Genome-wide association study in 79366 European-ancestry individuals informs the genetic

12 Pages·2017·1.13 MB·English

by  Xia Jiang

ARTICLE
OPEN
DOI:10.1038/s41467-017-02662-2
Genome-wide association study in 79,366
European-ancestry individuals informs the genetic
architecture of 25-hydroxyvitamin D levels
Xia Jiang et al.#
1234567890():,; Vdiisteaamsiens.DPirsevaiosutesroGidWhAoSrmoofnseerpurmec2u5rs-ohrydthroaxtyisviatasmsoicniaDtecdowncitehntararatinognesohfavheumidaennttirfiaeitdsfaonudr
genome-wide significant loci (GC, NADSYN1/DHCR7, CYP2R1, CYP24A1). In this study, we
expand the previous SUNLIGHT Consortium GWAS discovery sample size from 16,125 to
79,366 (all European descent). This larger GWAS yields two additional loci harboring
genome-widesignificantvariants(P=4.7×10−9atrs8018720inSEC23A,andP=1.9×10−14at
rs10745742inAMDHD1).Theoverallestimateofheritabilityof25-hydroxyvitaminDserum
concentrationsattributabletoGWAScommonSNPsis7.5%,withstatisticallysignificantloci
explaining38%ofthistotal.Furtherinvestigationidentifiessignalenrichmentinimmuneand
hematopoietictissues,andclusteringwithautoimmunediseasesincell-type-specificanalysis.
Larger studies are required to identify additional common SNPs, and to explore the role of
rare or structural variants and gene–gene interactions in the heritability of circulating 25-
hydroxyvitamin D levels.
CorrespondenceandrequestsformaterialsshouldbeaddressedtoE.H.(email:[email protected])ortoP.K.(email:[email protected])
ortoD.P.K.(email:[email protected])
#Afulllistofauthorsandtheiraffliationsappearsattheendofthepaper
NATURECOMMUNICATIONS| (2018) 9:260 |DOI:10.1038/s41467-017-02662-2|www.nature.com/naturecommunications 1
ARTICLE
NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02662-2
V
itamin D is an essential fat soluble vitamin and steroid control for population stratifications and confounders. We
pro-hormone that plays an important role in muscu- identified six susceptibility loci harboring genome-wide sig-
loskeletalhealth.VitaminDdeficiencyhasbeenlinkedto nificant SNPs, confirming four previously reported loci at GC
autoimmune1,2 and infectious disease3, cardiovascular disease4, (P=4.7×10−343atrs3755967),NADSYN1/DHCR7(P=3.8×10−62
cancer5, and neurodegenerative conditions6. Serum 25- atrs12785878),CYP2R1(P=2.1×10−46atrs10741657),CYP24A1
hydroxyvitamin D, a primary circulating form of vitamin D (P=8.1×10−23 at rs17216707), and two novel loci at AMDHD1
andameasurethatbestreflectsvitaminDstores,isinfluencedby (P=1.9×10−14 at rs10745742) and SEC23A (P=4.7×10−9 at
many factors including sun exposure, age, body mass index7, rs8018720) (Table 1; Manhattan plots and QQ-plots for overall
dietaryintakeofcertainfoodssuchasfortifieddairyproductsand samplesarepresentedinFig.1,andregionalassociationplotsare
oilyfish,supplements,andgeneticfactors8.Theconcentrationof presented in Supplementary Fig. 2). The associations at both
25-hydroxyvitamin D has been reported to be highly heritable, novel loci were confirmed in the two independent in-silico
with heritability estimates of 50–80% from classical twin replication cohorts (EPIC: P=1.21×10−8 at rs10745742, P=
studies9,10. 5.24×10−4atrs8018720;SOCCS:P=0.03atrs10745742,P=0.04
A genome-wide association study (GWAS) meta-analysis of atrs8018720)withconsistentdirectionofeffectbutslightlylarger
serum 25-hydroxyvitamin D11 in 4501 participants of European effect sizes and wider confidence intervals observed in SOCCS
ancestry and replication in 2221 samples identified variants in which could be due to the reduced sample size. When analyzing
threeloci(groupcomponent(GC),7-dehydrochlesterolreductase thetworeplicationdatasetstogetherwiththediscoverydataset,
(NADSYN1/DHCR7), and 25-hydroxylase (CYP2R1)). A larger theP-valuesinpooledsamplesbecamemoresignificant(P =
GWAS conducted by the SUNLIGHT consortium in 16,125 2.10×10−20 at rs10745742, P =1.11×10−11 at rs80p1o8ol7ed20)
pooled
European ancestry individuals, with a replication sample of (Table 1). We also found more than one distinct signal arising
17,871, replicated these three loci and discovered one additional from variants at the GC, CYP2R1, and AMDHD1 loci through
locus (CYP24A1)8. However, despite these loci being in or near conditionalandjointanalysis,butnotfortheNADSYN1/DHCR7,
genes encoding proteins involved in vitamin D synthesis, the CYP24A1, and SEC23A loci where only one primary associated
associatedvariantscollectivelyexplainonlyasmallfractionofthe SNP was identified (Supplementary Table 2).
variance in 25-hydroxyvitamin D concentrations (~5%)8,11,12.
Therefore,toextendourpreviousfindingsandbetterunderstand
SNP by dietary vitamin D intake interaction. In addition to
the genetic architecture underlying serum 25-hydroxyvitamin D,
performing the marginal effect meta-analysis using all samples,
as well as test for interactions between dietary vitamin D intake
wealsotestedamodelwithSNPsanddietaryvitaminDintakeas
and genetic factors, we conducted a large-scale GWAS meta-
main effects and a term for their interaction in a subset of
analysis on this important vitamin.
samples. Diet questionnaires, including vitamin D intake, were
Our GWAS with a 79,366 discovery sample and a 40,562
replicationsamplereplicatesfourpreviouslociandidentifiestwo available for a subset of 13 cohorts and an additional 2 cohorts
thatwerenotincludedintheoverallmeta-GWASanalysis(atotal
nfuerwthgerenfientidc leovciidefonrcesefrourmalsehvaelrsedofg2en5-ehtiycdrboavsixsyvbietatwmeiennDc.irWcue- of 15 cohorts, N=41,981). We performed a GWAS explicitly
allowing for an interaction between vitamin D intake and SNP
lating 25-hydroxyvitamin D and autoimmune diseases. Our
genotypes,inwhichdietaryvitaminDwascodedasacontinuous
analyses suggest a relatively modest SNP-heritability rate of 25-
variable. We performed two tests: (i) a 1 degree-of-freedom
hydroxyvitamin D when considering only common variants.
interactiontestbetweeneachSNPandvitaminDintake,and(ii)
Largerstudiesarerequiredtoidentify additionalcommonSNPs,
a 2 degree-of-freedom joint test of main genetic and interaction
andtoexploretheroleofrareorstructuralvariants.Thegenetic
instruments identified by our results could be used in future effects. However, for comparison purposes, we also performed,
(iii) a standard test of marginal genetic effect after adjusting for
Mendelian Randomization analyses of the association between
vitaminDintakeinthesamesub-samples(SupplementaryFig.3,
vitamin D and complex traits.
and Supplementary Fig. 4). The marginal genetic effect analyses
confirmedexistingassociationsignalsatGC(leadSNPrs2282679,
in complete linkage disequilibrium with the lead SNP rs3755967
Results
identified from meta-GWAS using all individuals), CYP2R1,
Study description and GWAS. This study represents an expan-
NADSYN1/DHCR7 (lead SNP rs4944062, in complete linkage
sion of our previous SUNLIGHT consortium GWAS8. Here, we disequilibrium with the lead SNP rs12785878 identified from
combine the 5 discovery cohorts and 5 in-silico replication
meta-GWAS using all individuals) and CYP24A1, as well as the
cohorts from that study, and augment these with 21 additional novel association at AMDHD1 (P=5.7×10−9, Supplementary
cohorts that have joined the SUNLIGHT consortium since 2010 Table 3). The joint analysis also identified the five genes above,
(study characteristics are described in Supplementary Table 1, butwithlesssignificantP-values.Forinstance,theassociationat
SupplementaryNote1).Incontrasttothepreviousmeta-analysis AMDHD1achievedonlysuggestivegenome-widesignificance(P
wstotahtgaiclehso,fwiunepvoptloevre7fdo9r,3md6ie6sdcionavdefiirvyris,dtuisantl-assgiaelnicddoisrceaopnvldiecraydteemd-nenotoavv-oaenlgafilenynsdoiistnygopsnininga sv=iag1rn.i2aifi×nc1tsa0n−at7t,ignTeanbmolmeare2g-i)wn.aiTdleheefsfieigncnttiefirtcaeacsnttisco,enthleteevsetlle.daAdidmSnoNonPtgitdihneenCt5iYfySPN2aRnP1ys
two independent separate in-silico data sets (40,562 individuals showednominalsignificanceforinteractionwithdietaryvitamin
pcaoslosleltesscstaeandnddbgyceonEnoPtmrIoCilcfaocnordnpt2or1po9ul5lianitnfliodantiivosidtnruafatailcfistcocaortsliloefnoctr,ewedaecbehyxaScmoOniCntrCeidbSu)Q.tiTQnog- 0Do.b7s4ei;nrvtraeskd1e0f7o4(rr5ts7h14e027o,4tP1h6e=5r70S,.N64PP;s=r(s0r1s.7022221886)2,7607b79u,,tPP=n=o00..44i65n);t.errsRa4ec9pt4ieo4an0t6sin2g,wPtehr=ee
cohort prior to meta-analysis. We did not observe evidence for
widespread inflation (median λ =0.92; only 1/31 samples with analysis using a tertile coding for vitamin D instead of a con-
λ >1.01), indicating that ourGGCWAS results were not inflated tinuous coding did not qualitatively change the results.
