Evidence for large-scale gene-by-smoking interaction effects on pulmonary function

Tobacco E idence o la ge-scale gene-by-smoking in e ac ion e ec s on pulmona y unc ion Hugues Ascha d, 1,2* Ma in D Tobin, 3,4 Dana B Hancock, 5 Da id Sku nik, 6 Akshay Sood, 7 Alan James, 8,9 Albe Ve non Smi h, 10,11 Ani W Manichaikul, 12,13 A chie Campbell, 14,15 B am P P ins, 16 Ca oline Haywa d, 17 Daan W Lo h, 18 Da id J Po eous, 14,15 Da id P S achan, 19 Ele he ia Zeggini, 16 Geo ge T O’Conno , 20,21 Guy G B usselle, 18,22,23 H Ma ike Boezen, 24,25 Holge Schulz, 26,27 Ian J Dea y, 28,29 Ian P Hall, 30 Igo Rudan 31 Jaakko Kap io, 32,33,34 James F Wilson, 31,17 Jemma B Wilk, 20 Jenni e E Hu man, 17 Jing Hua Zhao, 35,36 Kim de Jong, 24,25 Leo-Pekka Lyy ik€ ainen, 37,38 Louise V Wain, 3,4 Ma jo-Rii a Ja elin 39,40,41,42 Mika K€ aho¨ nen, 43 My iam Fo nage, 44 Oz en Polasek 31,45 Pa icia A Cassano, 46,47 R G aham Ba , 48 Rajesh Rawal 49,50,51 Sa ah E Ha is, 14,28 Sina A Gha ib, 52 S e an En o h, 53 Susan R Heckbe , 55 Te ho Leh im€ aki, 37,38 Ul Gyllens en, 53 Unde s anding Socie y Scien i ic G oup, Vic o ia E Jackson, 3 Vilmundu Gudnason, 10,11 Wenbo Tang, 46,55 Jose´e Dupuis, 20,56 Ma  ıa Sole A igas, 3 Ami D Joshi, 1,2,57 S ephanie J London 58† and Pe e K a 1,2† 1 Depa men o Epidemiology, Ha a d TH Chan School o Public Heal h, Bos on, MA, USA, 2 P og am in Gene ic Epidemiology and S a is ical Gene ics, Ha a d TH Chan School o Public Heal h, Bos on, MA, USA, 3 Gene ic Epidemiology G oup, Depa men o Heal h Sciences, Uni e si y o Leices e , Leices e , UK, 4 Na ional Ins i u e o Heal h Resea ch, Leices e Respi a o y Biomedical Resea ch Uni , Glen ield Hospi al, Leices e , UK, 5 Beha io al and U ban Heal h P og am, Beha io al Heal h and C iminal Jus ice Resea ch Di ision, Resea ch T iangle Ins i u e (RTI) In e na ional, Resea ch T iangle Pa k, NC, USA, 6 Di ision o In ec ious Diseases, B igham and Women Hospi al, Ha a d Medical School, Bos on, MA, USA, 7 Di ision o Pulmona y, C i ical Ca e and Sleep Medicine, Depa men o In e nal Medicine, Uni e si y o New Mexico School o Medicine, Albuque que, NM, USA, 8 Depa men o Pulmona y Physiology and Sleep Medicine, Si Cha les Gai dne Hospi al, Nedlands, Aus alia, 9 School o Medicine and Pha macology, Uni e si y o Wes e n Aus alia, C awley, Aus alia, 10 Icelandic Hea Associa ion, Kopa ogu , Iceland, 11 Facul y o Medicine, Uni e si y o Iceland, Reykja ik, Iceland, 12 Cen e o Public Heal h Genomics, Uni e si y o Vi ginia, Cha lo es ille, VA, USA, 13 Depa men o Public Heal h Sciences, Di ision o Bios a is ics and Epidemiology, Uni e si y o Vi ginia, Cha lo es ille, VA, USA, 14 Cen e o Genomic & Expe imen al Medicine, Ins i u e o Gene ics & Molecula Medicine, Uni e si y o Edinbu gh, Edinbu gh, UK, 15 Gene a ion Sco land, Cen e o Genomic and Expe imen al Medicine, Uni e si y o Edinbu gh, Edinbu gh, UK, 16 Depa men o Human Gene ics, Wellcome T us Sange V CThe Au ho 2017. Published by Ox o d Uni e si y P ess on behal o he In e na ional Epidemiological Associa ion 894 This is an Open Access a icle dis ibu ed unde he e ms o he C ea i e Commons A ibu ion License (h p://c ea i ecommons.o g/licenses/by/4.0/), which pe mi s un es ic ed euse, dis ibu ion, and ep oduc ion in any medium, p o ided he o iginal wo k is p ope ly ci ed. In e na ional Jou nal o Epidemiology, 2017, 894–904 doi: 10.1093/ije/dyw318 Ad ance Access Publica ion Da e: 12 Janua y 2017 O iginal a icle Downloaded om h ps://academic.oup.com/ije/a icle-abs ac /46/3/894/2894383/E idence- o -la ge-scale-gene-by-smoking by Tampe e Uni e si y Lib a y use on 09 Oc obe 2017 Ins i u e, Hinx on, UK, 17 MRC Human Gene ics Uni , Ins i u e o Gene ics and Molecula Medicine, Uni e si y o Edinbu gh, Edinbu gh, UK, 18 Depa men o Epidemiology, E asmus Medical Cen e , Ro e dam, The Ne he lands, 19 Popula ion Heal h Resea ch Ins i u e, S Geo ge’s Uni e si y o London, London, UK, 20 The Na ional Hea , Lung, and Blood Ins i u e’s F amingham Hea S udy, F amingham, MA, USA, 21 The Pulmona y Cen e , Depa men o Medicine, Bos on Uni e si y School o Medicine, Bos on, MA, USA, 22 Depa men o Respi a o y Medicine, Ghen Uni e si y Hospi al, Ghen , Belgium, 23 Depa men o Respi a o y Medicine, E asmus Medical Cen e , Ro e dam, The Ne he lands, 24 Uni e si y o G oningen, Uni e si y Medical Cen e G oningen, Depa men o Epidemiology, G oningen, The Ne he lands, 25 Uni e si y o G oningen, Uni e si y Medical Cen e G oningen, G oningen Resea ch Ins i u e o As hma and COPD, G oningen, The Ne he lands, 26 Ins i u e o Epidemiology I, Helmhol z Zen um Mu¨nchen, Ge man Resea ch Cen e o En i onmen al Heal h, Neuhe be g, Ge many, 27 Comp ehensi e Pneumology Cen e Munich (CPC-M), Membe o he Ge man Cen e o Lung Resea ch, Munich, Ge many, 28 Cen e o Cogni