Asp92Asn Polymorphism in the Myeloid IgA Fc Receptor Is Associated With Myocardial Infarction in Two Disparate Populations
CARE and WOSCOPS
Objective— Statins reduce inflammation and risk of myocardial infarction (MI). Because the myeloid IgA Fc receptor encoded by FCAR mediates inflammation, we hypothesized that the FCAR Asp92Asn polymorphism is associated with risk of MI and that this risk would be modified by pravastatin.
Methods and Results— In the placebo arm of the Cholesterol and Recurrent Events (CARE) study, male carriers of the 92Asn allele had an adjusted hazard ratio for incident MI of 1.68 (95% CI 1.10 to 2.57); relative risk reduction by pravastatin was 69% in carriers and 12% in noncarriers (Pinteraction=0.007). In the placebo arm of the all-male West of Scotland Coronary Prevention Study (WOSCOPS), carriers had an adjusted odds ratio for incident coronary heart disease (CHD) of 1.46 (90% CI 1.05 to 2.03); for pravastatin compared with placebo treatment, the adjusted odds ratios were 0.55 (95% CI 0.32 to 0.93) in carriers and 0.65 (95% CI 0.51 to 0.83) in noncarriers (Pinteraction=0.55).
Conclusions— Carriers of 92Asn had increased risk of MI in CARE and increased odds of CHD in WOSCOPS. Pravastatin significantly reduced risk in carriers in both CARE and WOSCOPS. A genotype by treatment interaction was observed in CARE but not in WOSCOPS.
Genetic polymorphisms may help to explain why some patients experience an MI or respond to preventive therapy whereas others, with the same conventional risk factor profile, do not.1,2 This differential risk of disease and response to preventive treatment may be explained by polymorphisms in genes that mediate progression of atherosclerosis, including genes involved in thrombosis, plaque instability, and inflammation.3 Many of the inflammatory processes that occur during atherosclerosis involve monocytes and macrophages.4 The activation of these cells can be regulated by cell surface receptors for immunoglobulin, such as the myeloid IgA Fc receptor (encoded by FCAR) and IgG Fc receptors.5 Recently, a polymorphism in an Ig Fc receptor, FcγRIIIA, has been shown to be associated with angiographically defined CHD.6
The relationship between the several IgA Fc receptors and CHD is not well understood. However, the involvement of the myeloid IgA Fc receptor in monocyte activation suggests that this receptor may play a role in the pathogenesis of cardiovascular disease,5,7 and in a preliminary survey of 551 cardiovascular candidate genes we observed an association of the Asp92Asn single nucleotide polymorphism (SNP) in FCAR with myocardial infarction (see Methods). The myeloid IgA Fc receptor, otherwise known as FcαRI or CD89, is expressed only on myeloid cells, including monocytes and macrophages. These cells can be activated by FcαRI when this receptor clusters on the cell surface.7 FcαRI is encoded by FCAR, which is located on human chromosome 19q13.4 within the leukocyte Ig-like receptor complex (LRC). This receptor complex comprises structurally related genes, which include NCR1, NALP7, and NKp46 and genes encoding the killer-cell Ig-like receptors (KIRs). FCAR has five exons and 10 alternative transcripts (OMIM #147045).
FcαRI is a member of the Ig gene superfamily and has two Ig-like extracellular domains, a transmembrane domain, and a 41-residue carboxy-terminal cytoplasmic domain.8 The membrane–distal Ig-like domain of FcαRI binds a single IgA molecule, and the FCAR Asp92Asn SNP is 5 amino acid residues away from this IgA binding site. Therefore, we hypothesized that the Asp92Asn SNP affects risk of MI and CHD and that the risk associated with this SNP would be modified by pravastatin treatment, a treatment shown to reduce vascular inflammation.9,10
We conducted genetic association studies of the FCAR Asp92Asn SNP in the male patients of the Cholesterol and Recurrent Events (CARE) and in the all-male West of Scotland Coronary Prevention Study (WOSCOPS). CARE, a secondary prevention trial, has been described elsewhere11 and in both the online Methods and supplemental Table I (available online at http://atvb.ahajournals.org). We used a composite MI end point of confirmed nonfatal MI (86%) or fatal MI (14%).
