Genome-wide meta-analysis associates HLA-DQA1/DRB1 and LPA and lifestyle factors with human longevity

ARTICLE Genome-wide me a-analysis associa es HLA- DQA1/DRB1 and LPA and li es yle ac o s wi h human longe i y Pe e K. Joshi e al. # Genomic analysis o longe i y o e s he po en ial o illumina e he biology o human aging. He e, using genome-wide associa ion me a-analysis o 606,059 pa en s’su i al, we disco e wo egions associa ed wi h longe i y (HLA-DQA1/DRB1 and LPA). We also alida e p e ious sugges ions ha APOE,CHRNA3/5,CDKN2A/B,SH2B3 and FOXO3A influence longe i y. Nex we show ha gi ing up smoking, educa ional a ainmen , openness o new expe ience and high-densi y lipop o ein (HDL) choles e ol le els a e mos posi i ely gene ically co ela ed wi h li espan while suscep ibili y o co ona y a e y disease (CAD), ciga e es smoked pe day, lung cance , insulin esis ance and body a a e mos nega i ely co ela ed. We sugges ha he e ec o educa ion on li espan is p incipally media ed h ough smoking while he e ec o obesi y appea s o ac ia CAD. Using ins umen al a iables, we sugges ha an inc ease o one body mass index uni educes li espan by 7 mon hs while 1 yea o educa ion adds 11 mon hs o expec ed li espan. DOI: 10.1038/s41467-017-00934-5 OPEN Co espondence and eques s o ma e ials should be add essed o P.K.J. (email: [email p o ec ed]) #A ull lis o au ho s and hei a flia ions appea s a he end o he pape NATURE COMMUNICATIONS |8: 910 |DOI: 10.1038/s41467-017-00934-5 |www.na u e.com/na u ecommunica ions 1 Longe i y is o in e es o us all, and philosophe s ha e long specula ed on he ex en o which i is p e-de e mined by a e. He e we ocus on a na owe ques ion— he ex en and na u e o i s gene ic basis and how his in e - ela es wi h ha o heal h and disease ai s. In wha ollows, we shall use longe i y as an umb ella e m. We shall also mo e specifically e e o li espan ( he du a ion o li e) and long-li edness (li ing o ex eme old age, usually defined by a h eshold, such as 90 yea s). Up o 25% o he a iabili y in human li espan has been es ima ed o be gene ic1, bu gene ic a ia ion a only h ee loci (nea APOE, FOXO3A and CHRNA3/5)2–5ha e so a been demons a ed o be obus ly associa ed wi h li espan. P ospec i e genomic s udies o li espan ha e been hampe ed by he ac ha subjec pa icipa ion is o en only ecen , allowing insu ficien ollow-up ime o a well-powe ed analysis o pa icipan su i al. On he o he hand, case-con ol s udies o long-li edness ha e had success2,3,6and some echnical appeal ( ocussing on he uly ema kable), bu such s udies can be limi ed and cos ly in hei ec ui men . We ecen ly showed ha he ex ension o he kin-coho me hod7 o pa en al li espans, beyond age 40, o geno yped subjec s could be used o de ec gene ic associa ions wi h li espan wi h some powe in genomically B i ish pa icipan s in UK Biobank (UKB)4. He e we ex end ha app oach in a genome-wide associa ion me a-analysis (GWAMA) o disco e y ac oss UKB Eu opean- and A ican-ances y popula ions and 24 u he popula ion s udies (Li eGen), mainly om Eu ope, Aus alia and No h Ame ica, o sea ch o u he gene ic a ian s influencing longe i y. We hen use hose GWAMA esul s o measu e gene ic co ela ions and ca y ou Mendelian andomisa ion (MR) be ween o he ai s and li espan seeking o elucida e he unde lying e ec s o disease and socio-economic ai s on longe i y, in a amewo k less hampe ed by con ounding and e e se causali y han obse a ional epidemiology. Resul s Genome-wide associa ion s udy. In o al, 606,059 pa en al li espans we e a ailable o analysis, o which 334,974 we e al eady comple e (Table 1). In ou GWAS o 586,626 Eu opean pa en al li espans, we find ou egions HLA-DQA1/DRB1,LPA,CHRNA3/5 and APOE,in which he lead SNPs s34831921, s55730499, s8042849 and s429358, espec i ely, associa e wi h su i al a genome-wide significance (p<5×10 −8) (Table 2, Fig. 1a, b, Fig. 2a–d). The wo p e iously un epo ed loci, s34831921 (HLA-DQA1/DRB1) and s55730499 (LPA), bo h showed s a is ically significan , di ec ionally consis en , e idence o associa ion a he p oxy SNPs in s onges LD in he la ges (5406 cases, 15,112 con ols) publicly a ailable se o GWAS summa y s a is ics o ex eme long-li edness (CHARGE-EU 90+)6, wi h p<0.0035 o bo h SNPs. As ou GWAS esul s we e o he obse ed e ec o o sp ing geno ype on pa en pheno ype and he ac ual e ec o ca ying an allele o he indi idual conce ned ( a he han hei pa en ) is wice ha obse ed in a pa en -o sp ing kin-coho s udy4, all epo ed e ec sizes (and hei s anda d e o s) h oughou his manusc ip ha e been doubled o gi e he es ima ed e ec size in he allele ca ie s hemsel es. The haza d a ios o one copy o he mino alleles we e 0.942 and 1.074 o s34831921 (HLA-DQA1/DRB1) and s5573049 (LPA), espec i ely, co esponding o an inc ease/dec ease in li espan o ~ 0.6/0.7 yea s o a ca ie o one addi ional copy o he mino allele. We me a-analysed ou esul s wi h he CHARGE-EU 90+ longe i y GWAMA6summa y s a is ics using Z-sco es and equal weigh s o each s udy, eflec ing hei simila s a is ical powe . We ound s eng hened signals, subs an ially a APOE ( s4420638, p=5.4 × 10−41) and sligh ly in he LPA egion ( s1045587, p=2.05 × 10−11). No imp o emen o s a is ical significance was obse ed in he HLA-DQA1/DRB1 egion, whe e he e we e no SNPs in s ong LD wi h he lead Li eGen SNP, no was he e an inc ease in significance nea CHRNA3/5. Howe e , in his me a-analysis one u he egion nea AKAP7/ EPB41L2 on ch omosome 6 jus eached genome-wide signifi- cance ( s1919453, A allele equency =0.36, p=4.34 × 10−8; Fig. 1c, Supplemen a y Fig. 1), and he obse ed haza d a io (SE) o he mino allele was 0.976 (0.0056) in Li eGen alone. In ou s udy o 9359 a he and 10,074 mo he li espans in pa icipan s wi h A ican ances y, no SNPs we e genome-wide (GW) significan in he analysis o bo h pa en s combined. Howe e , we ound one GW significan signal ( s10198124, G allele equency 0.39 in A ican subjec s), in an in e genic egion o ch omosome 2 associa ing wi h li espan o a he s (HR (SE) o G allele =1.22 (0.0354), p=1.66 × 10−8), wi h a consis en di ec ion o associa ion in all 9 coho s s udied. No associa ion was obse ed a his SNP in A ican mo he s, o a he s and mo he s o Eu opean ances y (HR (SE) =0.97 (0.038), 1.01 (0.007) and 1.00 (0.008), p=0.51, 0.21 and 0.77, espec i ely (Fig. 1d, Supplemen a y Fig. 2A−D). C oss- alida ion o candida e genes. We nex a emp ed o alida e 13 candida e genes iden ified in p e ious longe i y s udies. In ou s udy, only h ee o hese genes showed s a is ically significan , di ec ionally consis en e idence (p<0.0003, wo-sided es ) o associa ion; CDKN2A/B,SH2B3 and FOXO3A (Fig. 3, Supplemen a y Fig. 3and Supplemen a y Da a 3). Fo SH2B3 and FOXO3A ou es ima ed e ec sizes a e conco dan wi h hose epo ed om he mos obus (i.e., na owes 95% confidence in e al (CI)) p e ious s udy. Howe e , o CDKN2A/B, he 95% CI o ou es ima e is en i ely below ha om he mo e obus o he wo s udies conside ed. Table 1 Summa y o he Li eGen pa en al li espans Ances y Pa en Coun Mean age Ali e Dead To al Ali e Dead All A ican Fa he 2435 6924 9359 72.4 70.4 70.9 A ican Mo he 4185 5889 10,074 73.1 70.7 71.7 Eu opean Fa he 113,611 178,017 291,628 62.9 71.2 68 Eu opean Mo he 150,854 144,144 294,998 66.2 75.1 70.5 ALL 271,085 334,974 606,059 Summa y s a is ics o he 606,059 pa en al li espans ha passed pheno ypic QC (in pa icula , pa en age >40) and we e analysed he e. In p ac ice, ewe li es han hese we e analysed o some SNPs, as a SNP may no ha e passed QC in all coho s (in pa icula wi hin coho MAF >1%). The mean age o ali e pa en s ac oss Eu opean coho s was educed by he la ge iPSYCH coho , o ela i ely younge subjec s and hus pa en s, who we e p edominan ly ali e (mean a he /mo he age among he ali e pa en s in iPSYCH was 52.4/50.4) ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00934-5 2NATURE COMMUNICATIONS |8: 910 |DOI: 10.1038/s41467-017-00934-5 |www.na u e.com/na u ecommunica ions No s a is ically significan (p>0.22, wo-sided es ) e idence o associa ion was ound o he o he 10 genes. In all cases (wi h he possible excep ions o ABO and 5q33) ou es ima es o he odds a io we e close o 1 and ou 95% CI did no include p e ious es ima es, sugges ing, a leas o he emaining 8 SNPs (a o nea CAMK4,C3o 21,GRIK2,IL6,RGS7,CADM2,MINPP1 and ANKRD20A9P), ha ou non- eplica ion did no a ise solely om lack o powe . Consis en wi h ou p e ious epo s4, we ound age-specific and sex-specific e ec s o he lead SNPs in he APOE and CHRNA3/5 loci. Fo APOE, he haza d a io (SE) o he lead SNP was 1.07 (.01) o men and 1.13 (.01) o women, whe eas o CHRNA3/5 i was 1.07 (.01) o men and 1.04 (.01) o women (Fig. 4a). Con e sely, o APOE, haza d a ios s a ified by age we e 1.06 (.01) o ages 40−75 and 1.14 (.01) o ages 75+, whe eas o CHRNA3/5 hey we e 1.08 (.01) o 40−75 and 1.03 (.01) o age 75+ (Fig. 4b), wi h simila pa e ns when s a i ying by age and sex a he same ime, (Fig. 4c), al hough he dis inc- ions be ween men and women o CHRNA3/5 disappea ed beyond age 75. Fo LPA,CDKN2B and SH2B3, he e was no s a is ically significan e idence o age-specific o sex-specific e ec s, while he HLA and FOXO3 a ian s showed age bu no sex-specific e ec s (Fig. 4a, b), wi h he HLA locus ha ing a g ea e e ec a younge ages (40−75) while, con e sely, he FOXO3 locus had g ea e e ec a olde ages(75+). We es ed he ou SNPs iden ified in he disco e y phase (Table 2) o associa ion wi h o he ageing ai s, using PhenoScanne 8, an on-line ool which sea ches 88 complex ai GWAMAs and h ee GWAS ca alogues. Fo he SNP in he LPA egion, associa ions we e ound wi h blood lipids and co ona y ai s. Fo he SNP in he HLA egion, we ound associa ions wi h heuma oid a h i is and C ohn’s disease. Fo he CHRNA3/5 egion, we ound associa ions wi h ai s which associa e wi h smoking beha iou : nico ine dependence, lung cance , ch onic obs uc i e pulmona y disease and schizoph enia. Finally, o he APOE egion, we saw associa ions wi h Alzheime ’s disease, age- ela ed macula degene a ion, blood lipids, adiposi y, ca diac and cogni i e ageing ai s (Supplemen a y Da a 4). Gene ic co ela ion o complex ai s wi h li espan.We es ima ed he gene ic co ela ion be ween 113 complex quan i- a i e and disease suscep ibili y ai s and li espan using LD Sco e eg ession9: 46 showed meaning ul gene ic co ela ions ( g) wi h li espan (s a is ically significan , | g|>0.15). The mos s ongly co ela ed wi h mo ali y we e co ona y a e y disease (CAD) and ciga e es smoked pe day, g (SE) =0.66 (0.05) and 0.58 (0.11), espec i ely. Those mos nega i ely co ela ed we e yea s o schooling and o me s. cu en smoke , g (SE) =−0.47 (0.05) and −0.64 (0.09), espec i ely (Supplemen a y Fig. 4, Supplemen a y Da a 5). Lung cance , ype 2 diabe es and insulin esis ance also co ela ed ela i ely s ongly wi h ea lie mo ali y, while inc eased age a fi s bi h, openness o expe ience (a pe sonali y ai eflec ing cu iosi y s. cau ion, de e mined by ques ionnai e) and high-densi y lipop o eins (HDL) choles e ol we e co ela ed wi h la e dea h. Es ima es o g be ween 9 ai s and mo ali y and hei 95% CI ell wholly wi hin he ange [−0.15, 0.15], which we ha e labelled no meaning ully co ela ed wi h li espan. These we e emo al neck and lumba spine bone mine al densi y, se um c ea inine, ex eme heigh , heigh , bipola diso de , schizoph enia, au ism spec um diso de and pla ele coun . Fo he emaining 55 ai s, he e was insu ficien s a is ical powe o dis inguish whe he he g ell wi hin o ou side [−0.15, 0.15]. Gi en he simila i y in defini ion o many ai s (e.g., obesi y classes) and he s ong co ela ions be ween o he s, we clus e ed he 46 ai s which showed a significan and meaning ul g in o nine clus e s. Posi i e gene ic co ela ions wi h mo ali y o he clus e s anged om 0.68 (smoking) o 0.17 ( heuma oid a h i is and b eas cance ), whils nega i e co ela ions a ied om −0.50 (educa ion) o −0.15 (age a mena che); (Fig. 5, Supplemen a y Da a 5). We ound ha he beneficial ai clus e s o educa ion and happiness g oup oge he , as do a co e g oup o ac o s (obesi y, dyslipidemia/wais -hip