i e Ageing and Cogni i e Epidemiology, Uni e si y o Edinbu gh, Edinbu gh, UK, 29 Depa men o Psychology, Uni e si y o Edinbu gh, Edinbu gh, UK, 30 Di ision o Respi a o y Medicine, Uni e si y o No ingham, Queen’s Medical Cen e, No ingham, UK, 31 Cen e o Global Heal h Resea ch, Ushe Ins i u e o Popula ion Heal h Sciences and In o ma ics, Uni e si y o Edinbu gh, Edinbu gh, UK, 32 Depa men o Public Heal h, Uni e si y o Helsinki, Helsinki, Finland, 33 Ins i u e o Molecula Medicine, Uni e si y o Helsinki, Helsinki, Finland, 34 Na ional Ins i u e o Heal h and Wel a e, Depa men o Heal h, Helsinki, Finland, 35 MRC Epidemiology Uni , Uni e si y o Camb idge School o Clinical Medicine, Camb idge, UK, 36 Ins i u e o Me abolic Science, Biomedical Campus, Camb idge, UK, 37 Depa men o Clinical Chemis y, Fimlab Labo a o ies, Tampe e, Finland, 38 Depa men o Clinical Chemis y, Uni e si y o Tampe e School o Medicine, Tampe e, Finland, 39 Depa men o Epidemiology and Bios a is ics, MRC–PHE Cen e o En i onmen & Heal h, School o Public Heal h, Impe ial College London, UK, 40 Cen e o Li e Cou se Epidemiology, Facul y o Medicine, Uni e si y o Oulu, Oulu, Finland, 41 Biocen e Oulu, Uni e si y o Oulu, Oulu, Finland, 42 Uni o P ima y Ca e, Oulu Uni e si y Hospi al, Oulu, Finland, 43 Depa men o Clinical Physiology, Uni e si y o Tampe e and Tampe e Uni e si y Hospi al, Tampe e, Finland, 44 B own Founda ion Ins i u e o Molecula Medicine, Uni e si y o Texas Heal h Science Cen e a Hous on, Hous on, TX, USA, 45 Facul y o Medicine, Uni e si y o Spli , Spli , C oa ia, 46 Di ision o Nu i ional Sciences, Co nell Uni e si y, I haca, NY, USA, 47 Depa men o Heal hca e Policy and Resea ch, Weill Co nell Medical College, NY, NY, USA, 48 Depa men s o Medicine and Epidemiology, Columbia Uni e si y Medical Cen e , 49 Ins i u e o Gene ic Epidemiology, Helmhol z Zen um Mu¨nchen, Ge man Resea ch Cen e o En i onmen al Heal h, Neuhe be g, Ge many, 50 Resea ch Uni o Molecula Epidemiology, Helmhol z Zen um Mu¨nchen, Ge man Resea ch Cen e o En i onmen al Heal h, Neuhe be g, Ge many, 51 Ins i u e o Epidemiology II, Helmhol z Zen um Mu¨nchen, Ge man Resea ch Cen e o En i onmen al Heal h, Neuhe be g, Ge many, 52 Compu a ional Medicine Co e a Cen e o Lung Biology, Di ision o Pulmona y & C i ical Ca e Medicine, Uni e si y o Washing on, Sea le, WA, 53 Depa men o Immunology, Gene ics and Pa hology, Uppsala Uni e si e , Science o Li e Labo a o y, Uppsala, Sweden, 54 Ca dio ascula Heal h Resea ch Uni and Depa men o Epidemiology, Uni e si y o Washing on, Sea le, WA, USA, 55 Boeh inge Ingelheim Pha maceu icals, Inc., Ridge ield, CT, USA, 56 Depa men o Bios a is ics, Bos on Uni e si y School o Public Heal h, Bos on, MA, USA, 57 Di ision o Gas oen e ology, Massachuse s Gene al Hospi al, Bos on, MA, USA. Human Se ices, Resea ch T iangle Pa k, NC, USA and 58 Epidemiology B anch, Na ional Ins i u e o En i onmen al Heal h Sciences, Na ional Ins i u es o Heal h, US Depa men o Heal h and Human Se ices, Resea ch T iangle Pa k, NC, USA *Co esponding au ho . Depa men o Epidemiology, Ha a d School o Public Heal h, Building 2, Room 205, 665 Hun ing on A enue, Bos on, MA 02115, USA. E-mail: [email p o ec ed] † These au ho s con ibu ed equally o his wo k. Accep ed 10 Oc obe 2016 In e na ional Jou nal o Epidemiology, 2017, Vol. 46, No. 3 895 Downloaded om h ps://academic.oup.com/ije/a icle-abs ac /46/3/894/2894383/E idence- o -la ge-scale-gene-by-smoking by Tampe e Uni e si y Lib a y use on 09 Oc obe 2017 Abs ac Backg ound: Smoking is he s onges en i onmen al isk ac o o educed pulmona y unc ion. The gene ic componen o a ious pulmona y ai s has also been demon- s a ed, and a leas 26 loci ha e been ep oducibly associa ed wi h ei he FEV 1 ( o ced expi a o y olume in 1 second) o FEV 1 /FVC (FEV 1 / o ced i al capaci y). Al hough he main e ec s o smoking and gene ic loci a e well es ablished, he ques ion o po en ial gene-by-smoking in e ac ion e ec emains unanswe ed. The aim o he p esen s udy was o assess, using a gene ic isk sco e app oach, whe he he e ec o hese 26 loci on pulmona y unc ion is in luenced by smoking. Me hods: We e alua ed he in e ac ion be ween smoking exposu e, conside ed as ei he e e s ne e o pack-yea s, and a 26-single nucleo ide polymo phisms (SNPs) gene ic isk sco e in ela ion o FEV 1 o FEV 1 /FVC in 50 047 pa icipan s o Eu opean ances y om he Coho s o Hea and Aging Resea ch in Genomic Epidemiology (CHARGE) and Spi oMe a conso ia. Resul s: We iden i ied an in e ac ion (b in ¼–0.036, 95% con idence in e al, –0.040 o –0.032, P¼0.00057) be ween an unweigh ed 26 SNP gene ic isk sco e and smoking s a- us (e e /ne e ) on he FEV 1 /FVC a io. In in e p e ing his in e ac ion, we showed ha he gene ic isk o alling below he FEV 1 /FVC h eshold used o