GC
bypopulationstratificationorcrypticrelatedness(Supplementary
Fig. 1). Despite the slightlydeflatedλ observed in someof the SNP-heritabilityof25-hydroxyvitaminD.Wefurtherevaluated
GC
constituent cohorts most probably due to over-correction of test the SNP-heritability, defined as the heritability explained by
statistics, our λ of 0.99 in all samples indicated appropriate GWASSNPsof25-hydroxyvitaminD,usingLDscoreregression
GC
2 NATURECOMMUNICATIONS| (2018) 9:260 |DOI:10.1038/s41467-017-02662-2|www.nature.com/naturecommunications
ARTICLE
NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02662-2
Table1Singlenucleotidepolymorphismsidentifiedingenome-wideanalysesforcirculating25-hydroxyvitaminDconcentrations
Gene SNP Chromosome:Position Effect/reference Allelefrequency Meta-GWASestimates
allele
Effect(Beta) StandardError P-value
Firststagediscoverymeta-GWAS(N=79,366)
GC rs3755967 4:72828262 T/C 0.28 −0.089 0.0023 4.74E–343
NADSYN1/DHCR7 rs12785878 11:70845097 T/G 0.75 0.036 0.0022 3.80E–62
CYP2R1 rs10741657 11:14871454 A/G 0.4 0.031 0.0022 2.05E–46
CYP24A1 rs17216707 20:52165769 T/C 0.79 0.026 0.0027 8.14E–23
AMDHD1 rs10745742 12:94882660 T/C 0.4 0.017 0.0022 1.88E–14
SEC23A rs8018720 14:38625936 C/G 0.82 −0.017 0.0029 4.72E–09
Replicationdataset1:samplescollectedbyEPIC(N=40,562)
AMDHD1 rs10745742 12:94882660 T/C 0.41 0.041 0.0071 1.21E–08
SEC23A rs8018720 14:38625936 C/G 0.83 −0.032 0.0093 5.24E–04
Replicationdataset2:additionalcontrolsamplescollectedbySOCCS(N=2195)
AMDHD1 rs10745742 12:94882660 T/C 0.37 0.045 0.021 0.03
SEC23A rs8018720 14:38625936 C/G 0.81 −0.051 0.026 0.04
Pooledanalysis(discoverymeta-GWAS+replication1+replication2)(N=122,123)
AMDHD1 rs10745742 12:94882660 T/C 0.39 0.019 0.002 2.10E–20
SEC23A rs8018720 14:38625936 C/G 0.82 −0.019 0.0027 1.11E–11
Inthepooledanalysis,Pheterogeneity=0.003forAMDHD1rs10745742;Pheterogeneity=0.14forSEC23Ars8018720.
a
GC NADSYN1/DHCR7 Know
30 CYP2R1 Novel
25 CYP24A1
e)
u
p – val 20 AMDHD1
g (10 15
–lo 10 SEC23A
5
1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6789012
1 1 1 1 1 1 1111222
Chromosome
b
e
u
al
d p-v 200 (cid:2) = 0.99
e
ect
orr
0 c
1
g
o 100
–l
d
e
erv 50
bs
O
0
0 1 2 3 4 5 6
Expected –log10 corrected p-value
Fig.1Genome-wideassociationofcirculating25-hydroxyvitaminDgraphedbychromosomepositionsand−log10P-value(Manhattanplot),andquantile-
quantileplotofallSNPsfromthemeta-analysis(QQ-plot).aManhattanplot:TheP-valueswereobtainedfromthesinglestagefixed-effectsinverse
varianceweightedmeta-analysis.TheYaxisshows−log P-values,andtheXaxisshowschromosomepositions.Horizontalgraydashlinerepresentsthe
10
thresholdsofP=5×10−8forgenome-widesignificance.Knownlociwerecoloredcodedasred,andnovellociwerecolorcodedasgreen.bQQ-plot:TheY
axisshowsobserved−log P-values,andtheXaxisshowstheexpected−log P-values.EachSNPisplottedasablackdot,andthedashlineindicatesnull
10 10
hypothesisofnotrueassociation.DeviationfromtheexpectedP-valuedistributionisevidentonlyinthetailarea,withalambdaof0.99,suggestingthat
populationstratificationwasadequatelycontrolled
(see Methods). The overall observed heritability of 25- loci,theheritabilitydecreasedto4.70%(SE:0.72%).Theestimate
hydroxyvitamin D estimated by using all common SNPs further decreased to 1.73% (SE: 0.32%) after excluding all SNPs
(2,579,296 after QC) was 7.54% (standard error (SE): 1.88%). that reached nominal significance (156,675 SNPs with P≤0.05).
After excluding genome-wide significant SNPs (533 SNPs with These results indicate that common variants tagged by GWAS
P≤5×10−8) from six loci and all SNPs within ±500kb of those chipsexplainamodestfractionofoverallvariabilityincirculating
NATURECOMMUNICATIONS| (2018) 9:260 |DOI:10.1038/s41467-017-02662-2|www.nature.com/naturecommunications 3
ARTICLE
NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02662-2
Table2 Resultsfrom the SNP-by-dietaryvitaminDintakeinteraction analysis
Gene SNP Chromosome: Effect/ Allele SNP-by-dietaryvitaminDintakeInteractionanalysis
Position Reference Frequency
Main Interaction P-valuefor P-value
Allele
Genetic Effect interaction forjoint
Effect test
Effect Standard Effect Standard
(Beta_G) Error (Beta_Int) Error
Firststagediscoverymeta-GWAS(N=79,366)
GC rs3755967 4:72828262 T/C 0.28 −0.082 0.0042 −2.01E–05 1.67E–05 0.23 2.92E–171
rs2282679* 4:72827247 T/G 0.28 0.085 0.004 1.20E–05 1.60E–05 0.45 1.40E–187
NADSYN1/ rs12785878 11:70845097 T/G 0.75 0.033 0.0039 6.61E–06 1.62E–05 0.68 3.52E–29
DHCR7
rs4944062* 11:70864942 T/G 0.75 0.034 0.004 5.30E–06 1.60E–05 0.74 1.90E–31
CYP2R1 rs10741657 11:14871454 A/G 0.4 0.03 0.0035 3.21E–05 1.46E–05 0.028 2.23E–38
CYP24A1 rs17216707 20:52165769 T/C 0.79 0.025 0.0048 1.39E–05 1.88E–05 0.46 1.32E–14
AMDHD1 rs10745742 12:94882660 T/C 0.4 0.016 0.0036 −7.05E–06 1.49E–05 0.64 1.20E–07
SEC23A rs8018720 14:38625936 C/G 0.82 −0.013 0.0051 −2.40E–05 2.06E–05 0.24 1.94E–05
*TopSNPsidentifiedintheSNP-by-dietaryvitaminDintakeinteractionanalysis,performedinasubsetofindividuals.ForGCandNADSYN1/DHCR7,thetopSNPsidentifiedthroughthemarginaleffect
regressionmeta-analysisusingallindividualswereinhighlinkagedisequilibriumwiththetopSNPsidentifiedthroughtheSNP-by-dietaryvitaminDintakeinteractionanalysisusingasubsetofindividuals
(r2forrs3755967andrs2282679:1.0;r2forrs12785878andrs4944062:1.0).Beta_GindicatesthemaineffectoftheSNP,Beta_IntindicatestheinteractioneffectofSNP-by-dietaryvitaminDintake
DHS,8.5-foldenrichment,P=0.02),transcriptionfactorbinding
sites (5.7-fold enrichment, P=0.048), super-enhancers (1.9-fold
Type enrichment, P=0.04), and all four histone marks were enriched
NPs 2244aannnnoottaattiioonnss_500 bp (bothversionsofH3K27ac(oneversionprocessedbyHniszetal.,
S
op. 20 and another version used by the Psychiatric Genomics Con-
Pr sortium (PGC), H3K4me1, H3K4me3 (500bp), H3K29ac (500
2/
p.H bp)).Wealsoobserveddepletionforrepressedregions(0.06-fold
Pro enrichment, P=0.006). However, none of those annotations
= 10
nt ** withstood multiple-testing corrections (Bonferroni corrected P-
e
hm ** threshold: 0.05/24) except for the active enhancer histone mark
nric H3K27ac (PGC) (4.2-fold enrichment, P=8×10−4) and
E
0 H3K4me1 (1.8-fold enrichment, P=0.0019).
Weak enhancer Conserved TSS Fetal DHS TFBS H3K27ac (PGC) H3K27ac H3K4me1 H3K4me3 Super enhancer H3K27ac (Hnisz) Repressed ef1eonn0lrrWdiibccreehhonmmsarudieebcnnhscttemes,qllwueP-nteey=ntrp,et0elPy.i0ng=p0rt1e2ohr7.uef2)op,×irsm1m.a0mnAe−dd5us)nc,sCeehglNloaa-snwSttydrnpo(he3ii-nen.s6mtpe-Tfesaotcatiloibndfiplaceoleai3nteni,trsaiisctcluhyhteseimissssteou(bn4pey.ts4,ut-(hsf4Poir.n3el=dge-
Functional categories 0.0039). There was also significant enrichment for liver, kidney,
Fig.2Heritabilityenrichmentofthetop12genomicfunctionalelements. andconnectiveandbonetissues,buttheseresultsdidnotsurvive
WepartitionedtheSNP-heritabilityofserum25-hydroxyvitaminD multiple-testing corrections. When further analyzing 220 cell-
concentrationsinto24publiclyavailablegenomicfunctionalelementsusing type-specific annotations, we observed the most significant
LD-scoreregression.Weplottedtheenrichment(Yaxis)foreachofthe12 enrichment in CD19 cells (approximately 8-fold enrichment,
topannotations(asshowninXaxis)intoabarchart.Graybarsandblue P~0.001),followedbyCD20cells(6.4-foldenrichment,P=0.003)
barsrepresenttheannotationswithandwithoutthe500base-pair and CD3 cells (7.8-fold enrichment, P=0.01) (Supplementary
windows.Theheightofeachbarrepresentsmagnitudeofenrichment. Data 1).