WOSCOPS, a primary prevention trial, has been described elsewhere12 and in the online Methods. The setting of the genetic study of WOSCOPS presented here was a previously reported prospective nested case–control study that included as cases all WOSCOPS patients with on-trial CHD.13 Institutional Review Committees approved these studies. Informed consent was given by all CARE and WOSCOPS enrollees included in these studies.
The association of the FCAR Asp92Asn SNP with MI in CARE was initially observed in the context of a genetic study of 551 cardiovascular candidate genes. The 1000 SNPs tested (supplemental Table II) were predicted to affect protein function or expression levels, and 67 of these SNPs were associated with MI (supplemental Table III) in an analysis of the White patients of CARE (2,523 men and 390 women). Because risk factors for complex diseases such as MI are likely to have modest effects,14 the statistical power to detect the effect of a SNP in a single study while accounting for multiple testing is low. Therefore, we pursued a strategy of replication by testing the prespecified hypotheses that the risk alleles associated with MI in CARE would be associated with CHD in WOSCOPS. We selected 24 of the 67 SNPs based on their risk estimates in CARE and the biological plausibility of their association with MI (the 24 SNPs are indicated by footnote symbols in supplemental Table III) and asked if the same risk alleles were associated with CHD in WOSCOPS. Only the FCAR 92Asn risk allele was associated with CHD in WOSCOPS.
The Asp92Asn SNP is in exon 3 of FCAR and corresponds to the first nucleotide of codon 113 in the full-length transcript (NM_002000). Codon 113 corresponds to residue 92 in the mature protein that is formed by cleavage of the leader sequence. In build 36.1 of the NCBI SNP database, the FCAR Asp92Asn SNP (rs11666735) is located on human chromosome 19 genomic, contig accession NT_011109.15 at contig position 27,665,103. The Asp92Asn SNP was genotyped using two different genotyping technologies (see online Methods). Using a tagging SNP approach,15 we investigated FCAR and two regions of high linkage disequilibrium that flank the FCAR gene16 (see online Methods).
Unless noted otherwise, statistical analyses were performed with SAS version 9. For CARE, we used Cox proportional hazard models (Wald tests) to assess the association of incident MI with FCAR Asp92Asn genotype and also to assess the effect of pravastatin treatment on incident MI in the FCAR 92Asn carriers and noncarriers. Time-to-event was defined as the time from the date of randomization to the date of the first on-trial MI. Patients were censored at the time of their last visit (if lost to follow-up) or at the end of the study. Potential interaction between each conventional risk factor and the FCAR Asp92Asn genotype was tested in separate regression models that included an interaction term between the risk factor and the Asp92Asn genotype (likelihood ratio tests). Continuous variables tested were age, body mass index (BMI), baseline LDL-C, and baseline HDL-C levels. Categorical variables tested were smoking (current versus noncurrent), history of hypertension, and history of diabetes. When adjusting Cox models for on-trial change in lipid levels, we used time-dependent covariates. In CARE, we excluded 8 patients (0.3%) from the analysis because of missing risk factor information and 94 patients (3%) because of inadequate quantity or quality of DNA. For WOSCOPS, we used conditional logistic regression models because the controls had been matched to the cases for age and current smoking. Thus, covariates included in the adjusted model were the same covariates used in the Cox analysis of CARE except for age and current smoking. In WOSCOPS, 8 patients (0.5%) were excluded from the analysis because of inadequate quantity or quality of DNA. We assessed differences between continuous variables by t tests and between categorical variables by χ2 tests. We assessed deviation from Hardy–Weinberg equilibrium by exact tests17 in male patients of the CARE cohort and also in the control patients of WOSCOPS. All probability values are two-sided except that for the analysis performed to confirm the prespecified risk allele in WOSCOPS, we report one-sided probability values. For a description of the analysis of the association between haplotypes and disease, see the online Methods.
In CARE, we estimated MI-free survival separately in placebo- and pravastatin-treated patients using Kaplan–Meier analysis within subgroups defined by FCAR genotypes. We used Kaplan–Meier estimates of MI-free survival at 5 years of follow-up to calculate absolute risk reduction. Differences in survival distributions were assessed with the log-rank test. For WOSCOPS, absolute risk reduction was projected for the first 4.9 years of follow-up using the actual FCAR Asp92Asn genotype frequencies for all cases and an estimate of the genotype frequencies in event-free patients. This estimate assumed that the FCAR Asp92Asn genotype frequencies determined in the controls of the nested case–control study were the same as the genotype frequencies in the rest of the event-free patients in the cohort.
