high case fatality rate.1–3 A
(AAAs), and thoracic aortic aneurysms (TAAs) are 3 co-occurrence of AAA and TAA is known,4,5 and a weak
different forms of arterial vessel wall dilatations, which can co-occurrence of IA and AAA has also been suggested.6,7
Theauthors’affiliationsareprovidedonpage16ofthearticle.
*AcompletelistoftheAneurysmConsortiumandtheVascularResearchConsortiumofNewZealandcanbefoundinanAppendixattheendofthearticle.
Correspondenceto:PaulI.W.deBakker,PhD,DepartmentsofMedicalGeneticsandofEpidemiology,UniversityMedicalCenterUtrecht,Universiteitsweg100,
3584CGUtrecht,TheNetherlands.E-mail:[email protected]
ReceivedSeptember23,2015;acceptedMarch16,2016.
ª2016TheAuthors.PublishedonbehalfoftheAmericanHeartAssociation,Inc.,byWileyBlackwell.ThisisanopenaccessarticleunderthetermsoftheCreative
CommonsAttribution-NonCommercialLicense,whichpermitsuse,distributionandreproductioninanymedium,providedtheoriginalworkisproperlycitedandis
notusedforcommercialpurposes.
DOI:10.1161/JAHA.115.002603 JournaloftheAmericanHeartAssociation 1
SharedGeneticRiskforAneurysms van’tHofetal
O
R
Furthermore, IA, TAA, and AAA cluster within affected controls were extracted from the Helsinki Birth Cohort Study IG
I
N
families,8,9 suggesting a shared genetic background of these (HBCS).21 Additionally, 651 controls were extracted from A
L
3 diseases. anonymous donors from Kuopio University Hospital and R
E
Genome-wideassociationstudies(GWASs)haverevealeda Helsinki and from the Health 2000 study (H2000).22 All SE
A
handful of genetic risk factors for IA,10–12 AAA,13–16 and cases were genotyped on Illumina CNV370 Duo BeadChips, RC
H
TAA,17 andoftheseidentified loci,locus 9p21 near CDKN2A, and controls were genotyped on Illumina HumanHap550
CDKN2B, and CDKN2BAS appears to be shared by IA and BeadChips (HBCS) and on Illumina CNV370 Duo BeadChips
AAA.11,18 However, the heritability explained by the risk loci (anonymous donors and H2000).
identified to date is low for all 3 diseases, suggesting that The Dutch and the Finnish IA cohort both included cases
many aneurysm risk loci remain to be discovered. Among with ruptured and unruptured IA. Ruptured IA cases were
these unknown and known risk loci, there could be shared definedbysymptomssuggestiveofsubarachnoidhemorrhage
risk loci for multiple types of aneurysms. (SAH) combined with subarachnoid blood on a computed
In this study, we searched for shared genetic risk factors tomography (CT) scan and a proven IA at angiography
for aneurysms. We combined individual participant GWAS (conventional angiogram, CT- or magnetic resonance [MR]
Do data from a Dutch and a Finnish IA cohort,12 a Dutch AAA angiogram).Unruptured IAcaseswereidentifiedbyCTorMR
w
nlo cohort,14 and a TAA cohort from the United States17 for a angiography or conventional angiography in the absence of
a
d GWAS mega-analysis acrossthesetraits.Inacomplementary clinicalorradiologicalsignsofSAH.PatientswithfusiformIA,
e
d
fro approach, we performed a polygenic analysis to test groups possible traumatic SAH, and polycystic kidney disease were
m
h of single-nucleotide polymorphisms (SNPs) for a joint effect excluded.
ttp onriskacross diseases.Finally,wetestedtheeffectofbona- The AAA cohort consisted of 859 cases, predominantly
://ja fide risk SNPs from previously published IA, AAA, and TAA with unruptured AAA. These cases were recruited from 8
h
a
.a GWASs on the other aneurysm types by meta-analyzing medical centers in The Netherlands, mainly when individuals
h
a
jo summarystatistics intheGWAS cohortsofIA,AAA,andTAA, visited their vascular surgeon in the outpatient clinic or, in
u
rna extended by association results of additional IA and AAA some cases, during hospital admission for elective or
ls
.o GWAS cohorts. emergencyAAAsurgery.AnAAAwasdefinedasaninfrarenal
rg
a/ aorta diameter of ≥30 mm. Mean AAA diameter was
t U
n 58.4 mm. Of these patients, 530 had undergone surgery,
iv
e Methods including 43 with rupture. Genotyping was performed on
rs
ite Illumina HumanHap610 chips.14 As controls, we included
its Study Populations
bib 3110 Dutch subjects who were recruited as part of the
lio FortheGWASmega-analysisandpolygenicanalysis,weused Nijmegen Biomedical Study (n=1832) and the Nijmegen
th
ee data of subjects genotyped in previously published GWAS Bladder Cancer Study (n=1278).19,20 These controls were
k
U cohortsofaneurysmcasesandcontrols:2IAcohorts,1from genotyped on Illumina CNV370 Duo BeadChips.