diagnose ch onic ob- s uc i e pulmona y disease is highe among e e smoke s han among ne e smoke s. A eplica ion analysis in wo independen da ase s, al hough no s a is ically signi ican , showed a simila end in he in e ac ion e ec . Conclusions: This s udy highligh s he bene i o using gene ic isk sco es o iden i ying in e ac ions missed when s udying indi idual SNPs and shows, o he i s ime, ha pe sons wi h he highes gene ic isk o low FEV 1 /FVC may be mo e suscep ible o he dele e ious e ec s o smoking. Key wo ds: FEV 1 /FVC, smoking, gene–en i onmen in e ac ion, gene ic isk sco e In oduc ion Spi ome ic measu es o pulmona y unc ion, such as he o ced expi a o y olume in 1 second (FEV 1 ) o i s a io wi h he o ced i al capaci y (FEV 1 /FVC), o m he basis o he diagnosis o ch onic obs uc i e pulmona y disease (COPD). 1–3 Pulmona y unc ion measu es a e also used clin- ically o moni o se e i y and con ol o as hma and o he e- spi a o y diseases and a e independen isk ac o s o mo ali y. 1–3 Pulmona y unc ion is s ongly in luenced by ciga e e smoking and by mul iple low-pene ance gene ic a ian s. Indeed, genome-wide associa ion s udies (GWAS) o ma ginal gene ic e ec s (i.e. no including in e ac ion e - ec s be ween gene ic a ian s and smoking) ha e iden i ied a leas 26 loci associa ed wi h FEV 1 o FEV 1 /FVC in he gene al popula ion. 4 Howe e , he in e play be ween gene ic ac o s and en i onmen al exposu es has no been well Key Messages •Spi ome ic measu es o pulmona y unc ion a e in luenced by bo h smoking and gene ics. This pape epo s a gen- e ic isk sco e-by-e e smoking in e ac ion on FEV 1 /FVC ( o ced expi a o y olume in 1 second/ o ced i al capaci y). •In indi iduals o Eu opean ances y, he educ ion in FEV 1 /FVC as a esul o smoking was g ea e among indi iduals who a e gene ically p edisposed o lowe FEV 1 /FVC a io. •Gene ic isk sco e-by-e e smoking in e ac ion can allow he iden i ica ion o subg oups in he popula ion whose gen- e ic backg ound makes hem mo e suscep ible o he dele e ious e ec s o smoking. 896 In e na ional Jou nal o Epidemiology, 2017, Vol. 46, No. 3 Downloaded om h ps://academic.oup.com/ije/a icle-abs ac /46/3/894/2894383/E idence- o -la ge-scale-gene-by-smoking by Tampe e Uni e si y Lib a y use on 09 Oc obe 2017 es ablished o pulmona y unc ion o i s associa ed ai s. Mo e b oadly, al hough conside able e o s ha e been made o iden i y in e ac ion e ec s be ween gene ic a ian s and en i onmen al exposu es ac oss he wide ange o human ai s and diseases, 5,6 such in es iga ions ha e been mos ly unsuccess ul in de ec ing obus gene–en i onmen in e - ac ions. 5,7 The well-es ablished e ec o ciga e e smoking on nume ous human heal h ou comes 8 makes i a se ious candida e o iden i ica ion o no el gene–en i onmen in e - ac ions, especially o pulmona y ai s. Hypo hesizing he p esence o single nucleo ide polymo ph- ism (SNP)-by-smoking in e ac ion, Hancock e al. 9 pe o med a genome-wide in e ac ion s udy o pulmona y unc ion, mod- elling single SNP main e ec s and hei in e ac ions wi h smoking in 50 047 pa icipan s o Eu opean ances y ac oss 19 s udies wi hin he Coho s o Hea and Aging Resea ch in Genomic Epidemiology (CHARGE) 10 and Spi oMe a con- so ia 11 — he la ges genome-wide in e ac ion s udy o pul- mona y unc ion as modi ied by smoking o da e. Howe e , a he han ocusing on he in e ac ion e ec s pe se, hey pe - o med a me a-analysis o he join es o SNP main e ec s and SNP-by-smoking in e ac ion e ec s o imp o e powe o iden i ying gene ic a ian s associa ed wi h pulmona y unc- ion. 12,13 Al hough hey epo ed new candida e a ian s based on his join es , he s udy did no iden i y any SNPs wi h genome-wide signi ican in e ac ion wi h smoking. He e, we explo ed gene-by-smoking in e ac ion e ec s limi ed o gene ic a ian s p e iously ound o be associ- a ed wi h pulmona y unc ion in s anda d ma ginal e ec s GWAS, 4 he e o e no including he new a ian s epo ed by Hancock e al. 9 based on he join es o main e ec s plus in e ac ion. Speci ically, we aimed o de e mine whe he smoking modi ies he e ec o es ablished gene ic a ian s when conside ed singly o in combina ion using a gene ic isk sco e summa izing he gene ic p edisposi ion o abno mal pulmona y unc ion. The p ima y mo i a ion o using gene ic isk sco e is s a is ical powe . 14,15 Indeed, se e al gene ic isk sco e-by-exposu e in e ac ions ha e al- eady been iden i ied in cases whe e single SNPs did no show e idence o s a is ically signi ican in e ac ions. 16–21 Gene ic isk sco e-by-exposu e in e ac ion es ing expands on he p inciple o omnibus es while le e aging he as- sump ion ha , o a gi en choice o coded alleles, mos in e ac ion e ec s will ha e he same di ec ion. This is simila o bu den es s ha ha e been widely used o a e a ian analysis 22 whe e a single pa ame e can accumula e e idence o associa ion wi hou inc easing he numbe o deg ees o eedom. When in e ac ion e ec s a e null on a e age (i.e. i in e ac ion e ec s a e bo h nega i e and posi i e so ha he sum o in e ac ion coe icien s end o ze o), he single SNP app oach will gene ally ou pe o m he isk sco e-based app oach. Con e sely, i in e ac ion e ec s end o be in he same di ec ion, he isk sco e- based app oach can ha e d ama ically highe powe . 