In the placebo arm of CARE, during preliminary analysis of the association between the FCAR Asp92Asn SNP and MI, we observed a significant (P=0.03) interaction between genotype and sex for carriers of the 92Asn risk allele: the adjusted hazard ratio was 1.68 (95% CI 1.10 to 2.57) in men and 0.20 (95% CI 0.02 to 1.66) in women. Therefore, we conducted subsequent analyses in sex subgroups, but present only the results of the male subgroup because there were not enough women in the CARE trial to provide meaningful risk estimates for women (only 1 female carrier of the FCAR 92Asn risk allele had an MI during the CARE trial). The baseline characteristics of the male patients genotyped for the FCAR Asp92Asn SNP in CARE and WOSCOPS are presented in supplemental Table IV; this table also indicates which of these baseline characteristics were associated with the end points. The baseline characteristics of carriers of the FCAR 92Asn risk allele in both treatment arms in CARE and WOSCOPS are presented in supplemental Table V. The frequency of the 92Asn allele of FCAR was 7.5% in the male CARE patients (14.3% were carriers) and 7.8% in the WOSCOPS control group (14.9% were carriers). The distribution of FCAR Asp92Asn genotypes did not deviate from Hardy–Weinberg expectations in either CARE (P=0.06) or WOSCOPS (P=0.52). For the other 23 SNPs tested in both CARE and WOSCOPS, the Hardy–Weinberg equilibrium probability values ranged from 0.09 to 1.00 in CARE and from 0.04 to 1.00 in WOSCOPS. FCAR Asp92Asn was not associated with the conventional risk factors tested in CARE or WOSCOPS (supplemental Table VI).
Association of the FCAR 92Asn Allele With MI and CHD
The risk of MI among men in the placebo arm of CARE was significantly greater for carriers of the FCAR 92Asn allele than for noncarriers. For carriers, the hazard ratio for MI was 1.68 (95% CI 1.10 to 2.57, Table 1) in a model adjusted for age, smoking, history of hypertension, history of diabetes, BMI, baseline LDL-C, and baseline HDL-C levels. Interestingly, only the 92Asn risk allele and smoking remained significant predictors of MI after adjustment for the dichotomized risk factors shown in Figure 1.
We then investigated whether the FCAR 92Asn risk allele was associated with CHD in the placebo arm of the all-male WOSCOPS nested case–control study. We found that the odds ratio for CHD for 92Asn allele carriers was 1.46 (90% CI 1.05 to 2.03) in a model adjusted for history of hypertension, history of diabetes, BMI, baseline LDL-C, and baseline HDL-C levels (Table 2).
The simplest explanation for the association of the FCAR Asp92Asn SNP with MI and CHD is that the Asp92Asn variant is involved in the pathogenesis of MI. Alternatively, the Asp92Asn SNP may be associated with MI because it is in linkage disequilibrium with some other SNP involved in the pathogenesis of MI. Therefore, to search for other possible causative SNPs in FCAR or in flanking regions, we asked whether 21 SNPs in these regions (19 tagging SNPs and 2 additional missense SNPs) were associated with MI in CARE and with CHD in WOSCOPS (supplemental Figure I). We found that only FCAR Asp92Asn (which was one of the 19 tagging SNPs) was associated with both MI in CARE and CHD in WOSCOPS (supplemental Table VII). In addition, no haplotype in this region was found to be associated with both MI in CARE and CHD in WOSCOPS (data not presented).
FCAR Asp92Asn and Benefit From Pravastatin
We then asked whether pravastatin modified the risk of MI in CARE and the odds of CHD in WOSCOPS in those who carried the FCAR 92Asn risk allele. In addition, we asked whether the effect of pravastatin on the risk of MI in CARE and the odds of CHD in WOSCOPS differed between carriers and noncarriers of the risk allele.
