An interaction map of circulating metabolites, immune gene networks, and their genetic regulation

RESEARCH Open Access An in e ac ion map o ci cula ing me aboli es, immune gene ne wo ks, and hei gene ic egula ion A ika P. Na h 1,2 , Sco C. Ri chie 2,3 , Sean G. Bya s 3,4 , Liam G. Fea nley 3,4 , Aki S. Ha ulinna 5,6 , Anni Joensuu 5 , An i J. Kangas 7 , Pasi Soininen 7,8 , Annika Wenne s öm 5 , Lili Milani 9 , And es Me spalu 9 , Sa u Männis ö 5 , Pe e Wü z 7,10 , Johannes Ke unen 5,7,8,11 , Emma Rai oha ju 12 , Mika Kähönen 13 , Ma kus Juonala 14,15 , Aa no Palo ie 6,16,17,18 , Mika Ala-Ko pela 7,8,11,19,20 , Samuli Ripa i 6,21 , Te ho Leh imäki 12 , Gad Ab aham 2,3,4 , Olli Rai aka i 22,23 , Veikko Salomaa 5 , Ma kus Pe ola 5,6,9 and Michael Inouye 1,2,3,4* Abs ac Backg ound: Immunome abolism plays a cen al ole in many ca diome abolic diseases. Howe e , a obus map o immune- ela ed gene ne wo ks in ci cula ing human cells, hei in e ac ions wi h me aboli es, and hei gene ic con ol is s ill lacking. He e, we in eg a e blood ansc ip omic, me abolomic, and genomic p o iles om wo popula ion-based coho s ( o al N = 2168), including a subse o indi iduals wi h ma ched mul i-omic da a a 7-yea ollow-up. Resul s: We iden i y opologically eplicable gene ne wo ks en iched o di e seimmune unc ions including cy o oxici y, i al esponse, B cell, pla ele , neu ophil, and mas cell/basophil ac i i y. These immune gene modules show complex pa e ns o associa ion wi h 158 ci cula ing me aboli es, including lipop o ein subclasses, lipids, a y acids, amino acids, small molecules, and CRP. Genome-wide scans o module exp ession quan i a i e ai loci (mQTLs) e eal i e modules wi h mQTLs ha ha e bo h cis and ans e ec s. The s onges mQTL is in ARHGEF3 ( s1354034) and a ec s a module en iched o pla ele unc ion, independen o pla ele coun s. Modules o mas cell/basophil and neu ophil unc ion show empo ally s able me aboli e associa ions o e 7-yea ollow-up, p o iding e idence ha hese modules and hei cons i uen gene p oduc s may play cen al oles in me abolic in lamma ion. Fu he mo e, he s onges mQTL in ARHGEF3 also displays clea empo al s abili y, suppo ing widesp ead ans e ec s a his locus. Conclusions: This s udy p o ides a de ailed map o na u al a ia ion a he blood immunome abolic in e ace and i s gene ic basis, and may acili a e subsequen s udies o explain in e -indi idual a ia ion in ca diome abolic disease. Backg ound O e he pas decade inc easing e idence has implica ed in lamma ion as a p obable causal ac o in me abolic and ca dio ascula diseases. Consequen ly, esea ch has begun o ocus on he in e play be ween immuni y and me abol- ism, o immunome abolism. While i is in ol ed in di e se pa hophysiologies, immunome abolism is pa icula ly ele an o diseases o immense global heal h bu den, such as ype 2 diabe es (T2D) and a he oscle osis. Fo T2D, immune o e ac i a ion in adipose issue has been implica ed as a key d i e [1, 2]. S udies ha e shown ha mac ophage in il a ion and subsequen o e exp ession o p oin lamma o y cy okines, such as TNF-α, in adipose issues is associa ed wi h insulin e- sis ance [1, 2]. Mo eo e , e idence o me abolic in lam- ma ion has been shown in o he issues whe e, in blood, ele a ed glucose and ee a y acid le els po en ia e IL- 1β-media ed des uc ion o panc ea ic ß cells and subse- quen T2D p og ession [3–5]. While ci cula ing me abo- li es a e known o be associa ed wi h ca dio ascula disease [6], in lamma ion is an inc easingly ecognized ac o in pa hogenesis. In a he oscle osis, lipid-induced in lamma o y esponse mechanisms ha e also been * Co espondence: minouye@bake .edu.au 1 Depa men o Mic obiology and Immunology, The Uni e si y o Melbou ne, Pa k ille 3010, Vic o ia, Aus alia 2 Sys ems Genomics Lab, Bake Hea and Diabe es Ins i u e, Melbou ne, Vic o ia, Aus alia Full lis o au ho in o ma ion is a ailable a he end o he a icle © The Au ho (s). 2017 Open Access This a icle is dis ibu ed unde he e ms o he C ea i e Commons A ibu ion 4.0 In e na ional License (h p://c ea i ecommons.o g/licenses/by/4.0/), which pe mi s un es ic ed use, dis ibu ion, and ep oduc ion in any medium, p o ided you gi e app op ia e c edi o he o iginal au ho (s) and he sou ce, p o ide a link o he C ea i e Commons license, and indica e i changes we e made. The C ea i e Commons Public Domain Dedica ion wai e (h p://c ea i ecommons.o g/publicdomain/ze o/1.0/) applies o he da a made a ailable in his a icle, unless o he wise s a ed. Na h e al. Genome Biology (2017) 18:146 DOI 10.1186/s13059-017-1279-y implica ed in p og ession o myoca dial in a c ion [7]. In a he ogenic lesions, oxidized phospholipids a e known o lead o a new mac ophage pheno ype [8], and choles- e ol loading in mac ophages p omo es p oin lamma o y cy okine sec e ion [9]. Pe haps su p isingly, ew la ge-scale s udies ha e sys- ema ically assessed in e ac ions be ween he human im- mune sys em and me aboli es. Recen s udies ha e in es iga ed ma ched blood ansc ip omic and me abo- lomic p o iles o unde s and hei in e play [10–16]. Howe e , hese s udies had modes sample sizes and hus ha e no had he powe o ocus on he di e se ange o immune p ocesses ha in e ac wi h ci cula ing