a io (DL/ WHR), ype 2 diabe es, CAD and smoking) which show s onge co ela ion no only o mo ali y bu also among each o he , while albuminu ia and blood p essu e seem o o m hei own isk clus e . We nex conside ed whe he and o wha ex en he obse ed co ela ions be ween mo ali y and he ai clus e s a e media ed h ough o he clus e s, using pa ial co ela ions. In mos cases, he e was ela i ely li le di e ence be ween co ela ions and pa ial co ela ions wi h mo ali y (Supplemen- a y Table 1) and he di ec ion o e ec s emained he same. On he whole, he co ela ion o each isk clus e is he e o e no mainly media ed ia o he clus e s. Howe e , he en i e co ela ion o he DL/WHR clus e wi h li espan was 0.41, whe eas i s pa ial co ela ion was −0.18, implying ha one o mo e o he o he clus e s influenced he gene ic co ela ion, likely CAD wi h which i is s ongly co ela ed and whose pa ial co ela ion did no all in he same manne . Simila ly, he en i e co ela ion o he educa ion clus e wi h li espan ell om −0.50 o −0.18 as a pa ial co ela ion, in his case appa en ly due o media ion h ough smoking beha iou . Blood p essu e and age a mena che also showed educ ions in pa ial g, o nea ze o o age a mena che, consis en wi h media ion by o he ai s. Causal ela ionships wi h li espan. Finally, we used MRbase10 and u he summa y s a is ics o b eas cance (BCAC11) and C- eac i e p o ein (CHARGE-CRP12) made a ailable o us o Table 2 Fou egions associa ed wi h li espan a genome-wide significance and eplica ion ia p oxy SNPs in CHARGE sid Gene a1 F eq a1 N(000) pa en HR a1 SE P- alue Yea s P oxy 2CHARGE P Di . s34831921 HLA-DQA1 /DRB1 A 0.09 481 0.942 0.011 4.18 E-08 0.6 s3129720 0.39 0.003 + s55730499 LPA T 0.083 563 1.074 0.011 8.67 E-11 −0.7 s10455872 0.97 0.002 − s8042849 CHRNA3/5 C 0.356 567 1.046 0.006 3.75 E-14 −0.4 s9788721 0.98 0.951 − s429358 APOE C 0.142 556 1.091 0.008 1.44 E-27 −0.9 s6857 0.69 2E-20 − a1 he e ec allele, CHARGE, CHARGE Eu opean GWAS o su i o ship beyond age 90 s. younge con ols,6CHARGE P, he p- alue o he wo-sided es o associa ion be ween p oxy and long- li edness in CHARGE, Di . di ec ion o e ec o a1 in CHARGE: “+”means long-li edness inc easing, “−“means long-li edness dec easing, F eq. equency, N(000) coun ( housands o pa en s wi h li espan and subjec geno ype in o ma ion), HR, Haza d Ra io, Pp- alue o he Wald es o associa ion be ween impu ed dosage o a1 and li espan, P oxy, he closes p oxy SNP in CHARGE, 2 he linkage disequilib ium be ween he disco e y SNP and i s CHARGE p oxy, in he 1000 genomes EU panel, SE, S anda d E o , Yea s he numbe o addi ional yea s o li espan expec ed o a ca ie o one addi ional copy o a1. The e a e ou o e lapping coho s be ween he wo s udies; EGCUT, NTR, PROSPER and RS1, bu only RS1 con ibu ed cases o he CHARGE: ou o all 5406 cases analysed in CHARGE, 892 cases ( om RS1) o e lapped he 300,000 geno yped subjec s s udied in disco e y and he pheno yped indi iduals we e in any case no he same NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00934-5 ARTICLE NATURE COMMUNICATIONS |8: 910 |DOI: 10.1038/s41467-017-00934-5 |www.na u e.com/na u ecommunica ions 3 pe o m wo-sample Mendelian andomisa ion o in es iga e causal influences on li espan. O mo e han 90 es ed pheno ypes, se en isk ac o s (ciga e es smoked pe day, HDL choles e ol, LDL choles e ol, as ing insulin, sys olic blood p essu e and CRP) and six disease suscep ibili ies (Alzheime ’s disease, b eas cance , CAD, ischaemic s oke, squamous cell lung cance and ype 2 diabe es)significan ly associa ed wi h mo ali y (Table 3). Smoking causally educed li espan by 6.8 yea s o li elong smoking o one pack o 20 ciga e es a day, BMI educed li e by 7 mon hs pe uni , while educa ion causally inc eased li espan by 11 mon hs o each u he yea spen s udying. In con as o he gene ic co ela ions ( g CRP: mo ali y =0.35), gene ically aised CRP seems o ha e a li e-leng hening e ec : 5.5 mon hs o inc eased li espan pe log mg/L. We compa ed he ela i e s eng hs o hese di e en pheno ypic e ec s on li espan using a measu e independen o scale: ex apola ing he gene ic e ec s ac oss he in e qua ile pheno ypic ange. Va ia ion in smoking and sys olic blood p essu e had he s onges causal li e-sho ening e ec s (5.3 and 5.2 yea s, espec i ely), ollowed by as ing insulin, body mass index and CAD, while yea s o educa ion showed by a he mos beneficial e ec (4.7 yea s), when compa ing he es ima ed e ec o mo ing om he fi s o he hi d qua ile o he pheno ype dis ibu ion. Simila ly, we es ima e mo ing om he bo om o he op o he in e qua ile pheno ypic ange o CRP inc eases li espan by 0.7 yea s. Discussion We eplica ed p e ious findings o genome-wide significan associa ions be ween longe i y and a ian s a CHRNA3/5 and APOE and disco e ed wo u he associa ions, a LPA and HLA- DQA1/DRB1, wi h eplica ion o he u he associa ions in a long-li edness s udy. We ound no e idence o ou lead SNPs a he CHRNA3/5,LPA and HLA-DQA1/DRB1 loci associa ing wi h ai s o he han smoking beha iou , ca dio-me abolism and heuma oid a h i is, espec i ely, while finding mo e pleio opy a APOE. We also obus ly eplica ed p e ious wo k sugges ing associa ions wi h longe i y a CDKN2A/B,SH2B3/ATXN2 and FOXO3A. We ound no e idence o associa ion be ween li espan and he o he 10 loci p e iously ound o sugges i ely associa e wi h li espan, despi e appa en powe o do so. We showed s ong nega i e gene ic co ela ion be ween CAD, smoking and ype 2 diabe es and li espan, while educa ion and openness o expe ience we e posi i ely gene ically co ela ed. Using MR, we ound ha mo ing om he 25 h o 75 h pe cen ile o ciga e es pe day, sys olic blood p essu e, as ing insulin and BMI causally educed li espan by 5.3, 5.2, 4.1 and 3.8 yea s, espec i ely, and simila ly mo ing om he 25 h o 75 h pe cen ile o educa ional a ainmen causally ex ended li espan by 4.7 yea s. S ikingly, we also ound ha inc eased CRP inc eases li espan, as a causal e ec , he e e se o i s co ela ion. Lipop o ein(a) is a sphe ical lipop o ein ca ying choles e ol