doi: 10.1111/j.1529-8817.2005.00210.x

Association of Common CRP Gene Variants with CRPLevels and Cardiovascular Events

D. T. Miller1,3,∗, R. Y. L. Zee2,3,∗, J. Suk Danik2,3, P. Kozlowski1,3, D. I. Chasman2,3, R. Lazarus4,N. R. Cook2,3, P. M. Ridker2,3 and D. J. Kwiatkowski1,3,§

1Division of Hematology, Brigham and Women’s Hospital, Boston, MA2Division of Preventive Medicine and Division of Cardiovascular Disease Prevention, Brigham and Women’s Hospital, Boston, MA3Donald W. Reynolds Cardiovascular Clinical Research Center on Atherosclerosis at Harvard Medical School, Boston, MA4Channing Laboratory, Brigham and Women’s Hospital, Boston, MA

Summary

C-reactive protein (CRP) is a well-documented marker of atherosclerotic cardiovascular disease risk. We resequencedCRP to identify a comprehensive set of common SNP variants, then studied and replicated their association withbaseline CRP level among apparently healthy subjects in the Women’s Health Study (WHS; n = 717), PravastatinInflammation/CRP Evaluation trial (PRINCE; n = 1,110) and Physicians’ Health Study (PHS; n = 509) cohorts.The minor alleles of four SNPs were consistently associated in all three cohorts with higher CRP, while theminor alleles of two SNPs were associated with lower CRP (p < 0.05 for each). Single marker and haplotypeanalysis in all three cohorts were consistent with functional roles for the 5′-flanking triallelic SNP −286C>T>Aand the 3′-UTR SNP 1846G>A. None of the SNPs associated with higher CRP were associated with risk ofincident myocardial infarction (MI) or ischemic stroke in a prospective, nested case-control study design from thePHS cohort (610 case-control pairs). One SNP, −717A>G, was unrelated to CRP levels but associated withdecreased risk of MI (p = 0.001). Taken together, these data imply significant interactions between both geneticand environmental contributions to the increased CRP levels that predict a greater risk of future atherothromboticevents in epidemiological studies.

Keywords Association Study, Atherosclerosis, C-reactive protein, CRP, Haplotype analysis, Myocardialinfarction.

Introduction

Atherosclerosis is a common high-impact disease influ-enced by genetic and environmental factors, and there-fore a prototypical complex human disease (Boerwinkleet al. 1996; Frohlich et al. 2001). In numerous prospec-tive studies, elevated baseline C-reactive protein (CRP)levels have been associated with a higher incidence offuture cardiovascular events at all levels of LDL choles-

∗These authors contributed equally to this work.§Corresponding author: David J. Kwiatkowski, MD, Ph.D., Divi-sion of Hematology, Brigham and Women’s Hospital, 75 FrancisSt., CHRB 6, Boston, MA 02115. Tel: (617) 355-9005; Fax:(617) 355-9016; E-mail: dk@rics.bwh.harvard.edu

terol. For example, in both the Physician’s Health Study(PHS) and Women’s Health Study (WHS), baseline lev-els of CRP were associated with a two-fold increase inrisk of future vascular events, after adjustment for all tra-ditional cardiovascular risk factors (Ridker et al. 1997,2002). Other large case-control studies have confirmedthe association between CRP and cardiovascular events(Danesh et al. 2004).

CRP may act as a marker of inflammation, or mayplay a direct role in atherogenesis, but demonstrating adirect effect of CRP has been difficult, as outlined ina recent report (Clapp et al. 2005). This 110kDa pen-traxin protein appears to be an important componentof the innate immune system. CRP promotes agglu-tination, bacterial capsular swelling, phagocytosis, and

C© University College London 2005 Annals of Human Genetics (2005) 69,623–638 623

D. T. Miller et al.

complement fixation (Pepys & Hirschfield, 2003;Volanakis, 2001). CRP levels are influenced by in-flammatory cytokine levels, and interleukin-6 (IL6) isthe primary inducer and regulator of CRP expression(Vickers et al. 2002; Volanakis, 2001). Levels of CRPmay increase by as much as two or three orders of mag-nitude during the acute phase of infection or injuryresponse (Gabay & Kushner, 1999; Volanakis, 2001).For this reason, CRP levels used for cardiovascular riskprediction are ideally assessed during periods of relativehealth, and are thus referred to as baseline levels.

CRP levels are a continuous trait, demonstrating alog-normal distribution with a mean of about 1.6 mg/Lwith a wide variation (Koenig et al. 1999). The Centersfor Disease Control and the American Heart Associationhave published guidelines to facilitate the application ofCRP levels in the determination of cardiovascular risk,such that levels <1, 1 to 3, and >3 mg/L correspond tolower, moderate, and higher risk (Ridker, 2003). How-ever, CRP levels predict future vascular events acrossthe full range of levels seen in clinical practice (Ridker& Cook, 2004). Baseline CRP levels are influenced bya number of factors, including age, smoking status, BMIand ethnicity (Ford et al. 2003a, 2003b, 2004). They arealso a heritable genetic trait, with heritability estimates(hr

2) of 0.40 and 0.39 in two large studies (Pankow et al.2001; Vickers et al. 2002), suggesting that dissection ofgenetic factors regulating CRP levels is possible.

Genetic variation at the CRP locus has been shownto be associated with CRP levels in prior associationstudies. However, these studies did not assess geneticvariation within CRP in a comprehensive manner, nordid they assess large numbers of individuals in a repli-cation study design to ensure confidence in the find-ings. Three studies have reported no association between−717A>G and baseline CRP (Brull et al. 2003; Chenet al. 2004; Obisesan et al. 2004). −286C>T>A andIVS1+29A>T were reported to have no associationwith baseline CRP in one prior study each (Obisesanet al. 2004; Russell et al. 2004), but each have been asso-ciated with higher baseline CRP in more recent studies(Kovacs et al. 2005; Suk et al. 2005; Szalai et al. 2005).Subjects homozygous for a 20 repeat allele of an in-tronic GT-microsatellite had elevated CRP levels in oneprevious study (Szalai et al. 2002). 1059G>C was asso-ciated with lower baseline CRP in three prior studies

(Russell et al. 2004; Suk et al. 2005; Zee & Ridker,2002), but was not associated with CRP in three otherstudies (Brull et al. 2003; de Maat et al. 2004; Obis-esan et al. 2004). 1444C>T minor allele homozygosityhas been associated with higher baseline CRP (Brullet al. 2003) and higher acute phase CRP (D’Aiuto et al.2005). The minor allele of 1846G>A in the 3′-UTRhas been associated with lower baseline CRP (Russellet al. 2004).

Limited data is available regarding association of CRPSNPs and cardiovascular events. The minor allele of−717A>G was associated with decreased risk of coro-nary heart disease in a population of Han Chinese (Chenet al. 2004), though not associated with CRP levels inthe same population. Possibly as a result of ethnic dif-ferences, SNP discovery in that study failed to identifymany of the common CRP variants observed in otherpopulations.

Here, we report resequencing analysis to comprehen-sively identify variation within the CRP locus, followedby genotyping to examine the influence of such varia-tion on CRP levels in three independent populations.Several CRP SNPs are associated with higher or lowerCRP values in these individuals, and haplotype analy-sis permits tentative identification of two SNPs whichappear functional in this association. We also examinethe influence of genetic variation at CRP SNPs on therisk of future atherosclerotic cardiovascular events, andfind that CRP genetic variation which is associated withhigher CRP levels is not associated with atherothrom-botic events.