SignificantestimatesofenrichmentthatpassedBonferronicorrections
(P-valueforenrichment<0.05/24)aremarkedwithdoublestars.TSS
Genetic correlations between 25-hydroxyvitamin D and traits.
transcriptionstartsites,DHSDNaseIhypersensitivesites,TFBS
We continued to assess the genetic correlation between 25-
transcriptionfactorbindingsites,Repressedrepressedregions
hydroxyvitaminDandeachofthe37traitswithpubliclyavailable
GWASsummarystatisticsdata(SupplementaryTable5).Noneof
25-hydroxyvitaminD,andthatanappreciableproportionofthis
the genetic correlations remained significant after Bonferroni
SNP-heritability is explained by the six genetic regions of asso-
ciated SNPs identified through GWAS. correction (corrected P-threshold: 0.05/37, Fig. 3). Without
multiple-testing correction, there were some correlations with
nominal statistical significance. For example, ever smoking
Partitioningthetotalheritabilityof25-hydroxyvitaminD.We (r (SE): −0.17 (0.073), P=0.019), primary biliary cirrhosis
g
next partitioned the heritability by functional elements using (r (SE): −0.18 (0.076), P=0.019) and BMI adjusted waist-hip-
g
baseline model with 24 publicly available annotations (see ratio (r (SE): −0.10 (0.050), P=0.042) were observed to be
g
Methods), and observed large and significant enrichment for inversely correlated with 25-hydroxyvitamin D; whereas lung
severalfunctionalcategories(Fig.2,SupplementaryTable4).For function (r (SE): 0.14 (0.046), P=0.0036) showed a positive
g
example, we found the largest enrichment in weak enhancers, correlation with 25-hydroxyvitamin D. Subsequent directional
with 2.1% of SNPs explaining 42.3% of the overall heritability genetic correlation analysis did not reveal any apparent putative
(20-fold enrichment, P=0.02), followed by conserved regions causal relationship of 25-hydrovyvitamin D with other traits,
(13.8-foldenrichment,P=0.03),openchromatin(asreflectedby except for a potential link between 25-hydroxyvitamin D and
4 NATURECOMMUNICATIONS| (2018) 9:260 |DOI:10.1038/s41467-017-02662-2|www.nature.com/naturecommunications
ARTICLE
NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02662-2
Table3 Heritabilityenrichment of ten groupedcell types
Category ProportionofSNPs(%) Proportionofh2 (%) Enrichment(standarderrors) P-value
g
Kidney 4.26 27.27 6.4(2.44) 0.027
Liver 7.22 37.68 5.22(1.55) 0.01
Gastrointestinal 16.77 72.88 4.35(0.97) 0.0017
Immuneandhematopoietic 23.34 100.17 4.29(0.76) 2.20E-05
Centralnervoussystem 14.88 54.09 3.64(0.87) 0.0039
Cardiovascular 11.11 35.74 3.22(1.26) 0.078
Connectivetissue/bone 11.5 35.65 3.1(1.04) 0.037
Adrenal/pancreas 9.36 26.17 2.8(1.31) 0.18
Other 20.27 56.68 2.8(0.98) 0.076
SkeletalMuscle 10.38 14.29 1.38(1.25) 0.76
BlackboldfontindicatessignificantP-valuesaftermultiplecorrections(P<0.05/10)
HDL (Supplementary Table 6). However, with only six cause craniolenticulosutural dysplasia, a disease characterized by
25-hydroxyvitamin D associated SNPs included in the analysis, craniofacialandskeletalmalformationsuchasdelayinclosureof
we consider an overall null directional correlation as our main fontanels, sutural cataracts and facial dysmorphisms, due to
finding, and further well-designed large-scale Mendelian rando- defective collagen secrection13,14. The second novel locus is
mization analyses are warranted. AMDHD1 (amidohydrolase domain containing 1). This gene
Finally, we analyzed the 220 cell-type-specific annotations in encodes an enzyme involved in the histidine, lysine, phenylala-
each of the 37 traits and compared the cell-type-specific nine, tyrosine, proline and tryptophan catabolic pathway. Muta-
enrichments for 25-hydroxyvitamin D to the enrichments for tions in AMDHD1 are found to be associated with atypical
these traits. The enrichment pattern for 25-hydroxyvitamin D lipomatoustumor,acancerofconnectivetissuesthatresemblefat
differed notably from the patterns for psychiatric diseases and cells15.
metabolic related traits. Psychiatric diseases showed enrichment OurSNP-heritabilityresultssuggestthat25-hydroxyvitaminD
for histone marks specific to CNS cell types, and metabolic has a modest overall heritability due to common genome-wide
diseases showed enrichment for gastrointestinal cell types, while SNPs of 7.5%, and that an appreciable proportion (2.84% out of
these annotations were depressed in 25-hydroxyvitamin D. 7.5%,i.e.,38%)ofthistotalcouldbeexplainedbyknowngenetic
Conversely, 25-hydroxyvitamin D showed similar patterns with regionsidentifiedthroughGWAS.Ourfindingsareinlinewitha
autoimmuneinflammatorydiseases,wheremultipleimmunecell previouspublishedreport(byHirakietal.12)whichestimatedthe
types were enriched. We consistently observed that 25- variance in circulating 25-hydroxyvitamin D explained by SNPs
hydroxyvitamin D was clustered with autoimmune diseases in a total of 5575 individuals12. According to that report, by
(Supplementary Fig. 5). employing a linear mixed model fitting the additive genetic
matrix created from all genotyped and imputed SNPs, the pro-
Discussion portionofvarianceexplainedwas8.9%;byemployingapolygenic
VitaminDinadequacy hasbeenlinkedtomanydiseasessuch as score approach comprised of the then GWAS-discovered SNPs
cancer, autoimmune disorder and cardiovascular conditions in (GC, CYP2R1, DHCR7/NADSYN1), the proportion of variance
additiontomusculoskeletaldiseases,whichhasledtosubstantial explained was 5%. Both of these estimates were close to ours. In
interest in the determinants of vitamin D status, especially its Hiraki et al., the known 25-hydroxyvitamin D associated envir-
genetic components. We have performed a large 25- onmental factors such as age, BMI, season of blood drawn,
hydroxyvitamin D meta-GWAS involving 31 studies with a vitaminDdietaryintake,vitaminDsupplementintake,regionof
total of 79,366 individuals. Our results recapitulated several pre- residence and ethnicity, explained ~18% of the observed var-
viously reported findings. First of all, we confirmed the role for iance12. Our results, in agreement with these findings, suggest
common genetic variants in regulation of circulating 25- thatalthoughthereappears tobesomepolygenicsignals outside
hydroxyvitamin D concentrations. Our study validated three of the identified regions, the remaining common effects may be
loci, GC, NADSYN1/DHCR7, and CYP2R1, all were established small. There also may be low frequency variants with larger
25-hydroxyvitamin D risk loci identified through two earlier effects that were not investigated here. For example, while this
GWASs8,11. In addition, we were able to confirmthe association paper wasunderreview,arelated studyidentifiedlow-frequency
of a locus containing CYP24A1 with 25-hydroxyvitamin D con- (MAF=2.5%) synonymous coding variant rs117913124_A at
centrations using our large sample size, which highlights the CYP2R1conferringalargeeffecton25-hydroxyvimtainDlevels,
importance of this protein in the degradation of vitamin D whichwasfourtimesgreaterinmagnitudeandindependentofa
molecule, by catalyzing hydroxylation reactions at the side chain previously described association for a common variant
of 1,25-dihydroxyvitamin D, the physiologically active form (rs10741657) near CYP2R116.