trech the Netherlands12,19,20 and 1 from Finland,11,12,21,22 1 AAA The TAA cohort consisted of 765 cases with either an
t on cohort from the Netherlands14,18 and 1 TAA cohort from the ascending TAA without dissection (n=401) or with a type A
Se United States.17 All studies were approved by the relevant and/ortypeBaorticdissection(n=364).Allcasesweremore
p
tem medical ethical committees, and all participants provided than 30 years old, did not have a first-degree relative with
b
er 1 written informed consent. All study populations were previ- TAA or dissection, and had no evidence of a syndromic form
6, 2 ouslydescribedindetail.12,17,19,20Belowisabriefdescription of TAA or dissection. Samples were genotyped with Illumina
01 of each study population. CNV370-Quad BeadChips. Controls (n=2229) were included
6
IAcasesintheDutchcohort(n=786)wereadmittedtothe from the Wellcome Trust Case-Control Consortium (WTCCC)
UniversityMedicalCenterUtrecht,(Utrecht,TheNetherlands) 1958 Birth Cohort (n=1355), and from the US National
between 1997 and 2011. All cases were genotyped on Institute of Neurological Disorders and Stroke (NINDS)
Illumina CNV370 Duo BeadChips (Illumina, San Diego, CA). Repository’s Neurologically Normal Control Collection
Controls (n=2089) were ascertained by the Rotterdam Study, (n=874).
apopulation-basedcohortofsubjectsage45 yearsandolder
recruited from a district in Rotterdam (The Netherlands).
These controls were genotyped on Illumina HumanHap550 Quality Control
BeadChips.23
TheFinnishIAcohortconsistedof790casestreatedatthe We performed quality control (QC) in each of the 4 cohorts
Helsinki and Kuopio University hospitals and 2396 controls separately, using PLINK software (version 1.07).25 After
that were genetically matched to cases.24 Of these, 1666 removal of SNPs with A/T or C/G alleles and SNPs that
DOI:10.1161/JAHA.115.002603 JournaloftheAmericanHeartAssociation 2
SharedGeneticRiskforAneurysms van’tHofetal
O
R
were not called in any individual, we performed sample QC Using these common, high-quality SNPs, we performed IG
I
N
and SNP QC. principal components (PCs) analysis using EIGENSTRAT on A
L
SampleQCwasperformedaftermergingcasesandcontrols, the remaining study subjects and HapMap-CEU subjects. We R
E
using a subset of common, high-quality SNPs (as defined by excluded SNPs from 3 regions with known long-distance LD: SE
A
SNPswithoutdeviationfromHardy-Weinbergequilibrium[HWE; the major histocompatibility region (chr6: 25.8–36 Mbp); the RC
H
P>0.001], with high minor allele frequency [MAF; >20%], and chromosome 8 inversion (chr8: 6–16 Mbp); and a chromo-
with low rate of missing genotypes [<1%]). Linkage disequilib- some 17 region (chr17: 40–45 Mbp). We created PC plots
rium (LD) pruning (r2>0.5) was performed. Subjects were with the first 4 PCs, using R software (version 2.11; R
removed based on the following 3 criteria: missing genotypes Foundation for Statistical Computing, Vienna, Austria).26
(subjects with a call rate below 95% were removed); heterozy- Based on visual inspection of these plots, we excluded
gosity (subjects were excluded if the inbreeding coefficient subjects that appeared to be outliers with respect to the
deviated more than 3 SDs from the mean); and cryptic CEU (Utah residents with ancestry from northern and
relatedness (by calculating identity-by-descent [IBD] for each western Europe) or the study population. After outlier
pairofindividuals).IneachpairwithanIBDproportionof>20%, removal, we recomputed PCs for them to be included as
D
o a subject was excluded, if it exhibited distant relatedness with covariates in the logistic regression models. PC plots after
w
nlo morethan1individual.Forcase-controlpairs,weremovedthe outlier removal are shown in Figure 1.
