Evaluating the Causal Relation of ApoA-IV with Disease-Related Traits - A Bidirectional Two-sample Mendelian Randomization Study

1 Scien i ic RepoR s | 7: 8734 | DOI:10.1038/s41598-017-07213-9 www.na u e.com/scien i ic epo s E alua ing he Causal Rela ion o ApoA-IV wi h Disease-Rela ed T ai s - A Bidi ec ional Two-sample Mendelian Randomiza ion S udy Salome Mack1, S e an Coassin 1, Julien Vauche 2, Flo ian K onenbe g 1, Claudia Lamina1 & ApoA-IV-GWAS Conso ium Apolipop o ein A-IV (apoA-IV) has been obse ed o be associa ed wi h lipids, kidney unc ion, adiposi y- and diabe es- ela ed pa ame e s. To assess he causal ela ionship o apoA-IV wi h hese pheno ypes, we conduc ed bidi ec ional Mendelian andomiza ion (MR) analyses using publicly a ailable summa y-le el da ase s om GWAS conso ia on apoA-IV concen a ions (n = 13,813), kidney  unc ion (es ima ed glome ula   il a ion  a e (eGFR), n = 133,413), lipid  ai s (HDL choles e ol, LDL choles e ol, iglyce ides, n = 188,577), adiposi y- ela ed  ai s (body-mass-index (n = 322,206), wais -hip- a io (n = 210,088)) and  as ing glucose (n = 133,010). Main analyses consis ed in in e se- a iance weigh ed and mul i a iable MR, whe eas MR-Egge eg ession and weigh ed median es ima ion we e used as sensi i i y analyses. We ound ha eGFR is likely o be causal on apoA-IV concen a ions (53 SNPs; causal e ec  es ima e pe  1-SD inc ease in eGFR = −0.39; 95% CI = [−0.54, −0.24]; p- alue = 2.4e-07). T iglyce ide concen a ions we e also causally associa ed wi h apoA-IV concen a ions (40 SNPs; causal e ec  es ima e pe  1-SD inc ease in  iglyce ides = −0.06; 95% CI = [−0.08, −0.04]; p- alue = 4.8e-07), independen ly o  HDL-C and LDL-C concen a ions (causal e ec  es ima e  om mul i a iable MR = −0.06; 95% CI = [−0.10, −0.02]; p- alue = 0.0014). E alua ing he in e se di ec ion o  causali y  e ealed a possible causal associa ion o  apoA-IV on HDL-choles e ol (2 SNPs; causal e ec  es ima e pe  one pe cen  inc ease in apoA-IV = −0.40; 95% CI = [−0.60, −0.21]; p- alue = 5.5e-05). Apolipop o ein A-IV (apoA-IV) has been discussed as a bioma ke and /o isk ac o o ca dio ascula diseases, kidney disease and diabe es1–4. Whe he a measu able ai is a causal isk ac o o a ma ke ac ing as an indi- ca o o he p esence o a disease o i s se e i y, is one o he key ques ions in biomedical esea ch5, 6. To ac as a bioma ke o a ce ain disease, causali y is no s ic ly equi ed, since a bioma ke can be egula ed as a esponse o a disease and/o an associa ed ai al eady in a e y ea ly s age o he disease. I can he e o e ac as a disease p edic o wi hou being causally in ol ed in he pa hogenesis o he disease. Mo eo e , he associa ion be ween a bioma ke and a disease may be con ounded by measu ed o unmeasu ed ac o s ha dis o hei ela ionship. Howe e , es ablishing causali y be ween a measu able ai and a disease is c ucial o unde s and he pa hophys- iological mechanisms. Disease endpoin s a e o en de ec ed using con inuous su oga e ma ke s (e.g. as ing glucose as ma ke o impai ed glucose me abolism leading o diabe es) o a e in luenced by many o he in e media e pheno ypes, which migh be e ep esen he causal pa hway leading o a disease han he disease s a e i sel . Wi h his mo i- a ion in mind, we e alua ed he causal ela ion o apoA-IV wi h disease- ela ed, con inuous su oga e ma ke s and in e media e pheno ypes o which e idence om obse a ional associa ion s udies o unc ional expe i- men s p oposes a unc ional link wi h apoA-IV. ApoA-IV is a glycop o ein, which is a componen o iglyce ide- ich and high-densi y lipop o eins bu i also ci cula es ee in human plasma7. I s physiological ole is no ully elucida ed, ye . I pa icipa es in e e se 1Di ision o Gene ic Epidemiology, Depa men o Medical Gene ics, Molecula and Clinical Pha macology, Medical Uni e si y o Innsb uck, Innsb uck, 6020, Aus ia. 2Depa men o In e nal Medicine, Lausanne Uni e si y Hospi al, Lausanne, 1015, Swi ze land. A comp ehensi e lis o conso ium membe s appea s a he end o he pape . Co espondence and eques s o ma e ials should be add essed o C.L. (email: [email p o ec ed]) Recei ed: 14 Ma ch 2017 Accep ed: 21 June 2017 Published: xx xx xxxx OPEN www.na u e.com/scien i ic epo s/ 2 Scien i ic RepoR s | 7: 8734 | DOI:10.1038/s41598-017-07213-9 choles e ol anspo 8, 9 and plays an impo an ole in elie ing pe iphe al cells o an o e load o choles e ol10, 11. I has an i-a he ogenic p ope ies12, 13 and low concen a ions we e ound o be associa ed wi h ca dio ascu- la ou comes such as co ona y hea disease o acu e co ona y synd ome2, 14–16. Abou 30% o he pheno ypic a iance has been shown o be gene ically egula ed17. In a ecen genome-wide associa ion s udy (GWAS) me a-analysis we ha e iden i ied wo a ian s wi hin he APOA5-A4-C3-A1 clus e and one a ian in he KLKB1 gene o be associa ed wi h apoA-IV concen a ions17. Fu he mo e, he lead APOA4-SNP was shown o be asso- cia ed wi h HDL choles e ol (HDL-C) concen a ions. O he a ian s wi hin he APOA5-A4-C3-A1 clus e ha e also been shown o be associa ed bo h wi h HDL-C, bu also wi h iglyce ide (TG) concen a ions18–20. In addi- ion, a TG gene ic sco e was ound o be associa ed wi h apoA-IV concen a ions17. S udies in animal models, gene ic associa ion s udies and s udies e alua ing disease p og ession a gue o apoA-IV as he d i ing ac o in luencing