Serum Lipid Levels, Body Mass Index, and Their Role in Coronary Artery Calcification

DOI: 10.1161/CIRCGENETICS.114.000496

Serum Lipid Levels, Body Mass Index, and Their Role in Coronary Artery Calcification: A Polygenic Analysis Running title: van Setten et al.; Polygenic analysis of coronary calcification Jessica van Setten, PhD1; Ivana Išgum, PhD2*; Sonali Pechlivanis, PhD3*; Vinicius Tragante, PhD4; Pim A. de Jong, MD, PhD5; Joanna Smolonska, PhD6,7; Mathieu Platteel, BSc6, Per Hoffmann, PhD8-10; Matthijs Oudkerk, MD, PhD11; Harry J. de Koning, MD, PhD12; Markus M. Downloaded from http://circgenetics.ahajournals.org/ by guest on October 13, 2017

Nöthen, MD8,9; Susanne Moebus, PhD3; Raimund Erbel, MD13; Karl-Heinz Jöckel,, PhD3; Max 1,14 1,1 14 W. de Bakker, Bak akke ker, ke r, P PhD hD1, A. Viergever, PhD2; Willem P.Th.M. Mali, MD, PhD5; Paul I.W.

1

Department n off Medical nt Medi Me dica di call Genetics, ca Geenetics, ne ne Center for Molecular M Medicine, edicine, 2Image Sciences ed Scie iennces Institute, 4Departm ie Department m of 5 14 Cardiology, De Department Dep partment nt of Ra R Radiology, diol o oggy, y De Department epa p rtment of Ep Epidemiology, pidemio io olo ogy g , Julius J lius C Ju Center enter fo for or Health S Sciences c encc and ci Institute Instit tut utee fo forr Me Med Medical dicaa Primary Care, Medical Center Utrecht, Utrecht, The ry C ry are, University Universi s ty yM ed dic ical al C en nte t r Ut U trech ht,, Ut treecht, t, T h Netherlands; he Nethherllan nds; 3In 13 Clinic lin nic ooff Card Cardiology, rdiolo ogy gy, W West-German esst--Germ man H Heart eart C Centre, entree, University Uniivee Informatics, Biometry Epidemiology, cs, B cs, iometry and dE pid demio olo ogy, Cl s Es sen, E sen, se n, G erma maany ny;; 6D Hospital Essen, Essen, Germany; Department epaart r me ment n off Ge nt Genetics, eneti tics cs,, 7De cs Department epa p rt rtme m nt me n of of Ep Epid Epidemiology, id dem e io olo logy gyy, University Unniv verssi sity y Medical M Institute nst stitute of Human Genetics, Genee Center Groningen, University Groningen, r roningen, Universsit ity y of Gro rooni ningen, Groningen, n, The Netherlands; Netthe h rlands; 8In 9 Department n of Genomics, Life nt Life & B Brain rain ra in Ce Cent Center, n er nt er,, University Univ Un iver iv ersi sity si ty of B Bonn, onnn, n Bo Bonn Bonn, nn n, Ge G Germany; erm rman rm a y; 10Division of Medical M y Hospi p tal and Depa p rtment of Biomedicine,, University y of Basel,, Basel,, Switzerla a Genetics,, University Hospital Department Switzerland; 11 Department ntt of Radiology-Radiodiagnostics, Radiollog gy-Ra y-Radi yRaadi diod odiagn od gn gnos nosti ostics tiics cs,, Un Univ University niver iver ersi sity sity yM Medical ed edic di al C Center ente en ter te er Gr Gron Groningen, onin on in nge gen, n, U University nive ni vers ve rsity of Gro rs Groningen, Gron on ngen; n gen; ge n; 12De Groningen; Department Depa part rtme ment nt ooff Pu Publ Public blic ic H Health Health, ealt ea lthh E Erasmus rasm ra smus us M Medical edic ed ical al C Center, Center ente en terr Ro Rotterdam Rott Rotterdam, tter erda dam m T The he Netherland N Netherlands ethe et herl rlan andd *contributed equally

Correspondence: Paul de Bakker, PhD University Medical Center Utrecht Stratenum 1.305, P.O. Box 85090 3508 AB Utrecht The Netherlands Tel: +31-88-7550406 Fax: +31-88-7555410 E-mail: [email protected]

Journal Subject Codes: [89] Genetics of cardiovascular disease, [135] Risk factors, [58] Computerized tomography and magnetic resonance imaging 1

DOI: 10.1161/CIRCGENETICS.114.000496

Abstract: Background - Coronary artery calcification (CAC) is widely regarded as a cumulative lifetime measure of atherosclerosis but it remains unclear what is the relation between calcification and traditional risk factors for coronary artery disease (CAD) and myocardial infarction (MI). This study characterizes the genetic architecture of CAC by evaluating the overall impact of common alleles associated with CAD/MI and its traditional risk factors. Methods and Results - Based on summary association results from the CARDIoGRAMplusC4D study of CAD/MI, we calculated polygenic risk scores in 2,599 participants of the NELSON Downloaded from http://circgenetics.ahajournals.org/ by guest on October 13, 2017