a
d control subject. In the case-case or control-control pairs, the After sample QC, we excluded SNPs with more than 2%
e
d
fro subject with the lowest call rate was excluded. missing genotypes, MAF <1%, missing genotype rate higher
m
h
ttp
://ja
h
a
.a A B
h
a
jo
u
rn
a
ls
.o
rg
a/
t U
n
iv
e
rs
ite
its
b
ib
lio
th
e
e
k
U
tre
c
h
t o
n C D
S
e
p
te
m
b
e
r 1
6
, 2
0
1
6
Figure 1. Principalcomponentanalysis(PCA)plotsofIA,AAA,andTAAGWAScohorts.(A)DutchIAGWAScohort;(B)FinnishIAGWAScohort;
(C)AAAGWAScohort;(D)TAAGWAScohort.ThesefiguresshowPCA2valuesplottedagainstPCA1valuesforeachindividualinthe4GWAS
cohorts(DutchIAandFinnishIA,AAA,andTAA)oftheaneurysmmega-analysis, afterremovalofoutliersfromqualitycontrol.AAAindicates
abdominal aorticaneurysm; GWAS, genome-wideassociation study;IA, intracranial aneurysm; TAA,thoracic aorticaneurysm.
DOI:10.1161/JAHA.115.002603 JournaloftheAmericanHeartAssociation 3
SharedGeneticRiskforAneurysms van’tHofetal
O
R
than MAF, and HWE deviation (P<0.001). Because cases and We pruned the SNPs genotyped or imputed in the IG
I
N
controlshadbeengenotypedseparately,weperformedthese discovery sample, using an LD threshold of r2>0.1. For each A
L
QC steps in each study cohort separately and again after genomic region, we chose SNPs with the lowest P values in R
E
S
merging cases and controls. We also removed SNPs with a theGWAS ofthediscovery sample, inorder to retain asetof E
A
differential degree of missing genotypes between cases and independent, maximally associated SNPs. Next, we created RC
controls (P<1910(cid:1)5; chi-squared test). setsofSNPswithdiseaseassociationinthediscoverysample H
at 12 different significance thresholds, increasing from
P<5910(cid:1)8 to P<0.5. For each SNP set, we calculated a
Imputation
polygenic risk score in each individual of target sample as
For each case-control data set, we performed genotype follows:
imputation using the prephasing/imputation step-wise
Genetic score¼b x þb x þ(cid:3)(cid:3)(cid:3)þb x ;
approach implemented in IMPUTE2 and SHAPEIT (chunk size 1 1 2 2 n n
of 3 Mb and default parameters).27,28 The imputation refer- wherex istheestimatedalleledosage(between0and2)ina
i
ence set consisted of 2184 phased haplotypes from the full givenindividual,andb istheeffectsizefromtheGWASinthe
D i
o 1000GenomesProjectdataset(February2012;40 318 253 discovery sample for the ith SNP. We tested the association
w
nlo variants). All genomic locations are given in NCBI Build between these polygenic risk scores and case-control status
a
de 37/UCSC hg19 coordinates. After imputation, SNPs with an in the target sample using logistic regression, adjusting for
d
fro imputation accuracy score <0.6 or MAF <0.5% were thefirst4PCsinthetargetsample.Foranalysesinvolvingthe
m
h excluded. combined IA cohort as target sample, we also adjusted for
ttp://jah panodpuAlaAtiAonco(FhionrntisshasortaDrugtectha).nFdordisacnoavlyesreyssianmvopllvei,ngwethealsIAo
a GWAS Mega-Analysis Across IA, AAA, and TAA
.ah adjusted for the genotype of the known shared IA/AAA risk
a
jou We performed a mega-analysis on all 4 GWAS cohorts. SNPs on locus 9p21.11,18
rna Association testing was carried out in PLINK25 using
ls
.o imputed SNP dosages. We included as covariates the first
rg
at Un/ 4conPtCrosl adnadtaasnet.inSdNicPastowr ivthariPa<b5le9t1o0(cid:1)a8djwusetrefocronesaicdherecdasaes- TLoaobkleup1.ofOKvneorvwienwIAo,fASAtuAd,yaPnodpTuAlaAtioRnisskUSsNedPsfortheSpecific
ive genome-wide significant. We also performed a GWAS on
rs