Materials and Methods

Subjects

We evaluated the role of CRP SNPs and haplotypesas determinants of plasma CRP levels in three distinctpopulations, the Women’s Health Study (WHS), thePravastatin Inflammation/CRP Evaluation (PRINCE),and the Physician’s Health Study (PHS). In each pop-ulation sample participants were excluded from anal-ysis if they used post-menopausal hormone replace-ment therapy, as each of these is known to affectbasal CRP levels (Cushman et al. 1999; Ridker et al.1999). Descriptions of each of these three cohorts

624 Annals of Human Genetics (2005) 69,623–638 C© University College London 2005

CRP SNP Associations

Table 1 Baseline Characteristics of WHS, PRINCE, and PHS subjects.∗

WHS PHS

Characteristic Low CRP High CRP p PRINCE Controls Cases p

Number of subjects n = 359 n = 358 n = 1,110 n = 640 n = 640Age at enrollment 51.0 ± 6.0 54.8 ± 7.7 <0.0001 60.6 ± 8.5 60.9 ± 8.5 m.v.Male (%) n/a n/a 72.3% 100% 100%Female (%) 100% 100% 27.7% n/a n/aCurrent smoker (%) 8.1% 17.0% 0.0003 14.8% 16.7% 16.7% m.v.History of diabetes (%) 0 0 8.9% 3.3% 8.6% <0.0001Body mass 21.9 ± 2.4 31.5 ± 6.3 <0.0001 29.2 ± 5.3 24.8 ± 2.7 25.4 ± 3.1 <0.001

index (kg/m2)Systolic blood n/a n/a 130.2 ± 15.6 128.6 ± 12.2 132.5 ± 13.5 <0.0001

pressure - mm HgDiastolic blood n/a n/a 80.1 ± 9.2 79.7 ± 7.2 81.8 ± 7.6 <0.0001

pressure - mm HgaSystolic BP < 140 97.8% 78.2% <0.0001 n/a n/a n/aaDiastolic BP <90 98.6% 87.4% <0.0001 n/a n/a n/aTotal serum 194.4 ± 37.4 215.5 ± 43.8 <0.0001 229.8 ± 32.3 n/a n/a

cholesterol (mg/dL)LDL cholesterol (mg.dL) 114.6 ± 31.3 131.4 ± 35.8 <0.0001 143.0 ± 25.7 n/a n/abHyperlipidemia >220mg/dL 15.0% 22.9% <0.001Baseline CRP (mg/L)∗ ∗ 0.15 [0.11-0.18] 7.5 [6.2-20] <0.0001 1.6 [0.78, 3.4] 1.2 [0.6, 2.2] 1.6 [0.8, 2.7] <0.0001

∗Plus-minus values are means ±SD. CRP values are presented as median with 25th & 75th percentiles. Differences between WHS highor low CRP groups or between PHS matched cases and controls (m.v. = matching variable), are indicated by a p value generated byχ 2 for categorical variables, t-test for continuous variables, and Wilcoxon Rank sum test for CRP.aIn the WHS, absence of DM was a selection criterion, BP measures (self-reported) are available only in categorical format.bIn PHS, cholesterol values were recorded categorically without lipid fractions.

are summarized in Table 1, and have been publishedpreviously (Albert et al. 2001; Ridker et al. 1997, 2002).As described below, we also evaluated the role of CRPSNPs and haplotypes as determinants of incident my-ocardial infarction and ischemic stroke within one ofthese cohorts, the PHS.

The WHS cohort comprises an ongoing trial of ini-tially healthy U.S. women who were randomized to as-pirin and vitamin E in a two-by-two factorial designand followed over an average period of 10.7 years forincident cardiovascular events. Of these women, nearly28,000 provided baseline blood samples which had pre-viously been measured for CRP. For the current analy-sis, CRP SNP genotyping in the WHS was undertakenusing an extreme phenotype approach among 359 Cau-casian women selected for very low CRP levels (CRP0.03 - 0.20 mg/L), and 358 Caucasian women selectedfor high CRP levels (CRP> 5.0 mg/L).

The PRINCE cohort was a multi-centre,community-based trial of the effects of statintherapy on CRP levels, that included a group of 1,702individuals without known coronary artery disease

(primary prevention cohort). For the current analysis,all of these individuals underwent CRP measurementas well as CRP SNP genotyping. Of these subjects,1,437 were Caucasians who were used for studies oflinkage disequilibrium. Of the Caucasians, 1,110 werefree of HRT exposure and were used for studies ofCRP association.

Finally, the PHS comprises an ongoing cohort ofmore than 22,000 initially healthy US men who wererandomly assigned to aspirin or beta-carotene and fol-lowed over a 14 year period for the occurrence of firstever vascular events. For the current analysis we per-formed a prospective, nested case-control study geno-typing each CRP SNP among 610 PHS participantswho subsequently developed either myocardial infarc-tion (n = 346) or thromboembolic stroke (n = 264)during the follow-up period (cases), and among an equalnumber of study participants who remained free of in-cident cardiovascular disease during follow-up and werematched to the case participants on the basis of age,smoking status, and time to randomization (controls).An additional 86 unmatched controls, making a total

C© University College London 2005 Annals of Human Genetics (2005) 69,623–638 625

D. T. Miller et al.

of 696 controls, were available for association testing.Of the control subjects 509 also had CRP levels ascer-tained and thus served as a third reference group for thecross-sectional analyses.

All human subject research in this study was ap-proved by the Institutional Review Board of Brighamand Women’s Hospital, Boston, MA. In all studies CRPlevels were ascertained by validated high-sensitivityassays.

Resequencing Analysis

Resequencing was performed on a 3.5kb contiguoussegment of the CRP gene, starting 1000bp upstream ofthe transcription initiation site and extending 450bp 3′

of exon 2. Primers were designed to provide a minimum180bp overlap between amplicons. Sequencing wasperformed using standard ABI PRISMTM BigDyeTM

Terminator (Applied Biosystems), with electrophore-sis on an Applied Biosystems 3730 Sequence Analyzer.Automated reads were analyzed using Phred, Phrap,Polyphred, and Consed bioinformatics tools (Ewing &Green, 1998; Ewing et al. 1998). Analysis of proteinstructure-function for the rare missense SNP identifiedin exon 2 was performed according to published meth-ods (Lau & Chasman, 2004).

Genotyping

SNPs were genotyped by either Sequenom® matrix-assisted laser desorption and ionization-time of flight(MALDI-TOF) mass spectrometry or ABI PRISM®

TaqMan® assays, according to standard protocols (Leush-ner & Chiu, 2000; Livak et al. 1995). A small fraction ofthese SNPs have been genotyped in previously reportedstudies: IVS1+29A>T and 1059G>C in PRINCE byTaqman (Suk et al. 2005), and 1059G>C in PHS (Zee& Ridker, 2002) by previously described methods (Cao& Hegele, 2000). Mean completion rates were 96% forWHS; 99% for PRINCE; and 96% for PHS. Qualitycontrol assays were performed on all sets of samples byblinded repeat genotyping of 5% of samples. In everycase the discordance rate was ≤1%. Genotyping of theCRP dinucleotide repeat was performed by PCR usinga fluorescently labelled primer, with size separation ofamplification products on an Applied Biosystems 3730

Sequence Analyzer, and comparison to the size markerGeneScan 500 Liz (Applied Biosystems).

Statistical Analysis and Haplotype Inference

Although these three cohorts contain ethnically di-verse subjects, and non-Caucasians were included inthe sample selection for genotyping, there were insuf-ficient numbers of each non-Caucasian group to per-form haplotype inference and association testing, withthe exception of limited SNP association testing amongPRINCE African-Americans, as presented in the Dis-cussion. The WHS, PRINCE, and PHS cohorts are94.0%, 85.7%, and >94% Caucasian, respectively. Se-lecting only Caucasians to avoid confounding from eth-nic differences resulted in the following group sizes forassociation testing: WHS (n = 717), PRINCE (n =1,110), and PHS control (n = 696 where 509 hadCRP levels measured). Calculation of allele frequencies,Hardy-Weinberg Equilibrium (HWE) using a χ 2-test,and pairwise linkage disequilibrium using the r2 and D′

measures were performed in SAS/Genetics® v8.2 and9.0 (SAS Institute Inc., Cary, NC).

SNP genotypes of WHS subjects (n = 717) withlow CRP (0.03 − 0.20 mg/L; n = 359) and highCRP (>5.0 mg/L; n = 358) were examined by lo-gistic regression to permit calculation of odds ratios forthe effect of minor alleles on CRP, both unadjusted andadjusted for BMI, age at randomization, and smokingstatus. In PRINCE (n = 1,110) and PHS controls (n =509), tests of individual SNP associations with CRPwere performed with linear regression with an additivemodel of log-transformed baseline CRP against indi-vidual SNP genotypes, unadjusted and adjusted for age,BMI, smoking status and gender (for PRINCE). Re-gressions were performed within the R software systemVersion 2.0.1 (Copyright 2004, The R Foundation forStatistical Computing). Genotype frequencies and me-dian CRP were determined for each SNP genotype inthe PRINCE and PHS. The effect of genotype on CRPin the PRINCE was also displayed by boxplots, becausePRINCE includes samples from both men and women,and across the full range of baseline CRP. Tests for alleleand genotype frequency differences for the CRP dinu-cleotide repeat between subjects with low or high CRPwere performed as a χ 2-test and Armitage trend test.