and iglyce ides in he bloods eam13. Va ia ion in LPA has 14 ab cd 25 20 15 10 5 0 02 Expec ed –log 10 (p) 46 12 10 –log 10 (p) Obse ed –log 10 (p) 8 6 4 2 0 14 12 10 –log 10 (p) 8 6 4 2 0 14 12 10 –log 10 (p) 8 6 4 2 0 123456 Ch omosome 7 9 11 13 16 19 123456 Ch omosome 7 9 11 13 16 19 123456 Ch omosome 7 9 11 13 16 19 Fig. 1 Genome-wide associa ions wi h pa en al li espan. Associa ion analysis was ca ied ou using impu ed allelic dosages. aManha an plo o Li eGen Eu opean ances y, wi h bo h pa en s combined; bQ−Q plo compa ing he expec ed (unde he null hypo hesis) and ac ual (obse ed) –log 10 p- alues o esul s in a;cManha an plo o me a-analysis o Li eGen Eu opeans (bo h pa en s combined) wi h CHARGE-EU 90+ published summa y s a is ics6. The me a-analysis used Z-sco es and equal weigh s, as sugges ed by he nea equali y (9.5/9.4, Li eGen, CHARGE) o Z- es s a is ics a s4420638. The addi ional (jus ) GW significan SNP lies be ween he wo ch omosome 6 hi s in a;dManha an plo o Li eGen A ican a he s only. In Manha an plo s, he y-axis has been es ic ed o 15 o aid legibili y ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00934-5 4NATURE COMMUNICATIONS |8: 910 |DOI: 10.1038/s41467-017-00934-5 |www.na u e.com/na u ecommunica ions been ex ensi ely s udied14, and ound o influence ca dio ascula disease15 and ype 2 diabe es16. A close p oxy o ou lead SNP ( s10455872, 2=0.97) has been s ongly associa ed wi h dec eased Lp(a) size and inc eased Lp(a) plasma concen a ion and is one o he s onges p edic o s o co ona y hea disease isk wi h an odds a io o 1.7 pe allele, consis en ac oss popula ions17, all sugges ing ha s55730499 a ec s mo ali y by inc easing Lp(a) le els and suscep ibili y o ca dio ascula e en s. The la ge majo his ocompa ibili y complex (MHC) encompasses HLA-DQA1/DRB1. MHC class II genes encode componen s o he an igen-p esen ing appa a us and a e he mos polymo phic egion o he human genome. Genes wi hin he MHC ha e p e iously been associa ed wi h many au oimmune condi ions and o he ai s, including pso iasis18, heuma oid a h i is19, mul iple scle osis20 and T1D21. In a ecen in o med GWAS o longe i y, Fo ney e al.22 iden ified, bu ailed o eplica e, wo a ian s close o he HLA-DRA locus22. The FOXO3A locus has been epea edly epo ed by o he s udies3,23 as associa ing wi h ex eme longe i y. Va ian s3800231, which exhibi s he s onges associa ion in ou da a, seems o exe i s beneficial e ec on people aged abo e 75 bu may ha e a neu al, o dele e ious e ec a younge ages, suppo ing he consensus ha FOXO3A plays a pu a i e ole in ex eme longe i y and gene al heal h in o old age. This con as s ou findings o he CHRNA3/5,LPA, HLA-DQA1/DRB1 loci, whe e e ec s appea o be specific o disease suscep ibili y, a he han gene al ageing. The CDKN2A/B locus a 9p21 has p e iously been associa ed wi h CAD24, while he missense allele s3184504-T we iden ified wi hin he SH2B3/ATXN2 locus has been p e iously associa ed wi h inc eased isk o ype 1 diabe es25, dias olic blood p essu e26 and se e al au oimmune condi ions27–29. The ailu e o eplica e p e ious findings o li espan inc eases a ABO and 5q33.3/EBF1 may be due o a combina ion o limi ed powe in ou s udy, despi e i s size, and a deg ee o winne ’s cu se in p e ious findings. Howe e , o CAMK4,C3o 21,GRIK2,IL6, RGS7,CADM2,MINPP1 and ANKRD20A9P, ou findings appea inconsis en wi h he p e ious wo k, sugges ing hose findings we e ei he alse posi i e associa ions, o di e ences in e ec s a e due o he di e ences be ween he ypes o li es s udied by us and o he s udies. The use o di e en coho s om a di e se ange o coun ies wi h common sha ed ances y is common in GWAMA and po en ially gi es ise o he e ogenei y in e ec sizes, wha e e he ai unde conside a ion. Howe e , a s udy o li espan is pe haps pa icula ly suscep ible o such e ec s, as mean li espans a y by coho (Supplemen a y Da a 2) and gene ic e ec s migh a y by en i onmen . None heless, such he e ogenei y is no ele an unde he null hypo hesis (e ec size =0 in all coho s) and so will no ha e induced alse posi i es. On he o he hand, 10 ab cd 100 Recombina ion a e (cM/Mb) 80 60 40 20 0 100 Recombina ion a e (cM/Mb) 80 60 40 20 0 100 Recombina ion a e (cM/Mb) 80 60 40 20 0 100 Recombina ion a e (cM/Mb) 80 60 40 20 0 15 10 5 0 8 s34831921 s55730499 s429358 s8042849 0.8 0.6 0.4 0.2 2 0.8 0.6 0.4 0.2 2 0.8 0.6 0.4 0.2 2 0.8 0.6 0.4 0.2 2 6 4 –log10(p- alue)–log10(p- alue) 2 0 10 8 6 4 –log10(p- alue)–log10(p- alue) 2 0 30 25 20 15 10 5 0 HCG23 DNAJA4 WDR61 CRABP1 IREB2 HYKK PSMA4 CHRNA5 CHRNA3 CHRNB4 LOC646938 ADAMTS7 HLA–DRB5 HLA–DRB6 HLA–DRB1 HLA–DQA1 HLA–DQB1 HLA–DQA2 HLA–DQB2 HLA–DOB TAP2 PSMB8 PSMB8–AS1 TAP1 PSMB9 SLC22A3 PVR CEACAM19 CEACAM16 BCL3 MIR8085 CBLC BCAM PVRL2 APOC1 TOMM40 APOE APOC1P1 CLPTM1 RELB CLASRP ZNF296 GEMIN7 NKPD1 PPP1R37 APOC4 APOC4–APOC2 APOC2 LPAL2 LPA PLG BTNL2 HLA–DRA 32.4 32.5 32.6 Posi ion on ch 6 (Mb) 32.7 32.8 160.8 160.9 161 Posi ion on ch 6 (Mb) 161.1 161.2 45.2 45.3 45.4 Posi ion on ch 19 (Mb) 45.5 45.6 78.6 78.7 78.8 Posi ion on ch 15 (Mb) 78.9 79 Fig. 2 Locus zoom plo s o ou genome-wide significan associa ions wi h li espan. Resul s om he me a-analysis o subjec s o Eu opean ances y analysis, o bo h pa en s combined. The displayed p- alue co esponds o ha o a wo-sided es o associa ion be ween he SNP and pa en li espan unde he Cox model. aThe s34831921 a ian , a he HLA-DQA1/DRB1 locus, P=4.18E-08. bThe s55730499 a ian , a he LPA locus, P=8.67E-11. cThe s8042849 a ian , a he CHRNA3/5 locus, P=3.75E-14. dThe s429358 a ian , a he APOE locus, P=1.44E-27 NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00934-5 ARTICLE NATURE COMMUNICATIONS |8: 910 |DOI: 10.1038/s41467-017-00934-5 |www.na u e.com/na u ecommunica ions 5 he e ogenei y may ha e educed powe and es ima ed e ec sizes should pe haps be conside ed as (sample-weigh ed) a e ages o e he coho s pa icipa ing. The lack o obse ed gene ic co ela ion be ween mo ali y and schizoph enia is pe haps su p ising, gi en he known inc eased isk o ea ly dea h due o