14 Me hods S udy sample The p esen analysis elies on he Hancock e al. 9 genome- wide me a-analysis o main gene ic e ec s plus in e ac ion e ec s wi h smoking in ela ion o pulmona y unc ion among 50 047 pa icipan s (56% women) o Eu opean an- ces y om 19 s udies. The mean age was 53 yea s a he ime o pulmona y unc ion es ing. App oxima ely 15% we e cu en smoke s and 56% we e e e smoke s. Among e e smoke s, he a e age pack-yea s o smoking was 21. Supplemen a y Table 3 (a ailable as Supplemen a y da a a IJE online) p o ides he main cha ac e is ics o he s ud- ies included; comple e de ails o s udy-speci ic pulmona y unc ion es ing p o ocols ha e been published. 4 Fo s ud- ies wi h spi ome y a a single isi , we analysed FEV 1 / FVC and FEV 1 measu ed a ha isi . Fo s udies wi h spi ome y a mo e han one isi , measu emen s om he baseline isi o he mos ecen examina ion wi h spi om- e y da a was used. Smoking his o y (cu en , o me and ne e smoking) was asce ained by ques ionnai e a he ime o pulmona y unc ion es ing. Pack-yea s o smoking we e calcula ed o cu en and pas smoke s by mul iply- ing smoking amoun (packs pe day) and du a ion (yea s smoked). App oxima ely 2.5 million au osomal SNPs we e es ed o in e ac ion wi h smoking s a us (e e smoking s ne e smoking) and pack-yea s, o wo ou comes: FEV 1 and FEV 1 /FVC (see nex sec ion). We also used wo inde- penden da ase s o indi iduals o Eu opean ances y o es o eplica ion. The i s eplica ion da ase included 8859 un ela ed indi iduals, and he second da ase included 9457 amily-based indi iduals. The look-up was done in he GWAS o ma ginal gene ic e ec s done sepa - a ely in e e and ne e smoke as pa o a ecen me a- analysis o FEV 1 and FEV 1 /FVC. 23 Single SNP-by-smoking in e ac ion The analysis pe o med in his s udy used summa y s a is ics da a om he a o emen ioned me a-analysis o 19 s udies pe o med by Hancock e al. 9 In b ie , each o he 19 s udies de i ed he esiduals o FEV 1 and FEV 1 /FVC a e eg essing ou age, age 2 , sex, s anding heigh , p incipal componen eigen ec o s o geno ypes and ec ui men si e i applicable. The esiduals we e no malized using a ank-based in e se no mal ans o ma ion. Single SNP in e ac ion e ec s we e assessed using he ollowing model (see Supplemen a y No e, a ailable as Supplemen a y da a a IJE online): In e na ional Jou nal o Epidemiology, 2017, Vol. 46, No. 3 897 Downloaded om h ps://academic.oup.com/ije/a icle-abs ac /46/3/894/2894383/E idence- o -la ge-scale-gene-by-smoking by Tampe e Uni e si y Lib a y use on 09 Oc obe 2017 Yb0þbGGþbGEkGEkþXl¼1...3bElEl;(1) whe e b G and bEla e he main e ec o he SNP Gand ex- posu e E l ,bGEkis he in e ac ion e ec be ween Gand ex- posu e E k , and b 0 he in e cep . De ailed desc ip ion o s udies used in he eplica ion analysis can be ound in Sole A igas e al. 23 In b ie , lin- ea eg ession o age, age 2 , sex, heigh and p incipal com- ponen s o popula ion s uc u e was unde aken on FEV 1 and FEV 1 /FVC sepa a ely o e e smoke s and ne e smoke s. The esiduals we e no malized using a ank- based in e se no mal ans o ma ion, again sepa a ely in e e smoke s and ne e smoke s. These ans o med e- siduals we e hen used as he pheno ype o associa ion es ing unde an addi i e gene ic model in each exposu e s a a. In e ence o he in e ac ion e ec s om he exposu e-s a i ied analyses a e desc ibed in he Supplemen a y No e (a ailable as Supplemen a y da a a IJE online). Mul i a ia e in e ac ion analysis o e iew Fi s , we conside ed an unweigh ed gene ic isk sco e-by- smoking in e ac ion whe e he isk sco e simply sums he numbe o isk alleles (i.e. alleles associa ed wi h a lowe pulmona y unc ion). This unweigh ed gene ic isk sco e is mos powe ul when he in e ac ion e ec s ha e he same di ec ion as ma ginal SNP e ec s (i.e. he ha m ul e ec s o smoking a e magni ied in indi iduals wi h a gene ic p e- disposi ion o educed pulmona y unc ion). Second, we used a weigh ed gene ic isk sco e whe e SNPs we e weigh ed by he absolu e alue o hei ma ginal e ec es i- ma es ob ained om s age 1 sc eening o FEV 1 and FEV 1 / FVC om Sole A igas e al. 4 (Supplemen a y Table 1, a ailable as Supplemen a y da a a IJE online). This weigh ing scheme is mos powe ul when he magni ude o in e ac ion e ec s is p opo ional o he SNP ma ginal e - ec s. Finally, o ou hi d mul i a ia