Insight into the nature of the CRP–coronary event association using Mendelian randomization

Abstract

Background It is unclear wheather the association between C-reactive protein (CRP) and incident coronary events is free from bias and confounding. Individuals homozygous for a +1444C>T polymorphism in the CRP gene have higher circulating concentrations of CRP. Since the distribution of this polymorphism occurs at random during gamete formation, its association with coronary events should not be biased or confounded.

Methods We calculated the weighted mean difference in CRP between individuals with variants of the +1444C>T polymorphism in the CRP gene among 4659 European men from six studies (genotype-intermediate phenotype studies). We used this difference together with data from previous observational studies to compute an expected odds ratio (OR) for non-fatal myocardial infarction (MI) among individuals homozygous for the T allele. We then performed four new genetic association studies (6201 European men) to obtain a summary OR for the association between the +1444C>T polymorphism and non-fatal MI (genotype-disease studies).

Results CRP was 0.68 mg/l [95% confidence interval (95% CI) 0.31–1.10; P = 0.0001] higher among subjects homozygous for the +1444-T allele, with no confounding by a range of covariates. The expected ORs among TT subjects for non-fatal MI corresponding to this difference in CRP was 1.20 (95% CI 1.07–1.38) using the Reykjavik Heart study data and 1.25 (1.09–1.43) for all observational studies to 2004. The estimate for the observed adjusted-OR for non-fatal MI among TT subjects was 1.01 (95% CI 0.74–1.38), lower than both expected ORs.

Conclusions A common CRP gene polymorphism is associated with important differences in CRP concentrations, free from confounding. The null association of this variant with coronary events suggests possible residual confounding (or reverse causation) in the CRP–coronary event association in observational studies, though the confidence limits are still compatible with a modest causal effect. Additional studies of genotype (or haplotype) and coronary events would help clarify whether or not the link between CRP and coronary events in observational studies is causal.

Prospective studies indicate a robust association, in healthy individuals, between levels of the acute phase reactant C-reactive protein (CRP) and later coronary events,^1–3 and it has been proposed that measurement of CRP may be a useful adjunct to coronary risk assessment and that CRP could be causally involved in atheroslerosis.⁴ However, CRP concentrations are correlated with smoking status, blood pressure, obesity, diabetes, physical activity, social class, low birth weight, and other products of the inflammatory response and are also higher in individuals with clinical cardiovascular disease.^5–8 Despite statistical adjustment, this association may, therefore, be subject to residual confounding or bias due to reverse causation, where the disease even in its sub-clinical state leads to elevation in the level of CRP.^5–8

A randomized controlled clinical trial of a selective intervention to reduce CRP would provide an unbiased insight into the nature of the association. Unfortunately, no such selective intervention currently exists. Although inhibitors of HMG-CoA reductase (statins) and glitazones reduce CRP,^9,10 they also have major effects on lipid profile and glucose metabolism. An alternative approach to control for confounding and reverse causality is to identify a common polymorphism in the CRP gene reliably associated with differences in circulating CRP concentration. The inheritance of such a variant should be subject to the random assortment of maternal and paternal alleles at the time of gamete formation, according to Mendel's second law.¹¹ If CRP actually increases the risk of coronary events then carriage of an allele that exposes individuals to a long-term elevation in CRP should confer an increased risk of coronary events proportional to the difference in CRP attributable to the allele. This relationship should be largely unconfounded and free of reverse causality bias.^12,13 Therefore, if non-genetic observational studies were unbiased, the increase in risk estimated from these studies should be consistent with the increase in risk conferred by carriage of the allele.^12,13 This approach, known as ‘Mendelian randomization’, has been used recently to investigate the link between homocysteine, fibrinogen, and cardiovascular disease.^14–17

CRP concentration is a heritable trait.^18,19 In three small studies we found that homozygosity for the rarer T allele of a +1444C>T polymorphism in the 3′-untranslated region of the CRP gene was associated with higher basal and/or stimulated CRP concentrations.^20,21 We have now genotyped a large number of Caucasian men to test whether the increase in risk of coronary events in individuals homozygous for this polymorphism is close to that predicted under the assumption that the CRP–coronary event association from previous non-genetic observational studies is free from residual confounding and bias due to reverse causation.