schizoph enia30, howe e , he e we s udy li espan a e he age o 40, whe e he e ec o schizoph enia ela i e o o he causes o mo ali y is less p onounced. We conjec u e ha a s udy o ea ly mo ali y migh show a di e en pa e n, bu belie e he pa en -o sp ing kin-coho me hod would be less sui able, as pa en s would ha e o su i e beyond ep oduc ion o be a ailable o s udy. The albuminu ia clus e , which co ela ed wi h mo ali y, is unde s ood o be a consequence o poo glome ula fil a ion a ising om ch onic kidney disease, o en a ibu able o diabe es o high blood p essu e31. Ou finding ha he happiness clus e (dep essi e symp oms and subjec i e well-being) has a beneficial co ela ion wi h li espan ( g =0.24), is in line wi h a ecen me a-analysis which has shown a li e-leng hening e ec o subjec i e well-being on li espan32. Simila ly, dep ession has been shown o inc ease mo ali y, and is one o he s onges quali y-adjus ed li e expec ancy losses, wice as much as be e -s udied isk ac o s such as smoking, hea disease, s oke and diabe es33. Ou esul s hus ein o ce he impo ance o public policy ocusing no only on physical heal h bu also on gene al well-being in o de o inc ease li e expec ancy and quali y34. In gene al he esul s o he MR analyses appea consis en wi h hose o he LD sco e eg ession es ima es. This migh be expec ed since he main di e ence is ha MR compa es wo pheno ypes using jus a small numbe o SNPs which he unde lying GWAS we e powe ed o find, and LD sco e eg ession uses he whole genome. Ne e heless, as a esul he la e may indica e a sha ed he i able con ounding ac o , a he han a causal e ec , which appea s o be he case o ou CRP esul s, as he measu ed e ec o CRP on li espan is in he opposi e di ec ion o he gene ic co ela ion. CRP’s e ec s pe se a e no well unde s ood, bu ou esul s lead us o specula e i may ha e a p o ec i e unc ion, ising in he p esence o disease, a he han causing i , despi e obse a ional associa ions wi h disease and consequen a emp s o de elop a d ug o educe i 35. I ue, his pa e n is somewha analogous o findings o he N- e minal agmen o p o-BNP, which is a p o ec i e molecule, bu obse a ionally posi i ely associa es wi h ca diac ailu e and ad e se ca dio ascula ou comes36. Ou finding ha a educ ion in one BMI uni leads o a 7-mon h ex ension o li e expec ancy, appea s b oadly consis en wi h hose ecen ly published by he Global BMI Mo ali y Collabo a ion, whe e g ea e o was made o exclude con ounding and e e se causali y37. We also ound each yea longe spen in educa ion ansla es in o app oxima ely a yea longe li espan. When compa ed using he in e qua ile dis ance, isk ac o s gene ally exhibi ed s onge e ec s on mo ali y han disease suscep ibili y. Al hough bo h CAD and ciga e e smoking show a e y simila gene ic co ela ion wi h li espan, he measu ed e ec o smoking is wice as la ge as ha o CAD, pe haps because smoking influences mo ali y h ough mul iple pa hways. Ou esul s show ha longe i y is pa ly de e mined by he p edisposi ion o common diseases and, o an e en g ea e ex en , by modifiable isk ac o s. The gene ic a chi ec u e o li espan appea s complex and di e se and he e appea s o be no single gene ic elixi o long li e. Me hods Genome-wide associa ion. As is con en ional in GWAMA, analysis was ca ied ou locally a each coho and hen me a-analysed cen ally. Ini ial pheno ype and geno ype quali y con ol we e ca ied ou in acco dance wi h local s anda ds, wi h a ian s impu ed o 1000 Genomes ( ypically phase 1, e sion 3). Coho cha - ac e is ics, including geno yping and impu a ion me hods and summa y s a is ics o he pa en al li es analysed a e desc ibed in Supplemen a y Da as 1and 2. S udy p o ocols we e app o ed by he ele an commi ees o each o he local coho s. W i en in o med consen was ob ained om each pa icipan in each s udy. We conduc ed an associa ion es be ween pa en al su i al (age and ali e/dead s a us) and o sp ing geno ype. To do so, su i al ai s we e ans o med in o esiduals, pe mi ing analysis as quan i a i e ai s. To acili a e s anda disa ion ac oss he GWAS conso ium, esiduals o GWAS we e calcula ed in acco dance wi h he analysis plan se ou below using a common R p o ocol dis ibu ed o all g oups. These esidual ai s we e hen es ed o associa ion in a GWAS o e he impu ed SNP panel. Pa en s who died below he age o 40 we e excluded. Analysis was hus o su i o ship beyond he age o 40. Associa ion es ing was conduc ed unde he ollowing Cox P opo ional Haza ds Model38, hxðÞ¼h0xðÞeβXþγ1Z1þ¼þγkZk h0is he baseline, β he haza d log e a io associa ed wi h X( he e ec allele coun ) and Z 1 ,…., Z k he o he a iables fi ed i.e., subjec sex, and he fi s 10 PCs o gene ic s uc u e along wi h each s udies’usual u he co a ia es, such as ba ch o assessmen cen e. Ra he han fi he ull model in one s ep, we calcula ed Ma ingale esiduals o he Cox model (excluding X). Ma ingale esiduals39 a e, b Mi¼δi b Λ0τi ðÞe b γ1Z1þ¼þbγkZk whe e δ i and τ i a e he pa en s a us (1—dead/, 0—ali e a assessmen da e) and age o he i h indi idual, bγ1¼bγka e e ec es ima es o , Z 1 ,…., Z k. Whe e he allele coun , X, has an e ec , b Mihas a linea associa ion wi h i 39. Howe e , al hough hese esiduals a e associa ed p opo iona ely wi h he haza d a io and hus pe mi s a is ical hypo hesis es ing, hei ela ionship wi h CDKN2A/B: s1333049-C CDKN2A/B: s4977756-A S udy Disco e y Li eGen SH2B3: s3184504-T FOXO3: s3800231-G FOXO3: s2802292-T FOXO3: s10457180-A ABO: s514659-C 5q33.3/EBF1: s2149954-C CAMK4: s10491334-T C3o 21: s9825185-A GRIK2: s954551-G GRIK2: s1416280-G IL6: s2069837-G RGS7: s4611001-G RGS7: s4443878-T CADM2: s9841144-A MINPP1: s9664222-A ANKRD20A9P: s2440012-G 0.25 0.50 0.75 Longe i y OR 1.00 Fig. 3 Valida ion o associa ions epo ed elsewhe e by lookup in Li eGen. A sea ch o ecen li e a u e sugges ed he gene egions shown he e we e mos likely o ha bou associa ions wi h li espan, beyond he ou loci iden ified in Table 2, which a e u he explo ed in he Discussion. The mos powe ul Li eGen analysis (i.e., Eu opean ances y, a he and mo he