e analysis, we de i ed a s anda d omnibus es o all in e ac ion e ec s. This es will e ain powe in he p esence o e ec s in bo h di ec ions o o di e en magni udes. Al hough he e is s ong co ela ion among he 12 es s pe o med ( hese h ee models, conside ing in e ac ion wi h wo smoking me ics, e e /ne e smoking o pack-yea s, o he wo pul- mona y unc ion me ics FEV1 and FEV1/FVC), we used a s ingen Bon e oni P- alue co ec ion h eshold o 410 –3 o accoun o mul iple es ing. When aw da a a e a ailable, he weigh ed gene ic isk sco e (GRS) is usually exp essed as GRS ¼R m [w i G i ], whe e mis he numbe o SNPs included in he gene ic isk sco e and w¼(w 1 ,..w m ) a e he weigh s a ibu ed o each single SNP. Following p e ious no a ion, he es o in e - ac ion be ween he GRS and he exposu e E k can be applied using he ollowing model: Yc0þcGRS GRS þcINT GRS EkþXl¼1...3cEl El; (2) whe e c 0 ,c GRS ,cEland c INT a e he in e cep , he main e - ec o he GRS, he main e ec o he exposu e E l and he in e ac ion e ec be ween E k and he GRS, espec i ely. Howe e , because indi idual-le el da a we e no di ec ly a ailable, we pe o med he es o c INT om summa y s a is ics o in e ac ion e ec s using an in e se- a iance weigh ed sum as p oposed by Ascha d. 14 The chi-squa e o he in e ac ion e m c INT was de i ed as ollows: 2 in ¼Xi¼1...m wi^ bGiEk ^ 2 bGiEk ! 2 Xi¼1...m w2 i ^ 2 bGiEk ;(3) whe e ^ bGiEkand ^ 2 bGiEk a e he es ima ed e ec s and a i- ance o he in e ac ion be ween he exposu e E k and he SNP G i ob ained om Equa ion (1) and w i is he weigh applied o SNP G i .Unde he null hypo hesis o no in e - ac ion e ec , 2 in ollows a chi-squa ed dis ibu ion wi h one deg ee o eedom. The s anda d omnibus es o all in e ac ion e ec s con- sis ed o e alua ing join ly aGEk¼ðaG1Ek;...;aGmEkÞ om he model: Ya0þXi¼1...m½aGiGi þXi¼1...m½aGiEkGiEkþXl¼1...3aElEl; (4) whe e a0,aGi,aEland aGiEka e he in e cep , he main e - ec s o SNP G i and he exposu e E l , and he in e ac ion e - ec be ween G i and E k .Le e aging he independence be ween he SNPs conside ed (a single SNP was selec ed o each independen locus), we also de i ed he omnibus es using summa y s a is ics. Unde his independence as- sump ion, he G i E k in e ac ion e ms would also be in- dependen s, 14 so ha i can be pe o med by summing he chi-squa e om each uni a ia e in e ac ion es o o m a chi-squa e wi h mdeg ees o eedom as ollows: 2 omnibus ¼Xi¼1...m ^ b2 GiEk ^ 2 bGiEk ;(5) whe e ^ bGiEkand ^ 2 bGiEk a e he es ima ed e ec s and 898 In e na ional Jou nal o Epidemiology, 2017, Vol. 46, No. 3 Downloaded om h ps://academic.oup.com/ije/a icle-abs ac /46/3/894/2894383/E idence- o -la ge-scale-gene-by-smoking by Tampe e Uni e si y Lib a y use on 09 Oc obe 2017 a iance o he in e ac ion be ween he exposu e E k and he SNP G i ob ained om Equa ion (1). Rela i e isk in e e smoke s s ne e smoke s GRS in e ac ion e ec s can u he be ansla ed in e ms o isk p edic ion. Fo pulmona y unc ion, low FEV 1 o FEV 1 / FVC inc eases he isk o dea h 24 and oge he hey o m he basis o he diagnosis o COPD. 1–3 COPD s age 2 o highe a e de ined by he Global Ini ia i e o Ch onic Obs uc i e Lung Disease (GOLD) as FEV 1 /FVC <0.70 and FEV 1 <80% o he p edic ed alue. Acco ding o ecen s ud- ies, 2,25 be ween 5% and 20% o Eu opean ances y adul s a eexpec ed oha eFEV 1 /FVC <0.70, depending on smok- ing cha ac e is ics and age dis ibu ion. Se e al s udies a gue o a mo e s ingen h eshold o de ine COPD 25,26 based on lowe limi o no mal p edic ed alue, a he han a ixed ab- solu e alue, o p e en disease misclassi ica ion. To explo e he impac o in e ac ion e ec on he isk o disease, we de i ed he ela i e isk (RR) o ha ing FEV 1 /FVC below a gi en h eshold (1%, 5% and 20%) in e e smoke s s ne e smoke s condi ional on he un- weigh ed GRS. This quan i y is de ined as he join p ob- abili y o ha ing bo h FEV 1 /FVC in he in e al [–1, FEV 1 /FVC up ] and he GRS in he in e al [GRS low ,GRS up ]. This can be exp essed as he ollowing in eg al: ð FEV1=FVCup 1 ð GRSup GRSlow 1ðyjg;eÞ 2ðgjeÞdy dg;(6) whe e y,eand ga e FEV 1 /FVC, smoking s a us and he GRS, espec i ely, and 1 and 2 a e he p obabili y densi y unc ion o y and g. The de ailed de i a ion o he abo e in eg al is a ailable as Supplemen a y da a a IJE online. Resul s We selec ed 26 loci p e iously ound o be associa ed wi h FEV 1 o FEV 1 /FVC a genome-wide signi icance (P<510 –8 ) in ma ginal associa ion es s 4,11,27 (i.e. no including in e ac ion e ec s wi h smoking exposu es) and eplica ed in he GWAS by Sole A igas e al., 4 he la ges me a-analysis o ma ginal gene ic e ec conduc ed o hese wo ai s in he gene al popula ion. Addi ional loci o hese wo pheno ypes ha e been iden i ied in wo ecen s udies. 