626 Annals of Human Genetics (2005) 69,623–638 C© University College London 2005

CRP SNP Associations

Haplotypes in all three cohorts were inferred onlyfor Caucasians with complete genotyping data using thehaplo.em expectation-maximization algorithm allowingfor incorporation of covariate data (Schaid et al. 2002).Only confidently inferred haplotypes (posterior proba-bility >0.95) with frequency >0.01 in all cohorts wereused, leading to a slightly smaller sample size as com-pared to the SNP association analysis (WHS, n = 550;PRINCE, n = 1,071; PHS, n = 446). Tests of haplo-type association were performed in haplo.score (Schaidet al. 2002) to test all haplotypes simultaneously, and thenby regression analysis, both unadjusted and adjusted asdescribed for SNP association tests, in the PRINCE andPHS to provide an estimate of the proportion of variancein CRP levels explained by each haplotype. Haplotyperegression used H1 (T-G-C-A-G-C-G) as a reference,because it is common and consistently not associatedwith CRP levels in haplo.score.

In PHS case-control pairs (n = 610), odds ratios,95% confidence intervals, and p values are given forthe association of genotypes, in the additive mode, withcombined incidence of MI and stroke, and each eventconsidered separately, in a conditional logistic regres-sion analysis. Cases and controls were matched by con-ditioning upon the matching by age, smoking status,and length of follow-up since randomization, and fur-ther controlling for randomized treatment assignment,history of hypertension, presence or absence of diabetesand BMI. In addition, the relationship between hap-lotypes and MI and/or ischemic stroke was examinedusing a haplotype-based conditional logistic regressionanalysis with stepwise selection procedure conditionalon age and smoking, with further adjustments for BMI,hypertension, diabetes, and randomized aspirin/beta-carotene treatment assignment after haplotype inferenceby haplo.em as described.

Results

CRP Resequencing

To comprehensively ascertain both common and rarevariants in the CRP gene that might influence CRPlevels, 192 individuals from the Women’s Health Study(WHS) population were resequenced. These subjectshad extreme discordant baseline CRP levels: WHS

Group 1 (0-5th percentile; 0.03 - 0.20 mg/L; n = 48),WHS Group 2 (86-96th percentile; 5.0 - 10.0 mg/L;n = 48), and WHS Group 3 (>98th percentile; 19.85- 174.9 mg/L; n = 96). Initially WHS Group 3 wasnot going to be studied, to exclude individuals whoseCRP levels might be high due to an unrecognized un-derlying inflammatory state or acute illness. However,during the course of this study newer analyses showedthat even the highest levels of baseline CRP correspondto a graded increase in cardiovascular risk (Ridker &Cook, 2004), prompting us to include the 96 subjectsfrom WHS Group 3 for resequencing and genotyping.

Seven common CRP SNPs for which the minor allelefrequency (MAF) was > 0.05 were identified, consist-ing of three 5′-flanking, one intronic, one exonic, andtwo 3′-UTR SNP (Figure 1 and Table 2A). Notably,−286C>T>A is a triallelic SNP occurring at position−286 relative to the transcriptional initiation site. Ini-tial analyses on the192 resequenced subjects indicatedthat the minor allele of 6 out of the 7 CRP SNPs wasassociated with higher or lower CRP values (data notshown), prompting further examination of these associ-ations in the WHS and other populations. An intronicmicrosatellite polymorphism was also identified by re-sequencing, as previously described (Szalai et al. 2002).

Multiple novel rare variants were also identified byresequencing, including: four single nucleotide variants

1059G>C

-286C>T>A

1846G>A IVS1+29A

1444C>T

-757T>C

-717A>G CRP (GT)n

3.5kb

Figure 1 The CRP Gene.Structure of human CRP, located at chromosome band 1q23.Human CRP consists of two exons (black/gray boxes). Thecoding regions of exons 1 and 2 (black) correspond to a 204amino acid peptide, separated by a single ∼280bp introncontaining a microsatellite repeat denoted CRP (GT)n. A short(104bp) 5′-untranslated, and relatively longer (∼1.2kb)3′-untranslated region (gray) flank the gene. Approximately1060bp of 5′-flanking and 175bp of 3′-flanking sequences wereincluded in the resequencing for SNP discovery. Common SNPsare indicated.

C© University College London 2005 Annals of Human Genetics (2005) 69,623–638 627

D. T. Miller et al.

Table 2 CRP variants identified by resequencing(A) Common CRP SNPs

dbSNP GeneVariant Originally Identified as identifier Position Alleles∗ MAF Flanking sequence

−757T>C n/a rs3093059 5′-flanking T>C 0.112 TGTACTCAAT[T/C]NGCTGAGAAA−717A>G −717G>A rs2794521 5′-flanking A>G 0.296 CAAACACCGC[A/G]TGTTCTCACT

(Brull et al. 2003)−286C>T>A CRP 1 rs3091244 5′-flanking C>T 0.306 GATGGCCACT[C/T]GTTTAATATG

(Russell et al. 2004)C>A 0.113 GATGGCCACT[C/A]GTTTAATATG

IVS1+29A>T +29T>A rs1417938 IVS1 A>T 0.296 TATGGGAGAG[A/T]TTTGATCTGA(Obisesan et al. 2004)

1059G>C 1059G/C rs1800947 Exon 2; L184L G>C 0.042 ACTTTGTGCT[G/C]TCACCAGATG(Zee and Ridker 2002)

1444C>T +1444C>T rs1130864 3′-UTR C>T 0.286 TGNTGGGAAA[C/T]GGTCCAAAAG(Brull et al. 2003)

1846G>A CRP 4 rs1205 3′-UTR G>A 0.226 GNGAGAGACT[G/A]TGAGGACAGA(Russell et al. 2004)

(B) Rare CRP variants

Rare Gene Nucleotide WHSVariant Position Position Alleles n MAF Group Flanking sequence

1 5′-flanking −912 C>T 1 0.003 1 CCTGACTC[C/T]TGCCTGAA2 5′-flanking −603 A>G 3 0.008 3 AGCATTAG[A/G]AGATATAC3 5′-flanking −424 del 1 0.003 3 TTTCCCAG[T]CTGTAAAT4 5′-flanking −217 del 1 0.003 3 AGAGATTT[C]TTCATTTT5 Exon 2 c.238 C>T (T45M) 1 0.003 3 ACCGTTAA[C/T]GAAGCCTC6 3′-UTR c.810 C>G 2 0.005 2 GTACCTCC[C/G]GGTTTTTT7 3′-UTR c.1356 G>A 1 0.003 3 AATACCAC[G/A]CAGTCCCT8 3′-UTR c.1424 T>C 2 0.005 3 AGATCTGC[T/C]TTTTTTCC9 3′-UTR c.1596 C>T 1 0.003 3 TTCATCCC[C/T]GCATTCCC

∗SNP major alleles in our study were assigned according to the allele present with respect to the “sense” orientation of the CRP gene.In some cases this changes the designation from the way it was originally reported.

each in the 5′-flanking region and 3′-untranslated re-gion, and a single nonsynonymous variant in exon 2(Table 2B). The nonsynonymous SNP causes a T45Mamino acid substitution and is predicted to be deleteri-ous to protein function according an algorithm designedto predict functional protein variants (Lau & Chasman,2004). Each rare variant was found in, at most, 3 outof 196 subjects, precluding our ability to examine theireffect on baseline CRP in the general population. Theeffects of these rare variants may be pursued in futurestudies.