(hormonal form) of vitamin D. Significant finding at this locus Resultsoftwinandfamilialstudieshaverevealedasubstantial
was only shown in the pooled analyses involving both discovery geneticbasisinthevariabilityofcirculating25-hydroxyvitaminD
and replication samples in an earlier GWAS8. levels, with estimates of heritability reaching as high as
We extended previously reported findings by identifying two 86%9,10,17–19.Theseestimates,however,seemtobeinfluencedby
additional new loci. SEC23A (Sec23 Homolog A, coat protein environmental conditions.Forexample,ina studyconducted by
complex II (COPII) component) encodes a member of Orton et al. with 40 monozygotic and 59 dizygotic twin pairs,
SEC23 subfamily. In eukaryotic cells, secreted proteins are syn- bloods were collected at the end of winter and a heritability of
thesized in the endoplasmic reticulum (ER), packaged into 77% was reported10. Similarly, the study conducted by Karohl
COPII-coatedvesicles,andtraffictotheGolgiapparatus.Aspart et al. with 310 monozygotic and 200 dizygotic male twins
of COPII complex, SEC23 plays a role in promoting ER-Golgi observedaheritabilityof70%duringwinter,whereasinsummer,
protein trafficking. SEC23A mutations have been reported to serum 25-hydroxyvitamin D concentrations appeared to be
NATURECOMMUNICATIONS| (2018) 9:260 |DOI:10.1038/s41467-017-02662-2|www.nature.com/naturecommunications 5
ARTICLE
NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02662-2
Genetic correlation
Traits (95%CI) p-value Significance
Autoimmune/inflammatory diseases
Age-related macular degeneration –0.02 (–0.1, 0.06) 0.57
Celiac Disease –0.02 (–0.17, 0.15) 0.89
Crohn’s Disease –0.07 (–0.19, 0.05) 0.25
Lupus 0 (–0.15, 0.14) 0.98
Multiple Sclerosis –0.17 (–0.45, 0.1) 0.21
Primary biliary cirrhosis –0.18 (–0.33, –0.03) 0.019 *
Rheumatoid Arthritis 0.05 (–0.08, 0.19) 0.44
Ulcerative Colitis 0.04 (–0.1, 0.18) 0.59
UKBiobank Asthma –0.01 (–0.11, 0.1) 0.9
UKBiobank Eczema –0.01 (–0.12, 0.1) 0.91
Inflammatory Bowel Disease –0.02 (–0.13, 0.09) 0.71
Alzheimer’s –0.08 (–0.28, 0.11) 0.39
Psychiatric disorder/traits
Anorexia –0.03 (–0.15, 0.1) 0.7
Autism 0.03 (–0.11, 0.17) 0.64
Bipolar Disorder 0.04 (–0.11, 0.18) 0.63
Depressive Syndrome 0.04 (–0.08, 0.17) 0.49
Neuroticism 0.02 (–0.07, 0.11) 0.61
Schizophrenia 0 (–0.08, 0.08) 1
Subject Well Being 0.14 (–0.1, 0.38) 0.25
Metabolism related traits
Coronary Artery Disease –0.03 (–0.16, 0.1) 0.66
Fasting Glucose 0 (–0.12, 0.12) 0.98
HDL 0.07 (–0.04, 0.19) 0.21
LDL 0.06 (–0.07, 0.18) 0.36
Triglycerides –0.06 (–0.16, 0.03) 0.18
Type 2 Diabetes –0.08 (–0.22, 0.06) 0.26
UKBiobank Hypertension 0.02 (–0.05, 0.1) 0.54
Others
Ever Smoked –0.17 (–0.32, –0.03) 0.019 *
UKBiobank Age at Menarche –0.07 (–0.02, 0.15) 0.13
UKBiobank Age at Menopause –0.07 (–0.19, 0.05) 0.25
UKBiobank BMI –0.03 (–0.11, 0.05) 0.45
UKBiobank Diastolic 0.02 (–0.06, 0.09) 0.63
UKBiobank FEV1FVC –0.01 (–0.08, 0.06) 0.69
UKBiobank FVC 0.14 (0.04, 0.23) 0.0036 *
UKBiobank Heel TScore –0.02 (–0.09, 0.05) 0.64
UKBiobank Height 0.06 (–0.01, 0.12) 0.078
UKBiobank Systolic –0.01 (–0.08, 0.06) 0.8
UKBiobank Waist-Hip Ratio –0.1 (–0.2, 0) 0.042 *
–0.5 –0.25 0 0.25 0.5
Fig.3Geneticcorrelationsbetween25-hydroxyvitaminDand37traits.WecollectedGWASsummarystatisticsof37diseasesandtraitsspanningawide
rangeofphenotypes(autoimmuneinflammatorydiseases,psychiatricdisorders,metabolictraits,andanthropometricindex)frompubliclyavailable
resources,andestimatedtheirsharedgeneticsimilaritieswithserum25-hydroxyvitaminDlevels.Weplottedthegeneticcorrelationtogetherwith95%
confidenceintervalsusingabluesquareandgrayhorizontallines.Redverticallineindicatesnogeneticcorrelation(r =0).Statisticalsignificancewas
g
definedasP-value<0.05.NoneofthepairwisecorrelationspassedBonferronicorrections
entirely determined by non-genetic factors (heritability: 0%)9. studies9,10, but our estimate of 7.5%, calculated using common
Comparable estimates were also identified in a slightly larger genome-wide SNPs, is far lower than reported heritability from
studyconductedbyMillsetal.(winter:90%vs.summer:56%)18. twin and family based studies. In addition to potentially inflated
Consistent with season dependency, sex differences were also estimates from twin studies, the difference may reflect the pro-
observed (males: 86% vs. females: 17%)17. While these estimates portion of heritability explained by rare SNPs or structural var-
should be treated with caution due to small samples and related iants thatwere not included in our data,and the potential gene-
imprecision, they confirm the substantial variation in 25- geneinteractionsthatremaintobeidentified.Thecombinationof
hydroxyvitamin D levels by season (as shown previously20) and our samples from all seasons is also likely to decrease the prob-
illustrate that heritability estimates derived from a homogenous ability of finding genetic variants, and hence deflate heritability
source may be highly inflated. In a relevantly well-powered twin estimates.
study with a total of ~2100 female twins, the heritability of 25- Through partitioning the SNP-heritability of serum 25-
hydroxyvitamin D was calculated to be 40%, indicating a larger hydroxyvitamin D levels, we observed a significant enrichment
proportion of variance explained by non-genetic factors21. Her- in immune and hematopoietic tissues; likewise, the cell-type-
itability estimates obtained using GWAS SNPs have typically specific analysis revealed clustering of 25-hydroxyvitamin D and
beenfound tobeapproximately half ofthosefromclassicaltwin autoimmunediseases,indicatingthatthesetraitsshareamajority
6 NATURECOMMUNICATIONS| (2018) 9:260 |DOI:10.1038/s41467-017-02662-2|www.nature.com/naturecommunications
ARTICLE
NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02662-2
ofcommoncelltypes.ThelinkbetweenvitaminDdeficiencyand samplesinourpreviousGWASpublication(the1958BritishBirthCohort
increased risk for autoimmune inflammatory diseases has long (1958BC),theCardiovascularHealthStudy(CHS),theFraminghamHeartStudy
been recognized by epidemiological investigations22,23. Although (FHS),theGothenburgOsteoporosisandObesityDeterminantsstudy(GOOD),
theHealth,Aging,andBodyCompositionstudy(HealthABC),theIndiana
theunderlyingmechanismsremainunclear,itisnowevidentthat
Womencohort,theNorthFinlandBirthCohort1966(NFBC1966),theOldOrder
vitamin D is involved in many biological processes that regulate AmishStudy(OOA),theRotterdamStudy(RS),andtheTwinsUK),andan
both innate and adaptive immune responses, through ligand- additional21cohortswereincludedforthecurrentanalysis(theAlpha-Toco-
receptor binding, activation, interaction with response elements pherol,Beta-CaroteneCancerPreventionStudy(ATBC),theAtherosclerosisRisk
inCommunitiesStudy(ARIC),theAtheroGeneregistry,B-vitaminsforthePre-
inthepromoterregionsofdifferentgenes,andeventuallyleadto
ventionOfOsteoporoticFractures(B-PROOF),theEpidemiologyofDiabetes
functional changes in a wide variety of immune cells including InterventionsandComplications(EDIC),theCase-ControlStudyforMetabolic
Th1, Th2, Th17, T regulatory and natural killer T cells22,23. The Syndrome(GenMets),theHelsinkiBirthCohortStudy(HBCS),theHealthPro-
sharedcelltypeenrichmentsbetweenvitaminDandautoimmune fessionalFollow-upStudy(HPFS,nestedacoronaryheartdiseasecase-control
study),theInvecchiareinChiantiStudy(InChianti),theCooperativeHealth
diseasesobservedinourstudy,furthersuggestthatvitaminDnot
ResearchintheregionAugsburg(KORA),theLeidenLongevityStudy(LLS),the
only affects autoimmune diseases through its direct effect (as a LudwigshafenRiskandCardiovascularHealthStudy(LURIC),theMulti-Ethnic
ligand), but also through their shared genetic etiology. Thus, StudyofAtherosclerosis(MESA),theNijmegenBiomedischeStudie(NBS),the
individualswithvitaminDdeficiencymaybemoresusceptibleto Nurses’HealthStudy(NHS,nestedabreastcancercase-controlstudy,andatype2
diabetescase-controlstudy),theOrkneyComplexDiseaseStudy(ORCADES),the
these disorders, both because of environmental and genetic
influences. Prostate,Lung,Colorectal,andOvarianCancerScreeningTrial(PLCO),the
PROspectiveStudyofPravastatinintheElderlyatRisk(PROSPER),theStudyof
Ourgenome-wideinteractionanalysisbetweengeneticvariants HealthinPomerania(SHIP),theScottishColorectalCancerStudy(SOCCS),the
and dietary intake of vitamin D did not identify new signals. All CardiovascularRiskinYoungFinnsStudy(YFS),andmoresamplesfromtheRS
significant associationsobserved in thejoint test of maingenetic (RSI,RSII,andRSIII)).Fulldescriptionsofallparticipatingcohorts,detailsof
and interaction effects were of equal or higher significance level genotypingplatformsused,numberofSNPs,andthemeasurementsofserum25-
hydroxyvitaminDconcentrationsineachcohortareshowninSupplementary
(i.e., lower P-values) in the GWAS of marginal genetic effect Table1andSupplementaryNote1.Writteninformedconsentwasobtainedfrom
performed in the same individuals, indicating no major con- allparticipantsintheincludedcohorts,andthestudyprotocolswerereviewedand
tribution of interaction effects at these loci. Indeed, only one of approvedbylocalinstitutionalreviewboards.
thetop5locifromtheoverallmarginalGWASshowednominally
significant interaction effect, and none passed Bonferroni cor- Powercalculation.Ourlargesamplesizeprovidedgoodstatisticalpowerfor
associationanalysis.Atthegenome-widesignificancethresholdof5×10−8,witha
rections.Whilesmallergene-dietinteractioneffectsremaintobe
discoverysamplesizeof75,000,ourstudyhad85%powertodetectagenetic
discovered, our results provide some evidence against large
variant(singlenucleotidepolymorphism,SNP)accountingfor0.06%ofthetotal
interactions between common SNPs and dietary vitamin D varianceinserum25-hydroxyvitaminDconcentrations,and99%powertodetecta
intake. Still, one cannot completely rule out the possibility of variantthatexplained0.1%ofthetotalvariance.Wealsohadpowertodetectgene-
interaction, but only conclude that genetic effects appear stable environmentinteractioneffectsevensmallerthantheobservedmarginaleffects.In
thecasewhereaSNPhasnomarginaleffectoncirculating25-HydroxyvitaminD
withinvitaminDintakerangeinthepopulationsstudied.Indeed,
concentrations(andsocouldnothavebeendiscoveredviathemarginalGWAS),
asforanygene-environmentinteractiontests,statisticalpoweris wehad80%powertodetectaninteractionthatexplained0.07%ofthetotal
highly dependent on the variance of exposure in the samples variancein25-hydroxyvitaminDconcentrations.
analyzed24, and interactions would remain unobserved if the
exposure is homogeneous among individuals. Also, we were not Associationanalysis.Genome-wideanalyseswereperformedwithineachcohort
abletocapturevitaminDsupplementationadequatelytoinclude accordingtoauniformanalysisplan.Wefitadditivegeneticmodelsusinglinear
regressiononnatural-logtransformed25-hydroxyvitaminD,andadjustedthe
this in the dietary intake variable, and were not able to estimate
modelsformonthofsamplecollection(12categories),age,sex,andbodymass
sunlightexposureasasourceofvitaminDproductionintheskin.