Study, in whom quantitative CAC levels (Agatston scores) were determined from chest CT imaging data. The most significant polygenic model explained almost 14% 14% of tthe he oobserved bseerve bs v d CAC variance (P = 1.6 x 10-11), which points to a residual effect due to many ny as yyet et uunknown nkno nk nown no wn lloci oc that overlap between weeen CA wee CAD/ CAD/MI D/MI D/ MII and CAC. In addition, w wee cconstructed onstructed risk sscores cores based on publi published i atio at ons for traditional trad adit ad ittio onaal cardiovascular card ca rdio rd iova io vasccullar risk riisk factors facto ors aand nd te ttested sted st e th ed hese see sco co ore ress for foor as asso s c SNP associations these scores association with CAC. We found found nominally nom minalllyy significant sign gnificcan a t aassociations ssoociatio ionss for for ge genetic enet etic ri risk iskk scores scoores of sc o LD LDL-rol, ro l, andd bbody odyy ma od ass s index, inddex e , which wh were w ree suc we ucce cess sssfu fullyy replicatedd in 22,182 cholesterol,, total cholestero cholesterol, mass successfully individuals of the Heinz Nixdorf Niixddorff Recall R calll S Re Study. tudy d . dy Conclusions ns - Pe Perv Pervasive rvas asiv i e po iv ppolygenic poly lyge ly yggeni niic sh shar sharing arin ing in g be bbetween twee tw eenn CA CAC C an andd CA C CAD/MI D/MI D/ MI ssuggests ugge ug gggest stts th that at a substantial fraction of the heritable risk for CAD/MI is mediated through arterial calcification. We also provide evidence that genetic variants associated with serum lipid levels and body mass index influence CAC levels.

Key words: coronary artery calcification, genome-wide analysis, risk score, coronary artery disease, myocardial infarction

2

DOI: 10.1161/CIRCGENETICS.114.000496

Introduction Over the last few years, a number of collaborative genome-wide association studies (GWAS) have identified many loci associated with coronary artery disease (CAD) and myocardial infarction (MI).1, 2 The CARDIoGRAMplusC4D Consortium reported 46 loci at genome-wide significance in an association analysis of as many as 64,000 cases and 131,000 controls, where the lead variants at these loci collectively explain 10.6% of CAD heritability.2 In parallel, several studies have also focused on the genetic basis of known CAD risk factors, including circulating Downloaded from http://circgenetics.ahajournals.org/ by guest on October 13, 2017

lipid levels,3, 4 hypertension,5-7 type 2 diabetes (T2D),8 body mass index ex (BMI),9, 10 and arterial arte ar t calcification,11-13 altogether pinpointing hundreds of loci across the genome. nome. A An n iimportant mp porta tant ta nt ow is is to what wha hat extent ex genetiic vvariants ariants overlap aacross cross traits and whic question now the identified genetic which biological mechanisms mecchanisms ar are re shar shared. red d. his study stu tudy tu dy we focus f cu fo cuss on coronary coron on nary ar rte tery y cal calcification allci cifi ifi f caati tionn ((CAC), CA AC), C) a st C) sstrong roong n andd indepen iindependent ndep nd depen e In this artery 14-18 4-18 18 risk factor for f cardiovascul cardiovascular lar events.14 T ddate, To ate,, onl only ly two lo lloci cii ((CDKN2A/CDKN2B CDKN CD KN2A KN 2A//CD 2A C KN2B and

PHACTR1)) hha have a e bbeen consistentl consistently istentll associated iatedd with ith ith h CAC CAC att genome-wide genome iide de significance ignifi ifi iin n th three h independent studies, and these loci are also linked to CAD and MI risk.2, 11-13 Going beyond these two bona fide loci, we and others have demonstrated a significant concordance in direction of effect for 25 SNPs associated with CAD/MI11, 13 (identified by the initial CARDIoGRAM Study1). It was subsequently suggested that there is strong etiological overlap between vascular calcification and cardiovascular events, even though the associated SNPs discovered to date explain only a modest fraction of the heritability of CAC, CAD or MI.19 Here, we test the hypothesis that CAD/MI SNPs associated below genome-wide significance influence the degree of calcification in the coronary arteries beyond the overlapping associations already identified. Since power is limited to detect small effects for single SNPs, we

3

DOI: 10.1161/CIRCGENETICS.114.000496

evaluated the collective – polygenic – impact of SNPs published by the CARDIoGRAMplusC4D consortium2 on CAC. We also investigated whether SNPs associated with serum lipid levels, type 1 diabetes (T1D), type 2 diabetes (T2D), height, BMI, and blood pressure have a measurable influence on CAC levels.