CRP Genotyping

The seven common CRP SNPs were genotyped in 717WHS samples with low (0.03 - 0.20 mg/L; n = 359) or

high (>5.0 mg/L; n = 358) baseline CRP. Two addi-tional replication populations were then genotyped: thePravastatin Inflammation/CRP Evaluation (PRINCE)cohort (n = 1,702), and case-control pairs from thePhysicians’ Health Study (PHS) cohort, in which caseshad either MI or ischemic stroke (n = 610 cases and696 controls). All seven SNP were in Hardy-Weinbergequilibrium (p > 0.05 for χ 2 test) in the full set ofsamples assayed within each of the three cohorts, aswell as in Caucasians only. Minor allele frequencies forthe seven CRP SNP were similar among Caucasians inthe three populations, although significant differencesin MAF were observed between the low CRP and highCRP WHS groups (Table 3). Genotyping the CRP in-tronic GT dinucleotide polymorphism in a smaller setof WHS samples showed three common alleles among

628 Annals of Human Genetics (2005) 69,623–638 C© University College London 2005

CRP SNP Associations

Table 3 Minor allele frequencies for 7 common CRP SNPs

WHSPRINCE PHS Mean,

Low CRP High CRP Primary Controls all CohortsVariant (n = 359) (n = 358) p (n = 1,110) (n = 696) (n = 2,523)

−757T>C 0.028 0.092 <1.0 × 10−6 0.063 0.068 0.064−717A>G 0.309 0.263 0.055 0.277 0.277 0.280−286C>T 0.221 0.336 3.0 × 10−6 0.309 0.316 0.302−286C>A 0.030 0.096 1.0 × 10−6 0.064 0.070 0.065IVS1+29A>T 0.214 0.325 4.0 × 10−6 0.310 0.313 0.2991059G>C 0.112 0.049 2.4 × 10−5 0.070 0.063 0.0711444C>T 0.218 0.339 <1.0 × 10−6 0.308 0.312 0.3011846G>A 0.416 0.325 0.0005 0.337 0.326 0.344

Caucasians: GT(16), GT(20), and GT(21) with frequenciesof 0.61, 0.07, and 0.23, respectively.

Linkage Disequilbrium (LD) Patterns AmongVariants in CRP

Pairwise LD statistics were calculated for seven CRPSNPs among Caucasians in the WHS, PRINCE andPHS cohorts (Figure 2). Non-Caucasian samples wereexcluded due to known ethnic differences in LD pat-terns; the small number of these samples precluded sep-arate analysis among non-Caucasians. In Caucasians,3 out of the 7 CRP SNPs shared a pattern of pair-wise LD with −286C>T>A, the triallelic SNP (blackboxes in Figure 2). The minor allele of −757T>Cwas in LD with the A allele of −286C>T>A. Theother minor allele of −286C>T>A (T) was in LDwith IVS1+29A>T and 1444C>T. These high pair-wise LD relationships were observed in all three co-horts. −717A>G, 1059G>C, and 1846G>A showedonly weak LD by the r2 measure. None of the allelesof the CRP microsatellite showed strong LD with anyof the 7 CRP SNP (data not shown, r2 < 0.3 for eachcomparison).

Association Between CRP Variants andBaseline CRP

Association of CRP genotypes with CRP level was ex-amined with logistic regression (in WHS) and linearregression (in PRINCE and PHS) methods both beforeand after adjustment for the effects of age, smoking,and BMI. Highly significant and consistent associationswere seen between the minor allele of SNPs−757T>C,

−286C>T>A, IVS1+29A>T and 1444C>T andhigh CRP levels across populations with p values rang-ing as low as 10−7 (Table 4). Following adjustment forcovariates, similar highly significant and strong associa-tions were observed, with some modest weakening ofthe association in the WHS and strengthening of theassociation in the PRINCE. In addition, highly signifi-cant and consistent associations were also seen betweenthe minor allele of 1059G>C and 1846G>A and lowCRP levels in all three populations in both the unad-justed and adjusted analyses (Table 4). An association inthe PHS population was not seen for 1059G>C, pos-sibly due to the relatively low minor allele frequencyfor this SNP (0.06) and the smaller sample size in com-parison to the other cohorts. None of the 9 alleles atthe intronic microsatellite were associated with baselineCRP (χ 2 for trend 11.95 with 8 degrees of freedom; pfor trend 0.15).

The correlation between genotypes at the CRP SNPsand CRP levels in the PRINCE primary preventionsubjects is shown in Figure 3. Due to the discordantphenotype design, such results cannot be presented forthe WHS; the PHS control population is much smallerand includes only males.−757C>T was associated withan increase in median CRP of 0.39mg/L compared tothe major allele homozygote (p = 0.008); for −286TTand TA genotypes the increase in median CRP levelswas 0.40 and 1.03 mg/L (p = 0.009 and 0.002, respec-tively). In contrast, the 1059CC and 1846AA geno-types were associated with a reduction of median CRPlevels of 0.35 and 0.51 mg/L, respectively (p = 0.007and 0.00001, respectively). Thus, several common CRPSNPs were associated with CRP levels.

C© University College London 2005 Annals of Human Genetics (2005) 69,623–638 629

D. T. Miller et al.

0.026 0.027 0.024

-757T>C

-717A>G

-286C>T

IVS1+29A

1059G>C

1444C>T

1846G>A

-286C>A

0.026 0.032 0.024

0.154 0.176 0.160

1.000 1.000 0.975

0.027 0.027 0.024

n/a n/a n/a

0.025 0.032 0.025

0.149 0.165 0.156

0.984 0.945 0.982

0.026 0.032 0.035

0.005 0.005 0.004

0.035 0.028 0.026

0.034 0.030 0.030

0.006 0.004 0.005

0.033 0.033 0.029

0.025 0.028 0.032

0.150 0.176 0.174

0.974 0.966 0.907

0.026 0.028 0.033

0.982 0.917 0.911

0.034 0.029 0.030

0.015 0.034 0.029

0.166 0.189 0.181

0.130 0.216 0.218

0.016 0.034 0.035

0.132 0.199 0.219

0.086 0.141 0.110

0.125 0.207 0.212

-717

A>

G

-757

T>

C

-286

C>

T

-286

C>

A

IVS

1+29

A>

T

1059

G>

C

1444

C>

T

1846

G>

A

1.000 1.000 0.914

1.000 0.994 0.957

1.000 1.000 0.835

1.000 1.000 1.000

1.000 1.000 0.846

1.000 1.000 0.912

1.000 0.957 0.955

1.000 0.992 1.000

1.000 0.946 1.000

0.610 1.000 0.911

1.000 1.000 1.000

0.984 0.994 1.000

0.848 0.994 0.985

n/a n/a n/a

1.000 0.975 1.000

1.000 0.957 1.000

0.989 0.985 0.963

0.766 0.994 0.988

1.000 1.000 1.000

1.000 0.933 1.000

0.989 0.948 1.000

0.766 1.000 1.000

1.000 1.000 1.000

1.000 0.962 0.960

0.786 0.949 1.000

1.000 0.956 1.000

0.758 0.979 0.919

0.746 0.972 0.976

Figure 2 Pairwise linkage disequilibrium among CRP SNPs.LD was calculated for Caucasians from the WHS (n = 717), PRINCE Primary prevention cohort (n =1,437), and PHS controls (n = 696) using D′ (upper right) and r2 (lower left) statistics. Black boxes indicatehigh pairwise LD by r2. Within each box, the upper value is for the WHS, middle value for the PRINCE,and lower for the PHS. The minor alleles of −286C>T>A are in high pairwise LD (r2 > 0.9) with−757T>C, IVS1+29A>T, and 1444C>T (black boxes).

Haplotype Analysis in the WHS, PRINCE,and PHS Cohorts

Haplotypes formed by the 7 CRP SNPs were inferredin Caucasians from the WHS, PRINCE, and PHS pop-ulations using haplo.em (Schaid et al. 2002). The WHSand PRINCE subjects exposed to HRT, and any sub-jects with incomplete genotyping data, were excludedprior to haplotype inference. Six haplotypes, H1-H6,with a frequency of 0.01 or greater were identified inall three cohorts (Table 5).

Two of the six common haplotypes, H2 and H5,were associated with higher CRP in all three cohortsusing haplo.score (Table 6). Two others, H3 and H4,were associated with lower CRP across cohorts, withthe exception that H4 was not significantly associatedwith CRP in the PHS. The two common haplotypesassociated with higher CRP each contained a minor al-lele of the triallelic −286C>T>A. In contrast, the twocommon haplotypes associated with lower CRP eachcontained the minor allele of 1846G>A. In addition,the common H4 haplotype contained the minor allele of

630 Annals of Human Genetics (2005) 69,623–638 C© University College London 2005

CRP SNP Associations

Tab

le4

Ass

ocia

tion

ofC

RP

SNP

with

base

line

CR

P.In

WH

S,w

here

the

CR

Ptr

ait

isbi

nary

(low

vs.

high

CR

P)th

eef

fect

ofC

RP

isgi

ven

asan

odds

ratio

.In

PRIN

CE

and

PHS,

whe

reC

RP

isa

cont

inuo

ustr

ait,

geno

type

freq

uenc

ies

and

med

ian

CR

Ple

vels

are

prov

ided

for

each

geno

type

.Pva

lues

for

the

asso

ciat

ion

are

give

nun

adju

sted

and

adju

sted

for

cova

riat

esas

desc

ribe

din

the

Met

hods

.