sa ie y, adiposi y, glucose concen a ions and ch onic kidney disease21–28. On he o he hand apoA-IV concen a ions ha e also been shown o change ollowing ood in ake, gas ic bypass su ge y o weigh loss29, 30. Those indings wa an u he in es iga ion in causal associa ions and especially in he di ec ion o he e ec s be ween apoA-IV and lipid le els, adiposi y, diabe es- ela ed pa ame e s and kidney unc ion. Recen de elop- men s in gene ic epidemiology using gene ic ma ke s as p oxies o a ai help es ima e causal associa ions wi h a disease ou come h ough a Mendelian andomiza ion app oach31. Such causal e ec s a e usually e y small. The e o e, single s udies a e o en unde powe ed. Howe e , esul s o genome-wide associa ion s udies a e inc easingly made publicly a ailable. Ha nessing summa y-le el da a, Mendelian andomiza ion analyses hen each su icien s a is ical powe o yield mo e p ecise causal e ec es ima es32. Fo all o he al eady men ioned apoA-IV-associa ed and disease- ela ed ai s, la ge genome-wide associa- ion s udies a e a ailable, in which dozens o gene ic a ian s ha e been iden i ied18, 33–36. Fu he mo e, a GWAS me a-analysis on apoA-IV concen a ions has been published ecen ly17. The e o e, we conduc ed a bidi ec ional wo-sample Mendelian andomiza ion analysis o assess he causal ela ionship o apoA-IV concen a ions wi h HDL-C, LDL choles e ol (LDL-C), TG, as ing glucose (FG) as well as body-mass-index (BMI), wais -hip- a io adjus ed o BMI (WHR), and es ima ed glome ula il a ion a e (eGFR). He ein, lipid le els, adiposi y- ela ed ai s, glucose and eGFR we e conside ed as isk ac o s on apoA-IV concen a ions as ou come a iable. We complemen ed he analysis by in es iga ing he in e se associ- a ion o whe he apoA-IV is causally associa ed wi h he same ai s. We applied di e en sensi i i y analyses o accoun o p esence o po en ial pleio opy. Resul s Validi y o ins umen al a iables and powe analysis. A g aphical o e iew o he design and s udies used in ou in es iga ion is gi en in Fig.1. To assess he s eng h o he ins umen al a iables (i.e. he single-nu- cleo ide polymo phisms [SNPs]) ha we e used o he Mendelian Randomiza ion, he explained a iance (R²) de ined by he SNPs is gi en o each o he in es iga ed po en ial isk ac o s in Tables1 and 2. R² a ies be ween 0.0060 (WHR) and 0.0727 (LDL-C). Table1 shows, which ac ion o de ec able causal R² o he exposu e ai on he ou come (i.e. which magni- ude o causal e ec o he ai on he ou come) can be de ec ed o he calcula ed R² o he gene ic a ian s on he exposu e ( o HDL-C o as ing glucose as he exposu e ai , apoA-IV is he exposu e and o apoA-IV as he exposu e ai , he o he in es iga ed ai s unc ion as he ou come). I a ies be ween 0.0127 o LDL-C and 0.1544 o WHR, assuming a signi icance le el o α = 0.00625 and a powe o 80%. Tha is, o de ec a causal e ec o WHR on apoA-IV, WHR would ha e o explain mo e han 15% o he pheno ypic a iance o apoA-IV. Fo he Figu e 1. G aphical o e iew o he s udy design. Se en bioma ke s ou o ou conso ia (GLGC18; CKDGen conso ium33; GIANT conso ium34, 60; MAGIC35, 36) we e included as well as esul s om a GWAS me a- analysis on apoA-IV17. www.na u e.com/scien i ic epo s/ 3 Scien i ic RepoR s | 7: 8734 | DOI:10.1038/s41598-017-07213-9 e e se causa ion (e ec o apoA-IV on he o he ac o s), causal e ec s anging be ween 0.002 and 0.0048 would be de ec able (Table2). Mendelian andomiza ion analyses. Causal associa ion o lipid ai s on apoA-IV. An o e iew abou he cha ac e is ics and he me a-analysis esul s o he 71 HDL-SNPs, he 58 LDL-SNPs and he 40 iglyce ide-SNPs can be ound in Supplemen a y Tables1–3. Using he MR-IVW me hod e ealed a signi ican associa ion o HDL-C le els (in SD) wi h he log- ans o med apoA-IV concen a ions esul ing in a causal es ima e o βMR-IVW = 0.0341 (95% CI = [0.0142, 0.0540], p- alue = 0.0008, Table3 and Fig.2). This means an inc ease in one SD in HDL-C le els (≈17 mg/dL) leads o an inc ease o he log- ans o med apoA-IV le els o app oxima ely 0.034, which co esponds o a el- a i e inc ease o 3.4%. A e adjus ing o LDL-C as well as TG in he mul i a iable Mendelian andomiza ion using all 71 SNPs, his causal e ec diminished and became non-signi ican (βadjus ed = 0.0195, 95% CI = [−0.0084, 0.0475], p- alue = 0.1706, see Table3). One SNP was iden i ied o be po en ially pleio opic by he g x-package in R ( s964184 wi hin he gene egion APOA5-A4-C3-A1). This is he same SNP ha we planned o exclude as a sen- si i i y analysis due o i s close p oximi y o he APOA4 gene (see Me hods). A e exclusion o his SNP, he e ec became non-signi ican (βMR-IVW_g x = 0.0271, 95% CI = [0.0067, 0.0474], p- alue = 0.0091, see Table3). By adjus - ing o LDL-C and TG, he be a es ima e emained app oxima ely he same and was non-signi ican , oo (Table3). Excluding he 23 SNPs ha we e in LD wi h SNPs associa ed wi h any o he o he ai s (Supplemen a y Table8) and applying he MR-IVW me hod o he emaining 48 SNPs did no change he causal es ima e ob ained using all 71 SNPs, bu i became non-signi ican (p = 0.0453, Supplemen a y Table15 and Supplemen a y Figu e4). The applica ion o he mul i a iable Mendelian andomiza ion app oach o he 61 SNPs no in LD wi h SNPs om he non-lipid ai s ga e compa able esul s o using he adjus men model on all 71 SNPs (p = 0.1212, Supplemen a y Table15 and Supplemen