In CARE, among male carriers of the FCAR 92Asn risk allele, pravastatin treatment reduced the risk of MI: the hazard ratio for MI for pravastatin treatment compared with placebo-treatment was 0.29 (95% CI 0.13 to 0.65, Table 3) and among noncarriers, the hazard ratio for pravastatin treatment was 0.90 (95% CI 0.67 to 1.20). This difference in risk reduction between carriers and noncarriers was significant (P=0.005 for genotype by treatment interaction). This difference in risk reduction is also consistent with the lack of association of FCAR Asp92Asn with MI in the pravastatin arm of CARE (supplemental Table VIII). The estimates of risk reduction by pravastatin were essentially unchanged after adjustment for conventional risk factors: the hazard ratios were 0.31 (95% CI 0.14 to 0.68) for carriers and 0.88 (95% CI 0.66 to 1.17) for noncarriers (Table 3). In models adjusted for conventional risk factors and also adjusted for on-trial changes in LDL-C and HDL-C levels, the hazard ratio for MI for pravastatin treatment compared with placebo treatment was 0.25 (95% CI 0.09 to 0.68) among carriers and 1.08 (95% CI 0.72 to1.62) among noncarriers (data not shown).
In CARE, MI-free survival in male carriers of the FCAR 92Asn risk allele was significantly better in the pravastatin-treated group than in the placebo-treated group (P=0.001, Figure 2). This difference was reflected in an absolute risk reduction of 10.3% by pravastatin in carriers of the 92Asn risk allele after 5 years of follow-up. In noncarriers, the absolute risk reduction was 1.5%.
In WOSCOPS, pravastatin treatment reduced the odds of CHD both in carriers of the FCAR 92Asn risk allele and in noncarriers. In carriers, the adjusted odds ratio for CHD was 0.55 (95% CI 0.32 to 0.93, Table 3) for pravastatin compared with placebo treatment; in noncarriers the adjusted odds ratio was 0.65 (95% CI 0.51 to 0.83). The difference in risk reduction between carriers and noncarriers was not significant (Pinteraction=0.55 for genotype by treatment interaction). In WOSCOPS, pravastatin treatment resulted in a projected absolute risk reduction of 4.2% in carriers and 2.2% in noncarriers.
We found that the 92Asn allele of FCAR is associated with increased risk of MI in CARE and increased odds of CHD in WOSCOPS. Male carriers of the 92Asn risk allele had an adjusted hazard ratio of 1.68 for incident MI in the placebo arm of CARE and an adjusted odds ratio of 1.46 for incident CHD in the placebo arm of WOSCOPS. These risk estimates are similar in magnitude to the risk estimate for current smoking in the placebo arm of this genetic study of CARE. We also found that in carriers of the 92Asn risk allele, pravastatin treatment reduced the risk of MI and the odds of CHD.
In CARE, the risk of MI in carriers of the FCAR 92Asn allele was reduced by pravastatin treatment, and this benefit from pravastatin treatment was greater in carriers than in noncarriers. However, this result does not suggest that noncarriers should not be treated because some noncarriers might benefit from pravastatin because of environmental or genetic factors that are independent of their Asp92Asn genotype. Because adjustment for on-trial changes in LDL-C and HDL-C levels did not diminish the apparent benefit of pravastatin in carriers of the 92Asn risk allele, the effect of pravastatin treatment on MI in carriers of the 92Asn risk allele might be explained, in part, by a mechanism unrelated to the LDL-C lowering effect of pravastatin.
In WOSCOPS, pravastatin treatment reduced the odds of CHD both in carriers of the FCAR 92Asn risk allele and in noncarriers. This reduction in incident CHD did not differ significantly between carriers and noncarriers. A similar observation was reported in the Scandinavian Simvastatin Study for the ε4 allele of APOE. The ε4 allele predicted risk of on-trial death, and in carriers of the ε4 risk allele, risk was significantly reduced by simvastatin treatment and the interaction between genotype and treatment was not significant (P=0.23).18 These results contrast with the significant interaction between genotype and treatment that we observed in CARE. The interaction observed in CARE is similar to the significant interaction between pravastatin treatment and the Taq1B1 polymorphism in CETP that was observed in a genetic study of the REGRESS trial.19 Nevertheless, for carriers of the FCAR 92Asn risk allele in both CARE and WOSCOPS, pravastatin reduced absolute risk: in CARE the absolute risk reduction was 10.3% (versus 1.5% in noncarriers) and in WOSCOPS the projected absolute risk reduction was 4.2% (versus 2.2% in noncarriers). To put these numbers in perspective, when male patients were not stratified by genotype, absolute risk reduction was 2.8% in CARE and 2.5% in WOSCOPS.