WH

S(n

=71

7)PR

INC

E(n

=1,

110)

PHS

(n=

509)

SNP/

geno

type

OR

95%

CI

p(u

nadj

)p

(adj

)Fr

eqC

RP

p(u

nadj

)p

(adj

)Fr

eqC

RP

p(u

nadj

)p

(adj

)

−75

7T>

CT

T0.

881.

490.

871.

11T

C3.

492.

08–5

.84

2.02

×10

−6

0.01

0.12

1.73

0.00

810.

0006

0.12

1.68

0.02

90.

0012

CC

n/a

n/a

0.01

2.93

−71

7A>

GA

A0.

531.

540.

531.

30A

G0.

790.

63–1

.00

0.04

90.

170.

381.

490.

550.

770.

391.

120.

520.

94G

G0.

081.

640.

081.

43−

286C

>T

>A

CC

0.40

1.45

0.39

0.97

CT

1.84

1.45

–2.3

58.

38×

10−

70.

0002

0.39

1.50

0.00

940.

0011

0.38

1.34

0.01

80.

044

TT

0.09

1.72

0.11

1.43

TA

0.04

2.14

0.04

1.57

CA

3.83

2.27

–6.4

74.

73×

10−

70.

011

0.08

1.72

0.00

228.

2×

10−

50.

081.

740.

0004

0.00

01A

An/

an/

a0.

012.

93IV

S1+

29A

>T

AA

0.48

1.47

0.41

1.42

AT

1.75

1.37

–2.2

25.

91×

10−

60.

0005

0.43

1.56

0.05

60.

011

0.49

1.09

0.04

20.

13T

T0.

091.

690.

111.

3410

59G

>C

GG

0.86

1.58

0.87

1.28

GC

0.42

0.27

–0.6

46.

14×

10−

50.

024

0.14

1.17

0.00

660.

0015

0.12

1.03

0.31

0.27

CC

0.00

1.28

0.00

0.47

1444

C>

TC

C0.

481.

480.

491.

07C

T1.

831.

44–2

.32

6.94

×10

−7

4.2×

10−

50.

431.

540.

067

0.01

30.

421.

360.

0022

0.01

3T

T0.

091.

720.

101.

5018

46G

>A

GG

0.42

1.68

0.44

1.47

GA

0.73

0.59

–0.9

10.

0042

0.03

80.

471.

341.

26×

10−

54.

7×

10−

70.

481.

077.

75×

10−

51.

5×

10−

6

AA

0.11

1.35

0.08

0.81

C© University College London 2005 Annals of Human Genetics (2005) 69,623–638 631

D. T. Miller et al.

Table 5 Common CRP Haplotypes in the WHS, PRINCE, and PHS

WHS PRINCE PHSLow CRP High CRP Combined Primary Controls Mean, all Cohorts

Before inference (n = 359) (n = 358) (n = 717) (n = 1,110) (n = 509) (n = 2,336)After inference (n = 269) (n = 281) (n = 550) (n = 1,071) (n = 446) (n = 2,067)

# Haplotype Frequency

H1 T-G-C-A-G-C-G 0.296 0.244 0.269 0.277 0.261 0.269H2 T-A-T-T-G-T-G 0.199 0.327 0.265 0.297 0.297 0.287H3 T-A-C-A-G-C-A 0.297 0.224 0.260 0.260 0.288 0.266H4 T-A-C-A-C-C-A 0.089 0.048 0.068 0.070 0.061 0.067H5 C-A-A-A-G-C-G 0.032 0.085 0.059 0.063 0.061 0.061H6 T-A-C-A-G-C-G 0.041 0.037 0.039 0.013 0.018 0.021

Figure 3 CRP levels according to genotype at CRP SNP.Variation in CRP level by SNP genotype was determined for Caucasian subjects from the PRINCE primary prevention cohort (n =1,110). Results are presented as boxplots to describe variation in log-transformed CRP, where boxes extend from the 25 percentile to75 percentile (interquartile range). The line within each box marks the median CRP value by genotype, indicated above the line asthe non-transformed value in mg/L. Whiskers extend to 1.5 times the interquartile range of the log-transformed CRP values. Valuesbeyond the outliers are indicated by circles. Significance values were calculated by linear regression on CRP level as a function ofgenotype unadjusted for covariates. −717A>G was not associated with CRP levels, in either unadjusted or adjusted analysis, and isnot presented as a boxplot.

632 Annals of Human Genetics (2005) 69,623–638 C© University College London 2005

CRP SNP Associations

Table 6 Influence of CRP haplotypes on baseline CRP in haplo.score

WHS PRINCE PHSBefore haplotype inference: n = 717 n = 1,110 n = 509After haplotype inference: n = 550 n = 1,071 n = 446

Haplotype Score p value Score p value Score p value

H1 −1.8276 0.0676 1.092 0.275 −0.7968 0.4256H2 4.7997 <0.0001 2.118 0.034 2.5735 0.0101H3 −3.0227 0.0025 −2.990 0.003 −3.7689 0.0002H4 −2.8406 0.0045 −2.781 0.005 −0.0693 0.9448H5 3.8901 0.0001 2.910 0.004 3.3861 0.0007H6 −0.5790 0.5626 1.179 0.238 0.3733 0.7090

Global p <0.001 <0.001 <0.001Max p <0.001 0.02 0.0014

the less common 1059G>C SNP. The association be-tween −286C>T>A minor allele haplotypes and CRPlevels were highly statistically significant in all three pop-ulations, including after adjustment for age, BMI, andsmoking in Caucasians, with p values < 10−5 in simu-lation testing. The association of the 1846G>A minorallele haplotype H3 with lower CRP was also very con-sistent in the three populations. The lack of consistencyof the H4 association may partially reflect that its fre-quency is lower than H3.

CRP SNP Association with AtherothromboticEvents in the PHS Population

Since multiple epidemiologic studies have indicated thatelevated CRP levels are associated with increased ratesof atherothrombotic events – MI or ischemic stroke –we examined the possibility that CRP genetic varia-tion might also be associated with atherothromboticrisk. We examined 610 PHS subjects with a historyof MI or ischemic stroke matched by age and smok-ing status with PHS controls (n = 346 MI pairs;n = 264 ischemic stroke pairs). Although increasedCRP is known to increase the risk of atherothrom-botic events, and several CRP genotypes are associatedwith increased CRP levels, none of the minor alleles of−757T>C,−286C>T>A, IVS+29A>T, 1059G>C,or 1444C>T were associated with development of ei-ther MI or ischemic stroke (Table 7). Instead, the mi-nor allele of −717A>G, not associated with CRP lev-els in any of the populations in this study, except fora marginal association in the WHS (O.R. 0.79, 95%CI 0.63–1.00, p = 0.049), was associated with reducedrisk of atherothrombosis (any event: O.R. 0.80, 95% CI

0.66–0.97, p = 0.020; MI: O.R. 0.66; 95% CI 0.52–0.85; p = 0.001) in a stepwise conditional logistic re-gression analysis. In addition, the haplotype defined bythe minor allele of −717A>G, H1, was associated witha decreased risk of MI (O.R. 0.64; 95% CI 0.52–0.90;p = 0.002) in a haplotype-based conditional analysiswith forward-stepwise selection procedure. Even moresurprising, the 1846G>A minor allele that had beenassociated with lower baseline CRP in the PHS andPRINCE was found to be associated with an increasedrisk of atherothrombotic events in the PHS in the addi-tive mode (O.R. 1.27; 95% CI 1.04-1.55; p = 0.022).Thus, the association between CRP variants and base-line CRP did not correlate with the effects of thosevariants on clinical cardiovascular events.

Discussion

We had two main goals in this study: 1) to comprehen-sively identify genetic variation within the CRP locusand examine its potential effects on baseline CRP, and2) to investigate potential associations of such CRP vari-ants with adverse cardiovascular events. The WHS andPRINCE cohorts were chosen in pursuit of the firstgoal, while the PHS was chosen in pursuit of the firstand second goals. The WHS study examined subjectswith extreme discordant phenotypes by resequencing,with the goal of identifying both rare and common vari-ants that might be associated with CRP levels. Severalrare variants in CRP, including a previously undescribedexon 2 missense change, T45M, were identified. Cod-ing region variation affecting the amino acid sequence ofCRP has not been previously described, and an analysis

C© University College London 2005 Annals of Human Genetics (2005) 69,623–638 633

D. T. Miller et al.

Table 7 Association of CRP SNPs with clinical cardiovascular endpoints in matched pairs of PHS cases and controls; adjusted for BMI,history of hypertension, and presence or absence of diabetes.