a y Figu e5). Using he MR-Egge eg ession me hod esul ed in a ma kedly highe and s a is ically signi ican causal e ec o βMR-Egge = 0.0611 (95% CI = [0.0254, 0.0969], p- alue = 0.0008, see Supplemen a y Table16, Supplemen a y Figu e15), bu also wi h a b oade con idence in e al. Howe e , when using a weigh ed median es ima ion app oach, he causal e ec diminished and became non-signi ican (Supplemen a y Table16). Fo LDL-C le els, we ob ained a signi ican causal es ima e o βMR-IVW = −0.0376 (95% CI = [−0.0572, −0.0179], p- alue = 0.0002, Table3 and Fig.3) by using he MR-IVW me hod. This means ha LDL-C le els and he log- ans o med apoA-IV concen a ions a e in e sely co ela ed and an inc ease in LDL-C will esul in a dec ease o apoA-IV le els. Adjus ing o HDL-C as well as TG in he mul i a iable Mendelian andomiza ion using all 58 SNPs educed he size o he causal e ec es ima e and i became non-signi ican (βadjus ed = −0.0225, 95% CI = [−0.0441, −0.0010], p- alue = 0.0406, see Table3). Using he g x-package in R, no he e ogeneous SNP could be de ec ed. A e exclusion o he abo e men ioned SNP s964184 and applica ion o he MR-IVW me hod based on he emaining 57 SNPs, we ob ained a signi ican causal es ima e compa able o he MR-IVW Pheno ype (Exposu e ai ) No. o SNPs N exposu e da ase N ou come da ase R² o SNPs on exposu e in % De ec able R2 o exposu e on ou come in %* HDL-C 71 188,577 13,813 6.76 1.37 LDL-C 58 188,577 13,813 7.27 1.27 TG 40 188,577 13,813 3.25 2.85 eGFR 53 133,413 13,813 3.22 2.88 BMI 77 322,206 13,813 2.41 3.84 WHR 40 210,088 13,813 0.60 15.44 Fas ing Glucose 36 133,010 13,813 3.57 2.60 Table 1. S eng h o ins umen al a iables used o he MR analyses: Explained a iance (R²) o SNPs on he di e en exposu e ai s and de ec able causal explained a iance (R2) o he exposu e ai s on apoA-IV as ou come. *Powe = 80%; signi icance le el α = 0.00625; sample size n = N ou come da ase . Pheno ype (Ou come ai ) No. o SNPs N exposu e da ase N ou come da ase R² o SNPs on exposu e in % De ec able R2 o exposu e on ou come in %* HDL-C 2 16,080 188,577 2.00 0.34 LDL-C 2 16,080 188,577 2.00 0.34 TG 2 16,080 188,577 2.00 0.34 eGFR 2 16,080 133,413 2.00 0.48 BMI 2 16,080 322,206 2.00 0.20 WHR 2 16,080 210,088 2.00 0.30 Fas ing Glucose 3 16,080 133,010 2.90 0.48 Table 2. S eng h o ins umen al a iables used o he MR analyses: Explained a iance (R²) o SNPs on apoA-IV as exposu e ai and de ec able causal explained a iance (R2) o apoA-IV as exposu e ai on he di e en ou comes. *Powe = 80%; signi icance le el α = 0.00625; sample size n = N ou come da ase . www.na u e.com/scien i ic epo s/ 4 Scien i ic RepoR s | 7: 8734 | DOI:10.1038/s41598-017-07213-9 es ima e we go using all 58 SNPs (Table3). By adjus ing o HDL-C and TG, he be a es ima e emained nea ly he same bu i became non-signi ican βadjus ed = −0.0259 (95% CI = [−0.0480, −0.0039], p- alue = 0.0213, see Table3). Excluding he 12 SNPs ha we e in LD wi h SNPs associa ed wi h any o he o he ai s (Supplemen a y Table9) and applying he MR-IVW me hod o he emaining 46 SNPs did no ma kedly change he causal es ima e ob ained using all 58 SNPs, bu i became non-signi ican (p = 0.0144, Supplemen a y Table15 and Supplemen a y Figu e6). The applica ion o he mul i a iable Mendelian andomiza ion app oach o he 57 SNPs no in LD wi h SNPs om he non-lipid ai s ga e compa able esul s o using he adjus men on all 58 SNPs (p = 0.0395, Supplemen a y Table15 and Supplemen a y Figu e7). Using MR-Egge eg ession and he weigh ed median es ima ion me hod, LDL-C emained associa ed wi h apoA-IV (Supplemen a y Table16 and Supplemen a y Figu e16). Taking he 40 SNPs associa ed wi h iglyce ide le els and using he MR-IVW me hod, a signi ican causal es ima e o βMR-IVW = −0.0600 (95% CI = [−0.0834, −0.0366], p- alue = 4.80 × 10−7, Table3 and Fig.4) was ob ained. As o LDL-C le els, an inc ease in TG le els will esul in a dec ease o apoA-IV concen a ions. Adjus ing o HDL-C and LDL-C in he mul i a iable MR esul ed in nea ly he same s a is ically signi ican Pheno ype (exposu e) Causal e ec o pheno ype on apoA-IV All SNPs Remo ing SNP s964184 o lipids No. o SNPs Be a*95% CI*P- alue No. o SNPs Be a*95% CI*P- alue Lipid ai s HDL-C, unadjus ed 71 0.0341 [0.0142, 0.0540] 0.0008 70 0.0271 [0.0067, 0.0474] 0.0091 HDL-C, adjus ed o LDL-C and TG (mul i a iable MR) 71 0.0195 [−0.0084, 0.0475] 0.1706 70 0.0262 [−0.0018, 0.0542] 0.0672 LDL-C, unadjus ed 58 −0.0376 [−0.0572, −0.0179] 0.0002 57 −0.0320 [−0.0519, −0.0121] 0.0016 LDL-C, adjus ed o HDL-C and TG (mul i a iable MR) 58 −0.0225 [−0.0441, −0.0010] 0.0406 57 −0.0259 [−0.0480, −0.0039] 0.0213 TG, unadjus ed 40 −0.0600 [−0.0834, −0.0366] 4.8e-07 39 −0.0498 [−0.0776, −0.0220] 0.0004 TG, adjus ed o HDL-C and LDL-C (mul i a iable MR) 40 −0.0597 [−0.0962, −0.0232] 0.0014 39 −0.0487 [−0.0883, −0.0090] 0.0161 Kidney unc ion eGFR 53 −0.3890 [−0.5367, −0.2413] 2.4e-07 Adiposi y- ela ed pa ame e s BMI 77 0.0067 [−0.0266, 0.0400] 0.6946 WHR 40 0.0761 [0.0093, 0.1429] 0.0255 Fas ing Glucose Fas ing Glucose 36 0.0146 [−0.0364, 0.0656] 0.5742 Table 3. MR-IVW es ima es o all pheno ypes assumed o causally a ec he apoA-IV concen a ions. He e, he pheno ypes ac as exposu es and apoA-IV as he ou come. A p- alue smalle han 0.05/8 = 0.00625 is conside ed o be signi ican (ma ked in bold). *In log(apoA-IV) pe