me aboli es. Fu he mo e, e en ewe ha e assessed e - ec s o exp ession quan i a i e ai loci (eQTLs) on immune gene ne wo ks. A obus in eg a ed map o immunome abolic ela ionships and hei gene ic egula- ion would p o ide a ounda ion o in es iga ing he di - e en ial ca diome abolic disease suscep ibili y amongs indi iduals while also iden i ying key a ge in e ac ions o mechanis ic in i o and in i o ollow-up. In his s udy, we p esen an in eg a ed immunome a- bolic map using ma ched blood me abolomic and an- sc ip omic p o iles om 2168 indi iduals om wo popula ion-based coho s. We pe o m gene coexp es- sion ne wo k disco e y and c oss-coho eplica ion o iden i y obus gene modules which encode immune- ela ed unc ions. Using a high- h oughpu quan i a i e NMR me abolomics pla o m ha can sepa a e lipids and lipop o ein sub- ac ions as well as quan i y a panel o pola me aboli es, we iden i y signi ican in e ac ions be ween immune gene modules and ci cula ing me abol- i e measu es. Genome-wide scans o QTLs a ec ing im- mune gene modules iden i y many cis and ans loci a ec ing module exp ession. Finally, we es he long- e m s abili y o gene modules, hei in e ac ions wi h me aboli e measu es, and gene ic con ol using a 7-yea ollow-up sampling o 333 indi iduals. Resul s and discussion Summa y o coho s and da a We analyzed genome-wide geno ype, whole blood an- sc ip omic, and se um me abolomics da a om wo popula ion-based coho s (“Me hods”and Fig. 1). In DILGOM07, 240 males and 278 emales aged 25–74 yea s we e ec ui ed ( o al N = 518). Da a we e a ailable o a subse o 333 pa icipan s om DILGOM07 who we e ollowed up a e 7 yea s (DILGOM14). In YFS, ele an da a we e a ailable o 755 males and 895 e- males aged 34–49 yea s ( o al N = 1650). DILGOM and YFS geno yping was pe o med using Illumina Human 610 and 670 a ays, espec i ely, wi h subsequen geno ype impu a ion pe o med using IM- PUTE2 [17] and he 1000 Genomes Phase I e sion 3 e e ence panel. Fo bo h coho s, whole blood an- sc ip ome p o iling was pe o med using Illumina HT-12 a ays and se um me abolomics p o iling was ca ied ou using he same se um NMR me abolomics pla o m (B ainshake L d) [18]. Indi iduals on lipid-lowe ing medica ion and p egnan women we e excluded om he me abolome analyses (“Me hods”). O he 159 me- aboli e measu es analyzed, 148 we e di ec ly quan i ied and 11 de i ed (Addi ional ile 1: Table S1). A e il e - ing, ma ched ansc ip ome and me abolome da a we e a ailable o 440 indi iduals in DILGOM07 and 216 o hese indi iduals (DILGOM14) who we e p o iled a 7- yea ollow-up. In YFS, 1575 indi iduals we e a ailable wi h simila da a (see “Me hods” o de ails). Robus immune gene coexp ession ne wo ks om blood We i s iden i ied ne wo ks o igh ly coexp essing genes in DILGOM07 and hen used a pe mu a ion ap- p oach, Ne Rep [19], o s a is ically es eplica ion pa - e ns o densi y and connec i i y o hese ne wo ks in YFS. Fo module de ec ion, we applied weigh ed gene coexp ession ne wo k analysis (WGCNA) o all 35,422 p obes in he DILGOM07 da a, iden i ying a o al o 40 modules o coexp essed genes (“Me hods”). Fo each module, we used Ne Rep o calcula e se en p ese a ion s a is ics in he YFS, gene a e empi ical null dis ibu ions o each o hese es s a is ics, and calcula e hei co e- sponding P alues [19, 20]. A module was conside ed s ongly p ese ed i he P alue was <0.001 o all se en p ese a ion s a is ics (Bon e oni co ec ion o 40 modules). O he 40 DILGOM07 modules, 20 we e s ongly p ese ed in YFS (Addi ional ile 2: Table S2). Fo each o he 20 eplica ed modules, we de ined co e gene p obes, hose which a e mos igh ly coexp essed YFS DILGOM07 DILGOM14 S udy Popula ions Molecula Da a Ma ched T ansc ip ome- Me abolome Ma ched T ansc ip ome- Geno ypes N=333 N=258 N=294 Me abolome T ansc ip ome Blood Samples Geno ypes Disco e Gene Modules in DILGOM07 Ne wo k analysis Replica e Gene Modules in YFS and DILGOM14 SNPs N=518 N=440 N=515 N=1650 N=1575 N=1400 Pe mu a ion s a is ics GO en ichmen analysis Immune-linked gene ne wo ks Immune-me aboli e associa ion Gene ic d i e s o immune modules Analysis S a egy T ansc ip ome Fig. 1 The s udy design. GO Gene On ology, SNP single nucleo ide polymo phism Na h e al. Genome Biology (2017) 18:146 Page 2 o 15 and hus obus o clus e ing pa ame e s, using a pe mu a ion es o module membe ship (“Me hods”; Addi ional ile 3: Table S3). To iden i y modules o pu a i e immune unc ion, we ca ied ou Gene On ology (GO) biological p ocess en- ichmen analysis using GO illa o he co e genes o each eplica ed module [21]. Signi ican GO e ms ( alse disco e y a e (FDR) <0.05) we e hen summa ized in o ep esen a i e e ms based on seman ic simila i y using REVIGO [22] (Addi ional ile 4: Figu e S1). A module was conside ed immune- ela ed i i was signi ican ly en iched o GO e ms “immune sys em p ocesses” (GO:0002376) and/o “ egula ion o immune sys em p ocesses”(GO:0002682) in he REVIGO ou pu . Six ou o 20 modules we e en iched o a leas one o hese e ms (Addi ional ile 5: Table S4). We also iden i ied wo addi ional modules which we e no en iched o any GO e ms bu ha e been p e iously linked o immune unc ions ela ed o mas cell and basophil unc ion [13] and pla ele agg