iteits each cohort separately and a combined analysis of only the Cohort Cases(n) Controls(n)
bib Dutch and Finnish IA cohort, which was needed for the IA
lioth polygenic analysis as described below. We calculated Netherlands 717 1987
eek genomic inflation factors (kGC) for each GWAS and the Finland 799 2317
U mega-analysis, defined as the ratio of the median of
tre Japan1 288 194
ch the empirically observed distribution of the test statistic to
t on the expected median.29 Japan2 1383 5484
Se We calculated the statistical power for detecting a USA+ 2617 2548
p
tem significant association (P<5910(cid:1)8) in the mega-analysis @neurIST 717 3296
b
er 1 using the genetic power calculator.30 In case of a risk Total 6548 16 843
6
, 2 allele frequency of 10%, the resulting power is 0.03% at a AAA
0
1 relative risk of 1.1 per allele, and 68.7% at a relative risk
6 Netherlands 812 2998
of 1.3 per allele, assuming additive effects. In case of a
Iceland 430 27 712
higher risk allele frequency of 20%, the power is 0.3% at a
USA 724 1604
relative risk of 1.1 and 98.9% at a relative risk of 1.3 per
allele. AneurysmConsortium 1846 5605
NewZealand1 608 612
NewZealand2 400 384
Polygenic Analysis
Total 4391 37 904
We performed polygenic analysis as previously described.31
TAA
WeusedtheIAcohorts(DutchandFinnishcohortscombined)
USA 760 2212
as a discovery sample and the AAA and TAA cohorts as 2
separate target samples, and vice versa, in all possible
AAAindicatesabdominalaorticaneurysm;IA,intracranialaneurysm;SNPs,single-
combinations (6 in total). nucleotidepolymorphisms;TAA,thoracicaorticaneurysm.
DOI:10.1161/JAHA.115.002603 JournaloftheAmericanHeartAssociation 4
SharedGeneticRiskforAneurysms van’tHofetal
O
R
LD Score Regression: Heritability Estimation and applied to the same 4 GWAS datasets as used in the GWAS IG
I
N
Genetic Correlation Analyses mega-analysisandpolygenicanalysisdescribedabove(Dutch A
L
IA and Finnish IA, AAA, and TAA), but for this analysis we R
E
S
We used LD score regression (LDSC) for heritability estima- only included genotyped SNPs (after QC), and not imputed E
A
tion and genetic correlation analysis.32 These analyses were SNPs. RC
H
GWAScohorts
IA(Netherlands) IA(Finland) AAA TAA
786 cases 790 cases 859 cases 765 cases
2089controls 2396controls 3110controls 2229controls
Genotyping
D
o
w IlluminaCNV370 IlluminaCNV370 Illumina IlluminaCNV370-
n
lo Duo/ Duo/ HumanHap610/ QuadBeadChip
a
d HumanHap550 HumanHap550 CNV370Duo
e
d fro BeadChip BeadChip BeadChip
m
h
ttp
://ja
h
a
.a
h
ajo SampleQC,SNPQCandimputationperGWAScohort
u
rn
a
ls
.o
rg
a/
t U 1.GWASmega-analysis
n
iv
ers 3094cases,9507 controls
ite
its Logisticregression:
bib Diseasestatus(aneurysm vscontrol)<-> SNPdosage+PCs+ GWAScohortindicatorvariable
lio
th
e
e
k U 2.Polygenicanalysis
trec IA(Netherlands+Finland) AAA TAA
h
t o
n
S
e
p
te
m Logisticregressionforeachcombination:
b
er 1 Diseasestatus(IA/AAA /TAAvscontrol)<-> polygenicscore+ PCs+ sex
6
, 2
0
1
6
3.Specificlook-up ofknownaneurysmrisk SNPsinIA,AAAandTAA
6 IAGWAScohorts 6 AAAGWAScohorts 1 TAAGWAScohort
6548cases,16843 controls 4391cases,37904 controls 760 cases,2212 controls
Extractionof14aneurysmriskSNPsfromGWAS results
(forIAandAAA: inverse-variancefixed-effectsmeta-analysis)
Figure 2. Work-flowfigure.Thisfiguregivesanoverviewofthestudymethods.AAAindicatesabdominal
aortic aneurysm; CNV, copy number variation; GWAS, genome-wide association study; IA, intracranial
aneurysm; PCs, principal components; QC, quality control; SNP, single-nucleotide polymorphism; TAA,
thoracic aorticaneurysm.