Methods

Study populations

After obtaining ethical approvals, genotyping for the CRP polymorphism and measurement of plasma CRP concentration were conducted in samples obtained from a number of cross-sectional or prospective studies, or randomized controlled trials summarized in Table 1 and in supplementary materials available online.^21–27 To test the association of +1444C>T genotype with CRP concentration, male subjects without clinically evident cardiovascular disease from six studies with available DNA and plasma samples were studied (Table 1).^{21–23,25–27} A second analysis was conducted to evaluate the association between genotype and risk of myocardial infarction (MI) in male subjects from four studies (Table 1).^22–25

Data collection

Data on sex, mean age, systolic and diastolic blood pressure, body mass index (BMI), smoking status, glucose, lipid profile, alcohol consumption, fibrinogen, and plasma CRP values were obtained from the original studies. Subjects were classified using unified definitions of hypertension, hypercholesterolaemia, type-2 diabetes mellitus, and obesity from the guidelines on primary prevention of the American Heart Association.²⁸ For the studies relating genotype and coronary events,^22–25 non-fatal MI according to the WHO criteria²⁹ was considered the primary outcome, as this end-point had been uniformly used across all studies. Analyses of genotype and CRP were limited to male subjects to preserve consistency with available studies of genotype and coronary events, which were all conducted in men.

Laboratory analyses

All studies included in the present report used high sensitivity assays to measure plasma CRP concentrations. For specific details of the assays used please see supplementary materials provided online.

CRP genotyping and examination of linkage disequilibrium

The CRP/+1444C>T single nucleotide polymorphism (SNP; rs1130864) was genotyped by PCR and RFLP analysis using primer pairs described previously²¹ and the restriction enzymes SduI or Bsp1286I that cleave the 181 bp PCR product into 23bp and 158bp fragments only in the presence of the common C allele. All DNA analysis was performed by staff unaware of the clinical status of the subjects. A public domain resequencing resource (Seattle SNPs; http://pga.mbt.washington.edu/) was accessed to investigate associations between the rs1130864 and other SNPs in the CRP gene, and to ascertain haplotype structure. MEDLINE was also searched to identify published studies reporting linkage disequilibrium (LD) between CRP SNPs and haplotype structure.

Statistical analysis

Genotype and CRP concentration

To quantify the effect of the CRP gene polymorphism on CRP concentration, we genotyped 4659 men from six studies (Table 1).^{21–23,25–27} We calculated the within-study mean difference in CRP concentration between individuals homozygous for the T allele and carriers of the C allele and then weighted each mean by the inverse of its variance to obtain an overall weighted mean difference (WMD). In the calculation of the WMD, we limited the analysis to subjects without known coronary or peripheral artery disease at the time of blood sampling in order to avoid the potential for established disease to modify the size of the genotype–CRP association.¹⁶ However, an analysis including all subjects was also conducted. Therefore, for prospective studies, baseline CRP data were used from all available subjects without clinically evident atherosclerosis. For case–control studies, genotype–CRP associations were analysed solely using control subjects. Because of its skewed distribution, CRP values were log-transformed before the analysis. For methods used to calculate the absolute geometric-WMD in CRP concentrations by genotype please see supplementary materials available at IJE online. Random and fixed effect models were used for these analyses.^30,31

Genotype and non-fatal myocardial infarction

To examine the effect of CRP genotype on risk of non-fatal MI, we genotyped 6201 men from four studies (Table 1).^22–25 We obtained the adjusted odds ratios (ORs) and 95% confidence intervals (95% CIs) for subjects homozygous for the T allele compared with carriers of the C allele from each study. In addition, for the two intervention trials^23,24 the potential interaction of the genotype–MI association with the active therapy was also evaluated. We then pooled the within-study ORs to obtain a summary adjusted-OR and 95% CI for non-fatal MI, under both fixed and random effect models. Fixed effect summary-ORs were calculated using the inverse variance-weighted method,³² and the DerSimonian and Laird Q test³³ was used to evaluate the degree of heterogeneity between studies.