index,andprincipalcomponentscapturinggeneticancestry.Furtheradjustments
Serum 25-hydroxyvitamin D concentrations are mainly includedcohort-specificvariables,suchasgeographicallocationandassaybatch,
determinedbymodifiableenvironmentalfactors,andcontraryto whererelevant.Forparticipatingstudieswithacase-controldesign,weanalyzed
estimates from previous twin studies, our large-scale analyses casesandcontrolsseparately.Weperformedafixed-effectsinversevariance
weightedmeta-analysisacrossthecontributingcohorts,asimplementedinthe
suggest a SNP-heritability rate that is relatively modest in mag- softwareMETAL26,withcontrolforpopulationstructurewithineachcohortand
nitude when considering common variants. Our study also qualitycontrolthresholdsofminorallelefrequency(MAF)>0.05,imputationinfo
showed that common genetic variants are unlikely to have a score>0.8,Hardy-Weinbergequilibrium(HWE)>1×10−6,andaminimumof
strong modifying effect on increases in 25-hydroxyvitamin D twostudiesand10,000individualscontributingtoeachreportedSNP-phenotype
association.WeregardedP-values<5×10−8asgenome-widesignificant.
followingtypicaldietaryintakes,suggestingthatconsiderationof
geneticbackgroundisnotrequiredwhendeterminingpopulation
Replicationstudy.Wereplicatedtheidentifiednovellociintwoindependentdata
based vitamin D intake recommendations. However, our results
setsforwhichgenotypedatawereavailable:theEuropeanProspectiveInvestigation
support the role of vitamin D in immunological diseases as we
intoCancerandNutrition(EPIC)studywith40,562individualsacrosstwonested
observedfromcell-type-specificanalysisforclusteringofvitamin case-controlstudies(EPIC-InterActandEPIC-CVD)andthecohort-wideEPIC-
D and autoimmune diseases, and the evidence for signal Norfolkstudy(SupplementaryNote1);andacohortof2195individuals(all
enrichment for immune and hematopoietic tissues. These find- controls)additionallycollectedaspartoftheSOCCSthatwerenotincludedinour
discoverystage.Asforthephenotype,EPICindividualswereassayedforplasma
ings are in line with previous Mendelian Randomization studies
25-hydroxyvitaminD andSOCCSindividualswereassayedfortotal25-
3
whichfoundaputativecausalassociationbetweenvitaminDand hydroxyvitaminD.Weperformedtheassociationanalysisinasimilarmanner,
autoimmune diseases such as multiple sclerosis1,2 and type 1 adjustedforage,sex,timeofsamplecollection,andstudycenterwhererelevant.
diabetes25. The additional genetic instruments identified by our WeregardedP-value<0.05inthereplicationsamples,andP-value<5×10−8inthe
pooledanalysisassuccessfulreplication.
results could also be used in future Mendelian Randomization
analysesoftheassociationbetweenvitaminDandcomplextraits.
Conditionalanalysis.Afteridentifyingtheprimaryassociatedvariantateachlocus
selectedaccordingtothestrengthofitsassociation,wefurthertestedwhetherthere
wereanyotherSNPssignificantlyassociatedwith25-hydroxyvitaminDafter
Methods accountingfortheeffectofleadSNP.Wethusperformedastepwisemodel
Studycohorts.WeexpandedourpreviousSUNLIGHTconsortiumGWAS,and selectionprocedureforthosechromosomeswhereasignificantvariantwaspre-
undertookalarge,multicenter,genome-wideassociationstudyof31cohortsin viouslyidentified.WestartedwiththemostsignificantlyassociatedSNP,scanning
Europe,CanadaandUSA.Ourfirststagediscoverymeta-analysisconsistedof throughthewholechromosome,selectingadditionalindependentlyassociated
79,366samplesofEuropeandescentdrawnfrom31epidemiologicalcohorts. SNPsusingastepwiseprocedure,oneatatime,basedontheirconditionalP-
Amongthose31cohorts,tenwereusedasdiscoveryandin-silicoreplication values.Finally,wefitallselectedSNPintoonemodeltoestimatetheirjointeffects.
NATURECOMMUNICATIONS| (2018) 9:260 |DOI:10.1038/s41467-017-02662-2|www.nature.com/naturecommunications 7
ARTICLE
NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02662-2
WeusedGCTA-COJOsoftwaretoaccommodateoursummarylevelGWAS GenomicsConsortium(PGC)38);openchromatin,asreflectedbyDNaseI
data27,andtheCancerGeneticMarkerofSusceptibility(CGEMS)GWASwith hypersensitivitysites(DHSsandfetalDHSs)33,obtainedasacombinationof
2287individualsofEuropeandescentand2,543,887genotypedandimputed EncyclopediaofDNAElements(ENCODE)andRoadmapEpigenomicsdata,and
(HapMap22)SNPsasreferencepanel. processedbyTrynkaetal.36;combinedchromHMMandSegwaypredictions
obtainedfromHoffmanetal.39,whichleverageonmanyannotationstopartition
thegenomeintosevenunderlyingchromatinstates(theCCCTC-bindingfactor
SNP-by-dietinteraction.Weperformedagenome-wideassociationscreeningof (CTCF),promoter-flanking,transcribedregion,transcriptionstartsite(TSS),
circulating25-hydroxyvitaminDwhileaccountingforpotentialinteractioneffect strongenhancer,weakenhancer,andtherepressedregion);regionsthatare
betweenSNPanddietaryvitaminDintake.Ourtestsincorporatinggene-diet conservedinmammals,providedbyLindblad-Tohetal.40andpost-processedby
interactionwerebasedonthefollowingmodel: WardandKellis41;super-enhancers,whicharelargegroupsofputativeenhancers
lnð25ðOHÞDÞ¼β þβ ´Gþβ ´Eþβ ´G´Eþβ ´Z withhighlevelsofactivity,providedbyHniszetal.37;FANTOM5enhancers
0 1 2 3 Z mappedbyusingcapanalysisofgeneexpressionintheFANTOM5panelof
samples,obtainedfromAnderssonetal.42;digitalgenomicfootprint(DGF)and
transcriptionfactorbindingsite(TFBS)annotationsdownloadedfromENCODE35
whereGisaSNPthatwascodedadditively,EistherawvitaminDintake,
measuredonacontinuousscale.Theparametersβ ,β ,β ,andβ arethe andpost-processedbyGusevetal.33.Weincluded500-bpwindowsaroundeachof
0 1 2 3 the24mainannotationsinthebaselinemodel,and100-bpwindowsaroundChIP-
intercept,themaineffectofSNP,themaineffectofdietaryvitaminDintake,and
seqwhenappropriate,topreventupwardbiasofestimatesgeneratedby
theinteractioneffectbetweenGandE.Themodelalsoincludedthesame
enrichmentinthenearbyregions.
covariatesZasforthemarginaleffectscreening,theeffectsofwhichwerecaptured
intheparameterβ .Weconsideredbothastandard1degree-of-freedomtestof Inadditiontothebaselinemodelusing24mainannotations,wealsoperformed
interactioneffect(iZ.e.,nullhypothesisofβ ¼0),andajoint2degree-of-freedom cell-type-specificanalysesusingannotationsofthefourhistonemarks(H3K4me1,
3 H3K4me3,H3K9acandH3K27ac).Eachcell-type-specificannotationcorresponds
testofmaingeneticeffectplusgene-by-dietinteraction(i.e.,nullhypothesisof
β ¼0andβ ¼0).Forcomparisonpurposes,wealsoconsideredamodel toahistonemarkinasinglecelltype(forexample,H3K27acinadiposenuclei
1 3 tissues),andtherewasatotalof220suchannotations.Wefurthersubdividedthese
adjustingforvitaminDintakebutnotmodelinginteraction(i.e.,notincludingthe
β ´G´Eterm)usingthesamesubsetofindividuals. 220cell-type-specificannotationsinto10categoriesbyaggregatingthecell-type-
3 specificannotationswithineachgroup(forexample,SNPsrelatedwithanyofthe
VitaminDintakewasavailablefor15cohortsonatotalof41,981individuals.It
fourhistonemodificationsinanyhematopoieticandimmunecellswereconsidered
includedboththepopulationbasedcohorts(ARIC,1958BC,B-PROOF,FHS,
asonebigcategory).Whengeneratingthecell-type-specificmodels,weaddedeach
HealthABC,MESA,NFBC,RS,RSIII,andYFSaspartoftheoverallMeta-GWAS,
annotationindividually(oneatatime)tothebaselinemodel,creatingseparate
plustwoadditionalcohorts,theProspectiveInvestigationoftheVasculaturein
modelstocontrolforoverlapwiththegenomicfunctionalelementsinthefull
UppsalaSeniors(PIVUS),andtheUppsalaLongitudinalStudyofAdultMen
baselinemodelbutnotoverlapwiththeothercelltypes.
(ULSAM),genotypedoncustomarraythatwerenotincludedintheoverallmeta-
WeadditionallyassembledthesummarystatisticsfromGWASof37traitsor
GWASbutwereincludedinthisSNP-by-dietinteractionanalysis),andcase-control
diseasesperformedinindividualsofEuropeandescent,whicharepublicly
studies(HPFS(HPFS_CHD),NHS(NHS_BRCA,NHS_T2D),andSOCCS).Forthe available38,43–55orappliedfromtheUKBiobank.Thesestudiesspanawiderange
latterstudies,allanalyseswereperformedseparatelyincasesandcontrols.The
ofphenotypes,fromanthropometricindicessuchasheight,weight,BMI,tomental
aforementionedinteractionmodelwasappliedtoeachoftheincludedcohorts,and
study-specificresultsweremeta-analyzedusinginverse-varianceweightedsumof disorders(forexampledepressivesyndromeandschizophrenia)toautoimmune
effectestimatesasimplementedinMETAL26.Forthe2degree-of-freedomtestwe andinflammatorydiseases(forexamplerheumatoidarthritisandceliacdiseases).
usedjointframeworkdescribedintwopreviouspublishedpapers28,29.Quality Wecalculatedthepairwisegeneticcorrelation(rg,crosstraitheritability)between
controlfilteringwasperformedoneachstudybeforemeta-analysis.OnlySNPswith 25-hydroxyvitaminDandeachofthe37traits.Wefurtherconductedthesame
imputationinfoscore>0.8,MAF>0.05,HWE>1×10−6,andaminimumoftotal cell-type-specificanalysisforeachtrait,andplottedbeta-coefficientz-scorematrix,
samplesizeinthemeta-analysis>10,000wereretained. constructedfromthetotal220annotationsby37traits,intofourheat-mapsbased
onthefourhistonemarks.