Materials and Methods Cohort characteristics Downloaded from http://circgenetics.ahajournals.org/ by guest on October 13, 2017

Details on sample collection, genotyping and measurement of phenotypes yp was described elsewhere.13 In brief, the Dutch and Belgian Lung Cancer Screening Trial (NELSON (NEL (N ELSO EL SON SO N trial) tria tr iaal) was designed too studyy the th he early e rlly detection of lung cancer iin ea n an at-risk popu population. ula lation. The study was approved by y th the he Ministry ry of of Health Heeal a th of of the the Netherlands, N th Ne her erlannds, and and written wriitt tteen informed inf nfoorme meed consent cons co n en ns entt was wa obtained from om al om aalll pa part participants. rttic i ip ipan an ants nts ts. Lo L Low-dose, w-do wdoose se, no nonnon-ECG n--EC E G sy synchronized, ync nchr hrron oniz izzed ed,, non-contrast n n--co no cont ntra nt rast ra st enhanced enhhan a ce cedd baseline chest available participants. Wee us computer-aided e CTs were av est avai a laabl ai b e fo fforr al alll pa part rtic rt iccip ipan an nts ts.. W used ed d a co omp mput u err-ai ut a ded detection system for aautomatic utom ut omat atic ic identification ide dent nttif ific icat atio tionn and and qu qquantification quan anti tifi fica cati tion ti on of C CAC. AC..20 Sc AC Scores Scor ores es were wer eree ma manu manually nual ally ly iinspected insp nsp ns and corrected when needed. CAC burden was expressed in terms of Agatston scores.21 All individuals were male smokers or former smokers. Genotype data and SNP imputation Genome-wide SNP genotype data was collected in 3,082 participants on the Illumina Human610-Quad BeadChip, and quality control was performed to remove low-quality SNPs and samples.13 After extensive quality control of the data (including principal components analysis), we kept 2,599 samples for all downstream analyses. We performed imputation of untyped SNPs with Minimac,22 splitting all samples into random batches of approximately 500 individuals. As reference panel, we used the 998 phased haplotypes from the Genome of the Netherlands Project release 4 encompassing 19,763,454 SNPs.23 4

DOI: 10.1161/CIRCGENETICS.114.000496

-0ikkkkkkAssociation testing framework We used a linear regression model to test genetic risk scores for association to the logtransformed Agatston scores (ln(Agatston score + 1)), including as covariates the first principal component of the genotype data (which was the only statistically significant component at univariate P < 0.05), age and smoking history (in pack years). The baseline model included only these three covariates. In the “3 SNP” model, we included the risk alleles of three SNPs (rs4977574 at 9p21, rs3825807 at ADAMTS7, rs12526453 at PHACTR1) as independent terms in Downloaded from http://circgenetics.ahajournals.org/ by guest on October 13, 2017

the regression model. These three SNPs reach genome-wide significance analysis nce in a combined an ana of the CHARGE consortium,11 the Heinz Nixdorf Recall Study,12 and the NELSON NEL ELSO SON SO N stud study. udyy.113 The ud m del e iincludes nclu nc lu ude d s th tthee risk alleles of all 45 C A /MI associated AD ed d SNPs identified by tthe “45 SNP” mod model CAD/MI CARDIoGRAMplusC4D RAM AMplusC4D DS Study tuddy2 ass iindependent ndeepend nden nt tterms erm ms in n the the regression reg greessio on mo model. Of Of th the he 46 igin ig inal in nal a ly rreported epor ep orte t d by CARDIoGRAMplusC4D, te CAR ARDIoG oG GRA RAMp M lusC Mp lusC sC4D 4D, one 4D onee SNP, SNP SN P, rs6903956 rs6690 9039 39566 att C6 39 C 6orff associationss orig originally C6orf105, was not included, c cluded, , because th this his S SNP NP was identified ide d ntif iffied in in the thhe Han C Chinese h nese ppopulation hi oppullation24 and hass not been replicated ated tedd iin nE Europeans ropeans th thus h s ffar far. a S Statistical tatisti i icall software soft oft f are R 33.0.2 0 2 was as used sed ed d for f the th h analysis. anal all sis i To calculate the explained variance of the 3 CAC SNPs and 45 CAD/MI SNPs, we subtracted the variance explained by the baseline model from that explained by the “3 SNP” and “45 SNP” models, respectively. Polygenic risk scores for CAD/MI We calculated polygenic risk scores for all individuals in the NELSON trial using publicly available association results for 79,128 SNPs from the CARDIoGRAMplusC4D consortium.2 We extracted all SNPs that were present in both the CARDIoGRAMplusC4D data and our NELSON GWAS data. After removing A/T and C/G SNPs we used PLINK25 to prune remaining SNPs based on LD, preferentially keeping SNPs with lower P-values in the

5

DOI: 10.1161/CIRCGENETICS.114.000496

CARDIoGRAMplusC4D results and leaving no pairs of SNPs with r2 > 0.05. We calculated multiple polygenic scores based on SNPs with P-values reported by the CARDIoGRAMplusC4D consortium2 that reached a predefined threshold (