combined) was used o alida ion. The odds a io (OR) o ex eme long- li edness is p esen ed o he epo ed li e-sho ening allele (i.e., he OR o long-li edness <1) in he o iginal s udy, bu no necessa ily in Li eGen. The Li eGen OR o being long-li ed was es ima ed empi ically on he assump ion ha he ela ionship be ween he Li eGen obse ed haza d a io (HR) and he OR is s able ac oss allelic e ec s, wi h APOE esul s om Li eGen and CHARGE-EU 90+ 6 being used o es ima e he a io o ln HR o ln OR (−4.7). These es ima es will only ully align wi h he published ORs i he shape o he e ec on li espan is simila o APOE, as is ue unde he p opo ional haza ds assump ion, none heless he pa e n is sugges i e. Fu he de ails a e shown in Supplemen a y Da a 3 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00934-5 6NATURE COMMUNICATIONS |8: 910 |DOI: 10.1038/s41467-017-00934-5 |www.na u e.com/na u ecommunica ions he haza d a io depends o he (pa en ) popula ion s uc u e, in pa icula he p opo ion dead. The Ma ingale esiduals we e he e o e scaled up by 1/ (p opo ion dead) sepa a ely o each pa en gende , o gi e a esidual ai wi h a 1:1 co espondence wi h he haza d a io39. This ans o med ai was hen es ed o associa ion wi h each SNP sepa a ely unde he ollowing (addi i e) model, P¼βXþe whe e βis he e ec size o he SNP (and an es ima e o he HR) and X he non- e e ence allele coun o he ma ke , wi h ebeing no mally dis ibu ed and independen . Despi e his e ficien app oach, un imes in UK Biobank we e s ill po en ially one ous, so RegScan 0.240 was used he e as i is ideally sui ed o mul iple, esidualised ai s in la ge da a se s. Fo coho s wi h significan ela edness, all bu one subjec amongs ela i es wi h coe ficien o kinship >5% we e excluded, o c ea e a (smalle ) un ela ed popula ion, in p e e ence o con en ional (po en ially mo e powe ul) mixed modelling among amily based s udies. This was done because he genomic ela ionship ma ix among o sp ing does no p ecisely eflec he gene ic co a iance among pa en al ai s. As an example, conside he o sp ing o wo b o he s: he co ela ion be ween gene ic alues o he a he ai is 0.5, bu o he mo he ai is 0, while he Gene ic Rela ionship Ma ix (GRM) en y would be 0.25 in bo h cases. The GRM hus does no ully exp ess co a iance among (pa en ) ai alues ac oss subjec s, so he smalle un ela ed popula ion was used. Excep ions o his we e o CILENTO, ERF, GeneSTAR, MICROS and OGP, whe e i was imp ac ical o exclude ela i es, and mixed modelling was used. A e p epa ing GWAS esul s locally, coho s submi ed hese o he cen al eam o me a-analysis. The cen al me a-analysis was ca ied ou in METAL, wi h QC ollowing ha o Easy-QC41, bu some imes mo e conse a i e, as ollows. UK Biobank da a we e ead in o METAL42 fi s , s anda dising all subsequen inpu alleles o ha impu a ion. SNPs wi h misma ching alleles in o he GWAS we e ejec ed. SNPs we e emo ed om a coho ’s GWAS i he mino allele equency o ha coho was <0.01. As all s udies had in excess o 500 li es, his mean ha mino allele coun exceeded 10 Alleles wi h an in o sco e (obse ed a iance in dosage/expec ed unde HWE) <0.3 we e excluded. Each GWAS was checked o sys ema ic e o s in allele coding/ equencies and es s a is ics o SNPs passing QC. A e QC, SNP coun s we e 13,689,868 o Eu opean a he s, 13,643,373 o Eu opean mo he s, 20,305,364, o A ican a he s and 20,296,065 o A ican mo he s. A ican and Eu opean ances ies we e me a-analysed sepa a ely, as we e he esul s o each pa en al sex, using in e se a iance me a-analysis in METAL. Double genomic con ol was applied. The median λ o 78 GWAS was 0.998 and he maximum was 1.048, sugges ing good con ol o s a ifica ion. The highes λ was o UK Biobank—genomically B i ish, he mos powe ed s udy. A e he fi s le el o genomic con ol, esul s we e me a-analysed by in e se a iance, while keeping con inen al ances y sepa a e and pa en al sex sepa a e. The λapplied was 1.034, 1.023, 1.027, 1.028, o Eu opean a he s, mo he s, A ican a he s, mo he s, espec i ely. Finally, wi hin con inen ac oss pa en in e se a iance me a-analysis was applied. As expec ed, due o en i onmen al co ela ion among spouses, he e was some infla ion: λo 1.107 and 1.094, o Eu opeans and A icans, espec i ely, gi ing wo final combined me a-analyses (A ican and Eu opean) o bo h pa en s combined, subjec o double genomic con ol. These GWAS esul s we e o he obse ed e ec o o sp ing geno ype on pa en pheno ype. The ac ual e ec o ca ying an allele o he indi idual conce ned ( a he han hei pa en ) is wice ha obse ed in a pa en -o sp ing kin-coho s udy4. All epo ed e ec sizes h oughou his manusc ip we e he e o e doubled o gi e he es ima ed e ec size in he allele ca ie s hemsel es. The e ec o haza d a ios on li espan was calcula ed om su i al cu es o he Cox model by each coho . The weigh ed a e age e ec o haza d a io on li espan ac oss all coho s and bo h sexes was ha a 1% educ ion in haza d ex ended expec ed li espan by 0.108 yea s. To a oid an undue sense o p ecision, and in acco dance wi h an ac ua ial ule o humb, whe e applicable, haza d a ios we e con e ed o es ima ed e ec s on li espan using a 10% HR: −1 yea o li espan a io. Genome-wide significan Eu opean lead SNPs a each QTL we e hen looked up in he la ges independen GWAS o li espan wi h published summa y s a is ics, o su i o ship beyond age 90 s. younge con ols (CHARGE-EU 90+)6. None o he lead SNPs we e p esen in ha da ase , so p oxy SNPs in s onges LD we e chosen using LDlink43, wi h Eu opean popula ions selec ed. The SNP showing he highes 2wi h each Li eGen lead SNP was ex ac ed om he CHARGE GWAS. The Ro e dam s udy was pa o bo h GWAMAs, bu he ai measu ed was in di e en people. In ou s udy, we conside ed he li espan o pa en s, whe eas he long-li edness analysis was in he o sp ing. The Li eGen and CHARGE-EU 90+ GWAMAs we e hen me a-analysed using p- alues and di ec ion o e ec (a e e e sing he sign o e ec o CHARGE o con e longe i y o mo ali y) wi h equal weigh s placed on each GWAMA, using METAL. The choice o equal weigh s was made, a