Two previous disease association studies of FCAR SNPs have been published. Jasek et al investigated the association of several variants, including Asp92Asn, with asthma and did not observe an association.20 Two SNPs in the promoter region of FCAR (−114T/C, rs12462181 and +56T/C, rs3816051) have been reported to be associated with IgA nephropathy and IgA levels.21 However, it is unlikely that the −114T/C and +56T/C SNPs are responsible for the association of the FCAR Asp92Asn SNP with MI and CHD because the linkage disequilibrium between these promoter SNPs and Asp92Asn is low (r2<0.03).
Among the SNPs and haplotypes tested in the FCAR gene and flanking regions, only the Asp92Asn SNP was associated with both risk of MI in CARE and odds of CHD in WOSCOPS. These results suggest that the FCAR Asp92Asn SNP may play a functional role in the pathogenesis of MI and CHD. The mechanism by which this SNP might increase inflammation and ultimately risk of CHD is not clear. However, the Asp92Asn SNP is located in the IgA-binding domain of the FCAR gene product (FcαRI) and is in close proximity to other residues shown to play important roles in IgA-binding (Arg82, His85, His87, and Phe88).22,23 Thus, the Asp92Asn SNP may influence IgA binding and FcαRI-dependent inflammatory signaling cascades; however, these possibilities remain to be addressed with functional studies.
This study was conducted as a follow-up of a survey of 551 candidate genes in CARE, a survey that raised the issue of how to account for multiple testing. To address this issue we pursued a strategy of replication. After finding that the 92Asn allele of FCAR was associated with risk of MI in CARE, we investigated whether this prespecified risk allele was associated with CHD in WOSCOPS. Although we observed a consistent association with risk in both studies, further confirmation is needed in other studies of CHD to address the generality of these associations.
This study has several limitations. First, there were disparities between the two study populations. These disparities include: CHD at enrollment (CARE was a secondary prevention trial whereas WOSCOPS was a primary prevention trial), the patients were recruited in the US and Canada for CARE and in Western Scotland for WOSCOPS, and the baseline LDL-C levels (the median LDL-C level was 192 mg/dL for the WOSCOPS cohort13 and 139 mg/dL for the CARE cohort11). Second, this genetic study included only White men; therefore, the association of the FCAR 92Asn allele with cardiovascular risk should be investigated in cohorts that are adequately powered for analysis in women and in other ethnic groups. Third, larger studies are needed to accurately determine the risk estimate for the minor homozygote of the FCAR Asp92Asn SNP. Finally, intermediate phenotypes related to IgA or other immune functions have not been demonstrated for the FCAR Asp92Asn SNP.
Hegele24 has suggested standards that should be met by genetic association studies. The study we describe herein met the following of these standards: optimized sampling and phenotyping, two independent genotyping methods, description of all SNPs tested, a biological rationale for the SNP studied, large sample sets, and replication in an independent study. However, this study did not provide physiologically meaningful evidence supporting the functional role of the FCAR Asp92Asn SNP in CHD.
The FCAR 92Asn allele predicted risk of MI in CARE and was associated with increased odds of CHD in WOSCOPS—two disparate studies. In both trials, carriers of the 92Asn risk allele benefited from pravastatin treatment, and a significant interaction between genotype and treatment was observed in CARE but not in WOSCOPS.
We express our gratitude to the CARE and WOSCOPS patients and to Drs Marc Pfeffer, Thomas J. White, Nicolas Dracopoli, John Sninsky, Koustubh Ranade, and Kit Lau for helpful comments on this manuscript, to Drs Goran Gogic and David Ross, and to Migdad Machrus, Daniel Civello, Christopher Charles, Alla Smolgovsky, and Joel Bolonick for their support in Computational Biology.
Sources of Funding
The CARE and WOSCOPS trials were supported by research grants from Bristol-Myers Squibb Pharmaceutical Research Institute; genotyping was funded by Celera; statistical analysis was funded by Celera and by Bristol-Myers Squibb.