change in SD o he lipid and obesi y ai s, in log(apoA-IV) pe change in log(eGFR) o kidney unc ion, in log(apoA-IV) pe change in mmol/L o as ing glucose. Figu e 2. Sca e plo showing he e ec es ima es o SNP-HDL-C (high-densi y lipop o ein choles e ol) associa ions (95% CI) on he x- and SNP-apoA-IV (apolipop o ein A-IV) associa ions (95% CI) on he y-axis o all 71 SNPs. The con inuous black line ep esen s he MR-IVW es ima e o HDL-C on apoA-IV (dashed lines ep esen co esponding 95% CI). The SNP s964184 (ZPR1, o me ly known as ZNF259), which was iden i ied o be po en ially pleio opic is ma ked in ed. www.na u e.com/scien i ic epo s/ 5 Scien i ic RepoR s | 7: 8734 | DOI:10.1038/s41598-017-07213-9 causal es ima e (βadjus ed = −0.0597, 95% CI = [−0.0962, −0.0232], p- alue = 0.0014, see Table3). By applying he g x-package in R, no he e ogeneous SNP could be de ec ed. Excluding he 23 SNPs ha we e in LD wi h SNPs associa ed wi h any o he o he ai s (Supplemen a y Table10) and applying he MR-IVW me hod o he emaining 17 SNPs did no change he causal es ima e ob ained using all 40 SNPs, bu i became non-signi ican (p = 0.1003, Supplemen a y Table15 and Supplemen a y Figu e8). The applica ion o he mul i a iable Mendelian andomiza ion app oach o he 33 SNPs no in LD wi h SNPs om he non-lipid ai s esul ed in a signi ican causal es ima e compa able o he es ima e ob ained using he adjus men on all 58 SNPs (p = 0.0047, Supplemen a y Table15 and Supplemen a y Figu e9). All o he sensi i i y analyses, i.e. he exclusion o SNP s964184 as well as he applica ion o he MR-Egge eg ession o he weigh ed median es ima ion me hod esul ed all in nea ly he same s a is ically signi ican causal es ima e (Table3 and Supplemen a y Table16). Looking a he unnel plo and he indi idual SNP-based con ibu ions o he MR analysis, he causal es ima e o one SNP ( s838880, wi hin SCARB1) a ac ed ou a en ion as i ha bo ed a po en ial bias due o i s low le el o p ecision (Supplemen a y Figu e17). Howe e , excluding his SNP om he MR-IVW analysis as well as om he MR-Egge eg ession analysis did no change he o e all causal es ima e as he p ecision o his one IV es ima e was e y low (da a no shown). Conce ning di ec ional pleio opy, MR-Egge in e cep s did no depa signi ican ly om he o igin o all in es iga ed lipid ai s, meaning ha no di ec ional bias could be de ec ed in all cases (Supplemen a y Table17). Assessmen o he NOME assump ion ga e I²GX = 0.98 o all in es iga ed lipid ai s, sugges ing an app oxima e 2% a enua ion o he causal es ima e owa ds ze o due o measu emen e o in he exposu e ai . The bias adjus men ia SIMEX did no change he causal MR-Egge es ima e no iceably (Supplemen a y Table18). Figu e 3. Sca e plo showing he e ec es ima es o SNP-LDL-C (low-densi y lipop o ein choles e ol) associa ions (95% CI) on he x- and SNP-apoA-IV (apolipop o ein A-IV) associa ions (95% CI) on he y-axis o all 58 SNPs. The con inuous black line ep esen s he MR-IVW es ima e o LDL-C on apoA-IV (dashed lines ep esen he co esponding 95% CI). The SNP s964184 (ZPR1, o me ly known as ZNF259), which was excluded in a sensi i i y analysis is ma ked in ed. Figu e 4. Sca e plo showing he e ec es ima es o SNP-TG ( iglyce ides) associa ions (95% CI) on he x- and SNP-apoA-IV (apolipop o ein A-IV) associa ions (95% CI) on he y-axis o all 40 SNPs. The con inuous black line ep esen s he MR-IVW es ima e o TG on apoA-IV (dashed lines ep esen he co esponding 95% CI). The SNP s964184 (ZPR1, o me ly known as ZNF259), which was excluded in a sensi i i y analysis is ma ked in ed. www.na u e.com/scien i ic epo s/ 6 Scien i ic RepoR s | 7: 8734 | DOI:10.1038/s41598-017-07213-9 Causal associa ion o kidney unc ion (eGFR) on apoA-IV. An o e iew abou he cha ac e is ics and he me a-analysis esul s o he 53 SNPs ound o be associa ed wi h eGFR can be ound in Supplemen a y Table4. Using he MR-IVW me hod o analyzing he in luence o kidney unc ion on he log- ans o med apoA-IV concen a ions esul ed in a s a is ically signi ican causal es ima e o βMR-IVW = −0.3890 (95% CI = [−0.5367, −0.2413], p- alue = 2.44 × 10−7, see Table3 and Fig.5). Consequen ly, an inc ease in eGFR will educe he apoA-IV concen a ions. Using he g x-package in R, no he e ogeneous SNP could be de ec ed. Excluding he 4 SNPs ha we e in LD wi h SNPs associa ed wi h any o he o he ai s (Supplemen a y Table11) and applying he MR-IVW me hod o he emaining 49 SNPs did no change he causal es ima e ob ained using all 53 SNPs a all (Supplemen a y Table15 and Supplemen a y Figu e10). The causal es ima e ob ained using MR-Egge eg ession ga e an e en highe signi ican e ec and he signi ican weigh ed median es ima e was be ween he o he wo es ima es in size (Supplemen a y Table16 and Supplemen a y Figu e18). Using he MR-Egge eg ession, no di ec ional pleio opy could be de ec ed o he analysis o eGFR on he log- ans o med apoA-IV concen a ions (Supplemen a y Table17). Assessmen o he NOME assump ion ga e I²GX = 0.82, sugges ing an app oxima e 15% o 20% a enua ion o he causal es ima e owa ds ze o due o meas- u emen e o in he exposu e ai . The e o e, he bias adjus men ia SIMEX did esul in a sligh ly s onge nega i e e ec o he co ec ed causal MR-Egge es ima e (Supplemen a y Table18). Causal associa ion o he adiposi y- ela ed ai s and as ing glucose on apoA-IV. An o e iew abou he