−757T>C −717A>G −286C>T −286C>A IVS1+29A>T 10559G>C 1444C>T 1846G>A

MAF Controls 0.07 0.28 0.32 0.07 0.31 0.06 0.31 0.33(n = 696)

Cases 0.06 0.26 0.29 0.07 0.29 0.06 0.30 0.38(n = 643)

Any Event O.R. 0.97 0.8 0.97 1.34 1.01 1.02 1.02 1.2795% CI 0.67–1.40 0.66–0.97 0.78–1.21 0.72–2.51 0.83–1.23 0.71–1.48 0.85–1.24 1.04–1.55p value 0.85 0.02 0.82 0.36 0.95 0.9 0.82 0.022

MI O.R. 1.16 0.66 1.2 1.73 1.11 1.48 1.11 1.2995% CI 0.75–1.81 0.52–0.85 0.9–1.59 0.84–3.58 0.87–1.42 0.87–2.52 0.87–1.41 0.99–1.67p value 0.51 0.001 0.19 0.14 0.39 0.15 0.39 0.06

Ischemic O.R. 0.62 1.11 0.69 0.49 0.85 0.7 0.9 1.2Stroke 95% CI 0.30–1.29 0.80–1.54 0.47–1.00 0.11–2.29 0.60–1.19 0.41–1.20 0.65–1.23 0.86–1.68

p value 0.2 0.53 0.05 0.36 0.34 0.19 0.5 0.29

of protein structure-function (Lau & Chasman, 2004)suggests that the T45M change would have significanteffects on protein function. However, the low frequencyof this and the other rare variants discovered makes anal-ysis of their association with CRP levels difficult.

This comprehensive resequencing and genotypinganalysis of CRP provides definitive information on thecommon genetic variants of CRP, augmenting severalmore limited analyses that have been published recently(Brull et al. 2003; Chen et al. 2004; D’Aiuto et al. 2005;Kovacs et al. 2005; Obisesan et al. 2004; Russell et al.2004; Suk et al. 2005; Zee & Ridker, 2002). Our studyalso adds information to the CRP resequencing per-formed in the Seattle SNPs PGA (SeattleSNPs - ac-cessed April, 2004) by looking at larger numbers of sub-jects and extreme phenotypes to identify rare variants.In addition, the initial analysis of CRP SNP allele fre-quencies in our resequenced extreme phenotype popu-lations (data not shown) immediately identified associa-tions with CRP levels, permitting three levels of repli-cation of these findings, one based on additional subjectswithin the same WHS population, and two within thePRINCE and PHS control populations. The associa-tions between the minor alleles of 5 of these CRP SNPsand higher or lower CRP levels were remarkably con-sistent among the three sets of genotyped individuals(Table 4).

The minor alleles of −757T>C, IVS1+29A>T,1444C>T, and both minor alleles of −286C>T>Awere all strongly associated with higher CRP levels

in all four analyses. Linkage disequilibrium and hap-lotype analyses suggest that −286C>T>A is the func-tional SNP among these, and that the associations seenwith the other SNPs are due to strong LD betweentheir minor alleles and one or the other minor alleleof −286C>T>A. There are similar effects and statis-tical significance in the association between the mi-nor allele of −757T>C and CRP, and the A alleleof −286C>T>A and CRP, in all three populations(Table 4). This observation correlates with the near per-fect LD relationship between these two alleles (Figure 2,r2 0.975-1.000). Similar observations can be made inregard to the association of each of the minor alleles ofIVS1+29A>T and 1444C>T, as well as the T alleleof −286C>T>A, with high CRP levels (Table 4). Inaddition, there is high LD among each of these minoralleles (Figure 2, r2≥0.91). Haplotype analysis indicatesthat two common haplotypes, H2 and H5, are associatedwith higher CRP in all three cohorts. H2 consists of theminor alleles of IVS1+29A>T and 1444C>T and theT allele of −286C>T>A. H5 consists of the minor al-lele of −757T>C and the A allele of −286C>T>A.Although it is possible that multiple CRP SNPs haveindependent effects on baseline CRP levels, possibly inan additive fashion, the similar results for single SNPand haplotype associations of −286C>T>A and CRPlevel lead us to favour the more parsimonious expla-nation that −286C>T>A is the functional SNP andthat both of the minor alleles lead to high baselineCRP levels through a transcriptional mechanism. In

634 Annals of Human Genetics (2005) 69,623–638 C© University College London 2005

CRP SNP Associations

addition, our resequencing analysis eliminated the pos-sibility that some other common variant, within CRPand in LD with −286C>T>A, is responsible for theassociation of −286C>T>A and CRP level. Indeed,recent functional studies support the conclusion that−286C>T>A is a functional SNP that modulates tran-scription factor binding (Szalai et al. 2005).

The minor alleles of two CRP SNPs, 1059G>C and1846G>A, were associated with lower baseline CRPlevels (Table 4). These findings are similar to those pre-viously reported in smaller series (Brull et al. 2003; Chenet al. 2004; Kovacs et al. 2005; Obisesan et al. 2004; Rus-sell et al. 2004; Suk et al. 2005; Zee & Ridker, 2002).Because of different minor allele frequencies, there isonly weak LD between these SNP as measured by r2,suggesting the possibility that each SNP contributes tothe effect. Haplotype analysis indicated that two haplo-types, H3 and H4, were strongly associated with lowerCRP levels in the WHS and PRINCE populations.However, the association of H4 with lower CRP wasnot seen in the PHS control population, possibly due toits much smaller size or random chance. H3 contains theminor allele of 1846G>A only, whereas H4 contains theminor allele of both 1059G>C and 1846G>A, consis-tent with the finding of high LD between 1059G>Cand 1846G>A as measured by D′ , a measure less sensi-tive to differences in MAF (D′ = 0.758 in WHS, 0.979in PRINCE, and 0.919 in PHS). Although it is pos-sible that both SNPs influence CRP values, we favourthe simpler explanation that 1846G>A influences CRPmRNA stability, thereby affecting CRP levels, and thatthe association of 1059G>C with CRP levels is dueto its inclusion on a relatively common haplotype withthe minor allele of 1846G>A. A similar hypothesis forthe effect of 1846G>A was made recently (Russell et al.2004); further study will be required to explore thispossibility.

Haplotype-based linear regression was also used toinvestigate the proportion of variance in baseline CRPexplained by CRP genotype. In the unadjusted analysis,the proportion of variance (adjusted R2 value) in log-transformed CRP levels explained by CRP haplotypeswas 0.021 in the PRINCE, and 0.049 in the PHS. Al-though overall these are small effects, they must be con-sidered relative to the many factors that likely influenceCRP levels, and are relatively large for a genetic effect in

a complex disease setting. Nonetheless, the proportionof variance due to age, BMI, and smoking – about 21%and 20% in the PRINCE and PHS – was significantlyhigher than that attributable to variation in the CRPgene – about 2% and 5% in the PRINCE and PHS,respectively. Moreover, the proportion of variance ex-plained by these CRP SNPs is small relative to the calcu-lated heritability of baseline CRP, estimated at 0.40 and0.39 in two large studies (Pankow et al. 2001; Vickerset al. 2002), suggesting the presence of additional, as yetunidentified, genetic variants that substantially influenceCRP levels.

One concern about any cohort-based study is howwell the results can be extended to the general popu-lation. The replication of these findings in three inde-pendent cohorts provides high confidence in such gen-eralization. However, although these cohorts containsubjects representing multiple ethnic groups, only theCaucasian subset provides enough subjects to performstatistically powerful analyses. The ability to general-ize these findings to other ethnic groups will dependupon the allele frequencies and patterns of LD in theother groups. In an effort to investigate these associa-tions in another ethnic group, we repeated the linearregression of ln(CRP) as a function of CRP genotypesin African-Americans from the PRINCE primary pre-vention cohort (n = 99). Too few African-Americanswere available from the other cohorts to perform similaranalyses. In this small sample, −286C>T was associatedwith higher CRP in the crude analysis (p = 0.048) and1846G>A with lower CRP (p = 0.050). This limitedanalysis supports the idea that the associations persist inother ethnic groups, but further investigation and repli-cation are warranted. They also highlight the likely im-portance of −286C>T>A and 1846G>A in drivingeffects on CRP levels.