ega ion ac i i y [23]. The eigh modules encoded di e se immune unc ions, including cy o oxic, i al esponse, B cell, pla ele , neu ophil, mas cell/baso- phil, and gene al immune- ela ed unc ions. Each immune module’s gene con en and pu a i e biological unc ion is summa ized in Table 1. Immune module associa ion analysis o eQTLs and me aboli e measu es Fo each gene module, we pe o med a genome-wide scan o iden i y module QTLs (mQTLs) ha egula e ex- p ession. In DILGOM07 and YFS, he module eigengene was eg essed on each SNP, hen mQTL es s a is ics we e combined in a me a-analysis (“Me hods”). Signi i- can mQTLs we e u he examined a indi idual gene exp ession le els. A genome-wide signi icance le el (P alue <5 × 10 −8 ) was used o iden i y mQTLs and signi i- can ans e ec s on indi idual gene exp ession (Fig. 2 and Table 2). Leukocy e and pla ele coun s we e a ail- able o YFS and we e used o es he obus ness o module associa ions wi h mQTLs and me aboli e mea- su es. Six modules showed s a is ically signi ican associ- a ion wi h pla ele o leukocy e coun s (P alue <0.05) (Addi ional ile 6: Table S5); howe e , adjus men o leukocy e coun s did no a ec mQTL no module- me aboli e measu e associa ions, wi h he excep ion o he pla ele module (PM) and cy o oxic cell-like module Table 1 Immune module gene con en and pu a i e biological unc ion based on GO e ms ( op h ee shown) and li e a u e Module Size GO e ms Li e a u e-based immune- ela ed unc ion o genes Cy o oxic cell-like module (CCLM) 130 (115) Immune sys em p ocess De ense esponse Immune esponse Cy o oxic e ec o s (GZMA,GZMB,GZMM,CTSW,PRF1 [66]); su ace ecep o s (IL2RB,SLAMF6,CD8A,CD8B,CD2,CD247,KLRD1,KLRG1 [66–68]); T and NK cell di e en ia ion (ID2 and EOMES [69,70]), ac i a ion (ZAP70 and CBLB [71,72]), and ec ui men (CX3CR1,CCL5,CCL4L2 [73]) Vi al esponse module (VRM) 95 (88) Response o i us Type I in e e on signaling pa hway Response o bio ic s imulus Type I in e e on-induced an i i al ac i i y (IFITM1,IFIT1,IFIT2,IFIT3,IFIT5,IFI44, IFI44L,IFI6,MX1,ISG15,ISG20,HERC5 [74,75]); i al RNA deg ada ion (OAS1, OAS2,OAS3,OASL,DDX60 [30]); ype 1 in e e on-signaling pa hway (IRF9, STAT1,STAT2 [76,77]) B-cell ac i i y module (BCM) 54 (49) Immune sys em p ocess Immune esponse B cell ac i a ion B cell su ace ma ke s (CD79A,CD79B,CD22 [33,78]); B cell ac i a ion (BANK1, BTLA,CD40,TNFRSF13B,TNFRSF13C [79]), de elopmen (POU2AF1,BCL11A, RASGRP3 [80]), mig a ion (CXCR5,CCR6 [80,81]), and hei egula ion (CD83, FCER2,FCRL5 [82]); an igen p esen a ion (HLA-DOA,HLA-DOB [83]) Pla ele module (PM) a 114 (106) Coagula ion Blood coagula ion Cell ac i a ion Pla ele ecep o signaling, ac i a ion, and coagula ion (GP6,GP9,ITGA2B, ITGB3,ITGB5,MGLL,MPL,MMRN1,PTK2,VCL,THBS1,F13A1,VWF,[84]); egula ing pla ele ac i i y (SEPT5,TSPAN9 [85,86]) Neu ophil module (NM) a 26 (26) Killing o cells o o he o ganism Cell killing Response o ungus An i-mic obial, - ungal, and - i al ac i i y (DEFA1,DEFA1B,DEFA3,DEFA4, ELANE,BPI,RNASE2,RNASE3 [87–90]); neu ophil-media ed ac i i y (AZU1, LCN2,MPO,CEACAM6,CEACAM8,OLFM4 [90,91]) and i s egula ion (LCN2, CAMP,OLR1 [49,92,93]) Lipid-leukocy e module (LLM) a 13 (13) Mas cell and basophil unc ion b Mas cell and basophil ela ed immune esponse and alle gic in lamma ion (FCER1A,HDC,GATA2,SLC45A3,CPA3,MS4A3 [13,94,95]) Gene al immune module A (GIMA) 509 (482) Immune sys em p ocess De ense esponse Regula ion o esponse o s imulus These modules con ain genes in ol ed in a b oad ange o immune p ocesses and hei egula ion such as signaling; cell dea h; de ense esponse o s ess, in lamma ion, and ex e nal s imuli; leukocy e ac i a ion, mig a ion, and adhesion Gene al immune module B (GIMB) 74 (69) Immune esponse-ac i a ing signal ansduc ion Posi i e egula ion o immune esponse Ac i a ion o immune esponse Size e e s o he numbe o co e genes in each module and he subse o hese co e genes wi h GO e m anno a ions a e lis ed in pa en heses. Func ions we e assigned o each o hese modules based on GO en ichmen s and li e a u e-based sea ches o genes in he modules. a Modules p e iously epo ed o ha e immune ela ed unc ion b The LLM was no signi ican ly en iched o any GO e m Na h e al. Genome Biology (2017) 18:146 Page 3 o 15 (CCLM) discussed below (Addi ional ile 7: Table S6). Since we did no ha e cell coun s a ailable o DIL- GOM07, all he associa ions be ween immune modules and me aboli e measu es discussed below, unless o he - wise no ed, ha e no been adjus ed o cell coun s. Cy o oxic cell-like module CCLM was associa ed wi h 24 me aboli e measu es, mainly consis ing o a y acids, in e media e densi y lipop o eins, and C- eac i e p o ein (CRP; Fig. 3; Addi ional ile 8: Table S7). O hese, he a e age deg ee Fig. 2 Module and exp ession QTL analysis. aManha an plo o me a-analyzed P alues om he DILGOM/YFS module QTL analysis. The lead SNP and i s closes genes a e no ed. Each signi ican mQTL locus is colo ed by module. The ho izon al dashed line ep esen s genome-wide (me a- P alue <5 × 10 −8 ) signi icance. b–dCi cula plo s summa izing he indi idual gene associa ions (me a-P alue <5 × 10 −8 ) o he lead module QTLs in he VRM, PM, and NM. Lead SNPs and cis genes a e labeled ou side he