DOI:10.1161/JAHA.115.002603 JournaloftheAmericanHeartAssociation 5
SharedGeneticRiskforAneurysms van’tHofetal
O
R
LDSC is a new approach that implements LD-score significance oftheseSNPsinseparate case-controldata sets IG
I
N
weighted linear regression methods to estimate the variance for IA, AAA and TAA beyond the individual-participant data A
L
(and covariance) explained by all SNPs on the whole genome sets already described above. We did not have access to all R
E
S
for a complex trait. This method can distinguish true genotypedataofthesedatasets,butweobtainedassociation E
A
polygenicityfromconfoundingeffectsattributedtopopulation resultsforthesecandidateSNPsonly.ForIA,weusedGWAS RC
H
structure and cryptic relatedness. Under a polygenic model, resultsof2Japanesecohorts,11acombinedcohortofseveral
themore-geneticvariantsanindex varianttags,themorethe previously IA studies that recruited subjects from mainly
probability that the index variant will be significant.32 Under NorthAmerica,butalsoPolandandAustralia (furtherreferred
this reasoning, the expected v2 statistic of variant j can be toas“USA+”),[email protected],weused
constructed as follows: GWAS results of a cohort from Iceland, the United Kingdom,
theUnitedStates,and2cohortsfromNewZealand.15,16,36,37
E½v2j‘j(cid:4)¼Nh2‘j=MþNaþ1; For TAA, no additional GWAS cohorts are currently available,
to our knowledge. The sample sizes and other details per
whereNisthesamplesize;MisthenumberofSNPs,sothat
cohort are listed in Table 1. In total, the target samples
D h2/M is the average heritability per SNP; a measures the
o available for this lookup analysis contained 6548 cases and
w contribution of confounding biases, like cryptic relatedness
nload and population stratification; and ‘j ¼Pkr2jk is the sum of 1A6AA8,4a3ndco7n6t0rolcsasfoersIaAn,d43292112cacsoenstraonlsdf3or7T9A0A4. controls for
ed the r2 values to all variants that a variant j tags, which is
http from ccaalllceudlattehde uLsDingscroerfeereonfcevapraiannetl jc.onTthaeiniLnDg wschoorlee gceannombee eexffteWecnetdsceodmmIeAbtaian-anenddatAlhyAesAiGs.cWoFAhoSorrrteesasuucshltisnSgfoNarPnt,hienwvseeerlfiseercs-tvteadcriSaalNnccPuelsaitfinexdtehdez-
://ja information of the population. For the analysis of the scoresfromtheprovidedPvaluesofeachGWAS,andsummed
h
a European population, we used the LD estimates that are
.a thezscoresacrossallstudiesusingtheeffectivesamplesizeof
h
a from the European-ancestry samples in the 1000 Genomes
jo eachstudyasweights.38Theresultingzscoreswereconverted
urna Project.33 Details were elaborated from the LDSC’s GitHub into chi-square values and 2-sided P values. We applied
ls repository.32
.o Bonferroni correction for performing 28 association tests (14
rg In addition to estimating heritability using LDSC, we also
a/ selected SNPs tested in 2 aneurysm types) and considered
t Univ upsaeirdoLfD3SdCisteoaesestsim(AaAteA,gIeAn,eatincdcToArrAe)l.a3t4ioWne(qin)vbeesttwigeaetnedeathche associations with P<1.8910(cid:1)3 (0.05/28) as significant.
e Figure 2givesanoverviewofthemethodsdescribedabove.
rsite Dutch and Finnish IA cohort separately in this analysis, and
its alsotestedthecorrelationbetweenthesetwoIAdatasets,to
b
iblio evaluate the genetic correlation within 1 disease.
th Results
e
e
k
U
tre Effects of Previously Established Risk SNPs Study Populations
c
ht o We investigated the effect of established IA,10–12 AAA,13–16 Table 2showsthenumbersofcases,controls,andSNPsofall
n
S and TAA17 risk SNPs from previous GWASs in each of the 4 cohorts after QC and imputation. Quantile-quantile (QQ)
e
p
te other aneurysm types. We looked up the effect sizes and plots for each GWAS per cohort are shown in Figure 3.
m
b
e
r 1
6
, 20 Table 2. Baseline Characteristics After Quality Control and Imputation of the Study Populations Used for the Mega-Analysis and
1
6 Polygenic Analysis of IA, AAA, and TAA GWAS Data
StudyCohort
Characteristics IA(Netherlands) IA(Finland) AAA TAA
Cases,n 717 799 818 760
Women,% 64.3 57.8 10.5 34.3
Controls,n 1988 2317 3004 2212
Women,% 56.2 57.2 37.7 53.0
SNPs,n 10 683 725 10 524 028 10 684 772 10 750 239
Genomicinflationfactor 1.10 1.06 1.04 1.05
AAAindicatesabdominalaorticaneurysm;GWAS,genome-wideassociationstudy;IA,intracranialaneurysm;SNP,single-nucleotidepolymorphism;TAA,thoracicaorticaneurysm.