For investigation of the relationship between CRPvariants and cardiovascular endpoints, we studied 610PHS case-control pairs for MI or ischemic stroke. Noneof the minor alleles of the CRP SNPs were associ-ated with risk of atherothrombotic events in the di-rection predicted by the association of SNP genotypewith CRP level. Instead, the minor allele of −717A>Gwas associated with reduced risk of atherothromboticevents in this analysis (p = 0.020 for any event),particularly for myocardial infarction (p = 0.001,

C© University College London 2005 Annals of Human Genetics (2005) 69,623–638 635

D. T. Miller et al.

Table 7). This relationship could also be observed inthe association between haplotype H1, bearing the mi-nor allele of −717A>G only, with reduced risk of MI(O.R. 0.67, 95% CI 0.51–0.88, p = 0.004). Further-more, the minor allele of 1846G>A, associated withlower baseline CRP and possibly a functional variantin CRP, was weakly associated with an increased riskof atherosclerotic events in the PHS population (O.R.1.27, 95% CI 1.04-1.55, p = 0.022 for any event).

Our results for −717A>G and cardiovascular eventsreplicate a recently reported association between thepresence of the −717A>G minor allele and decreasedrisk of coronary heart disease (Chen et al. 2004). Theother study reported a significant increase in coronaryheart disease among Han Chinese carrying the majorallele of −717A>G, with no effect on baseline CRP.That study was retrospective and used a less stringentclinical endpoint, which included coronary revascular-ization in addition to MI and stroke. Replication of thisassociation in an independent prospective cohort withdifferent ethnicity and a more clear clinical endpointsuggests a true association of −717A>G with cardio-vascular events, albeit one that cannot be explained onthe basis of altered CRP levels. Other studies have alsofound paradoxical associations between genotypes thatpredict increased levels of adverse serum markers butdecreased risk of cardiovascular events (Keavney et al.2004).

Conclusions

In summary, we clearly define several CRP SNPsstrongly associated with CRP levels, either positively ornegatively, in three separate cohorts of initially healthymen and women. Our prospective nested case-controlanalysis within the PHS identified an association be-tween the minor allele of −717A>G and decreased riskof cardiovascular events. Further studies will be neededto explore the mechanism for this association, as wewere not able to directionally link CRP SNPs that af-fect CRP levels to the risk of future atherothromboticevents. CRP SNPs explain a much smaller proportionof the variance in CRP (2–5%), as compared to envi-ronmental exposures and lifestyle factors as well as othergenetic loci, and thus the ability of these SNPs to in-fluence atherothrombotic risk may be obscured by the

strong effects of these other factors in aggregate. Whenother genetic loci that affect CRP levels are discovered,it will be of interest to assess their potential influence oncardiovascular risk.

Acknowledgements

The authors wish to thank all the subjects from WHS,PRINCE, and PHS for their participation in these prospec-tive cohort studies. We are indebted to many who con-tributed their technical expertise to this work, including: Al-ison Brown, Jessica Gould, Mel Hernandez, Mei Lin, SalMazza, Maura Regan, and Lynda Rose. We would like tothank our colleagues Nan Laird and Benjamin Raby forhelpful discussions. This work was supported by the Don-ald W. Reynolds Center for Cardiovascular Clinical Researchon Atherosclerosis at Harvard Medical School. Dr. Millerreceived fellowship support from the Donald W. ReynoldsFoundation (Las Vegas, NV) and NIH (1 F32 HL78274-01).The Women’s Health Study and the Physicians Health Studyare supported by grants HL43851, HL63293, and HL58755from the National Heart, Lung, and Blood Institute (Bethesda,Maryland) while the PRINCE cohort was initially fundedby Bristol Myers Squibb. Dr Ridker receives additional sup-port from the Leducq Foundation (Paris, FR), and a Distin-guished Clinical Scientist Award from the Doris Duke Char-itable Foundation (New York, NY).

Abbreviations

BMI – body mass indexCRP – C-reactive proteinHRT – hormone replacement therapyIVS – intervening sequenceLD – linkage disequilibriumLDL – low-density lipoproteinMAF – minor allele frequencyMI – myocardial infarctionNHLBI – National Heart Lung and Blood InstitutePCR – Polymerase chain reactionPGA – Programs for Genomic ApplicationsPHS – Physician’s Health StudyPRINCE – Pravastatin Inflammation/CRP EvaluationUTR – untranslated regionWHS – Women’s Health Study

References

Albert, M. A., Staggers, J., Chew, P. & Ridker, P. M. (2001)The pravastatin inflammation CRP evaluation (PRINCE):rationale and design. Am Heart J 141, 893–898.

636 Annals of Human Genetics (2005) 69,623–638 C© University College London 2005

CRP SNP Associations

Boerwinkle, E., Ellsworth, D. L., Hallman, D. M. & Bid-dinger, A. (1996) Genetic analysis of atherosclerosis: a re-search paradigm for the common chronic diseases. Hum MolGenet 5, 1405–1410.

Brull, D. J., Serrano, N., Zito, F., Jones, L., Montgomery, H.E., Rumley, A., Sharma, P., Lowe, G. D., World, M. J.,Humphries, S. E. & Hingorani, A. D. (2003) Human CRPGene Polymorphism Influences CRP Levels: Implicationsfor the Prediction and Pathogenesis of Coronary Heart Dis-ease. Arterioscler Thromb Vasc Biol 23, 2063–2069.

Calabro, P., Willerson, J. T. & Yeh, E. T. (2003) Inflammatorycytokines stimulated C-reactive protein production by hu-man coronary artery smooth muscle cells. Circulation 108,1930–1932.

Cao, H. & Hegele, R. A. (2000) Human C-reactive protein(CRP) 1059G>C polymorphism. J Hum Genet 45, 100–101.

Carlson, C. S., Aldred, S. F., Lee, P. K., Trocy, R. P., Schwartz,S. M., Rieder, M., Liu, K., Williams, O. D., Iribarren,C., Lewis, E. C., Fornage, M., Boerwinkle, E., Gross, M.,Jaquish, C., Nickerson, D. A., Myers, R. M., Siscovick, D.S. & Reiner, A. P. (2005) Polymorphisms within the C-Reactive Protein (CRP) Promoter Region Are Associatedwith Plasma CRP Levels. Am J Hum Genet 77, 64–77.

Chen, J., Zhao, J., Huang, J., Su, S., Qiang, B. & Gu, D. (2004)−717A>G polymorphism of human C-reactive proteingene associated with coronary heart disease in ethnic HanChinese: the Beijing atherosclerosis study. J Mol Med 28,28.

Clapp, B., Hirschfield, G., Storry, C., Gallimore, J., Stid-will, R., Singer, M., Deanfield, J., MacAllister, R., Pepys,M., Vallance, P. & Hingorani, A. (2005) Inflammation andendothelial function: direct vascular effects of human C-reactive protein on nitric oxide bioavailability. Circulation111, 1530–1536.

Cushman, M., Legault, C., Barrett-Connor, E., Stefan-ick, M. L., Kessler, C., Judd, H. L., Sakkinen, P. A. &Tracy, R. P. (1999) Effect of postmenopausal hormones oninflammation-sensitive proteins: the Postmenopausal Estro-gen/Progestin Interventions (PEPI) Study. Circulation 100,717–722.

D’Aiuto, F., Casas, J. P., Shah, T., Humphries, S. E., Hin-gorani, A. D. & Tonetti, M. S. (2005) C-reactive protein(+1444C>T) polymorphism influences CRP responsefollowing a moderate inflammatory stimulus. Atherosclero-sis 179, 413–417. Epub 2004 Dec 18.

Danesh, J., Wheeler, J. G., Hirschfield, G. M., Eda, S., Eiriks-dottir, G., Rumley, A., Lowe, G. D., Pepys, M. B. & Gud-nason, V. (2004) C-reactive protein and other circulatingmarkers of inflammation in the prediction of coronary heartdisease. N Engl J Med 350, 1387–1397.

de Maat, M. P., Bladbjerg, E. M., Hjelmborg, J. B., Bathum,L., Jespersen, J. & Christensen, K. (2004) Genetic influence

on inflammation variables in the elderly. Arterioscler ThrombVasc Biol 24, 2168–2173. Epub 2004 Sep 02.

Ewing, B. & Green, P. (1998) Base-calling of automated se-quencer traces using phred. II. Error probabilities. GenomeRes 8, 186–194.

Ewing, B., Hillier, L., Wendl, M. C. & Green, P. (1998) Base-calling of automated sequencer traces using phred. I. Ac-curacy assessment. Genome Res 8, 175–185.