ing. PM pla ele module, VRM i al esponse module, CCLM cy o oxic cell-like module, NM neu ophil module, BCM B-cell ac i i y module Table 2 QTLs o immune gene modules Module Top SNP Ch Hg19 pos. (Mb) Allele (mino /majo ) MAF (a g) P alue DILGOM07 (e ec size) P alue YFS (e ec size) Me a P alue VRM s182710579 4 19768086 G/T 0.012 2.01 × 10 –4 (0.05) 8.10 × 10 –6 (0.02) 9.23 × 10 –9 s151234502 7 148950168 T/C 0.012 2.59 × 10 –1 (0.01) 5.31 × 10 –9 (0.03) 2.46 × 10 –8 s147742798 11 70947761 T/C 0.016 1.51 × 10 –3 (0.04) 1.66 × 10 –6 (0.02) 9.43 × 10 –9 BCM s2523489 6 31348878 T/C 0.186 1.42 × 10 –1 (0.005) 5.29 × 10 –8 (0.006) 6.27 × 10 –8 PM s1354034 3 56849749 T/C 0.284 7.11 × 10 –14 (-0.02) 1.51 × 10 –16 (-0.008) 7.35 × 10 –28 s28367734 6 3128657 A/G 0.108 5.40 × 10 –4 (0.02) 2.02 × 10 –5 (0.006) 5.44 × 10 –8 NM s2485364 6 159512260 C/T 0.466 1.78 × 10 –3 (0.009) 6.05 × 10 –7 (0.004) 3.93 × 10 –9 s13297295 9 131659724 C/T 0.085 4.26 × 10 –2 (0.009) 8.39 × 10 –11 (0.01) 3.93 × 10 –11 s140929198 20 38555870 A/G 0.031 2.98 × 10 –2 (0.03) 8.47 × 10 –9 (0.01) 1.41 × 10 –9 GIMA s2185366 8 131342722 T/C 0.421 2.0 × 10 –2 (0.007) 1.52 × 10 –6 (0.004) 1.05 × 10 –7 Modules: VRM i al esponse module, BCM B-cell ac i i y module, PM pla ele module, NM neu ophil module, GIMA gene al immune module A. MAF mino allele equency Na h e al. Genome Biology (2017) 18:146 Page 4 o 15 o unsa u a ion in a y acids was he mos signi ican associa ion (me a-P alue = 7.23 × 10 –7 ). The immuno- modula o y e ec s o polyunsa u a ed a y acids a e well cha ac e ized; o example, omega-3 a y acids ha e been shown o induce cy o oxici y in in i o can- ce cell lines as well as animal models o umo inci- dence and g ow h [24, 25]. Adjus men o he associa ions be ween CCLM and me aboli e measu es o leukocy e coun s esul ed in he gain o 38 add- i ional associa ions and loss o ou (c ea inine, a io o polyunsa u a ed a y acids o o al a y acids, e y low densi y lipop o ein (VLDL) pa icle size, and CRP) exis ing associa ions (Addi ional ile 7: Table S6). Va y- ing p opo ions o leukocy e coun s can be co ela ed wi h ansc ip ion-le el a ia ion in human blood [26] bu no ac o con ound he la e 's associa ion wi h pheno ypes. I his is he case, hen adjus ing o leukocy e coun in he linea eg ession analysis can e- duce noise and hus boos s a is ical powe o de ec an associa ion, which may explain he addi ional associa- ions no ed wi h he CCLM module. CCLM had no sig- ni ican mQTLs. IDL_P L_LDL_P VLDL_D XL_HDL_P XXL_VLDL_P ACACE ACE ALA ALB APOA1 APOB APOB/APOA1 IDL_L L_LDL_L LDL_D XL_HDL_L XXL_VLDL_L BOHBUT HDL_D IDL_PL L_LDL_PL XL_HDL_PL XXL_VLDL_PL CIT CREA CRP IDL_C L_LDL_C XL_HDL_C XXL_VLDL_C DHA DHA/FA IDL_CE L_LDL_CE XL_HDL_CE XXL_VLDL_CE EST_C IDL_FC L_LDL_FC XL_HDL_FC XXL_VLDL_FC FAW3 FAW3/FA FAW6 FAW6/FA FREE_C IDL_TG L_LDL_TG XL_HDL_TG XXL_VLDL_TG GLC GLN GLOL GLY GlycA L_HDL_P M_LDL_P XL_VLDL_P HDL_C HDL_TG HDL2_C HDL3_C HIS L_HDL_L M_LDL_L XL_VLDL_L ILE L_HDL_PL M_LDL_PL L P_LDLV_LX L_HDL_C M_LDL_C C_LDLV_LX L_HDL_CE M_LDL_CE XL_VLDL_CE LA LA/FA LAC LDL_C LDL_TG LEU L_HDL_FC M_LDL_FC XL_VLDL_FC MUFA MUFA/FA L_HDL_TG M_LDL_TG XL_VLDL_TG L_VLDL_P M_HDL_P S_LDL_P L_VLDL_L M_HDL_L S_LDL_L PC PHE PUFA PUFA/FA PYR L_VLDL_PL M_HDL_PL S_LDL_PL L_VLDL_C M_HDL_C S_LDL_C REMNANT_C L_VLDL_CE M_HDL_CE S_LDL_CE SERUM_C SERUM_TG SFA SFA/FA SM L_VLDL_FC M_HDL_FC S_LDL_FC TG_PG TOT_CHO TOT_FA TOT_PG TYR L_VLDL_TG M_HDL_TG S_LDL_TG UNSAT M_VLDL_P S_HDL_P VAL VLDL_C VLDL_TG M_VLDL_L S_HDL_L M_VLDL_PL S_HDL_PL M_VLDL_C S_HDL_C M_VLDL_CE S_HDL_CE M_VLDL_FC S_HDL_FC M_VLDL_TG S_HDL_TG S_VLDL_P S_VLDL_L S_VLDL_PL S_VLDL_C S_VLDL_CE S_VLDL_FC S_VLDL_TG XS_VLDL_P XS_VLDL_L XS_VLDL_PL XS_VLDL_C XS_VLDL_CE XS_VLDL_FC XS_VLDL_TG FAW6/FAW3 LLM (124) NM (122) GIMA (98) GIMB (82) PM (56) CCLM (24) BCM (14) VRM (8) FDR adjus ed P- alue < 1.25 x 10 -04 1.25 x 10 -03 6.25 x 10 -03 > 6.25x10 -03 Signi icance h eshold 1.25 x 10 -04 Lipop o ein Subclasses HDL IDL LDL VLDL Lipop o ein pa icle size Apolipop o eins Choles e ol Glyce ides & Phospholipids Fa y Acids Small Me aboli es Amino Acids In lamma o y Ma ke s Fig. 3 Me aboli e measu e associa ions wi h immune gene modules. Ci cula hea map o associa ions be ween indi idual me aboli e measu e and he module eigengene o each module (colo ed by FDR-adjus ed P alues). Concen ic ci cles ep esen modules, wi h numbe s in pa en heses deno ing o al numbe o me aboli e measu es associa ed wi h ha module a FDR-adjus ed P alue <6.25 × 10 –3 .NM neu ophil module, LLM lipid leukocy e module, GIMA/GIMB gene al immune module A/B, PM pla ele module, CCLM cy o oxic cell-like module, BCM B-cell ac i i y module, VRM i al esponse module. See Addi ional ile 1: Table S1 o ull me aboli e desc ip ions Na h e al. Genome Biology (2017) 18:146 Page 5 o 15 Vi al esponse module Th ee genome-wide signi ican mQTLs we e iden i ied o he i al esponse module (VRM; Fig. 2a; Table 2). The s onges mQTL, s182710579 (me a-P alue = 9.22 × 10 –9 ), is wi hin a known lincRNA locus (RP11- 608O21.1) (Addi ional ile 4: Figu e S2a). Rs182710579 was a ans eQTL o h ee genes in he VRM (Fig. 2b; Addi ional ile 9: Table S8). The s onges associa ion was seen wi h CCL2 (me a-P alue = 