Consistency between the ORs from genetic and from non-genetic observational studies

We estimated the expected OR for non-fatal MI, corresponding to the WMD in CRP between TT subjects and C allele carriers, based on data from prior non-genetic observational studies that examined the CRP–coronary event association. To do this we used information from the most recent meta-analysis of observational studies of CRP and coronary events as well as data from the Reykjavik Heart Study, a very large study in which more extensive control for potential confounders was made. The first estimate, based on an OR of 1.58 (95% CI 1.48–1.68), for the top vs bottom tertiles of the CRP distribution, came from the recent meta-analysis³⁴ of 22 prospective studies with 7068 cases with different degrees of adjustment for traditional cardiovascular risk factors across the studies included and no adjustment for regression dilution bias. The second estimate was based on an OR of 1.45 (95% CI 1.26–1.68) between top and bottom tertiles of the CRP distribution reported in the Reykjavik Heart Study,³⁴ the largest single prospective observational study of CRP, which contributed 2459 cases to the meta-analysis and also undertook a comprehensive adjustment for potential confounders (age, sex, period of recruitment, smoking, systolic blood pressure, total cholesterol level, triglyceride levels, BMI, forced expiratory volume in one second, diabetes, and socioeconomic status). The third estimate from the Reykjavik Heart Study³⁴ was based on an OR of 1.92 (95% CI 1.68–2.18) for top vs bottom tertile of CRP, based on a more limited degree of adjustment (age, sex and period of recruitment). The final estimate also from the Reykjavik study³⁴ was based on the adjusted standard 10 year follow-up risk of 1.84 (95% CI 1.49–2.28) for top vs bottom tertile of CRP.

For these estimates we assumed equivalence between relative risk and OR, that the usual mean difference in CRP between the individuals in top and bottom tertiles was 1.4 mg/l⁶, and that the CRP–coronary event relationship was log-linear. The expected OR for TT homozygous subjects with reference to C allele carriers was calculated using the formula: expected

⁠. The 95% CI for this expected OR was obtained by simulation,¹⁶ and it was compared with that observed by an interaction test.³⁵ For details, please see supplementary materials available at IJE online. Consistency between the observed genetic and expected (non-genetic) ORs would suggest that the association between CRP and coronary events identified in prior observational epidemiological studies would be unlikely to be the result of residual confounding or reverse causality bias. Data were analysed using the Review Manager software (version 4.2) from the Cochrane Collaboration 2003 and Stata 8.2 (Stata Corporation, College Station, TX, 2003).

Results

Allele and genotype frequencies, and LD with other CRP SNPs

The allele and genotype frequencies of the +1444C>T polymorphism were in Hardy–Weinberg equilibrium for all studies included in the present report, apart from a marginal distortion in the NPHS-II study (please see Table 1 in supplementary materials available at IJE online). The frequencies of the rare allele in disease-free subjects from all studies were very similar (range: 26–33%). Data from the Seattle SNPs resequencing resource (http://pga.gs.washington.edu/data/crp) indicates that four common CRP haplotypes (approximate frequencies 0.3, 0.3, 0.3, and 0.05) account for >90% of chromosomes among subjects of European descent. The +1444T allele lies on one of the three common haplotypes (approximate frequency 0.3), which also contains the T allele of a common tri-allelic polymorphism (−286C>T/A; rs3091244). Prior studies also report a strong LD between minor alleles at +1444 and −286 polymorphisms; reported D′ values in Caucasians being 0.97,³⁶ 0.91,³⁷ 0.91 (unpublished data) and with an r²-value of 0.97.³⁸ In functional studies in vitro, the −286C>T/A SNP appears to account for increased transcriptional activity in studies using promoter–reporter assays. The unique haplotype containing minor alleles at −286 and +1444 has been associated with higher CRP values in two published studies,^36,39 and we have identified concordant population haplotype frequencies and associations with CRP in our own recent studies in the NPHS dataset of ∼2700 men (unpublished data).