cha ac e - is ics and he me a-analysis esul s o he 77 BMI-SNPs, he 40 WHR-SNPs and he 36 glucose-SNPs can be ound in Supplemen a y Tables5–7. All h ee ac o s did no show a signi ican associa ion on he log- ans o med apoA-IV concen a ions nei he by using he MR-IVW app oach based on all SNPs o a e excluding po en ially pleio opic SNPs using he p oxy sea ch (Table3, Supplemen a y Tables12–14 and Supplemen a y Figu es11–13) no by using he MR-Egge eg ession o he weigh ed median es ima ion app oach (Supplemen a y Table16, Supplemen a y Figu es1–3 and Supplemen a y Figu es19–21). Using he g x-package in R, no he e ogeneous SNPs could be de ec ed. The MR-Egge eg ession in e cep es ima e was non-signi ican o all o hese a iables, such ha no di ec ional pleio opy could be de ec ed (Supplemen a y Table17). Causal associa ion o apoA-IV on he o he ai s. The cha ac e is ics and me a-analysis esul s o he 3 SNPs ound o be associa ed wi h apoA-IV can be ound in Supplemen a y Table19. The in es iga ion o apoA-IV as exposu e on he o he ai s e ealed only one signi ican inding. Two o he h ee SNPs ha we e ound o be associa ed wi h apoA-IV we e p esen in he HDL-C da ase . Using his da a in a MR analysis using he MR-IVW me hod esul ed in a signi ican causal es ima e o βMR-IVW = −0.4017 (95% CI = [−0.5970, −0.2064], p- alue = 5.52e-05, see Table4). The e o e, an inc ease in apoA-IV will esul in a dec ease in HDL-C. Fo he o he ai s, none o he causal es ima es we e signi ican (p- alues anging be ween 0.0754 and 0.8930, Table4). Assessmen o he NOME assump ion ga e I²GX = 0.99, sugges ing an app oxima e 1% a enua ion o he causal es ima e owa ds ze o due o measu emen e o in he exposu e ai (Supplemen a y Table18). The e o e, bias due o measu emen e o in he exposu e ai is negligible. Discussion The e alua ion o he causal ela ions o apoA-IV wi h disease- ela ed ai s using Mendelian andomiza ion analyses e ealed h ee majo indings: The mos in e es ing inding was ha eGFR is likely o causally in luence apoA-IV concen a ions. Fu he mo e, TG was ound o a ec apoA-IV concen a ions, whe eas apoA-IV con- cen a ions seem o in luence HDL-C. The main Mendelian andomiza ion analysis as well as all sensi i i y analyses consis en ly showed ha an inc ease in eGFR is associa ed wi h a dec ease in apoA-IV concen a ions. The in es iga ion in he e e se di ec- ion, i.e. whe he apoA-IV le els a e causal on eGFR, did no esul in a s a is ically signi ican inding. This was Figu e 5. Sca e plo showing he e ec es ima es o SNP-eGFR (es ima ed glome ula il a ion a e) associa ions (95% CI) on he x- and SNP-apoA-IV (apolipop o ein A-IV) associa ions (95% CI) on he y-axis o all 53 SNPs. The con inuous black line ep esen s he MR-IVW es ima e o eGFR on apoA-IV (dashed lines ep esen he co esponding 95% CI). www.na u e.com/scien i ic epo s/ 7 Scien i ic RepoR s | 7: 8734 | DOI:10.1038/s41598-017-07213-9 a he i s glance su p ising since p io s udies ound apoA-IV as a p edic o o CKD p og ession independen o he GFR measu ed a baseline25. I has been sugges ed ha o he p ope ies o apoA-IV no ela ed o he il- a ion capaci y o he kidney migh explain he associa ion wi h CKD p og ession25. I apoA-IV concen a ions change al eady e y ea ly in he p og ession o kidney impai men , apoA-IV may s ill se e as a aluable ma ke o kidney disease and disease p og ession, which has been ound in ea lie c oss-sec ional s udies22–24. On he o he hand, knowing ha ch onic kidney disease is in ima ely associa ed wi h isk o ca dio ascula disease37, his inding is in e es ing o s imula e esea ch on biological pa hways ha link dec ease in kidney unc ion wi h apoA-IV le els. The second main inding was ha TG causally a ec s apoA-IV, e en a e adjus ing o HDL-C and LDL-C. Howe e , in he sensi i i y analysis excluding he po en ially pleio opic SNP s964184, he causal e ec calcu- la ed using he MR-IVW me hod esul ed in a s a is ically non-signi ican es ima e. In his se ing, his was no su p ising, as his SNP was he one wi h he highes e ec size on bo h TG and log(apoA-IV). Howe e , i is only manda o y o exclude his SNP om he analysis, i i iola es he ins umen al a iable assump ion. This would be he case, i i in luenced apoA-IV concen a ions di ec ly o h ough an endogenous a iable ha is di e - en om he exposu e. I was no iden i ied o be pleio opic by he applied s a is ical me hods (g x-package), hough. Al hough being loca ed wi hin he APOA5-A4-C3-A1 gene clus e , his SNP is a he independen om he APOA4-SNPs associa ed wi h apoA-IV concen a ions, as al eady discussed in Lamina e al.17. Rs964184 is loca ed in he 3′ UTR o ZRP1 ( o me ly known as ZNF259), app oxima ely 50 kb downs eam o APOA4, and shows 2 < 0.3 wi h SNPs in APOA4 (acco ding o 1000 Genomes, phase 3; Supplemen a y Figu e14). I p esen s, howe e , a mode a e linkage disequilib ium wi h SNPs in APOA5 (max. 2 = 0.52 o s2266788; Supplemen a y Figu e14), which is a majo egula o o plasma iglyce ide concen a ion38. Exp ession QTLs (eQTLs, i.e. SNPs associa ed wi h exp ession o a gene) epo ed in GTEx o BUD13, ZPR1 and all apolipop o ein genes in he APOA5-A4-C3-A1 gene clus e we e no in linkage disequilib ium (LD) wi h s964184 (maximum pai -wise LD was wi h eQTL s4225 o APOA1: 2 = 0.2; D′ = 0.08; single SNP da a no shown). Toge he his sugges s ha s964184 is nei he in LD