O.A.I., C.H.T., A.P.C., C.M.R., T.G.K., J.Z.L., L.M.P., J.J.C., D.U.L., B.A.Y., D.L., Z.T., M.M.L., K.E.Z., P.M.S., and J.J.D. have employment and ownership interest; M.S.S. has research grants from Bristol-Myers Squibb and AstraZeneca, he is on the speakers bureau and is a consultant for Bristol-Myers Squibb; C.J.P. has research grants from Bristol-Myers Squibb and AstraZeneca, he is on the speakers bureau for Bristol-Myers Squibb and is a consultants for AstraZenec; J.S. is on the speakers bureau for AstraZeneca, Pfizer, Sankyo and is a consultant for AstraZeneca, Pfizer and GSK; F.M.S. was an expert witness for Pfizer and is a consultant for Bristol-Myers Squibb; H.C. is a consultant for Bristol-Myers Squibb.
Original received April 13, 2006; final version accepted September 13, 2006.
Naghavi M, Libby P, Falk E, Casscells SW, Litovsky S, Rumberger J, Badimon JJ, Stefanadis C, Moreno P, Pasterkamp G, Fayad Z, Stone PH, Waxman S, Raggi P, Madjid M, Zarrabi A, Burke A, Yuan C, Fitzgerald PJ, Siscovick DS, de Korte CL, Aikawa M, Airaksinen KE, Assmann G, Becker CR, Chesebro JH, Farb A, Galis ZS, Jackson C, Jang IK, Koenig W, Lodder RA, March K, Demirovic J, Navab M, Priori SG, Rekhter MD, Bahr R, Grundy SM, Mehran R, Colombo A, Boerwinkle E, Ballantyne C, Insull W Jr, Schwartz RS, Vogel R, Serruys PW, Hansson GK, Faxon DP, Kaul S, Drexler H, Greenland P, Muller JE, Virmani R, Ridker PM, Zipes DP, Shah PK, Willerson JT. From vulnerable plaque to vulnerable patient: a call for new definitions and risk assessment strategies: Part II. Circulation. 2003; 108: 1772–1778.
Sacks FM, Pfeffer MA, Moye LA, Rouleau JL, Rutherford JD, Cole TG, Brown L, Warnica JW, Arnold JM, Wun CC, Davis BR, Braunwald E. The effect of pravastatin on coronary events after myocardial infarction in patients with average cholesterol levels. Cholesterol and Recurrent Events Trial investigators. N Engl J Med. 1996; 335: 1001–1009.
Packard CJ, O’Reilly DS, Caslake MJ, McMahon AD, Ford I, Cooney J, Macphee CH, Suckling KE, Krishna M, Wilkinson FE, Rumley A, Lowe GD. Lipoprotein-associated phospholipase A2 as an independent predictor of coronary heart disease. West of Scotland Coronary Prevention Study Group. N Engl J Med. 2000; 343: 1148–1155.
Monsuur AJ, de Bakker PI, Alizadeh BZ, Zhernakova A, Bevova MR, Strengman E, Franke L, van’t Slot R, van Belzen MJ, Lavrijsen IC, Diosdado B, Daly MJ, Mulder CJ, Mearin ML, Meijer JW, Meijer GA, van Oort E, Wapenaar MC, Koeleman BP, Wijmenga C. Myosin IXB variant increases the risk of celiac disease and points toward a primary intestinal barrier defect. Nat Genet. 2005; 37: 1341–1344.
Weir BS. Genetic Data Analysis II. Sunderland: Sinauer Associates Inc; 1996.
Gerdes LU, Gerdes C, Kervinen K, Savolainen M, Klausen IC, Hansen PS, Kesaniemi YA, Faergeman O. The apolipoprotein epsilon4 allele determines prognosis and the effect on prognosis of simvastatin in survivors of myocardial infarction : a substudy of the Scandinavian simvastatin survival study. Circulation. 2000; 101: 1366–1371.
Kuivenhoven JA, Jukema JW, Zwinderman AH, de Knijff P, McPherson R, Bruschke AV, Lie KI, Kastelein JJ. The role of a common variant of the cholesteryl ester transfer protein gene in the progression of coronary atherosclerosis. The Regression Growth Evaluation Statin Study Group. N Engl J Med. 1998; 338: 86–93.
Wines BD, Hulett MD, Jamieson GP, Trist HM, Spratt JM, Hogarth PM. Identification of residues in the first domain of human Fc alpha receptor essential for interaction with IgA. J Immunol. 1999; 162: 2146–2153.
Hegele RA. SNP judgments and freedom of association. Arterioscler Thromb Vasc Biol. 2002; 22: 1058–1061.