28,29 Howe e , hese new loci we e no included in ou analysis because bo h hese s udies used a la ge coho asce ained h ough smoking s a us. Fo each o he 26 se- lec ed loci, we choose he SNP wi h he s onges e idence o associa ion (i.e. smalles P- alue) wi h each o hese pheno ypes. The inal lis included 26 SNPs pe pheno ype, wi h only wo SNPs being di e en be ween FEV 1 and FEV 1 /FVC as p e iously epo ed 4 (Supplemen a y Table 1, a ailable as Supplemen a y da a a IJE online). Es ima ed in e ac ion e ec s o hese SNPs we e ex ac ed om he me a-analysis summa y s a is ics o he ou es s pe o med in he Hancock e al. 9 analysis: SNP-by- smoking s a us (e e smoking s ne e smoking) in e ac ion e ec on FEV 1 and FEV 1 /FVC; and SNP-by- smoking pack-yea s in e ac ion e ec on FEV 1 /FVC and FEV 1 . As shown in Supplemen a y Table 2 (a ailable as Supplemen a y da a a IJE online), nine SNPs showed nominal signi icance (P<0.05) ou o he 104 es s pe - o med; howe e , none emained signi ican a e accoun - ing o mul iple es ing (Bon e oni co ec ed P- alue h eshold o 5 10 –4 ). The minimum P- alue was obse ed o he in e ac ion be ween s993925, nea he TGFb2 gene, and smoking s a us on FEV 1 [b in ¼–0.036, 95% con idence in e al (CI), –0.009 o –0.032, P¼0.007]. Nex , using hese da a, we conduc ed h ee mul i a ia e (as opposed o single SNP) in e ac ion analyses, es ing join ly o he in e ac ion e ec s be ween hose SNPs and ei he smoking s a us o pack-yea s on he wo pheno ypes (FEV 1 and FEV 1 /FVC) o a o al o 12 es s. As shown in Table 1, none o he mul i a ia e in e ac ion es s wi h pack-yea s was signi ican . Howe e , ou o he six mul i- a ia e in e ac ion es s wi h smoking s a us (e e s ne e ) showed nominal signi icance, and wo es s o FEV 1 /FVC had a P- alue below he Bon e oni signi icance le el (12 es s, P <410 –3 ). The s onges signal was obse ed o he unweigh ed gene ic isk sco e-by-smoking s a us in e - ac ion e ec on FEV 1 /FVC (b in ¼–0.036, 95% CI –0.040 o –0.032, P¼0.00057). The Coch an’s Q es o he e o- genei y o he in e ac ion e ec ac oss s udies was no sig- ni ican (P¼0.97) and he o es plo o s udy-speci ic esul s did no display any ob ious ou lie (Supplemen a y Figu e 1, a ailable as Supplemen a y da a a IJE online). The con as be ween his signi ican isk sco e in e - ac ion and he absence o s ong single SNP in e ac ion e - ec s can be explained by looking a he dis ibu ion o he single SNP in e ac ion e ec es ima es. Figu e 1 shows his dis ibu ion o he alleles associa ed wi h dec eased FEV 1 / FVC. I highligh s ha , al hough he 95% CI o mos single SNP in e ac ion e ec s encompass he null (and he e o e he absence o signi ican single SNP in e ac ion e ec ), he e is an en ichmen o nega i e in e ac ion e ec s. Indeed, e en a binomial es can be used o con i m he unbalanced di ec ion o in e ac ion e ec s (18 o 26 in e - ac ions a e nega i e leading o a P- alue o 0.014 o a bi- nomial es wi h an expec ed equip obable dis ibu ion o 0.5). The gene ic isk sco e-based in e ac ion es exploi s such en ichmen by es ing o he a e age in e ac ion e - ec ac oss all SNPs. 14 As wi h any mul i a ia e app oach In e na ional Jou nal o Epidemiology, 2017, Vol. 46, No. 3 899 Downloaded om h ps://academic.oup.com/ije/a icle-abs ac /46/3/894/2894383/E idence- o -la ge-scale-gene-by-smoking by Tampe e Uni e si y Lib a y use on 09 Oc obe 2017 Table 1. Mul i a ia e in e ac ion es s o he 26 loci associa ed wi h pulmona y unc ion Ou come Exposu e Tes ˆ b in (CI) P- alue FEV 1 Smoking s a us a uGRS –0.0055 (–0.011, 2.7 10 –5 )0.051 wGRS –0.21 (–0.40, –0.033) 0.020 CHISQ ––0.49 FEV 1 Pack-yea s uGRS –1.6 10 –5 (–4.6 10 –5 , 1.4 10 –5 )0.30 wGRS –6.5 10 –4 (–1.6 10 –3 , 3.3 10 –4 )0.19 CHISQ ––0.46 FEV 1 /FVC Smoking s a us uGRS –0.0099 (–0.016, –0.0043) 0.00057 b wGRS –0.21 (–0.33, –0.073) 0.0022 b CHISQ ––0.026 FEV 1 /FVC Pack-yea s uGRS –4.4e-06 (–3.6 10 –5 , 2.7 10 –5 )0.78 wGRS –6.5 10 –5 (–8.0 10 –4 , 6.6 10 –4 )0.85 CHISQ – – 0.53 uGRS is he gene ic isk sco e using equal weigh s o all SNPs; wGRS is he gene ic isk sco e weigh ed by e ec es ima es om he ma ginal sc eening; CHISQ is he omnibus es o all in e ac ion e ec s; ˆ b in is he es ima ed in e ac ion e ec be ween he GRS and he ou come; and CI is he con idence in e al o ha es i- ma e. Nominally signi ican es s a e indica ed in bold. a Smoking s a us is de ined as ne e smoke s s e e smoke s. b Signi ican P- alue a e Bon e oni co ec ion. Figu e 1. Dis ibu ion o in e ac ion e ec s on FEV 1 /FVC. Single SNP isk allele-by-smoking s a us (e e /ne e ) in e ac ion e ec es ima es (b in ) and 95% con idence in e als a e plo ed by inc easing alues. The unweigh ed gene ic isk sco e-by-smoking s a us in e ac ion is plo ed a he bo om. 900 In e na ional Jou nal o Epidemiology, 2017, Vol. 46, No. 3 Downloaded om h ps://academic.oup.com/ije/a icle-abs ac /46/3/894/2894383/E idence- o -la ge-scale-gene-by-smoking by Tampe e Uni e si y Lib a y use on 09 Oc obe 2017 based on a composi e null hypo hesis, his esul indica es ha a leas a subse o hese 26 SNPs in e ac wi h smok- ing s a us, bu does no allow us o de e mine which o how many SNPs a e d i ing he gene ic isk sco e-by- smoking in e ac ion. The h ee o he se s o single SNP in e ac ion es s showed a simila (bu no signi ican a e co ec ion o mul iple es ing) end wi h en ichmen o nega i e in e ac ions (Supplemen a y Figu es 2–4, a ail- able as Supplemen a y da a a IJE online). We summa ized he con ibu ion o he unweigh ed gene ic isk sco e-by- smoking in e ac ion on FEV 1 /FVC in Table 2 and Figu e 2A. This indica es ha he dele e ious e ec o smoking is enhanced among ca ie s o he isk alleles o equi alen ly ha he dele e ious e ec o smoking is educed among subjec s ca ying he p o ec i e alleles. We used wo independen da ase s, one o 8859 un e- la ed indi iduals and ano he o 9457 ela ed indi iduals, o es o independen eplica ion o ou esul s (Supplemen a y No e, a ailable as Supplemen a y da a a IJE online). Al hough he in e ac ion e ec s we e no sig- ni ican , bo h eplica ion samples showed consis en nega- i e GRS-by-e e smoking in e ac ion e ec on FEV1/FVC (^ bin ¼–0.0025, 95% CI –0.0165, 0.0115, P¼0.72 and ^ bin ¼–0.0030, 95% CI –0.0214, 0.0154, P¼0.74, and o e all in e ac ion e ec in he combined eplica ion da a- se s ^ bin ¼–0.0027, 95% CI –0.0136, 0.0082 P¼0.63) and a Coch an’s Q es o he e ogenei y showed no signi ican di e ence in he h ee e ec es ima es (P¼0.51). To quan i y he impac o his esul om a public heal h pe spec i e, we es ima ed he impac o he gene ic isk sco e-by-smoking in e ac ion on ha ing FEV 1 /FVC below 1%, 5% and 20% in he lowe ails o he dis ibu ion in he popula ion. Speci ically, we de i ed he RR o ha ing FEV 1 /FVC below hese cu -o poin s (1%, 5% and 20%) in e e smoke s compa ed wi h ne e smoke s. Figu e 2B quan i ies he excess RR (i.e. he RR minus one) o indi id- uals ac oss i e GRS quin iles. I highligh s he highe isk associa ed wi h smoking among indi iduals ca ying isk alleles (i.e. alleles associa ed wi h poo e pulmona y unc- ion) as compa ed wi h indi iduals ca ying p o ec i e al- leles (i.e. alleles associa ed wi h be e pulmona y unc ion). Fo example, among indi iduals wi h a GRS abo e he 80 h pe cen ile, smoke s ha e on a e age a 26% excess RR o ha ing FEV 1 /FVC in he lowes 1% o he popula ion dis ibu ion, whe eas e e smoke s wi h a GRS below he 20 h pe cen ile ha e on a e age an 18% excess RR o all- ing in ha same FEV 1 /FVC ca ego y compa ed wi h ne e smoke s. Applying he same app oach o FEV 1 ,we obse ed a simila pa e n (Supplemen a y Figu e 5, a ail- able as Supplemen a y da a a IJE online). Howe e , as expec ed, he lowe magni ude o he gene ic isk sco e-by- e e smoking in e ac ion on FEV 1 implied a lowe di e - ence in RR be ween e e smoke s and ne e smoke s. Discussion Using he la ges da ase o da e o Eu opean ances y pa - icipan s om he gene al popula ion wi h pulmona y Table 2. Summa y o e ec es ima es o gene ic isk sco e- by-smoking s a us in e ac ion on FEV 1 /FVC P edic o s Be a SD P- alue F om he ma ginal exposu e model Pack-yea s –0.0030 0.00017 1.2 10 –71 Cu en smoking –0.040 0.0047 7.7 10 –18 Smoking s a us a –0.0023 0.0046 0.61 F om he in e ac ion model GRS –0.0363 0.0021 3.9 10 –64 GRS Smoking s a us a –0.0099 0.0029 5.7 10 –4 GRS is he unweigh ed gene ic isk sco e; be a is he e ec es ima es o each p edic o ; and SD he s anda d de ia ion o he each be a. a Smoking s a us was de ined as ne e smoke s s e e smoke s. Figu e 2. O e iew o he unweigh ed gene ic isk sco e-by-smoking in e ac ion e ec on FEV 1 /FVC. Uppe panel (A) p esen s he dis ibu ion o he unweigh ed gene ic isk sco e (GRS, g ey densi y plo ) and he ela ionship be ween he un- weigh ed GRS and s anda dized FEV 1 /FVC in e e smoke s (dashed line) and ne e smoke s (solid line). Lowe panel (B) shows he excess ela- i e isk (RR) o ha ing FEV 1 /FVC in he lowes 1%, 5% and 20% o he popula ion o e e smoke s compa ed wi h ne e smoke s, as s a i ied by GRS quin iles. In e na ional Jou nal o Epidemiology, 2017, Vol. 46, No. 3 901 Downloaded om h ps://academic.oup.com/ije/a icle-abs ac /46/3/894/2894383/E idence- o -la ge-scale-gene-by-smoking by Tampe e Uni e si y Lib a y use on 09 Oc obe 2017 unc ion (FEV 1 /FVC and FEV 1 ), smoking and gene ic da a, we iden i ied a gene-by-smoking in e ac ion e ec on FEV 1 /FVC by using a GRS composed o 26 SNPs iden i ied and eplica ed in a p io GWAS me a-analysis o ma ginal gene ic e ec s. To ou knowledge, ou s udy is he i s o epo a syne gis ic ac ion o genes and smoking on pul- mona y unc ion (i.e. he educ ion in FEV 1 /FVC as a e- sul o smoking is g ea e among