wi h SNPs in he APOA4 gene no wi h any known eQTL in he gene clus e , bu p esen s modes LD wi h SNPs in APOA5. I is he e o e unlikely ha s964184 has a di ec e ec on apoA-IV concen a ions and i would be oo conse a i e o exclude i om he analysis. I is also no su p ising ha he exclusion o he 23 SNPs, which a e included in he SNP se s o he o he ai s o in LD wi h SNPs o he o he ai s, esul ed in a non-signi ican causal es ima e. Among hese 23 SNPs we e he mos in o ma i e SNPs, i.e. SNPs wi h he highes e ec on TG and he e o e also on apoA-IV. No e en hal o he iglyce ide SNPs emain, which is p ima ily due o being in LD wi h HDL-C SNPs. The mul i a iable me hod is he mos use ul me hod o accoun o his in e ela ion wi h o he lipid ai s by simul aneously keeping he mos in o ma i e SNPs. The e o e, we addi ionally excluded only SNPs ha we e in LD wi h SNPs om he non-lipid ai s, bu adjus ed o HDL-C and LDL-C. Using his me hod, he causal es ima e om iglyce ides on apoA-IV emained s a is i- cally signi ican . Ne e heless, his ela ionship has o be in e p e ed wi h cau ion. Fu he s udies a e needed o con i m his po en ially causal associa ion o TG and apoA-IV. The in es iga ion o HDL-C and LDL-C as causal ac o s in luencing apoA-IV concen a ions yielded incon- sis en esul s. In p e ious gene ic s udies, se e al loci ha e been associa ed wi h apoA-IV as well as HDL-C and LDL-C17–19. Besides, a s udy in es iga ing he LDL-C lowe ing esponse o di e en die s ound ha he apoA-IV p o ein iso o ms modula e he LDL-C lowe ing esponse o a die 39. Howe e , we demons a ed ecen ly in a s udy wi h mo e han 13,000 indi iduals, ha he gene ic a ian s ha a e he molecula basis o he apoA-IV iso o ms do no ha e an e ec on apoA-IV concen a ions. In he da a a hand, he MR-IVW me hod and he MR-Egge eg ession e ealed a s a is ically signi ican posi i e causal es ima e, bu he e ec anished a e adjus men o he o he lipid ai s. I is he e o e concei able ha he e ec o bo h HDL-C as well as LDL-C on apoA-IV concen a ions is igge ed p ima ily by he associa ion wi h TG. Fu he mo e, 5 SNPs a e p esen in Pheno ype (ou come) Causal e ec o apoA-IV on pheno ype No. o SNPs Be a§95% CI§P- alue Lipid ai s HDL-C, unadjus ed 2−0.4017 [−0.5970, −0.2064] 5.52e-05 LDL-C, unadjus ed 2 0.1931 [−0.0197, 0.4059] 0.0754 TG, unadjus ed 2−0.0761 [−0.2677, 0.1154] 0.4360 Kidney unc ion eGFR 2 0.0207 [−0.0136, 0.0550] 0.2369 Adiposi y- ela ed pa ame e s BMI 2 −0.0325 [−0.1840, 0.1190] 0.6744 WHR 2 0.0119 [−0.1609, 0.1847] 0.8930 Fas ing Glucose Fas ing Glucose 3 0.0462 [−0.0708, 0.1631] 0.4392 Table 4. MR-IVW es ima es o all pheno ypes assumed o be causally a ec ed by apoA-IV concen a ions. §In SD pe change in log(apoA-IV) o he lipid and obesi y ai s, in log(eGFR) pe change in log(apoA-IV) o kidney unc ion, in mmol/L pe change in log(apoA-IV) o as ing glucose. He e, apoA-IV ac s as exposu e and he pheno ypes as he ou come ai s. A p- alue smalle han 0.05/8 = 0.00625 is conside ed o be signi ican (ma ked in bold). www.na u e.com/scien i ic epo s/ 8 Scien i ic RepoR s | 7: 8734 | DOI:10.1038/s41598-017-07213-9 all h ee lipid da ase s ( s12748152, s964184, s174546, s3764261, s2954029) which migh also in luence his esul . The explo a ion o he e e se causa ion, i.e. whe he apoA-IV in luences HDL-C, esul ed in a signi ican inding, showing an in e se causal di ec ion. Howe e , his esul is igge ed mainly by one SNP in he APOA4 gene ( s1729407). This SNP was associa ed wi h bo h apoA-IV concen a ions as well as HDL-C. Since i was he op signal in he GWAS on apoA-IV concen a ions17 bu no genome-wide signi ican ly associa ed wi h HDL-C18, al hough conduc ed in a much highe sample size, i is concei able ha i exe s an e ec p ima ily on apoA-IV. Howe e , a di ec e ec on HDL-C canno be excluded which would iola e he exclusion- es ic ion assump ion o Mendelian andomiza ion. Gi en ha apoA-IV is an impo an componen o he HDL-C pa i- cle21, 40–42, a causal e ec o apoA-IV on HDL-C le el is possible and pa ly suppo ed by ou da a, bu has o be conside ed wi h cau ion. In es iga ing he adiposi y- ela ed ai s esul ed in no signi ican indings. This is in con as o he li e a u e on obse a ional epidemiological and unc ional s udies: apoA-IV has been p oposed as a sa ie y ac o and ela ed o die -induced adiposi y, bo h in animal models21 and in humans26. In addi ion, he adminis a ion o apoA-IV esul ed in educ ion o ood in ake and gene ic associa ion s udies ha e ound associa ions be ween polymo phisms wi hin he APOA4 gene and adiposi y- ela ed ai s27, 28. Al hough hese indings a gumen in a o o apoA-IV as a isk ac o , i has also been shown ha apoA-IV is dec eased ollowing weigh loss and/ o gas ic bypass29, 30. Despi e he epidemiological and also unc ional s udies linking adiposi y- ela ed ai s o apoA-IV, we could no ind a causal e ec in one o he o he di ec ion. One eason o ha migh be ha – despi e he al eady huge da ase s and he high numbe o associa ed SNPs – he powe is s ill insu icien o de ec a causal e ec om especially WHR as exposu e on apoA-IV as ou come a iable. To be de ec able in ou analysis wi h a powe o 80%, he causal e ec o WHR on apoA-IV would ha e o be ha high, ha i explains a leas 15% o he pheno ypic a iance o apoA-IV (Table1). The de ec able explained a iance is in a mo e ealis ic ange o BMI (abou 3.8%), bu