CRP/+1444C>T polymorphism and CRP concentrations

Using data from 4659 men from six studies, the weighted mean CRP concentration in C-allele carriers without known cardiovascular disease was 2.01 mg/l (95% CI 1.94–2.07). Under a fixed effect model, the geometric-WMD in CRP concentration between individuals homozygous for the T allele compared with carriers of the C allele was 0.68 mg/l (95% CI 0.31–1.10; P = 0.0001). There was no significant between-study heterogeneity [P-value for heterogeneity (P_Het) = 0.47; Figure 1]. When alternative models of the effect of genotype on CRP were evaluated, only the TT vs CC comparison was significant [WMD = 0.78 mg/l (0.41–1.20); P < 0.0001], while heterozygosity (CT vs CC) was not [WMD = 0.06 (−0.24 − 0.44); P = 0.66]. When subjects with coronary or peripheral atherosclerosis at the time of blood sampling were included in this analysis (n = 5658), the WMD between individuals homozygous for the T allele compared with carriers of C was very similar [0.74 mg/l (95% CI 0.37–1.10); P < 0.0001]. The values we have obtained are concordant with those obtained from other studies examining associations of CRP SNPs with CRP concentration, taking into account known LD.^36–40

To evaluate potential confounding of the genotype–CRP association, we quantified the association between TT genotype and age, systolic and diastolic blood pressure, BMI, current smoking, glucose, alcohol intake, fibrinogen, triglyceride, total and HDL cholesterol. No significant difference was observed for any of these risk factors, with the exception of a slightly higher BMI among TT subjects (mean difference = 0.34 kg/m²; P = 0.02; Table 2). In view of the multiple comparisons made, however, this significance level is not particularly extreme. Moreover, this apparent association was not significant when subjects with and without coronary heart disease were considered together, and it is likely, but not yet certain, to reflect the play of chance.

OR estimated from observational studies and the expected mean difference in plasma CRP

Expected ORs for non-fatal MI among TT homozygotes compared with C allele carriers based on a between-genotype WMD difference in CRP of 0.68 mg/l (95% CI 0.31–1.10) were 1.37 (95% CI 1.14–1.68), 1.20 (95% CI 1.07–1.38), 1.34 (95% CI 1.12–1.67), and 1.25 (95% CI 1.09–1.43) based, respectively, on the Reykjavik Heart Study minimally and maximally adjusted models, the Reykjavik study 10 year estimate, and on the meta-analysis of studies up to 2004.³⁴

OR of non-fatal MI for the CRP/+1444C>T polymorphism

Data from four new studies involving 985 subjects with non-fatal MI and 5216 control subjects were pooled to obtain a summary adjusted OR. After combining the studies under a fixed effect model, subjects homozygous for the T allele compared with C allele carriers had no significant increase in the risk of non-fatal MI [summary adjusted-OR= 1.01 (95% CI 0.74–1.38); P = 0.95] (Figure 2). No significant inter-study heterogeneity was observed (P_Het = 0.14). When alalysis of the risk of non-fatal MI conferred by homozygosity for the T allele was restricted to prospective genetic studies, there was still no significant association with non-fatal MI [OR = 0.98 (95% CI 0.70–1.38); P = 0.93]. No significant increase in risk was observed in the individual studies, and there was no significant interaction between treatment and genotype–outcome association in the two clinical trials involved in this analysis: for LEADER, P = 0.53 and WOSCOPS, P = 0.44.

Consistency between the ORs estimated from the mean difference in CRP level and the OR from the genetic study

The estimate of the observed OR from the genetic study [1.01 (95% CI 0.74–1.38)] was mathematically smaller than the expected OR [1.37 (95% CI 1.14–1.68) or 1.34 (95% CI 1.12–1.67)], calculated using data from the minimally adjusted estimate or the 10 year estimate from the Reykjavik Heart Study.³⁴ It was, however, closer to estimates based on data from the maximally adjusted model of the large Reykjavik Heart Study [1.20 (95% CI 1.07–1.38)] and to an estimate based on an updated meta-analysis of studies to 2004 [1.25 (95% CI 1.09–1.43)]. Indeed, most of the 95% confidence limits of these latter two estimates lay within the upper bound of the 95% confidence limit of the genetic estimate (Figure 3). The P-value for interaction for the comparisons of the minimally and maximally adjusted estimates from the Reykjavik study, and from an updated meta-analysis of studies to 2004 with the observed genetic estimate is outlined inFigure 3.