Finally,inadditiontothegeneticcorrelationanalysiswhichreflectsshared
LinkageDisequilibriumscoreregression.Weperformedlinkagedisequilibrium geneticfactorsacrossdifferenttraitsbutdoesnotinformdirection,wealso
scoreregression(LDSC)analysistoestimatetheSNP-heritabilityofserum25- attemptedtoidentifydirectionsofsuchcorrelationusinganalgorithmproposedby
hydroxyvitaminDconcentrations30,31.Thismethodisbasedonavalidatedrela- Pickrelletal.56.ThemethodadoptsasimilarintuitionastheMendelian
tionshipbetweenLDscoreandχ2-statistics: Randomizationapproach,where,ifatraitXinfluencestraitY,thenSNPs
Ehχ2ji(cid:2)NMjh2gljþ1 itinnraflfliuutseennshcceionduglbdXybsmeheoccouhrldraenalialsstmeodsi.niFflnuudreetnphceeenr,dYse,innatcneodfYtXhd,eogSeeNsnPent-oisctpviencaflirfiiuacennetfscfeetchXta,tsibiznuefltsucfoeonrucltdeheYbtewdoo
h i notnecessarilyinfluenceX.Basedonthisintuition,themethodproposestwo
whereE χ2 denotestheexpectedχ2-statisticsfortheassociationbetween “causal”modelsandtwo“non-causal”models,andcalculatestherelativelikelihood
j ratioofthebestnon-causalmodelcomparedtothebestcausalmodel.We
onuutmcobmeresoanfdvaSrNianPtsj,aNndjisljtdheenosttuedsythseamLDplsecsoirzeeoafvSaNilaPbljedfeofirnSeNdPasj,ljM¼iPsthr2eðtj;oktaÞl. sdiegtneirfimciannetd(Psi<gn5ifi×c1a0n−t8)SNSNPPssfoarndeapchrugnievdenthteranitumbybeserslebctaesdedgeonnotmheei-rwLidDe-pattern
LDSCcalculatesheritabilityusingonlysummary-leveldatainsteadofinkdividual
intheEuropeanpopulationsinPhase1of1000GenomeProject.Wescanned
genotypes,andiscomputationallycosteffectiveatlargesamplesizes.Weusedthe
throughallpairsof25-hydroxyvitaminDandtraitstoidentifydirectional
sauvmailmabalreyinstaattisletiacsstf2rosmtud2ie5s-haynddroaxsyavmitapmleisnizDeomfeattal-eGasWt1A0S,0r0e0s.uWltse,wfiristhtaSnNalPyszed correlations.Weconsiderpairsoftraitswithlikelihoodrationon-causalvs.causal<
0.05ashavingevidenceofdirectionalcorrelations.
theSNP-heritabilitybyusing(1)SNPsacrosstheentiregenomethatpassedquality
control;(2)SNPsexcludingthetopassociations(SNPsreachinggenome-wide
significance,P≤5×10−8),aswellasallSNPswithin±500kbofthetophitsinthe
region;and(3)SNPsexcludingthenominallysignificantassociations(P≤0.05).We Dataavailability.TheGWASsummarystatisticsonserumcirculatingvitaminD
subsequentlypartitionedtheheritabilitythroughthreedifferentmodels:(1)afull concentrationsisavailableatdbGaphttps://drive.google.com/drive/folders/
baselinemodelincludingthe24publiclyavailablemainannotationsthatarenot 0BzYDtCo_doHJRFRKR0ltZHZWZjQ;allrelevantdataareavailablefromthe
specifictoanycelltype,the500-bpwindowsaroundeachannotation,aswellas authorsuponrequest.
100-bpwindowsaroundchromatinimmunoprecipitationandsequencingpeaks
(ChIP-seq)whenappropriate.Thisresultedinatotalof52overlappingfunctional
Received: 21 July2017 Accepted: 15December 2017
categoriesinthefullbaselinemodel;(2)acell-type-specificmodelincluding10cell
typegroups:adrenalandpancreas,centralnervoussystem(CNS),cardiovascular,
connectiveandbone,gastrointestinal,immuneandhematopoietic,kidney,liver,
skeletalmuscle,andother;(3)acell-type-specificmodelincluding220cell-type-
specificannotationsforthefourhistonemarkswithputativeenhancerorpromoter
functions,H3K4me1,H3K4me3,H3K9ac,andH3K27ac.
Detailsofthe24publiclyavailableannotations,the220cell-type-specific
References
annotations,aswellasthe10celltypegroupsweredescribedbyFinucaneetal.31.
Briefly,the24annotationsincludedcoding,UTR(3′UTRand5′UTR),promoter 1. Mokry,L.E.etal.VitaminDandriskofmultiplesclerosis:amendelian
andintronicregions,acquiredfromtheUCSCGenomeBrowser32andpost- randomizationstudy.PLoSMed.12,e1001866(2015).
processedbyGusevetal.33;thethreehistonemarks(mono-methylation 2. Rhead,B.etal.Mendelianrandomizationshowsacausaleffectoflowvitamin
(H3K4me1)ofhistoneH3atlysine4,tri-methylation(H3K4me3)ofhistoneH3at Donmultiplesclerosisrisk.Neurol.Genet.2,e97(2016).
lysine4,andacetylationofhistoneH3atlysine9(H3K9ac)processedbyTrynka 3. Martineau,A.R.etal.VitaminDsupplementationtopreventacuterespiratory
etal.34–36andtwoversionsofacetylationofhistoneH3atlysine27(H3K27ac,one tractinfections:systematicreviewandmeta-analysisofindividualparticipant
versionprocessedbyHniszetal.37,anotherversionusedbythePsychiatric data.Br.Med.J.356,i6583(2017).
8 NATURECOMMUNICATIONS| (2018) 9:260 |DOI:10.1038/s41467-017-02662-2|www.nature.com/naturecommunications
ARTICLE
NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02662-2
4. Pilz,S.,Verheyen,N.,Grübler,M.R.,Tomaschitz,A.&März,W.VitaminD 34.RoadmapEpigenomicsConsortium.etal.Integrativeanalysisof111reference
andcardiovasculardiseaseprevention.Nat.Rev.Cardiol.13,404–417(2016). humanepigenomes.Nature518,317–330(2015).
5. Garland,C.F.etal.TheroleofvitaminDincancerprevention.Am.J.Public 35.ENCODEProjectConsortium.AnintegratedencyclopediaofDNAelementsin
Health96,252–261(2006). thehumangenome.Nature489,57–74(2012).
6. FernandesdeAbreu,D.A.,Eyles,D.&Féron,F.VitaminD,aneuro- 36.Trynka,G.etal.Chromatinmarksidentifycriticalcelltypesforfinemapping
immunomodulator:implicationsforneurodegenerativeandautoimmune complextraitvariants.Nat.Genet.45,124–130(2013).
diseases.Psychoneuroendocrinology34Suppl1,S265–277(2009). 37.Hnisz,D.etal.Super-enhancersinthecontrolofcellidentityanddisease.Cell
7. Lagunova,Z.,Porojnicu,A.C.,Lindberg,F.,Hexeberg,S.&Moan,J.The 155,934–947(2013).
dependencyofvitaminDstatusonbodymassindex,gender,ageandseason. 38.SchizophreniaWorkingGroupofthePsychiatricGenomicsConsortium.
AnticancerRes.29,3713–3720(2009). Biologicalinsightsfrom108schizophrenia-associatedgeneticloci.Nature511,
8. Wang,T.J.etal.CommongeneticdeterminantsofvitaminDinsufficiency:a 421–427(2014).
genome-wideassociationstudy.LancetLond.Engl.376,180–188(2010). 39.Hoffman,M.M.etal.Integrativeannotationofchromatinelementsfrom
9. Karohl,C.etal.HeritabilityandseasonalvariabilityofvitaminD ENCODEdata.NucleicAcidsRes.41,827–841(2013).
concentrationsinmaletwins.Am.J.Clin.Nutr.92,1393–1398(2010). 40.Lindblad-Toh,K.etal.Ahigh-resolutionmapofhumanevolutionary
10.Orton,S.-M.etal.EvidenceforgeneticregulationofvitaminDstatusintwins constraintusing29mammals.Nature478,476–482(2011).
withmultiplesclerosis.Am.J.Clin.Nutr.88,441–447(2008). 41.Ward,L.D.&Kellis,M.Evidenceofabundantpurifyingselectioninhumans
11.Ahn,J.etal.Genome-wideassociationstudyofcirculatingvitaminDlevels. forrecentlyacquiredregulatoryfunctions.Science337,1675–1678(2012).
Hum.Mol.Genet.19,2739–2745(2010). 42.Andersson,R.etal.Anatlasofactiveenhancersacrosshumancelltypesand
12.Hiraki,L.T.etal.Exploringthegeneticarchitectureofcirculating25- tissues.Nature507,455–461(2014).
hydroxyvitaminD.Genet.Epidemiol.37,92–98(2013). 43.Boraska,V.etal.Agenome-wideassociationstudyofanorexianervosa.Mol.