s ill highe han o he o he in es iga ed pheno ypes. Fo he o he di ec ion o causali y (apoA-IV as exposu e o adiposi y), howe e , lack o powe should no be an issue. E en such small e ec s as 0.2–0.3% o explained a iance should ha e been possible o de ec . The e o e, i is a he unlikely, ha apoA-IV causally in luences BMI o WHR. I has o be no ed ha he choice o WHR and BMI as ac o s o sa ie y migh no be op imal. Howe e , hey a e he mos commonly used ma ke s o adiposi y, o which he e is also su icien in o ma ion on gene ic loci in luencing hese ai s and we e he e o e chosen o he e alua ion o adiposi y- ela ed ai s. Coun e -in ui i ely, we ound no causal associa ion o as ing glucose wi h apoA-IV and ice e sa. This should no be a p oblem o powe like o he adiposi y- ela ed ai s, as he e ec ha could be de ec ed wi h he gi en da a is e en smalle han o TG and eGFR. Al eady much is known ha links glucose wi h apoA-IV. Diabe ics ha e been ound o ha e signi ican ly highe apoA-IV le els26, 43. Apa om ha , apoA-IV was in e sely associ- a ed wi h p ediabe es (de ined by as ing glucose le els) and also wi h 2 h glucose le els4. Expe imen al s udies and APOA4 knockou mice ha e also shown ha apoA-IV has a glucose-lowe ing e ec 29, 44. The e o e, we would ha e expec ed o ind a causal associa ion o he apoA-IV SNPs (whe e all 3 SNPs we e p esen in he da ase om Manning e al.) on as ing glucose. In his case, apoA-IV would ha e o explain only mo e han 0.48% o he phe- no ypic a iance o as ing glucose o ha e a powe o 80%. S ill, we could no ind a causal associa ion wi h any o he me hods used in any o he in es iga ed di ec ions. No inding a signi ican associa ion does no exclude he possibili y ha he e s ill exis s one: apoA-IV and as ing glucose migh s ill be causally ela ed wi h each o he bu he ue e ec migh be smalle han wha can be de ec ed wi h ou da a. The o he possibili y o explain his non-signi ican inding migh be ha he obse ed associa ion o apoA-IV wi h glucose is con ounded by o he ac o s ha in luence bo h ai s. Using summa y-le el da a o all SNPs ha ha e been shown o be genome-wide signi ican ly associa ed wi h hei espec i e ai s as ins umen al a iables, is bo h a s eng h bu also a limi a ion o ou s udy. By using summa ized da a, a highe powe can be achie ed han in single s udies. The a ailabili y o dozens o associ- a ed SNPs ha we e iden i ied in mo e han 100,000 pa icipan s does no gua an ee o ge a high powe o he Mendelian andomiza ion analysis, hough. Fo he adiposi y- ela ed ai s as isk/p o ec i e ac o s o apoA-IV, he ins umen al a iables do no explain much o he a ia ion in hese ai s. The e o e, o be able o ind a causal associa ion o a easonable size, u he SNPs would ha e o be iden i ied ha can be used o inc ease he a iance explained on BMI and especially WHR. Fo he lipid ai s, howe e , ou applied me hods based on summa ized da a achie ed a high powe enabling he de ec ion o e y low causal e ec s. Ne e heless, no only high powe is necessa y o Mendelian andomiza ion analysis bu also he choice o alid ins umen s. To be a alid ins umen al a iable, a SNP should be associa ed only wi h he espec i e isk/p o ec i e ac o . Fo he co ela ed ai s HDL-C, LDL-C and TG, he e is subs an ial o e lap o he SNPs, which makes i ex emely di i- cul o disc imina e he ins umen al a iables o each o he lipid ai s co ec ly om each o he . The exclusion o all ins umen al a iables ha we e ound o be associa ed wi h a leas wo o he h ee lipid ai s would lead o a subs an ial loss o s eng h o he Mendelian andomiza ion analyses. Howe e , we checked o possible pleio opy o he SNPs using app op ia e s a is ical me hods and adjus ed o he e ec s o he o he lipid ai s o ind independen causal e ec s. S ill, he in e p e a ion o he iden i ied causal e ec s emains di icul . Ano he limi a ion which a ises om using summa y-le el da a is ha obse ed e ec sizes canno be de i ed om di ec ly and hus, i no ob ainable om he li e a u e, canno be compa ed wi h he es ima ed causal e ec s. Al hough we only in es iga ed causal ela ionships o which he e is al eady su icien e idence o ei he obse - a ional associa ion o unc ional expe imen s, di ec compa ison o obse a ional and causal es ima e is only possible i i is on he same scale. All GWAS me a-analysis da a we used o ou analyses a e de i ed om ei he log- ans o ma ion o in e se no mal ans o ma ion. Howe e , such ans o ma ions a e ha dly e e done in obse a ional epidemiological s udies. Resul s om unc ional we -lab expe imen s can ce ainly no be ans- e ed in o obse a ional es ima es. www.na u e.com/scien i ic epo s/ 9 Scien i ic RepoR s | 7: 8734 | DOI:10.1038/s41598-017-07213-9 To summa ize, ou da a e ealed an in e se causal associa ion o kidney unc ion (eGFR) on apoA-IV. Fu he mo e, in es iga ing he lipid ai s sugges ed a causal in ol emen o p ima ily iglyce ide le els on apoA-IV concen a ions. The causal e ec s o HDL-C and LDL-C on apoA-IV a e ha d o disc imina e om each o he and om he e ec s igge ed by TG. Ou esul s also a gue o a causal, in e se associa ion o apoA-IV concen a ions on HDL-C, which howe e ha e o be in e p e ed wi h cau ion due o po en ial p esence o plei- o opic e ec s o he gene ic ma ke s. Me hods Da a sou ces and SNP selec ion.  