6.78 × 10 –12 ), a p o- in lamma o y chemokine in ol ed in leukocy e ec ui - men du ing i al in ec ion [27, 28]. Also, adipocy e- de i ed CCL2 is known o play an impo an ole in obesi y-associa ed adipose issue in lamma ion and insulin esis ance [29]. The nex s onges mQTL, s151234502, esides wi hin in on 4 o he ela i ely un- s udied ZNF212, pa o a zinc inge gene clus e a 7q36 (Addi ional ile 4: Figu e S2b). Rs151234502 modu- la ed exp ession o 11 VRM genes in ans (Fig. 2b; Addi ional ile 9: Table S8). The s onges associa ion was wi h OAS2 (me a-P alue = 8.98 × 10 –10 ), an in e e on-induced gene encoding an enzyme p omo ing RNase L-media ed clea age o i al and cellula RNA [30]. The hi d mQTL, s147742798, was an in e genic SNP loca ed be ween SHANK2 and DHCR7 a 11q13.4 (Addi ional ile 4: Figu e S2c). Rs147742798 was a ans eQTL o wo genes in he VRM, BST2 and PARP9 (Fig. 2b; Addi ional ile 9: Table S8). BST2 encodes a ans-memb ane p o ein wi h in e e on-inducible an i- i al unc ion [31]. S udies ha e p e iously shown induc- ion o a y acid biosyn hesis by a ange o i uses [32]. VRM was associa ed wi h eigh me aboli e measu es, in- cluding amino acids (alanine, phenylalanine), a y acids (omega-6 a y acids, polyunsa u a ed a y acids, sa u- a ed a y acids, and o al a y acids), and choles e ol es e s in medium VLDL (Fig. 3; Addi ional ile 8: Table S7). Consis en wi h i s pu a i e ole in i al esponse, VRM was s ongly associa ed wi h CRP (me a P alue = 2.38 × 10 –10 ). B-cell ac i i y module The B-cell ac i i y module (BCM) was associa ed wi h 14 me aboli e measu es including CRP, his idine, lac a e, apolipop o iens, and mainly he medium high-densi y lipop o ein (HDL) subclass o lipop o eins (Fig. 3; Addi ional ile 8: Table S7). The s onges associa ion was seen wi h CRP (me a-P alue = 2.65 × 10 –8 ). His i- dine was he second s onges associa ion. This is in e - es ing gi en ha his idine is a subs a e o his amine, and bo h his amine elease and B-cell ac i i y a e cen al pa s o an alle gic eac ion. While no mQTLs o BCM eached genome-wide signi icance, he e was some e i- dence in he YFS o he MHC class I locus (Fig. 2a and Table 2). The op signal was loca ed be ween HLA-B/C and MICA ( s2523489, me a-P alue = 6.27 × 10 –8 ; Addi ional ile 4: Figu e S3). The HLA class I egion is well known o be associa ed wi h au oimmune diseases, whe e he ole o B cells is well ecognized. Rs2523489 was a ans eQTL o CD79B (me a-P alue = 1.16 × 10 –9 ), a sub- uni o he an igen-binding B-cell ecep o complex [33]. Pla ele module PM had he s onges mQTL o any gene module, an in- onic SNP o he ARHGEF3 gene a 3p14.3 ( s1354034; me a-P alue = 7.35 × 10 –28 , Fig. 2a; Table 2; Addi ional ile 4: Figu e S4a). ARHGEF3 encodes a Rho guanine nu- cleo ide exchange ac o , a ca alys o Rho GTPase con- e sion om inac i e GDP-bound o ac i e GTP-bound o m. Rs1354034 was an eQTL o he majo i y o genes in he PM, all o which we e in ans. An in e genic SNP, s2836773 (me a-P alue = 5.4 × 10 –8 ), a he HLA locus was also iden i ied as an mQTL o PM (Add- i ional ile 4: Figu e S4b). The ARHGEF3 mQTL ( s1354034) exhibi ed a s ong ans- egula o y e ec and was associa ed wi h 61 PM genes (65 unique p obes) (Fig. 2c; Addi ional ile 9: Table S8). The op ans eQTL was ITGB3 (me a-P alue = 5.09 × 10 –42 ), a gene encod- ing he β 3 subuni o he he e odime ic in eg in ecep o (in eg in α IIb β 3 ). This in eg in ecep o is mos highly exp essed on ac i a ed pla ele s and plays a key ole in media ing pla ele adhesion and agg ega ion upon bind- ing o ib inogen and Willeb and ac o [34, 35]. Ou da a a e consis en wi h p e ious obse a ions o he di- e se ans eQTL e ec s o s1354034 [23], including he pu a i e splice-QTL e ec s o s1354034 on TPM4, a signi ican eGene in he PM. ARHGEF3 i sel is o in ense in e es o pla ele biol- ogy. I has p e iously been shown ha silencing o ARH- GEF3 in zeb a ish p e en s h ombocy e o ma ion [36]. To es whe he ARHGEF3 exp ession had an e ec on PM genes, we eg essed ou ARHGEF3 le els and e- an he eQTL analysis. Adjus ing o ARHGEF3 did no a - enua e he ans-associa ions o s1354034, sugges ing ei he independence o downs eam unc ion o ARH- GEF3 and s1354034 o pos - ansc ip ional modi ica- ion o ARHGEF3. P e ious GWAS s udies ha e shown s1354034 is associa ed wi h pla ele coun and mean pla ele olume [36]; howe e , pe haps due o powe , we ound no signi ican ela ionship be ween pla ele coun s and s1354034 in YFS. While pla ele coun s we e posi- i ely associa ed wi h he PM (β= 0.29; P alue = 8.23 × 10 –30 ; Addi ional ile 6: Table S5), he associa ion be- ween s1354034 and he PM was s ill highly signi ican when condi ioning on pla ele coun s (β=−0.33; P alue = 1.40 × 10 –17 ). PM displayed di e se me abolic in e ac ions and was associa ed wi h 55 me aboli e measu es, la gely comp is- ing o lipop o ein subclasses and a y acids, as well as CRP (Fig. 3; Addi ional ile 8: Table S7). Choles e ol Na h e al. Genome Biology (2017) 18:146 Page 6 o 15 es e s in small HDL pa icles we e mos s ongly associ- a ed wi h he PM (me a-P alue = 9.45 × 10 –20 ). HDL has been shown o exhibi an i h ombo ic p ope ies by modula ing pla ele ac i a ion and agg ega ion and he coagula ion