13.Boyadjiev,S.etal.Cranio-lenticulo-suturaldysplasiaassociatedwithdefectsin Psychiatry19,1085–1094(2014).
collagensecretion.Clin.Genet.80,169–176(2011). 44.GlobalLipidsGeneticsConsortium.etal.Discoveryandrefinementofloci
14.Boyadjiev,S.A.etal.Cranio-lenticulo-suturaldysplasiaiscausedbyaSEC23A associatedwithlipidlevels.Nat.Genet.45,1274–1283(2013).
mutationleadingtoabnormalendoplasmic-reticulum-to-Golgitrafficking.Nat. 45.Bentham,J.etal.Geneticassociationanalysesimplicateaberrantregulationof
Genet.38,1192–1197(2006). innateandadaptiveimmunitygenesinthepathogenesisofsystemiclupus
15.Myung,J.K.etal.Well-differentiatedliposarcomaoftheoesophagus: erythematosus.Nat.Genet.47,1457–1464(2015).
clinicopathological,immunohistochemicalandarrayCGHanalysis.Pathol. 46.Okada,Y.etal.Geneticsofrheumatoidarthritiscontributestobiologyand
Oncol.Res.17,415–420(2011). drugdiscovery.Nature506,376–381(2014).
16.Manousaki,D.etal.Low-frequencysynonymouscodingvariationinCYP2R1 47.Okbay,A.etal.Geneticvariantsassociatedwithsubjectivewell-being,
haslargeeffectsonvitaminDlevelsandriskofmultiplesclerosis.Am.J.Hum. depressivesymptoms,andneuroticismidentifiedthroughgenome-wide
Genet.101,227–238(2017). analyses.Nat.Genet.48,624–633(2016).
17.Arguelles,L.M.etal.Heritabilityandenvironmentalfactorsaffectingvitamin 48.Jostins,L.etal.Host-microbeinteractionshaveshapedthegeneticarchitecture
DstatusinruralChineseadolescenttwins.J.Clin.Endocrinol.Metab.94, ofinflammatoryboweldisease.Nature491,119–124(2012).
3273–3281(2009). 49.Cross-Disorder,GroupofthePsychiatricGenomicsConsortiumIdentification
18.Mills,N.T.etal.Heritabilityoftransforminggrowthfactor-β1andtumor ofrisklociwithsharedeffectsonfivemajorpsychiatricdisorders:agenome-
necrosisfactor-receptortype1expressionandvitaminDlevelsinhealthy wideanalysis.LancetLond.Engl.381,1371–1379(2013).
adolescenttwins.TwinRes.Hum.Genet.Off.J.Int.Soc.TwinStud.18,28–35 50.Cordell,H.J.etal.Internationalgenome-widemeta-analysisidentifiesnew
(2015). primarybiliarycirrhosisrisklociandtargetablepathogenicpathways.Nat.
19.Livshits,G.,Karasik,D.&Seibel,M.J.Statisticalgeneticanalysisofplasma Commun.6,8019(2015).
levelsofvitaminD:familialstudy.Ann.Hum.Genet.63,429–439(1999). 51.Schunkert,H.etal.Large-scaleassociationanalysisidentifies13new
20.Yu,H.-J.,Kwon,M.-J.,Woo,H.-Y.&Park,H.Analysisof25-hydroxyvitamin susceptibilitylociforcoronaryarterydisease.Nat.Genet.43,333–338
Dstatusaccordingtoage,gender,andseasonalvariation.J.Clin.Lab.Anal.30, (2011).
905–911(2016). 52.Morris,A.P.etal.Large-scaleassociationanalysisprovidesinsightsintothe
geneticarchitectureandpathophysiologyoftype2diabetes.Nat.Genet.44,
21.Hunter,D.etal.Geneticcontributiontobonemetabolism,calciumexcretion,
981–990(2012).
andvitaminDandparathyroidhormoneregulation.J.BoneMiner.Res.Off.J.
Am.Soc.BoneMiner.Res.16,371–378(2001). 53.PsychiatricGWASConsortiumBipolarDisorderWorkingGroup.Large-scale
genome-wideassociationanalysisofbipolardisorderidentifiesanew
22.Agmon-Levin,N.,Theodor,E.,Segal,R.M.&Shoenfeld,Y.VitaminDin
systemicandorgan-specificautoimmunediseases.Clin.Rev.AllergyImmunol. susceptibilitylocusnearODZ4.Nat.Genet.43,977–983(2011).
45,256–266(2013). 54.Dubois,P.C.A.etal.Multiplecommonvariantsforceliacdiseaseinfluencing
immunegeneexpression.Nat.Genet.42,295–302(2010).
23.Yang,C.-Y.,Leung,P.S.C.,Adamopoulos,I.E.&Gershwin,M.E.The
55.Thorgeirsson,T.E.etal.SequencevariantsatCHRNB3-CHRNA6and
implicationofvitaminDandautoimmunity:acomprehensivereview.Clin.
Rev.AllergyImmunol.45,217–226(2013). 56.CPiYckPr2eAll,6J.afKfe.cettsaml.oDkeintegctbioenhaavniodr.inNteartp.rGeteanteiot.n4o2f,s4h4a8r–e4d5g3en(2e0ti1c0i)n.fluenceson
24.GAescnheta.rdE,pHid.emAiople.r4sp0e,c6t7iv8e–6o8n8in(2te0r1a6c)t.ioneffectsingeneticassociationstudies. 42humantraits.Nat.Genet.48,709–717(2016).
25.Cooper,J.D.etal.InheritedvariationinvitaminDgenesisassociatedwith
predispositiontoautoimmunediseasetype1diabetes.Diabetes60,1624–1631 Acknowledgements
(2011).
AfulllistofacknowledgementscanbefoundinSupplementaryNote2.
26.Willer,C.J.,Li,Y.&Abecasis,G.R.METAL:fastandefficientmeta-analysisof
genomewideassociationscans.Bioinforma.Oxf.Engl.26,2190–2191
(2010). Author contributions
27.Yang,J.etal.Conditionalandjointmultiple-SNPanalysisofGWASsummary C.P.,E.H.,M.I.M.,E.A.S.,P.L.L.,E.B.,D.A.,S.J.W.,N.D.F.,W.H.,L.C.P.G.M.D.G.,
statisticsidentifiesadditionalvariantsinfluencingcomplextraits.Nat.Genet. N.M.V.S.,N.v.V.,J.B.R.,B.K.,T.J.W.,D.P.K.,R.S.V.,C.O.,M.L.,J.G.E.,S.B.K.,M.B.,E.S.,
44,369–375(2012). I.H.D.B.,J.I.R.,S.S.R.,M.J.,P.K.,J.F.W.,L.L.,K.M.,L.Z.,H.C.,E.T.,S.M.F.,M.G.D.,T.L.,
28.Aschard,H.,Hancock,D.B.,London,S.J.&Kraft,P.Genome-widemeta- M.K.,O.T.R.,V.M.,M.A.I.,J.W.J.,N.J.W.,C.L.,N.G.F.,K.K.,A.B.,andJ.D.designedand
analysisofjointtestsforgeneticandgene-environmentinteractioneffects. managedindividualstudies.C.P.,E.H.,M.I.M.,E.A.S.,P.L.L.,D.A.,S.J.W.,W.H.,
Hum.Hered.70,292–300(2010). N.M.V.S.,A.E.,J.B.R.,R.S.V.,C.O.,L.V.,M.L.,J.G.E.,S.B.K.,M.B.,D.V.H.,I.H.D.B.,
29.Manning,A.K.etal.Meta-analysisofgene-environmentinteraction:joint J.I.R.,S.S.R.,J.F.W.,L.L.,E.I.,K.M.,H.C.,E.T.,S.M.F.,M.G.D.,T.L.,M.K.,O.T.R.,V.M.,
estimationofSNPandSNP×environmentregressioncoefficients.Genet. M.A.I.,N.S.,J.W.J.,N.J.W.,C.L.,N.G.F.,K.K.,A.B.,andJ.D.collecteddata.E.B.,A.E.,
Epidemiol.35,11–18(2011). C.O.,M.L.,Y.L.,M.B.,E.M.K.,J.I.R.,S.S.R.,J.F.W.,L.L.,E.I.,K.M.,S.T.,S.M.F.,M.G.D.,
30.Bulik-Sullivan,B.K.etal.LDscoreregressiondistinguishesconfoundingfrom T.L.,L.L.,J.W.J.,N.J.W.,C.L.,A.B.,andJ.D.performedthegenotyping.C.P.,D.B.,E.H.,
polygenicityingenome-wideassociationstudies.Nat.Genet.47,291–295 M.I.M.,W.T.,E.B.,N.M.V.S.,A.E.,J.D.,C.O.,L.V.,M.L.,K.K.L.,J.D.,E.M.K.,J.I.R.,
(2015). S.S.R.,J.F.W.,C.H.,E.I.,K.M.,S.T.,A.G.U.,F.R.,L.Z.,S.M.F.,M.G.D.,A.M.V.,T.L.,L.L.,
31.Finucane,H.K.etal.Partitioningheritabilitybyfunctionalannotationusing M.A.I.,N.S.,N.J.W.,C.L.,A.B.,andJ.D.preparedthegenotypedata.D.B.,E.H.,M.I.M.,
genome-wideassociationsummarystatistics.Nat.Genet.47,1228–1235(2015). E.A.S.,P.L.L.,A.E.,J.B.R.,S.B.,R.S.V.,C.O.,L.V.,M.L.,J.G.E.,D.K.H.,D.V.H.,E.M.K.,
32.Kent,W.J.etal.ThehumangenomebrowseratUCSC.GenomeRes.12, I.H.D.B.,A.C.W.,J.F.W.,L.L.,K.M.,M.C.Z.,A.G.U.,F.R.,L.Z.,E.T.,S.M.F.,M.G.D.,
996–1006(2002). A.M.V.,E.T.,L.L.,N.J.W.,C.L.,N.G.F.,A.B.,andJ.D.preparedthephenotypedata.D.B.,
33.Gusev,A.etal.Partitioningheritabilityofregulatoryandcell-type-specific E.H.,M.I.M.,J.B.R.,T.J.W.,D.P.K.,Y.H.,C.L.,A.C.W.,C.R.C.,P.F.O.,M.J.,X.J.,H.A.,
variantsacross11commondiseases.Am.J.Hum.Genet.95,535–552(2014). N.J.W.,C.L.,N.G.F.,A.B.,andJ.D.developedtheanalysisplan.A.Z.,D.B.,E.H.,M.I.M.,
NATURECOMMUNICATIONS| (2018) 9:260 |DOI:10.1038/s41467-017-02662-2|www.nature.com/naturecommunications 9
ARTICLE
NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02662-2
P.L.L.,W.T.,L.C.P.G.M.D.G.,N.v.V.,A.E.,J.B.R.,B.K.,T.J.W.,J.D.,D.P.K.,Y.H.,C.L., Competinginterests:Theauthorsdeclarenocompetingfinancialinterests.