DOI:10.1161/JAHA.115.002603 JournaloftheAmericanHeartAssociation 6
SharedGeneticRiskforAneurysms van’tHofetal
O
R
IG
I
A B NA
L
R
E
S
E
A
R
C
H
D
o
w
n
loa C D
d
e
d
fro
m
h
ttp
://ja
h
a
.a
h
a
jo
u
rn
a
ls
.o
rg
a/
t U
n
iv
e
rs
ite
its
b
ib
lio
the E
e
k
U
tre
c
h
t o
n
S
e
p
te
m
b
e
r 1
6
, 2
0
1
6
Figure 3. Quantile-quantile (QQ) plots of IA, AAA, and TAA GWAS and the aneurysm mega-analysis. A, Dutch IA
GWAS; (B) Finnish IA GWAS; (C) AAA GWAS; (D) TAA GWAS; (E) aneurysm mega-analysis. These QQ plots show the
observed distribution of P values (black dots) plotted against the expected distribution of P values (blue line) on a
negativelog scale,foreachofthe4GWAS(DutchIAandFinnishIA,AAA,andTAA)andforallcohortscombinedin
10
the aneurysm mega-analysis. Genomic inflation factors (k ) per study, defined as the ratio of the median of the
GC
empiricallyobserveddistributionoftheteststatistictotheexpectedmedian,arealsoshown.AAAindicatesabdominal
aorticaneurysm; GWAS, genome-wideassociation study;IA, intracranial aneurysm; TAA,thoracic aortic aneurysm.
DOI:10.1161/JAHA.115.002603 JournaloftheAmericanHeartAssociation 7
SharedGeneticRiskforAneurysms van’tHofetal
O
R
GWAS Mega-Analysis Across IA, AAA, and TAA (rs700651), which was previously found to be associated IG
I
N
withIAinaDutch,Finnish,andJapanesepopulation,12butdid A
In total, 3094 cases, 9507 controls, and 9 245 988 SNPs L
not reach genome-wide significance after adding other R
were available for the mega-analysis across all 4 aneurysm populations of IA patients.11 ESE
cohorts (k =1.06). The results of this mega-analysis are A
GC R
C
shown in a Manhattan plot (Figure 4). We found 4 genome- H
Polygenic Analysis
wide significant loci, though all these loci were previously
describedasrisklociforIA,AAA,and/orTAA.Thedirectionof Next, we investigated whether groups of SNPs associated
effect for these loci was consistent across all four aneurysm with 1 type of aneurysm (eg, IA) were also associated with
cohorts (see forest plots in Figure 5). First, SNPs at the the other types (eg, AAA or TAA). The results of these
known IA and AAA risk locus 9p21 near CDKN2A, CDKN2B, polygenicanalyses with the IA,AAA, and TAA GWAS cohorts
and CDKN2BAS were associated.11,18 The strongest associ- are shown in Tables 3 through 8 and in Figure 6. No SNP
ation at this locus was found for rs7866503, with sets with a significant joint effect on another aneurysm type
P=2.1910(cid:1)13. The second association was found for SNPs were observed, except for a small group of 7 SNPs
D at the known IA risk locus 18q11 near RBBP8,11 with the associatedwithIAwithP<10(cid:1)6.ThisSNPsetwasassociated
ow strongestassociationforrs8087799(P=1.6910(cid:1)9).Thethird with AAA with P=5910(cid:1)3 (Table 7). When taking a closer
n
loa association was found for SNPs at the known TAA risk locus look at this SNP set, it appears that the polygenic associ-
d
ed 15q21 near FBN1,17 with the strongest association for ation is driven by 2 SNPs (rs36071109 at 2q33;
from rs595222 (P=1.0910(cid:1)8). The fourth association was found PAAA=2.3910(cid:1)3; and rs4330012 at 18q11; P (cursive)AAA=
http for rs919433 (P=4.6910(cid:1)8), which is located at 2q33 near 1.3910(cid:1)2), both in very strong LD to genome-wide signif-
://ja ANKRD44. The same SNP was previously found to be icant SNPs in our mega-analysis described above. (The
ha associated with IA in a Finnish and Dutch population.24 This pleiotropic 9p21 locus shared by IA and AAA had already
.a
ha SNP is also in strong LD (r2=0.7) with a nearby SNP been adjusted for.)