Ford, E. S., Giles, W. H., Mokdad, A. H. & Myers, G. L.(2004) Distribution and correlates of C-reactive proteinconcentrations among adult US women. Clin Chem 50,574–581.

Ford, E. S., Giles, W. H., Myers, G. L. & Mannino, D.M. (2003a) Population distribution of high-sensitivity C-reactive protein among US men: findings from NationalHealth and Nutrition Examination Survey 1999-2000. ClinChem 49, 686–690

Ford, E. S., Giles, W. H., Myers, G. L., Rifai, N., Ridker, P.M. & Mannino, D. M. (2003b) C-reactive protein concen-tration distribution among US children and young adults:findings from the National Health and Nutrition Exami-nation Survey, 1999-2000. Clin Chem 49, 1353–1357.

Frohlich, J., Dobiasova, M., Lear, S. & Lee, K. W. (2001) Therole of risk factors in the development of atherosclerosis.Crit Rev Clin Lab Sci 38, 401–440.

Gabay, C. & Kushner, I. (1999) Acute-phase proteins andother systemic responses to inflammation. N Engl J Med340, 448–454.

Keavney, B., Palmer, A., Parish, S., Clark, S., Youngman, L.,Danesh, J., McKenzie, C., Delepine, M., Lathrop, M., Peto,R. & Collins, R. (2004) Lipid-related genes and myocardialinfarction in 4685 cases and 3460 controls: discrepancies be-tween genotype, blood lipid concentrations, and coronarydisease risk. Int J Epidemiol 33, 1002–1013. Epub 2004 Jul15.

Koenig, W., Sund, M., Frohlich, M., Fischer, H. G., Lowel,H., Doring, A., Hutchinson, W. L. & Pepys, M. B. (1999)C-Reactive protein, a sensitive marker of inflammation,predicts future risk of coronary heart disease in initiallyhealthy middle-aged men: results from the MONICA(Monitoring Trends and Determinants in CardiovascularDisease) Augsburg Cohort Study, 1984 to 1992. Circulation99, 237–242.

Kovacs, A., Green, F., Hansson, L. O., Lundman, P., Samne-gard, A., Boquist, S., Ericsson, C. G., Watkins, H., Ham-sten, A. & Tornvall, P. (2005) A novel common singlenucleotide polymorphism in the promoter region of theC-reactive protein gene associated with the plasma con-centration of C-reactive protein. Atherosclerosis 178, 193–198.

Lau, A. Y. & Chasman, D. I. (2004) Functional classificationof proteins and protein variants. Proc Natl Acad Sci U S A101, 6576–6581.

C© University College London 2005 Annals of Human Genetics (2005) 69,623–638 637

D. T. Miller et al.

Leushner, J. & Chiu, N. H. (2000) Automated mass spectrom-etry: a revolutionary technology for clinical diagnostics. MolDiagn 5, 341–348.

Li, S. P., Liu, T. Y. & Goldman, N. D. (1990) cis-acting el-ements responsible for interleukin-6 inducible C-reactiveprotein gene expression. J Biol Chem 265, 4136–4142.

Livak, K. J., Flood, S. J., Marmaro, J., Giusti, W. & Deetz, K.(1995) Oligonucleotides with fluorescent dyes at oppositeends provide a quenched probe system useful for detectingPCR product and nucleic acid hybridization. PCR MethodsAppl 4, 357–362.

Obisesan, T. O., Leeuwenburgh, C., Phillips, T., Ferrell, R. E.,Phares, D. A., Prior, S. J. & Hagberg, J. M. (2004) C-reactiveprotein genotypes affect baseline, but not exercise training-induced changes, in C-reactive protein levels. ArteriosclerThromb Vasc Biol 24, 1874–1879.

Pankow, J. S., Folsom, A. R., Cushman, M., Borecki, I. B.,Hopkins, P. N., Eckfeldt, J. H. & Tracy, R. P. (2001) Fa-milial and genetic determinants of systemic markers of in-flammation: the NHLBI family heart study. Atherosclerosis154, 681–689.

Pasceri, V., Willerson, J. T. & Yeh, E. T. (2000) Direct proin-flammatory effect of C-reactive protein on human endothe-lial cells. Circulation 102, 2165–2168.

Pepys, M. B. & Hirschfield, G. M. (2003) C-reactive protein:a critical update. J Clin Invest 111, 1805–1812.

Ridker, P. M. (2003) Clinical application of C-reactive proteinfor cardiovascular disease detection and prevention. Circu-lation 107, 363–369.

Ridker, P. M. & Cook, N. (2004) Clinical Usefulness of VeryHigh and Very Low Levels of C-Reactive Protein Acrossthe Full Range of Framingham Risk Scores. Circulation 29,29.

Ridker, P. M., Cushman, M., Stampfer, M. J., Tracy, R. P.& Hennekens, C. H. (1997) Inflammation, aspirin, and therisk of cardiovascular disease in apparently healthy men. NEngl J Med 336, 973–979.

Ridker, P. M., Hennekens, C. H., Rifai, N., Buring, J. E. &Manson, J. E. (1999) Hormone replacement therapy andincreased plasma concentration of C-reactive protein. Cir-culation 100, 713–716.

Ridker, P. M., Rifai, N., Rose, L., Buring, J. E. & Cook,N. R. (2002) Comparison of C-reactive protein and low-density lipoprotein cholesterol levels in the prediction offirst cardiovascular events. N Engl J Med 347, 1557–1565.

Russell, A. I., Cunninghame Graham, D. S., Shepherd, C.,Roberton, C. A., Whittaker, J., Meeks, J., Powell, R. J.,

Isenberg, D. A., Walport, M. J. & Vyse, T. J. (2004) Poly-morphism at the C-reactive protein locus influences geneexpression and predisposes to systemic lupus erythematosus.Hum Mol Genet 13, 137–147.

Schaid, D. J., Rowland, C. M., Tines, D. E., Jacobson, R. M.& Poland, G. A. (2002) Score tests for association betweentraits and haplotypes when linkage phase is ambiguous. AmJ Hum Genet 70, 425–434.

SeattleSNPs P ( - accessed April, 2004) NHLBI Program forGenomic Applications, UW-FHCRC, Seattle, WA (URL:http://pga.gs.washington.edu).

Suk, H. J., Ridker, P. M., Cook, N. R. & Zee, R. Y. L. (2005)Relation of polymorphism within the C-reactive proteingene and plasma CRP levels. Atherosclerosis 178, 139–145.

Szalai, A. J., McCrory, M. A., Cooper, G. S., Wu, J. & Kim-berly, R. P. (2002) Association between baseline levels ofC-reactive protein (CRP) and a dinucleotide repeat poly-morphism in the intron of the CRP gene. Genes Immun 3,14–19.

Szalai, A. J., Wu, J., Lange, E. M., McCrory, M. A., Langefeld,C. D., Williams, A., Zakharkin, S. O., George, V., Allison,D. B., Cooper, G. S., Xie, F., Fan, Z., Edberg, J. C. &Kimberly, R. P. (2005) Single-nucleotide polymorphismsin the C-reactive protein (CRP) gene promoter that affecttranscription factor binding, alter transcriptional activity,and associate with differences in baseline serum CRP level.J Mol Med 19, 19.

Vickers, M. A., Green, F. R., Terry, C., Mayosi, B. M.,Julier, C., Lathrop, M., Ratcliffe, P. J., Watkins, H. C. &Keavney, B. (2002) Genotype at a promoter polymorphismof the interleukin-6 gene is associated with baseline levels ofplasma C-reactive protein. Cardiovasc Res 53, 1029–1034.

Volanakis, J. E. (2001) Human C-reactive protein: expression,structure, and function. Mol Immunol 38, 189–197.

Wingender, E., Chen, X., Fricke, E., Geffers, R., Hehl, R.,Liebich, I., Krull, M., Matys, V., Michael, H., Ohnhauser,R., Pruss, M., Schacherer, F., Thiele, S. & Urbach, S. (2001)The TRANSFAC system on gene expression regulation.Nucleic Acids Res 29, 281–283.

Zee, R. Y. & Ridker, P. M. (2002) Polymorphism in thehuman C-reactive protein (CRP) gene, plasma concentra-tions of CRP, and the risk of future arterial thrombosis.Atherosclerosis 162, 217–219.

Received: 18 April 2005Accepted: 26 May 2005

638 Annals of Human Genetics (2005) 69,623–638 C© University College London 2005