pa hway [37]. Also, a ious LDL subclasses o lipop o eins we e associa ed wi h he PM, which is consis en wi h ou unde s anding ha LDL in luences pla ele ac i i y. Fo example, LDL has been shown o in luence pla ele ac i i y ei he by enhancing pla ele e- sponsi eness o agg ega ing s imuli o by inducing ag- g ega ion [38, 39]. Mo eo e , LDL-speci ic binding si es on pla ele s ha e also been epo ed [40, 41]. As no ed abo e, he PM was associa ed wi h pla ele coun s, and adjus men o pla ele coun s in he YFS esul ed in a - enua ion o app oxima ely hal o he weakes associa- ions be ween PM and me aboli e measu es; howe e , he s onges we e main ained (Addi ional ile 7: Table S6). Associa ion wi h VLDL pa icle size and h ee o he s we e gained ollowing he adjus men (Addi ional ile 7: Table S6). Neu ophil module Th ee loci we e iden i ied as mQTLs o he neu ophil module (NM; Fig. 2a and Table 2). The op mQTL was in onic o LRRC8A a 9q34.11 ( s13297295; me a-P alue = 3.93 × 10 –11 ; Addi ional ile 4: Figu e S5a). LRRC8A encodes a ans-memb ane p o ein shown o play a ole in B- and T-cell de elopmen and T cell unc- ion [42, 43]. Two addi ional in e genic mQTLs we e lo- ca ed a he TAGAP locus a 6q25.3 ( s2485364; me a-P alue = 3.93 × 10 –9 ) and a 20q12 ( s140929198; me a-P alue = 1.41 × 10 –9 ) (Addi ional ile 4: Figu e S5b, c). Rs13297295 was a s ong ans egula o o NM and was an eQTL o eigh NM genes ( en unique p obes), in pa icula he majo alpha de ensins (DEFA1-DEFA4), he genes o highes cen ali y in he module (Fig. 2d; Addi ional ile 9: Table S8). Rs13297295 was a cis-eQTL o ano he co e NM gene, LCN2 (pe mu ed me a-P alue = 1 × 10 –4 ) (Fig. 2d; Addi ional ile 9: Table S8). LCN2 is exp essed in neu ophils and inducible by TLR ac i a ion, ac ing as an an imic obial agen ia seques- a ion o bac e ial side opho es o p e en i on up ake [44–46]. LCN2’s ole in acu e phase esponse appea s o be ela ed o ca dio ascula diseases, such as hea ail- u e [47]. A he TAGAP locus, s2485364 was a ans- eQTL o eigh NM genes ( en p obes) and was also a s ong d i e o LCN2 (me a-P alue = 9.11 × 10 –17 ) (Fig. 2d and Addi ional ile 9: Table S8). Consis en wi h ou indings, neu ophils om LCN2-de icien mice ha e been shown o ha e impai ed chemo axis and phagocy ic capabili y and inc eased suscep ibili y o bac e ial and yeas in ec ions compa ed o wild ype [48, 49]. This sugges s a possible unc ional ole o TAGAP a ian s in egula ing neu ophil mig a ion h ough LCN2. NM was associa ed wi h 121 ci cula ing me aboli e measu es (~76% o all me aboli e measu es analyzed) as well as CRP (Fig. 3; Addi ional ile 8: Table S7). The s onges is he p e iously epo ed associa ion wi h in- lamma o y bioma ke GlycA (me a-P alue = 2.68 × 10 – 25 ) [10]; howe e , NM’s associa ion wi h a ious lipop o- ein subclasses, pa icle sizes o lipop o eins, a y acids, choles e ol, apolipop o eins, glyce ides and phospho- lipids, amino acids, and o he small molecules indica es i has a po en ially majo ole in linking neu ophil unc- ion o me abolism. Lipid-leukocy e module Toge he wi h NM, he lipid-leukocy e module (LLM) showed ex ensi e me abolic associa ions. O e all, 123 me aboli e measu es and CRP we e associa ed wi h LLM, wi h he s onges being he a io o iglyce ides o phosphoglyce ides (me a-P alue = 5.16 × 10 –138 ; Fig. 3; Addi ional ile 8: Table S7). Wi h he inclusion o he YFS, hese indings s ongly eplica e p e ious associa- ions be ween LLM and me aboli e measu es [14] as well as de ec ing addi ional associa ions. We also con i m he p e ious s ong nega i e associa ion be ween CRP and LLM (me a-P alue = 8.16 × 10 –20 ). Consis en wi h p e- ious s udies, no mQTLs we e de ec ed o LLM. Gene al immune modules A and B No mQTLs we e associa ed wi h gene al immune mod- ules A and B (GIMA and GIMB); howe e , hese modules we e associa ed wi h 97 and 82 me aboli e measu es, e- spec i ely (Fig. 3; Addi ional ile 8: Table S7). Choles e ol es e s in small HDL and he mean diame e o VLDL pa icles exhibi ed he s onges associa ions wi h GIMA (me a-P alue = 1.56 × 10 –30 ) and GIMB (me a-P alue = 1.83 × 10 –15 ), espec i ely. The GIMA was also associa ed wi h omega-3 a y acid le els (me a-P alue=4.1×10 –8 ) and CRP (me a-P alue = 5.7 × 10 –5 ) while GIMB was no , pe haps due o he sub ly di e ence pa hway en ichmen s o each module (Table 1). O he me aboli e measu es as- socia ed wi h hese wo modules include mainly he VLDL and HDL subclass o lipop o eins and a y acids; due o hei la ge size and he e ogeneous composi ion, howe e , in e p e a ion o me abolic ela ionships o GIMA and GIMB is limi ed. Long- e m s abili y o in e ac ions be ween me aboli e measu es, immune gene modules, and mQTLs The 216 indi iduals in bo h he DILGOM 2007 and 2014 ollow-up allowed in es iga ion o he long- e m s abili y o immunome abolic and mQTL ela ionships. Ac oss his se en-yea pe iod, he eigh immune gene coexp ession ne wo ks we e s ongly p ese ed (all p ese a ion s a is ics’pe mu a ion P alues <0.001; Addi ional ile 10: Table S9). The me aboli e– Na h e al. Genome Biology (2017) 18:146 Page 7 o 15 me aboli e co ela ion s uc u e was also la gely con- sis en be ween DIGOM07 and DILGOM14 (Addi ional ile 4: Figu e