jo
u
rn
a
ls
.o
rg
a/
t U
n
iv
e
rs
ite
its
b
ib
lio
th
e
e
k
U
tre
c
h
t o
n
S
e
p
te
m
b
e
r 1
6
, 2
0
1
6
Figure 4. ManhattanplotoftheaneurysmGWASmega-analysis.ThisManhattanplotshowsthePvalues
of all SNPs with an association with P<10(cid:1)4 to disease (IA, AAA, or TAA). P values on the y-axis are
presented on an inverse log scale. The x-axis represents the genomic position of each SNP. The red
horizontal line represents P=5910(cid:1)8, the cut-off value for genome-wide association. Index SNPs with
P<1910(cid:1)6aredepictedasdiamonds,whereasSNPsinthesameLDblockastheseSNPsaredepictedas
yellowandreddots.AAAindicatesabdominalaorticaneurysm;GWAS,genome-wideassociationstudy;IA,
intracranialaneurysm;LD,linkagedisequilibrium;SNP,single-nucleotidepolymorphism;TAA,thoracicaortic
aneurysm.
DOI:10.1161/JAHA.115.002603 JournaloftheAmericanHeartAssociation 8
SharedGeneticRiskforAneurysms van’tHofetal
O
R
IG
A I
N
A
L
R
E
S
E
A
R
C
H
B
D
o
w
n
lo
a
d
e
d
fro
m
h
ttp
://ja
h
a
.a
h
a C
jo
u
rn
a
ls
.o
rg
a/
t U
n
iv
e
rs
ite
its
b
ib
lio
th
e
e
k
U
tre
c D
h
t o
n
S
e
p
te
m
b
e
r 1
6
, 2
0
1
6
Figure 5. Forest plots of significant SNPs from the aneurysm mega-analysis. A, rs7866503 (risk
alleleT)atlocus9p21;(B)rs8087799(riskalleleA)atlocus18q11;(C)rs595244(riskalleleT)atlocus
15q21;(D) rs919433 (risk allele A) atlocus 2q33.These figures show forest plotsfor eachof the4
genome-wide significant SNPs from the mega-analysis of IA, AAA, and TAA GWAS cohorts. AAA
indicatesabdominalaorticaneurysm;f_ca(n),riskallelefrequencyincases(numberofcases);f_co(n),
riskallelefrequencyincontrols(numberofcontrols);het_P,Pvalueforheterogeneity;IA,intracranial
aneurysm;info,imputationaccuracyscore;ln(OR),naturallogofoddsratio;OR,oddsratio;P_value,P
valueforassociationofriskalleleinaneurysmcohort;SNP,single-nucleotidepolymorphism;STDerr,
standard error; TAA,thoracic aorticaneurysm.