Discussion

Using pooled data from 4659 individuals, we found that men homozygous for the T allele of the +1444C>T polymorphism of the human CRP gene had a circulating CRP concentration 0.68 mg/l (95% CI 0.31–1.10) higher than carriers of the C allele, confirming the results of earlier smaller studies.^20,21 CRP genotype, therefore, contributes to the variation in CRP concentration observed in the population. This association probably reflects LD with a putative functional SNP in the gene promoter (−286C>T/A; rs3091244), although an effect of +1444C>T on RNA stability cannot be ruled out.

We estimated that this difference in CRP of 0.68 mg/l (95% CI 0.31–1.10) would confer an OR for non-fatal MI of 1.25 (95% CI 1.09–1.43) or 1.20 (95% CI 1.07–1.38) using data from a meta-analysis of prospective cohort studies published in 2004,³⁴ and from the maximally adjusted model of the Reykjavik Heart Study,³⁴ respectively, assuming the observational associations have been unbiased and free from residual confounding. However, despite possessing a genotype that would have exposed them to a long-term elevation in CRP concentration, for men with the TT genotype, the point estimate of the OR for non-fatal MI was 1.01 (95% CI 0.74–1.38). This finding is important since, in contrast to the reported association between CRP and coronary events in previous studies, the association between CRP genotype and events should not be subject to reverse causality bias, regression dilution bias, or to confounding by other cardiovascular risk factors related to CRP, though some doubt about a marginal effect of the genotype on BMI remains to be evaluated.^{7,8,34,41–47} This is supported by the data in Table 2, which indicate that certain established cardiovascular risk factors and some inflammatory markers, that are known to be associated with CRP, were not associated with +1444C>T genotype.

The estimate of risk derived from the genetic association studies of CRP reported here should not be biased or confounded, but it is currently much less precise than the estimates obtained from classical observational epidemiology. It is likely that for this reason an interaction test, which has low power, did not provide formal evidence for heterogeneity between genetic and non-genetic ORs at the 5% level of significance (Figure 3). Approaches for evaluating consistency of genetic and non-genetic disease risk estimates from Mendelian randomization studies have been reviewed recently.⁴⁸ It is clear, however, from visual inspection of the Forest plot, that the genetic risk estimate is closer to more recent estimates of the CRP–coronary event association reported in an updated meta-analysis of observational studies published in 2004,³⁴ and in the maximally adjusted model of the large Reykjavik Heart Study.³⁴ Thus, recent results from prospective observational studies in which the random error has been reduced substantially by considering more studies in a meta-analysis, and in which more extensive statistical adjustment has been conducted, as exemplified in the Reykjavik Heart Study, indicate that the association of CRP with coronary heart disease risk has decreased substantially in comparison to earlier estimates.³⁴ The strong potential for confounding in observational studies of the measured phenotype (CRP) and the utility of genotype as a proxy for CRP was illustrated by a recent study⁴⁰ in which a range of covariates that exhibited strong correlations with CRP were distributed evenly among >3500 British women separated according to genotypes for a 1059G>C polymorphism that is in LD with the variant reported here.³⁶

Although a null association was identified between CRP genotype and coronary events in this study the width of the confidence limits is such that our observation could be compatible with a modest but potentially important causal link. It is estimated that a dataset of ∼10 000 cases and similar number of controls would be required in a genetic association study to exclude a small effect of CRP on coronary heart disease aetiology.⁴⁹ However, evidence from different sources indicates that, if anything, the discrepancy observed between the genetic and non-genetic ORs is likely to be greater, rather than smaller. First, the observational estimate of the CRP–coronary event association³⁴ on which we based the non-genetic OR, did not adjust for regression dilution bias and so may have been underestimated. However, this bias is unlikely to affect the genetic estimate.¹² Second, emerging evidence supporting the current null genetic estimate comes from three genetic studies in Caucasians, with an additional 1858 cases and 1347 controls,^37,50,51 where carriers of the C allele of the 1059G>C variant (in LD with the +1444/C>T variant) did not have an increased risk of cardiovascular events despite having a CRP concentration that was 0.3–0.6 mg/l higher than GG homozygous subjects. Moreover, recent data from two other genetic studies in Caucasians (703 cases and 1053 controls) found no association of the +1444/C>T variant with cardiovascular events despite a similar effect of this variant on CRP concentration.^37,38 An additional approach in future Mendelian randomization studies is to genotype several SNPs at the CRP locus to generate all common haplotypes and to ensure inclusion of SNPs with demonstrated functionality from in vitro studies in such analyses. Since there is extensive LD across the CRP gene, only a small number of ‘tagging’ SNPs (3–5) require typing to capture most of the genetic variation at this locus.^52,53 In addition to −286C>T/A, published and public domain data indicate that four common haplotypes identified by these tagging-SNPs account for >90% of chromosomes in subjects of European descent.⁵³