S6). Nex , we examined how me aboli e in e ac ions wi h immune gene modules changed o e he 7-yea ime pe iod (“Me hods”). The LLM–me aboli e measu e asso- cia ions we e he mos consis en o e ime wi h 90 and 79 me aboli e measu es eaching signi icance in DIL- GOM07 and DILGOM14, espec i ely, o which 74 we e signi ican a bo h ime poin s (Fig. 4a; Addi ional ile 11: Table S10). The di ec ion and e ec size o LLM– me aboli e measu e associa ions we e la gely main ained (Fig. 4b). Fo he neu ophil module, he py u a e asso- cia ion was signi ican ly main ained o e ime; howe e , he e was some e idence ha o he expec ed associa- ions wi h NM we e s able o e ime, including GlycA (Addi ional ile 11: Table S10). While no associa ions wi h me aboli e measu es we e signi ican ly main ained o he pla ele module, s1354034 was a empo ally s able mQTL o PM (mQTL P alue = 4.87 × 10 –7 ). No o he mQTLs eached signi icance o empo al s abili y. While we we e powe ed o opologically eplica e im- mune modules be ween ime-poin s, powe o de ec module–me aboli e associa ions and mQTLs was s ill limi ed, wi h only he s onges associa ions eaching signi icance. Fo he la e , he e ec sizes o module as- socia ions we e gene ally consis en be ween he ime poin s (Addi ional ile 4: Figu e S7), wi h he excep ion o GIMB. Wi h he pa icula ly s ong consis ency o as- socia ions o he LLM and NM, i may be ha smalle modules, which cap u e mo e de ined ansc ip ional p og ams, a e he mos empo ally s able in e ms o hei pheno ype associa ions. Howe e , gi en he obus - ness o hese associa ions be ween independen coho s, we an icipa e ha , as long- e m omics ollow-up o popula ion-based coho s inc ease in sample size, mo e o hese disco e ed associa ions will become s a is ically signi ican o e ime. Conclusions This s udy has u ilized o e 2000 indi iduals o map he immuno-me abolic c oss alk ope a ing in ci cula ion. We ha e iden i ied and cha ac e ized eigh obus im- mune gene modules, hei gene ic con ol, and in e ac- ions wi h di e se me aboli e measu es, including many o clinical signi icance (e.g., iglyce ides, HDL, LDL, b anched-chain amino acids). Also, se e al signi ican me aboli e measu es iden i ied he e, pa icula ly b anched chain amino acids and a y acids, ha e been p e iously shown o be p edic i e o ca dio ascula e en s and he de elopmen o T2D [6, 50]. Fu he - mo e, ou indings a e consis en wi h and build upon hose o p e ious s udies. In addi ion o i e newly iden- i ied gene modules, hei mQTLs and me aboli e in e - ac ions, we ha e eplica ed he p e iously cha ac e ized LL module and con i m i s associa ion wi h lipop o ein IDL_P L_LDL_P VLDL_D XL_HDL_P XXL_VLDL_P ACACE ACE ALA ALB APOA1 APOB APOB/APOA1 IDL_L L_LDL_L LDL_D XL_HDL_L XXL_VLDL_L BOHBUT HDL_D IDL_PL L_LDL_PL XL_HDL_PL XXL_VLDL_PL CIT CREA CRP IDL_C L_LDL_C XL_HDL_C XXL_VLDL_C DHA DHA/FA IDL_CE L_LDL_CE XL_HDL_CE XXL_VLDL_CE EST_C IDL_FC L_LDL_FC XL_HDL_FC XXL_VLDL_FC FAW3 FAW3/FA FAW6 FREE_C IDL_TG L_LDL_TG XL_HDL_TG XXL_VLDL_TG GLC GLN GLOL GLY GP L_HDL_P M_LDL_P XL_VLDL_P HDL_C HDL_TG HDL2_C HDL3_C HIS L_HDL_L M_LDL_L XL_VLDL_L ILE L_HDL_PL M_LDL_PL L_HDL_C M_LDL_C L_HDL_CE M_LDL_CE XL_VLDL_CE LA LA/FA LAC LDL_C LDL_TG LEU L_HDL_FC M_LDL_FC XL_VLDL_FC MUFA MUFA/FA L_HDL_TG M_LDL_TG XL_VLDL_TG L_VLDL_P M_HDL_P S_LDL_P L_VLDL_L M_HDL_L S_LDL_L PC PHE PUFA PUFA/FA PYR L_VLDL_PL M_HDL_PL S_LDL_PL L_VLDL_C M_HDL_C S_LDL_C REMNANT_C L_VLDL_CE M_HDL_CE S_LDL_CE SERUM_C SERUM_TG SFA SFA/FA SM L_VLDL_FC M_HDL_FC S_LDL_FC TG_PG TOT_CHO TOT_FA TOT_PG TYR L_VLDL_TG M_HDL_TG S_LDL_TG UNSAT M_VLDL_P S_HDL_P VAL VLDL_C VLDL_TG M_VLDL_L S_HDL_L M_VLDL_PL S_HDL_PL M_VLDL_C S_HDL_C M_VLDL_CE S_HDL_CE M_VLDL_FC S_HDL_FC M_VLDL_TG S_HDL_TG S_VLDL_P S_VLDL_L S_VLDL_PL S_VLDL_C S_VLDL_CE S_VLDL_FC S_VLDL_TG XS_VLDL_P XS_VLDL_L XS_VLDL_PL XS_VLDL_C XS_VLDL_CE XS_VLDL_FC XS_VLDL_TG Apolipop o eins Choles e ol Small Me aboli es Amino Acids In lamma o y Ma ke s Lipop o ein pa icle size Glyce ides & Phospholipids Fa y Acids HDL IDL LDL VLDL FDR adjus ed P- alue < 1.25 x 10 -04 1.25 x 10 -03 6.25 x 10 -03 > 6.25x10 -03 Signi icance h eshold 1.25 x 10 -04 Lipop o ein Subclasses Lipid Leukocy e Module (LLM) DILGOM14 (N=79) DILGOM07 (N=90) XL_VLDL_PL C_LDL V _LX FAW6/FA −0.50 −0.25 0 0.25 0.50 −0.50 −0.25 0 0.25 0.50 PUFA/FA FAW6/FA L_HDL_FC L_HDL_FC L_HDL_CE XL_HDL_PL UnSa XL_HDL_FC XL_HDL_CE XL_HDL_C HDL2_C L_HDL_PL XL_HDL_P TG_PG S_VLDL_TG M_HDL_TG VLDL_TG M_VLDL_TG L_VLDL_PL L_VLDL_TG L_VLDL_C HDL_TG MUFA S_LDL_TG SFA XXL_VLDL_PL ApoB Remnan _C XS_VLDL_P S_VLDL_C M_VLDL_CE Be a Es ima es DILGOM07 Be a Es ima es DILGOM14 Bo h DILGOM07 and DILGOM14 DILGOM07 only DILGOM14 only No signi ican B A Signi ican me aboli e measu e associa ions wi h LLM LA/FA XS_VLDL_L L_VLDL_CE Fig. 4 Tempo ally s able me aboli e measu e associa ions wi h he LLM. aCi cula hea map o associa ion be ween each me aboli e measu e and he LLM. bCompa ison o he e ec size es ima es o me aboli e measu e associa ion wi h LLM in DILGOM07 and DILGOM14 shows ha he o e all associa ion pa e ns a e consis en ac oss he wo ime-poin s. Colo s deno e me aboli es ha a e signi ican ly associa ed wi h he LLM in DILGOM07 only (o ange), DILGOM14 only (blue), and