D.K.H.,I.H.D.B.,A.C.W.,J.I.R.,S.S.R.,C.R.C.,P.F.O.,M.J.,X.J.,H.A.,M.C.Z.,A.G.U.,
F.R.,L.B.,N.J.W.,C.L.,andN.G.Freviewedtheanalysisplan.D.B.,E.H.,M.I.M.,L.Y., Reprintsandpermissioninformationisavailableonlineathttp://npg.nature.com/
W.T.,A.E.,J.B.R.,Y.H.,C.L.,Y.Z.,K.K.L.,J.D.,A.C.W.,P.F.O.,A.C.,X.J.,H.A.,P.K.J., reprintsandpermissions/
C.H.,S.T.,L.B.,L.Z.,E.T.,E.T.,L.L.,J.Z.,andM.Tanalyzedthedata.D.B.,Y.H.,P.F.O.,
andH.Aperformedthemeta-analysis.Y.H.andX.J.performedthepathwayandother Publisher'snote:SpringerNatureremainsneutralwithregardtojurisdictionalclaimsin
analyses.C.P.,E.H.,M.I.M.,E.A.S.,P.L.L.,L.Y.,W.T.,E.D.M.,E.B.,L.C.P.G.M.D.G., publishedmapsandinstitutionalaffiliations.
N.v.V.,A.E.,J.B.R.,T.J.W.,J.D.,D.P.K.,Y.H.,P.F.O.,M.J.,X.J.,H.A.,E.I.,K.M.,S.T.,
M.C.Z.,A.G.U.,F.R.,L.B.,L.Z.,E.T.,N.J.W.,andN.G.Finterpretedresults.E.H.,T.J.W.,
D.P.K.,L.A.C.,R.S.V.,Y.H.,I.H.D.B.,J.I.R.,S.S.R.,M.J.,P.K.,A.G.U.,F.R.,N.S.,and
J.W.J.supervisedtheoverallstudydesign.E.H.,J.B.R.,T.J.W.,D.P.K.,Y.H.,P.F.O.,M.J., Open Access This article is licensed under a Creative Commons
X.J.,andH.A.wrotethemanuscript.C.P.,E.H.,M.I.M.,E.A.S.,P.L.L.,L.Y.,W.T.,E.D.M., Attribution 4.0 International License, which permits use, sharing,
E.B.,D.A.,S.J.W.,N.D.F.,W.H.,L.C.P.G.M.D.G.,N.M.V.S.,N.v.V.,J.B.R.,B.K.,T.J.W., adaptation,distributionandreproductioninanymediumorformat,aslongasyougive
J.D.,D.P.K.,D.K.,S.B.,R.S.V.,Y.H.,C.L.,Y.Z.,C.O.,L.V.,M.L.,J.G.E.,M.K.S.,D.K.H., appropriatecredittotheoriginalauthor(s)andthesource,providealinktotheCreative
M.P.,M.J.E.,E.M.K.,S.P.,I.H.D.B.,A.C.W.,J.I.R.,S.S.R.,C.R.C.,P.F.O.,M.J.,P.K.,X.J., Commonslicense,andindicateifchangesweremade.Theimagesorotherthirdparty
H.A.,P.K.J.,J.F.W.,C.H.,L.L.,E.I.,K.M.,S.T.,H.V.,H.W.,L.Z.,E.T.,T.S.,E.T.,T.L.,L.L., materialinthisarticleareincludedinthearticle’sCreativeCommonslicense,unless
M.K.,O.T.R.,V.M.,M.A.I.,N.S.,J.W.J.,N.J.W.,C.L.,N.G.F.,J.Z.,T.G.,K.K.,J.L.,A.B., indicatedotherwiseinacreditlinetothematerial.Ifmaterialisnotincludedinthe
J.D.,E.S.,C.G.,W.M.,M.d.H.,andM.T.reviewedthemanuscript.E.H.,M.I.M.,E.A.S.,T. article’sCreativeCommonslicenseandyourintendeduseisnotpermittedbystatutory
J.W.,D.P.K.,L.A.C.,R.S.V.,L.F.,M.P.,M.J.,P.K.,M.C.Z.,A.G.U.,F.R.,L.B.,T.S.,andM. regulationorexceedsthepermitteduse,youwillneedtoobtainpermissiondirectlyfrom
A.I.overseetheconsortium. thecopyrightholder.Toviewacopyofthislicense,visithttp://creativecommons.org/
licenses/by/4.0/.
Additional information
SupplementaryInformationaccompaniesthispaperathttps://doi.org/10.1038/s41467-
©TheAuthor(s)2018
017-02662-2.
Xia Jiang1,2, Paul F. O’Reilly3, Hugues Aschard 1,4, Yi-Hsiang Hsu5,6,7, J. Brent Richards 8, Josée Dupuis9,10,
Erik Ingelsson11,12, David Karasik5, Stefan Pilz13, Diane Berry14, Bryan Kestenbaum15, Jusheng Zheng16,
JiananLuan16,EleniSofianopoulou17,ElizabethA.Streeten18,DemetriusAlbanes19,PamelaL.Lutsey20,LuYao20,
Weihong Tang20, Michael J. Econs21, Henri Wallaschofski22,23, Henry Völzke23,24, Ang Zhou25, Chris Power14,
MarkI.McCarthy 26,27,28,ErinD.Michos29,30,EricBoerwinkle31,StephanieJ.Weinstein19,NealD.Freedman19,
Wen-Yi Huang32, Natasja M. Van Schoor33, Nathalie van der Velde34,35, Lisette C.P.G.M.de Groot 36,
Anke Enneman34, L. Adrienne Cupples9,10, Sarah L. Booth37, Ramachandran S. Vasan10, Ching-Ti Liu 9,
Yanhua Zhou9, Samuli Ripatti38, Claes Ohlsson39, Liesbeth Vandenput39, Mattias Lorentzon40,
Johan G. Eriksson41,42, M. Kyla Shea37, Denise K. Houston43, Stephen B. Kritchevsky43, Yongmei Liu44,
Kurt K. Lohman45, Luigi Ferrucci46, Munro Peacock21, Christian Gieger47, Marian Beekman 48,
Eline Slagboom48, Joris Deelen 48,49, Diana van Heemst50, Marcus E. Kleber 51, Winfried März51,52,53,
Ian H. de Boer54, Alexis C. Wood55, Jerome I. Rotter56, Stephen S. Rich57,58, Cassianne Robinson-Cohen59,
Martin den Heijer60, Marjo-Riitta Jarvelin61,62,63,64, Alana Cavadino14,65, Peter K. Joshi 66,
James F. Wilson 66,67, Caroline Hayward 67, Lars Lind12, Karl Michaëlsson68, Stella Trompet50,69,
M. Carola Zillikens60, Andre G. Uitterlinden34,60, Fernando Rivadeneira 34,60, Linda Broer60, Lina Zgaga70,
HarryCampbell66,71,EvropiTheodoratou66,71,SusanM.Farrington 71,MariaTimofeeva71,MalcolmG.Dunlop71,
Ana M. Valdes72,73, Emmi Tikkanen74, Terho Lehtimäki75,76, Leo-Pekka Lyytikäinen 75,76, Mika Kähönen77,78,
Olli T. Raitakari79,80, Vera Mikkilä81, M. Arfan Ikram 34, Naveed Sattar82, J. Wouter Jukema 69,83,
Nicholas J. Wareham16, Claudia Langenberg16, Nita G. Forouhi16, Thomas E. Gundersen84, Kay-Tee Khaw17,
Adam S. Butterworth17, John Danesh17,85, Timothy Spector72, Thomas J. Wang86,
Elina Hyppönen14,25, Peter Kraft1 & Douglas P. Kiel5,6,7
1PrograminGeneticEpidemiologyandStatisticalGenetics.DepartmentofEpidemiology,HarvardT.H.ChanSchoolofPublicHealth,677Huntington
Avenue,Boston02115,MA,USA.2UnitofCardiovascularEpidemiology,InstituteofEnvironmentalMedicine,KarolinskaInstitutet,Nobelsvagen13,
Stockholm17177,Sweden.3DepartmentofSocialGenetic&DevelopmentalPsychiatry,King’sCollegeLondon,InstituteofPsychiatry,DeCrespigny
Park,LondonSE58AF,UK.4CentredeBioinformatique,BiostatistiqueetBiologieIntégrative(C3BI),InstitutPasteur,Paris75724,France.5Institute
forAgingResearch,HebrewSeniorLife,1200CentreStreet,Boston,MA02131,USA.6DepartmentofMedicine,BethIsraelDeaconessMedical
CenterandHarvardMedicalSchool,Boston,MA02115,USA.7BroadInstituteofHarvardandMassachusettsInstituteofTechnology,Boston,MA
02142,USA.8DepartmentsofMedicine,HumanGenetics,EpidemiologyandBiostatistics,3755CôteSte-CatherineRoad,SuiteH-413Montréal,
QuébecH3T1E2,Canada.9DepartmentofBiostatistics,BostonUniversitySchoolofPublicHealth,CrosstownCenter.801MassachusettsAvenue
10 NATURECOMMUNICATIONS| (2018) 9:260 |DOI:10.1038/s41467-017-02662-2|www.nature.com/naturecommunications