DOI:10.1161/JAHA.115.002603 JournaloftheAmericanHeartAssociation 9
SharedGeneticRiskforAneurysms van’tHofetal
O
R
Table 3. PolygenicAnalysis:AAA-AssociatedSNPsTestedfor Table 4. PolygenicAnalysis:TAA-AssociatedSNPsTestedfor IG
I
N
Association With TAA Case-Control Status Association With AAA Case-Control Status A
L
R
E
Threshold(P)* SNPs(n)† PValue‡ DirectionofEffect Threshold(P)* SNPs(n)† PValue‡ DirectionofEffect SE
A
5910(cid:1)8 0 NA NA 5910(cid:1)8 1 0.581 + R
C
H
1910(cid:1)7 1 0.269 + 1910(cid:1)7 1 0.581 +
1910(cid:1)6 4 0.262 + 1910(cid:1)6 1 0.581 +
1910(cid:1)5 21 0.481 + 1910(cid:1)5 25 0.853 +
1910(cid:1)4 181 0.979 + 1910(cid:1)4 148 0.971 (cid:1)
1910(cid:1)3 1417 0.815 + 1910(cid:1)3 1350 0.292 +
0.01 10 075 0.294 + 0.01 10 021 0.579 +
0.05 38 528 0.549 + 0.05 39 120 0.902 (cid:1)
0.1 67 187 0.614 (cid:1) 0.1 67 736 0.944 (cid:1)
D
o 0.2 113 644 0.823 (cid:1) 0.2 114 125 0.815 (cid:1)
w
n
lo 0.3 151 807 0.773 (cid:1) 0.3 151 712 0.956 (cid:1)
a
d
ed 0.4 183 891 0.765 (cid:1) 0.4 183 843 0.958 (cid:1)
from 0.5 211 723 0.611 (cid:1) 0.5 211 170 0.965 (cid:1)
h
ttp 10(cid:1)4to10(cid:1)3 1236 0.787 + 10(cid:1)4to10(cid:1)3 1202 0.252 +
://ja 10(cid:1)3to0.01 8658 0.273 + 10(cid:1)3to0.01 8671 0.877 (cid:1)
h
a
.a 0.01to0.05 28 453 0.972 (cid:1) 0.01to0.05 29 099 0.801 (cid:1)
h
a
jou 0.05to0.1 28 659 0.964 (cid:1) 0.05to0.1 28 616 0.953 (cid:1)
rn
als 0.1to0.2 46 457 0.115 (cid:1) 0.1to0.2 46 389 0.643 +
.o
rg 0.2to0.3 38 163 0.745 (cid:1) 0.2to0.3 37 587 0.255 (cid:1)
a/
t U 0.3to0.4 32 084 0.904 (cid:1) 0.3to0.4 32 131 0.997 (cid:1)
n
iv 0.4to0.5 27 832 0.0915 (cid:1) 0.4to0.5 27 327 0.422 +
e
rs
ite
its AAAindicatesabdominalaorticaneurysm;IA,intracranialaneurysm;NA,notapplicable; AAAindicatesabdominalaorticaneurysm;IA,intracranialaneurysm;SNPs,
b
ib SNPs,single-nucleotidepolymorphisms;TAA,thoracicaorticaneurysm. single-nucleotidepolymorphisms;TAA,thoracicaorticaneurysm.
lio *PvaluethresholdforinclusionofIA-,AAA-,orTAA-associatedSNPsineachpolygenic *PvaluethresholdforinclusionofIA-,AAA-,orTAA-associatedSNPsineachpolygenic
the model. model.
ek †NumberofSNPsincludedineachpolygenicmodel. †NumberofSNPsincludedineachpolygenicmodel.
U ‡PvalueforassociationofthepolygenicmodelwithIA,AAAorTAAcase-controlstatus. ‡PvalueforassociationofthepolygenicmodelwithIA,AAAorTAAcase-controlstatus.
tre
c
h
t o
n
S
e correlation was either close to zero or positive, but the SE
p
te LD Score Regression: Heritability Estimation and
m values were large. As expected, we observed the largest
ber 1 Genetic Correlation Analyses genetic correlation between the 2 cohorts of the same IA
6, 2 We attempted to analyze heritability of the 3 aneurysm disease (q=1.59). However, this correlation was also not
01 subtypes using LDSC. The estimated heritability was 0.15 significant (P=0.09; SE=0.93).
6
for AAA, 0.31 for the Finnish IA cohort, 0.34 for the Dutch
IA cohort, and 0.40 for TAA (Table 9). These estimates are
Effects of Previously Established Risk SNPs
smaller than the twin-based estimates as reported in the
literature (0.41 for IA,39 0.70 for AAA,16 and unknown for Table 11 shows the results of the lookup of previously
TAA), but the SEs are large, so the differences are not published aneurysm risk SNPs in IA, AAA, and TAA GWAS
statistically significant. Note that methods estimating results. Besides the 2 IA and AAA risk SNPs at 9p21,
heritability using SNP data can often underestimate the which are associated with IA and AAA but not TAA, we
heritability if the SNP set does not tag all underlying causal observed 2 SNPs with significant associations to another
variants. aneurysm type after multiple testing correction. First, the
We then performed genetic correlation analyses IA risk SNP rs11661542 at 18q11 near RBBP8 was
(Table 10). We found that all pairs did not show statistically associated with AAA (odds ratio [OR]=1.11; P=4.1910(cid:1)5).
significant genetic correlations (P>0.05). The direction of Second, the TAA risk SNP rs2118181 at 15q21 near FBN1
DOI:10.1161/JAHA.115.002603 JournaloftheAmericanHeartAssociation 10