ac oss bo h ime-poin s (g een). The g ey dashed line is he x = y line Na h e al. Genome Biology (2017) 18:146 Page 8 o 15 subclasses, lipids, a y acids, and amino acids [13, 14]. Associa ions be ween he co e genes in he LL module and isoleucine, leucine, and a ious lipids we e also iden i ied independen ly in he KORA coho [12]. Im- po an ly, we ha e shown he long- e m s abili y o LL and neu ophil module coexp ession and in e ac ions wi h me aboli e measu es, and we ha e g ea ly expanded he numbe o known bioma ke s associa ed wi h he NM om one (GlycA) o 123 [10]. Ou s udy has also expanded he widesp ead ans eQTL e ec s a he ARHGEF3 locus [23], shows hem o be s ongly main- ained wi hin indi iduals o e ime, and u he iden i ies ex ensi e in e ac ions wi h lipop o ein measu es ha may be a consequence o hese ans e ec s. Taken oge he , ou analyses illus a e he apidly g owing body o e idence in ima ely linking he immu- noin lamma o y esponse o he blood me abolome. Wi h ine - esolu ion maps o hese in e ac ions, new bioma ke s o ch onic and acu e in lamma o y s a es a e likely o eme ge. Wi h in i o and in e en ional s udies, modula ion o hese me aboli e–immune in e ac ions h ough exis ing lipid-lowe ing medica ions, gu mic obe e ec s, o die a y changes may p o ide new ways he immune sys em i sel can be u ilized o lessen he bu - den o ca diome abolic disease. Me hods S udy popula ions This s udy used da a om wo popula ion-based co- ho s, he Die a y, Li es yle, and Gene ic de e minan o Obesi y and Me abolic synd ome (DILGOM; N = 518) and he Ca dio ascula Risk in Young Finns S udy (YFS; N = 1650), which ha e been desc ibed in de ail elsewhe e [13, 51]. All subjec s en olled in hese s udies ga e w i - en in o med consen . The DILGOM s udy is a subsample o he FINRISK 2007 c oss-sec ional popula ion-based su ey, which e- c ui ed a andom sample o 10,000 indi iduals be ween 25 and 74 yea s o age, s a i ied by sex and 10-yea age g oups, om i e s udy a eas in Finland. All 6258 indi- iduals who pa icipa ed in he FINRISK 2007 baseline heal h examina ion we e in i ed o a end he DILGOM s udy (N = 5024), 630 o whom unde wen a leas one o he geno yping, ansc ip omics, o me abolomics p o iling conside ed he e. In 2014, a ollow-up s udy was conduc ed, o which 3735 indi iduals om he o iginal s udy e-pa icipa ed. Samples collec ed in 2007 and 2014 a e e e ed o as DILGOM07 and DILGOM14, espec i ely. The YFS is a longi udinal p ospec i e coho s udy ha s a ed in 1980, wi h ollow-up s udies ca ied ou e e y 3 yea s, o moni o ca dio ascula disease isk ac o s in child en and adolescen s om i e majo egions o Finland (Helsinki, Kuopio, Tu ku, Oulu, and Tampe e). A o al o 3596 child en and adolescen s in age g oups 3, 6, 9, 12, 15, and 18 yea s pa icipa ed in he baseline s udy; hese child en we e andomly selec ed om he na ional public egis e and hei de ails a e desc ibed in [51]. In his cu en s udy, da a collec ed om he 2011 ollow-up s udy (pa icipan s aged 34, 37, 40, 43, 46, and 49 yea s) we e analyzed. Sample collec ion Venous blood was collec ed ollowing an o e nigh as in all h ee s udies. Samples we e cen i uged and he esul ing plasma and se um samples we e aliquo ed in o sepa a e ubes and s o ed a −70 °C o analyses. P o o- cols o he blood sampling, physiological measu e- men s, and clinical su ey ques ions we e simila ac oss he YFS and DILGOM s udies and a e desc ibed ex en- si ely in [13, 52]. Geno yping and impu a ion Whole blood genomic DNA ob ained om bo h coho s was geno yped using he Illumina 610-Quad SNP a ay o DILGOM07 (N = 555) [13] and a cus om gene a ed 670 K Illumina BeadChip a ay o YFS (N = 2443) [53]. The 670 K a ay sha es 562,643 SNPs wi h he 610-quad a ay. The 670 K a ay emo es poo ly pe o ming SNPs om he 610-quad a ay and imp o es copy numbe a ia ion co e age [53]. Geno ype calling was pe o med wi h he Illuminus clus e ing algo i hm [54]. Quali y con ol was as p e iously desc ibed in [13] and [53] o DILGOM and YFS, espec i ely. Geno ypes we e im- pu ed o he 1000 Genomes Phase 1 e sion 3 e e ence panel using IMPUTE2 in bo h DILGOM and YFS [17]. Poo ly impu ed SNPs based on low call- a e (<0.90 o DILGOM, <0.95 o YFS), low-in o ma ion sco e (<0.4), mino allele equency <1%, and de ia ion om Ha dy– Weinbe g equilib ium (P<5×10 –6 ) we e hen emo ed. A o al o 7,263,701 SNPs in DILGOM and 6,721,082 in YFS passed quali y con ol, wi h 6,485,973 common be- ween he wo. A o al o N = 518 samples in DILGOM and N = 2443 samples in YFS indi iduals passed quali y con ol il e s. Me abolomics p o iling Me aboli e concen a ions o DILGOM07 (N = 4816), DILGOM14 (N = 1273), and YFS (N = 2046) we e quan i- ied om se um samples u ilizing a high- h oughpu NMR me abolomics pla o m (B ainshake L d, Helsinki, Finland) [18, 55]. De ails o he expe imen al p o ocol, in- cluding sample p epa a ion, NMR spec oscopy, and me- aboli e iden i ica ion, ha e been p e iously desc ibed in [13, 18]. A o al o 159 me aboli e measu es we e assessed, o which 148 we e di ec ly measu ed and 11 we e de i ed (Addi ional ile 1: Table S1). The 148 measu es include he cons i uen s o 14 lipop o ein subclasses (98 Na h e al. Genome Biology (2017) 18:146 Page 9 o 15