indi iduals who a e gene ically p edisposed o lowe FEV 1 /FVC a io). Ou s udy also highligh s he impo ance o de eloping and applying al e na i e s a egies o e alua e in e ac ion e - ec s o lung pheno ypes along wi h o he complex ai s and diseases. The gene ic isk sco e-based app oach enabled us o iden i y an in e ac ion when he s anda d uni a ia e es (i.e. e alua ing each single gene ic a ian o in e ac ion independen ly) ailed o iden i y any in e ac ions. Replica ion s udies showed in e ac ion e ec es ima es in he same di ec ion as he disco e y s udy bu we e no signi ican , and he magni ude o in e ac ion e ec s we e subs an ially smalle . We acknowledge ha , despi e ca e ul e alua ion o he in e ac ion e ec s in he disco e y sam- ple, he obse ed signal migh be o e es ima ed o con- ounded by unmeasu ed complex ac o s. Howe e , we can a p io i ule ou a sys ema ic bias o he single SNP in e ac ion e ec s in he disco e y s udy, because he gen- omic in la ion ac o k, de ined as he a io o he median o he empi ically obse ed dis ibu ion o he es s a is ic o he expec ed median, 30 was no subs an ially di e en om 1 (k¼1.044 o FEV1/FVC and smoking s a us). Ins ead, di e ences in signi icance and e ec es ima es migh be pa ly explained by he limi ed sample size in he eplica ion s udy and di e ences in he analy ical design. Indeed, he disco e y analysis was pe o med using a sa u- a ed model including h ee smoking exposu es and expli- ci ly modelled he in e ac ion e ec . In compa ison, he eplica ion analysis was no adjus ed o cu en smoking s a us and pack-yea , and he in e ac ion e ec was app oxima ed om analyses s a i ied by smoking s a us ou come, which has some limi a ions (see Supplemen a y No e and Supplemen a y Figu e 6, a ailable as Supplemen a y da a a IJE online). P e ious wo k has shown ha combined analyses a e mo e powe ul when e - ec s exis in bo h s a a, 31 as obse ed in disco e y s udy. Fu he , e en wi h N¼18 316 indi iduals in he combined eplica ion popula ion, we a e unde powe ed. This sample size p o ides less han 50% powe , a nominal signi icance o 5%, o de ec in e ac ion e ec s wi h he GRS. Gene ic isk sco e-by-exposu e in e ac ion can ha e highe clinical alue han he iden i ica ion o single SNP- by-exposu e in e ac ion by cap u ing a weal h o in o ma- ion in a single measu e o iden i y subg oups in he popula ion whose gene ic backg ound makes hem mo e suscep ible o he dele e ious e ec s o smoking. 19,32,33 Indeed, i single SNP-by-smoking in e ac ions a e dis ib- u ed uncondi ionally on he ma ginal gene ic e ec (i.e. in e ac ion e ec s a e equally likely o be posi i e o nega- i e gi en ha he coded alleles a e he isk alleles), he gene ic e ec is expec ed o be simila be ween e e and ne e smoke s. The en ichmen o nega i e in e ac ions we iden i ied h ough ou GRS app oach e eals a s onge gene ic componen among he e e smoke subg oup in he popula ion and can allow he implemen a ion o mo e e i- cien implemen a ion o p e en ion s a egies. Fo ex- ample, in he public heal h se ing, p og ammes a ge ing smoking cessa ion campaigns o indi iduals who a e gene - ically p edisposed o low pulmona y unc ion may ha e a s onge impac in p e en ing COPD. Ou esul s may also elucida e biological mechanisms unde lying he in e play be ween genes and smoking in pulmona y unc ion. In pa icula , he highe s a is ical powe o he gene ic isk sco e-based in e ac ion es poin s owa ds he po en ial p esence o an unmeasu ed in e media e bioma ke media ing he e ec o he 26 loci on FEV 1 /FVC. As shown in Figu e 3, he mos pa simoni- ous model (i.e. he less complex ollowing Occam’s azo ) ha would explain mul iple in e ac ions going in he same di ec ion (Figu e 1) implies ha he gene ic a ian s Figu e 3. Unde lying causal model. Po en ial causal diag ams unde lying he gene and smoking in e ac ion e ec s on FEV 1 /FVC. Panel (A) p esen s a scena io whe e each gene ic a ian in luences he ou come h ough a SNP-speci ic pa hway, and in e ac ions wi h he en i onmen al exposu e ake place along hese pa hways. Panel (B) p esen s an al e na i e (and simple ) model whe e mul iple gene ic a ian s in luence an unmeasu ed in e media e bio- ma ke U, which e ec on FEV 1 /FVC depends on smoking. In scena io (A), he single SNP-by-smoking in e ac ion es is he op imal app oach, whe eas, in scena io (B), he single SNP-by-smoking in e ac ion es can become ine icien , and in e ac ion would be easie o de ec using a gene ic isk sco e-by-smoking in e ac ion es , because i summa izes all in e ac ion e ec s in a single es . 902 In e na ional Jou nal o Epidemiology, 2017, Vol. 46, No. 3 Downloaded om h ps://academic.oup.com/ije/a icle-abs ac /46/3/894/2894383/E idence- o -la ge-scale-gene-by-smoking by Tampe e Uni e si y Lib a y use on 09 Oc obe 2017