he han weigh s eflec ing sample size, because (i) he CHARGE ex eme case-con ol app oach is mo e powe ul pe sample han pa en li espan Cox modelling, and compa ison o nis no s aigh o wa d, (ii) he Z- es s a is ics o s4420638 ( he mos significan SNP o e lapping in bo h s udies) we e simila : 9.4 and 9.5 o CHARGE and Li eGen, espec i ely, o he same n, indica ing simila o e all powe . We used PhenoScanne 8 o sea ch o o he ai associa ions wi h ou lead SNPs. Pheno Scanne se ings we e: sid, Ca alogue =GWAS, p- alue cu o =.001, p oxies =1000 G, 2=0.6. A e iew o ecen li e a u e was conduc ed o SNPs ha ha e been associa ed wi h longe i y and li espan by o he esea che s, o see i we could alida e hei esul s. Nine pape s published since 2008 we e selec ed: B oe e al.3, Deelen e al.6, Emanuele e al.44, Flachsba e al.45, Fo ney e al.22, Malo ini e al.46, Newman e al.47, Willcox e al.23 and Zeng e al.48. Va ian s wi h MAF abo e 1% in 1000 Genomes, om hese pape s we e aken o wa d, i hey exhibi ed genome-wide significance (p<5×10 −8) o hey had sugges i e associa ions ha we e eplica ed. APOE a b c APOE 40–75 CHRNA3/5 40–75 LPA 40–75 HLA 40–75 CDKN2A/B 40–75 SH2B3 40–75 FOXO3 40–75 CDKN2A/B 75+ SH2B3 75+ FOXO3 75+ LPA 75+ HLA 75+ CHRNA3/5 75+ APOE 75+ CHRNA3/5 LPA HLA CDKN2B Gene Gene SH2B3 FOXO3 APOE CHRNA3/5 LPA HLA CDKN2B Gene SH2B3 FOXO3 1.00 1.05 1.10 Haza d a io 1.15 S udy Fa he s Mo he s S udy Fa he s Mo he s 40–75 75+ S udy 1.00 1.05 1.10 Haza d a io 1.15 1.00 1.05 1.10 Haza d a io 1.15 1.20 Fig. 4 Age-specific and sex-specific e ec s o he 4 GWS associa ions in Li eGen and he alida ed candida e loci. The ou GWS and h ee sugges i e eplica ed loci we e analysed o age-specific and sex-specific e ec s on li espan. aThe a ian s a APOE and CHRNA3/5 exhibi sexually dimo phic e ec s on pa en al mo ali y, while all o he a ian s exhibi mo e modes o en non-significan sex-specific di e ences. bThe e ec s o each gene on male and emale li espan we e me a-analysed and s udied in he cases ha died aged be ween 40 and 75 o a e 75. APOE exe s a much g ea e e ec in he olde age g oup, while mos o he o he genes exhibi he opposi e e ec . FOXO3 appea s neu al, i no posi i e, in he ea lie age g oup. cE ec s on mo ali y we e s udied in bo h age g oups o bo h sexes. APOE has he s onges e ec on emales aged 75+, CHRNA3/5 ac s on males aged 40−75 and all o he genes display mo e ambiguous ends NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00934-5 ARTICLE NATURE COMMUNICATIONS |8: 910 |DOI: 10.1038/s41467-017-00934-5 |www.na u e.com/na u ecommunica ions 7 1 0.25 0.11 –0.23 –0.09 –0.19 –0.01 –0.25 –0.32 –0.23 –0.28 –0.48 –0.5 1 0.12 0.03 –0.17 0.01 –0.11 0.16 0 –0.17 0.06 0.05 –0.14 –0.18 0.12 1 0.03 0.12 –0.06 0.03 0.16 0.07 –0.1 0.12 0.11 0.02 –0.19 0.03 0.03 1 0.03 0.16 0.06 –0.07 –0.08 0.11 –0.01 –0.02 –0.14 –0.01 –0.17 0.12 0.03 1 0.11 –0.09 –0.13 0.03 –0.11 –0.02 0.02 0.03 0.09 0.01 –0.06 0.16 0.11 1 –0.02 0.01 –0.09 0.01 –0.01 –0.23 –0.07 0.23 –0.11 0.03 0.06 –0.09 –0.02 1 0.17 –0.17 –0.13 0.16 0.12 –0.05 0.16 0.16 0.16 –0.07 –0.13 0.01 0.17 1 –0.09 0.22 –0.18 –0.15 –0.15 0.34 0 0.07 –0.08 0.03 –0.09 –0.17 –0.09 1 0.46 –0.04 –0.12 –0.09 0.24 –0.17 –0.1 0.11 –0.11 0.01 –0.13 0.22 0.46 1 0.38 0.26 0.19 –0.19 0.06 0.12 –0.01 –0.02 –0.01 0.16 –0.18 –0.04 0.38 1 –0.13 –0.15 0.33 0.05 0.11 –0.02 0.02 –0.23 0.12 –0.15 –0.12 0.26 –0.13 1 –0.34 0.62 –0.14 0.02 –0.14 0.03 –0.07 –0.05 –0.15 –0.09 0.19 –0.15 –0.34 1 0.61 –0.18 –0.19 –0.01 0.09 0.23 0.16 0.34 0.24 –0.19 0.33 0.62 0.61 1 0.25 1 0.06 0.01 –0.11 –0.03 0.05 –0.08 –0.13 –0.04 –0.08 –0.22 –0.24 0.11 0.06 1 0 0.13 0.01 –0.07 –0.12 –0.04 –0.04 –0.08 –0.21 –0.15 –0.23 0.01 0 1 0.14 –0.03 –0.12 0.06 –0.02 0.03 0.08 0.17 0.17 –0.09 –0.11 0.13 0.14 1 0.05 0.06 –0.04 –0.01 0.05 –0.06 0.11 0.17 –0.19 –0.03 0.01 –0.03 0.05 1 0.27 –0.06 0.05 0.26 0.35 0.17 0.39 –0.01 0.05 –0.07 –0.12 0.06 0.27 1 0.09 0.21 0.11 0.21 0.13 0.33 –0.25 –0.08 –0.12 0.06 –0.04 –0.06 0.09 1 0.56 0.29 0.23 0.28 0.37 –0.32 –0.13 –0.04 –0.02 –0.01 0.05 0.21 0.56 1 0.48 0.37 0.34 0.41 –0.23 –0.04 –0.04 0.03 0.05 0.26 0.11 0.29 0.48 1 0.33 0.27 0.48 –0.28 –0.08 –0.08 0.08 –0.06 0.35 0.21 0.23 0.37 0.33 1 0.28 0.66 –0.48 –0.22 –0.21 0.17 0.11 0.17 0.13 0.28 0.34 0.27 0.28 1 0.68 –0.5 Edu 1 0.5 0 –0.5 –1 1 0.5 0 –0.5 –1 Happiness AM RA BC BP Albuminu ia Obesi y DS/WHR T2D CAD Smoking Mo ali y Edu Happiness AM RA BC BP Albuminu ia Obesi y DS/WHR T2D CAD Smoking Mo ali y Edu Happiness AM RA BC BP Albuminu ia Obesi y DS/WHR T2D CAD Smoking Mo ali y Edu Happiness AM RA BC BP Albuminu ia Obesi y DS/WHR T2D CAD Smoking Mo ali y –0.24 –0.15 0.17 0.17 0.39 0.33 0.37 0.41 0.48 0.66 0.68 1 Fig. 5 Gene ic co ela ions be ween ai clus e s ha associa e wi h mo ali y. The uppe panel shows whole gene ic co ela ions, he lowe panel, pa ial co ela ions. T2D, ype 2 diabe es; BP, blood p essu e; BC, b eas cance ; CAD, co ona y a e y disease; Edu, educa ional a ainmen ; RA, heuma oid a h i is; AM, age a mena che; DL/WHR Dyslipidaemia/Wais -Hip a io; BP, blood p essu e ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00934-5 8NATURE COMMUNICATIONS |8: 910 |DOI: 10.1038/s41467-017-00934-5 |www.na u e.com/na u ecommunica ions In agg ega e, 18 a ian s in 13 gene egions we e iden ified, o which ou we e genome-wide significan in he o iginal s udy, while nine we e sugges i e (Fig. 3). These lead SNPs we e hen looked up in ou esul s and compa ed wi h he p e iously epo ed associa ions (Supplemen a y Da a 3and Fig. 3). Whils compa able p- alues we e di ec ly appa en , we also wished o compa e e ec sizes, in e alia o unde s and whe he non- eplica ion in e ms o p- alue a ose om lack o powe o inconsis ency in obse ed e ec . Howe e , his was no s aigh o wa d due o he di e en s udy designs, p incipally ha we obse ed haza d a ios o mo ali y, while o he s udies obse ed odds a ios o ex eme long-li edness (o en o sligh ly di e en defini ions o cases and con ols). We he e o e p oceeded as ollows. The mos significan longe i y associa ion, APOE, was