It is important to note that there is evidence for an additional functional CRP promoter SNP found at high frequency in subjects of African origin.^36,54

It is important to make clear that our study was limited to male Caucasian subjects and that additional studies will be required to ascertain whether similar findings pertain to women and also to other ethnic groups. However, recent evidence suggests that a similar effect of the +1444C>T polymorphism on CRP concentrations would be expected to occur in women as recently reported in British and American women.^38,55 This is also supported by the effect of the 1059/G>C variant on CRP levels, and the 0.42 mg/l difference in CRP concentrations among C carriers compared with women homozygous to the G allele is very similar to the effect observed here in men.^38,40

Our finding using a genetic approach also needs to be set in the context of recent mechanistic studies that have investigated a potential pro-atherogenic effect of CRP on vascular cells and tissues.^56–61 It is becoming clear now that many of the pro-atherogenic effects demonstrated in vitro in initial studies could have been due to the presence of pro-inflammatory substances contaminating commercially sourced recombinant CRP from bacterial sources, or to the sodium azide preservative.^62–70 More recent experimental studies, using pure CRP preparations, have failed to confirm the proposed pro-atherogenic actions of CRP on vascular tissues and would be concordant with our own data from genetic epidemiology,^62–70 as would the recent experiments in atherosclerosis-prone mice, which showed that in contrast to earlier studies transgenic overexpression of human CRP did not enhance development of atherosclerotic lesions.^71–73

In conclusion, Caucasian men with a genotype that would have exposed them to a long-term elevation in CRP concentration [0.68 mg/l (95% CI 0.31–1.10)] were not at increased risk of non-fatal MI (OR 1.01; 95% CI 0.74–1.38), though the confidence limits currently encompass substantial uncertainty. Taking this into account, it is at the very least likely that unbiased and non-confounded estimates of the effect of CRP on coronary events are smaller than earlier studies estimated, a finding that is in agreement with the latest evidence derived from prospective observational studies and also with more refined and carefully conducted mechanistic studies.⁷⁴ If our null point estimate of the genetic effect is stable with the addition of further studies, it might lead to the re-evaluation of CRP both as a risk marker⁴ and as a potential therapeutic target.⁵ Additional studies with data on CRP genotype, CRP levels, and coronary events will be essential to obtain more precise genetic risk estimates and to help clarify whether or not CRP plays a causal role in atherothrombosis.

Declaration of interest

The authors have declared no conflict of interest.

This work was partially supported by a BHF-PhD studentship [FS/02/086/14760] awarded to T.S. J.W.S. was supported by Diabetes UK. A.D.H. holds a British Heart Foundation Senior Fellowship. S.H. holds the British Heart Foundation Chair of Cardiovascular Genetics (RG 2000/015). The HIFMECH study was supported by the Commission of the European Community: Hifmech Study contract BMH4-CT96-0272. The LEADER trial was supported by grants from the British Heart Foundation and the Medical Research Council. The NPHS-II study was supported by the British Medical Research Council, the US National Institute of Health (NHLBI 33014), and DuPont Pharma, Wilmington, USA. The WOSCOPS genotype analysis was partly funded by a research grant from Chest, Heart and Stroke Scotland. The different funding organizations named above had no role in the design and conduct, data management and analysis, manuscript preparation, or in the review or authorization for submission. The authors acknowledge the critical reading and suggestions of Prof George Davey-Smith.

References