Fo all e alua ed disease- ela ed ai s and apoA-IV, SNPs we e selec ed and summa ized esul s we e aken om he mos ecen and mos comp ehensi e GWAS, which we e conduc ed p ima ily in coho s o Eu opean ances y. A g aphical o e iew o he design and s udies used in ou in es iga- ion is gi en in Fig.1. A ac ion o he i e s udies which ha e been included in he apoA-IV-GWAS me a-anal- ysis ha e also been pa o he GWAS me a-analyses o he o he pa ame e s (lipids, eGFR, adiposi y- ela ed pa ame e s and as ing glucose). The o e lap o samples anges be ween 19% and 80%. De ails on he da a sou ces and SNP selec ion can be ound in he Supplemen a y ma e ial. Measu emen o apoA-IV. Fo all pa icipa ing s udies, quan i ica ion o plasma apoA-IV was done in he same labo a o y (Di ision o Gene ic Epidemiology, Medical Uni e si y o Innsb uck, Aus ia). I was based on a double-an ibody enzyme-linked immunoso ben assay using an a ini y-pu i ied polyclonal abbi an i-human apoA-IV an ibody o coa ing and he same an ibody coupled o ho se adish pe oxidase o de ec ion. Plasma wi h a known concen a ion o apoA-IV was used as he calib a ion s anda d45. Fou con ol se a wi h di e en concen a ions we e un on each pla e in double measu emen s o con ol pu poses h oughou he en i e p o- jec . The in a- and in e assay coe icien s o a ia ion we e 2.7% and 6.0%, espec i ely45. S a is ical Me hods Validi y o ins umen al a iables and powe analysis. In Mendelian andomiza ion analyses, ins u- men al a iables should mee ce ain equi emen s o minimize weak ins umen bias. The s eng h o he ins umen al a iables (SNPs) used o he Mendelian andomiza ion analysis was assessed using he explained a iance (R²). R² was calcula ed acco ding o Pa a o e al.33. Using his o mula, he pe cen age o pheno ypic a iance explained by he ins umen al a iables (SNPs) can be es ima ed as =∑= R R i ki 212 , whe e β=R a SNP a X()/() ii i 22 is he coe icien o de e mina ion o all k SNPs associa ed wi h he po en ial isk ac o / exposu e X , βi is he es ima ed e ec o he i h SNP on he isk ac o X , =× ×− () a SNPMAF MAF() 21 iSNP SNP ii and a X() is he a iance o he po en ial isk ac o X ( = a X() 1 o he lipid and adiposi y ai s, since he be a es ima es e e o change in 1 s anda d de ia ion (SD)). The F-s a is ic is ypically used o judge on he alidi y o ins umen al a iables. Howe e , i canno be calcu- la ed in his se ing, whe e only summa y-le el da a a e a ailable. Howe e , only genome-wide signi ican SNPs ha a e independen om each o he (pai wise LD be ween all SNPs: 2 < 0.1) we e included in his analysis (p- alue < 5 × 10−8). This co esponds o an F s a is ic > 30 o each single a ian 46. In he Mendelian andom- iza ion li e a u e a h eshold o F < 10 has ypically been used o de ine a “weak IV” ( he S aige -S ock ule47, 48). Since we a e using a combina ion o se e al genome-wide signi ican SNPs, weak ins umen bias is negligble. Powe calcula ions we e ca ied ou using he online ool h ps://sb452.shinyapps.io/powe /. The powe can- no be calcula ed di ec ly as di e en uni s we e used o he di e en ai s and only summa y-le el da a a e a ailable. The e o e, a ough app oxima ion o all in es iga ed ai s was pe o med based on s anda dized al- ues. The applied online ool was a he mean o Mendelian andomiza ion me hods based on indi idual-le el da a. Howe e , as shown in e . 49, analyses based on indi idual le el da a and summa ized da a me hods a e compa able wi h espec o powe . The sample size assumed o he powe analysis is se o he sample size o he ou come da ase , since he a io es ima e in ol es he a iance o he ou come da ase , bu no he a iance o he exposu e da ase . The e o e, he sample size was se o 13,800 o apoA-IV as he ou come a iable. Fo he powe -analysis o he e e se causa ion (apoA-IV as exposu e on he a ious pheno ypes), a sample size o 188,577 is assumed o he lipid analysis e c. (Table2). Gi en R2 o he gene ic a ian s on he exposu e, he sample size, he desi ed powe (80%) and he sig- ni icance le el, i can be calcula ed which causal e ec es ima e can be de ec ed wi h he gi en da a, i i is uly he e. This de ec able causal es ima e βcausal is based on change in one SD o he ou come pe one SD change in exposu e. In his se ing βcausal 2 is equal o he a iance explained ( Rcausal 2 ) o he exposu e a iable on he ou come. The e o e, we can answe he ques ion, which s eng h o causal associa ion we can mos likely de ec wi h he gi en da a. Mendelian andomiza ion me hods. The Mendelian andomiza ion analyses we e pe o med bidi- ec ional. In he i s un, apoA-IV was conside ed as an ou come a iable whe eas lipids, eGFR, BMI, WHR (adjus ed o BMI) and as ing glucose le els we e conside ed as exposu es. In he second un, apoA-IV was he exposu e a iable and he o he pheno ypes (lipids, eGFR, BMI, WHR (adjus ed o BMI) and as ing glucose le els) we e used as ou comes. Be o e any analysis, he SNP-exposu e and SNP-ou come associa ion es ima es ha e all been o ien ed owa ds an inc ease in he exposu e ai . Fo he main Mendelian andomiza ion analysis, he SNP-exposu e and SNP-ou come es ima es we e com- bined using he in e se- a iance weigh ed (MR-IVW) me hod as p oposed by Bu gess e al.46. Causal es ima es based on his me hod a e no a ed as βMR-IVW. The MR-IVW es ima e can also be in e p e ed as a weigh ed eg es- sion om he e ec es ima es o he exposu e SNPs on he es ima es o he ou come o he same SNPs ( emo ing he in e cep ).