used o es ima e he ela ionship be ween OR obse ed in case-con ol s udies and HR obse ed by us, as ollows. Fo APOE a ian s4420638(G) log e OR o su i al beyond age 90 has been es ima ed elsewhe e as −0.336and ou obse ed HR was 0.07, gi ing an empi ical ac o o −4.7 o es ima e ORs o case-con ol ex eme long-li edness om li espan HRs. This ac o was applied o ou obse ed HR o all he candida e SNPs in Fig. 3, gi ing an empi ical es ima e om ou da a o he OR o ex eme long-li edness. Some s udies did no epo s anda d e o s o hei ORs, me ely p- alue and e ec es ima e. We in e ed s anda d e o s, assuming ha a wo-sided es wi h a no mally dis ibu ed es ima o had been used. Gene ic co ela ions. We es ima ed gene ic co ela ions be ween mo ali y and o he ai s using ou bo h pa en s Eu opean ances y GWAMA summa y s a is ics and he LDHub web po al (h p://ldsc.b oadins i u e.o g/)49. As he pa en pheno ype-o sp ing geno ype GWAS hal es he gene ic e ec s4, bo h he gene ic co a iance and sq (he i abili y) es ima es a e hal ed, esul ing in 1:1 es ima ion o he o sp ing-o sp ing gene ic co ela ion( g), om pa en al GWAS-o sp ing GWAS based es ima es o g. LDHub es ima es g be ween one es GWAS and ~ 200 ai s om me abolomics o common diseases such as ca dio ascula disease and lung cance , using LD sco e eg ession9. Gi en hei edundancy and numbe , he me abolomic ai s we e excluded om he analysis. We added dias olic and sys olic blood p essu e50, C- eac i e p o ein (CRP)12 and b eas cance 11 o he ai s p esen in LDhub49, using GWAMA summa y s a is ics o hese s udies p o ided o us. Each o hese was un h ough he LDHub se e in o de o es ima e he gene ic co ela ions wi h he o he ai s while he gene ic co ela ion wi h li espan was es ima ed by using a local un o LD sco e eg ession. The Benjamini and Hochbe g mul iple co ec ion es p ocedu e was applied o de e mine he s a is ical significance o he esul ing gene ic co ela ions. We hen defined h ee ca ego ies o ai s: (a) Meaning ully gene ically co ela ed o mo ali y i es ima ed g >=0.15 and FDR <0.05; (b) No meaning ully gene ically co ela ed o mo ali y i 95% CI o g ⊂[−0.15,0.15]; and (c) O he wise, insu ficien e idence. A e subse ing o only hose meaning ully gene ically co ela ed o mo ali y, we es ima ed all gene ic co ela ions among hose ai s; some pai s o ai s showed e y high co ela ions. Fo example, many we e gene ically co ela ed o BMI and obesi y, we hus used he ICLUST clus e ing algo i hm o clus e he mos simila ones. The numbe o clus e s was chosen empi ically, by isual inspec ion. The ICLUST algo i hm om he psych R package clus e s i ems hie a chically based on he loading o he i ems on he ac o s om ac o analysis. Two clus e s a e hen me ged oge he only i by hei joining hei in e nal consis ency inc eases. As o a ion ma ix o he ac o analysis we used “p omax”which is a high e ficiency algo i hm which allows co ela ion be ween he di e en ac o s51. O he han o define he ini ial lis , mo ali y was no included in he clus e ing analysis. A he same ime, some highly co ela ed ai s, which he clus e ing algo i hm sough o combine, appea ed o cap u e dis inc clinical aspec s and hese we e he e o e kep sepa a e. In pa icula , we spli an educa ion/smoking/ Table 3 Mendelian andomisa ion associa ions o he 19 ai s wi h li espan Exposu e SNPs in he IV Be a SE P- alue Egge pleio opy P SD Yea s pe exposu e uni In e qua ile e ec in yea s Risk ac o Body mass index SD (kg/m2) 65 0.279 0.04 2.26 × 10−12 0.4 4.77 0.584 3.8 Yea s o schooling SD (yea s) 64 −0.348 0.054 9.42 × 10−11 0.039 3.71 −0.937 −4.7 Ciga e es smoked pe day s12914385 0.034 0.005 6.47 × 10−10 −11.7 0.338 5.3 HDL choles e ol SD (mg/dL) 39 −0.106 0.044 0.017 0.793 15.5 −0.068 −1.4 LDL choles e ol SD (mg/dL) 17 0.101 0.042 0.017 0.82 38.7 0.026 1.4 Fas ing insulin log pmol/L 6 0.389 0.176 0.027 0.823 0.79 3.89 4.1 SBP mmHg s381815 0.02 0.009 0.031 −18.9 0.204 5.2 CRP log mg/L 39 −0.046 0.021 0.033 0.073 1.08 −0.458 −0.66 DBP mmHg 3 0.029 0.015 0.056 0.248 Omega-3 a y acids (SD) s145717049 −0.229 0.182 0.208 − To al choles e ol SD (mg/dL) 11 0.036 0.068 0.597 0.348 T iglyce ides SD (mg/dL) 18 0.034 0.093 0.72 0.185 Apolipop o ein B (SD) 3 0.013 0.067 0.846 0.918 Disease suscep ibili y Alzheime ’s disease 18 0.035 0.013 0.009 0.783 −− 0.77 B eas cance 109 0.034 0.007 7.11 × 10−60.318 −− 0.74 Co ona y a e y disease 26 0.13 0.02 3.22 × 10−11 0.125 −− 2.9 Ischaemic s oke s4984814 0.012 0.003 1.39 × 10−5−−− 0.26 Squamous cell lung cance 2 0.073 0.03 0.014 −−− 1.6 Type 2 diabe es 22 0.036 0.015 0.02 0.247 −− 0.79 The 19 ai s which we e significan in he fi s s ep analysis a e shown. Exposu e, lis o exposu es es ed ( o ai s in which he be as in he o iginal GWAS we e exp essed in s anda d de ia ions, SD has been added a e he name o he exposu e). Abb e ia ions/defini ions: SNPs in he IV, he numbe o a ian s in he ins umen al a iable, o he iden i y o he SNP i <2. Be a, e ec s o exposu e on li espan exp essed as he log haza d a io o he Cox model, i.e., pa en /o sp ing e ec sizes ha e been doubled. Fo ai s analysed in SD uni s, he be as e e o a a ia ion o one s anda d de ia ion. CRP, C- eac i e p o ein, DBP, dias olic blood p essu e, HDL, high-densi y lipop o ein, LDL, low-densi y lipop o ein, SE, he s anda d e o o be a. Egge pleio opy P e e s o he p- alue om he MR Egge eg ession. SD, s anda d de ia ion o he exposu e. Reduced yea s o li e pe exposu e uni , educ ion in li espan exp essed in yea s pe measu emen uni o he exposu e (no SD uni s, e en o ai s whe e be a is in SD uni s). A nega i e numbe indica es a longe li espan. In e qua ile e ec on mo ali y (yea s), ex apola ed di e ence in yea s o li e be ween someone a he 3 d and 1s qua iles o he pheno ypic dis ibu ion, i.e., a 1.34 SD di e ence o quan i a i e ai s and 2.2 poin s on he log(OR) scale o bina y ai s. SBP, sys olic blood p essu e NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-00934-5 ARTICLE NATURE COMMUNICATIONS |8: 910 |DOI: 10.1038/s41467-017-00934-5 |www.na u e.com/na u ecommunica ions 9