|Home | About | Journals | Submit | Contact Us | Français|
Environmental and genetic correlates of inflammatory marker variability are incompletely understood. In the family-based Framingham Heart Study, we investigated heritability and candidate gene associations of systemic inflammatory biomarkers.
In Offspring participants (n=3710), we examined 11 inflammatory biomarkers [CD40 ligand, C-reactive protein, intercellular adhesion molecule-1 (ICAM1), interleukin-6, urinary isoprostanes, monocyte chemoattractant protein-1, myeloperoxidase, P-selectin, tumor necrosis factor-alpha, tumor necrosis factor receptor II, fibrinogen]. Heritability and bivariate genetic and environmental correlations were assessed by Sequential Oligogenic Linkage Analysis routines (SOLAR) in 1012 family members. We examined 1943 tagging SNPs in 233 inflammatory pathway genes with ≥5 minor allele carriers using a general genetic linear model.
Clinical correlates explained 2.4% (CD40 ligand) to 28.5% (C-reactive protein) of the variability in inflammatory biomarkers. Estimated heritability ranged from 10.9% (isoprostanes) to 44.8% (P-selectin). Most correlations between biomarkers were weak although statistically significant. A total of 45 SNP-biomarker associations met the q-value threshold of 0.25. Novel top SNPs were observed in ICAM1 gene in relation to ICAM1 concentrations (rs1799969, p=1.32×10−8) and MPO in relation to myeloperoxidase (rs28730837, p=1.9×10−5).
Lowest p-values for trans-acting SNPs were observed for APCS with monocyte chemoattractant protein-1 concentrations (rs1374486, p=1.01×10−7) and confirmed for IL6R with interleukin-6 concentrations (rs8192284, p=3.36×10−5). Novel potential candidates (APCS, MPO) need to be replicated.
Our community-based data support the relevance of clinical and genetic factors for explaining variation in inflammatory biomarker traits.
Chronic inflammation predisposes to long-term morbidity and mortality from cardiovascular disease, chronic pulmonary disease, chronic kidney disease, osteoporosis, dementia and the aging process.1–6 Chronic inflammation is associated with abdominal obesity, smoking, and physical inactivity.7,8 Furthermore, the modern epidemic of metabolic syndrome, and its sequelae insulin resistance and type 2 diabetes, have been attributed to an elevated proinflammatory state, with adipose tissue being considered the main source of pro-inflammatory cytokines.9 Environmental and lifestyle factors are likely to contribute to increased low-grade inflammatory activity, as well as an individual’s genetic predisposition.
Systemically measurable inflammatory mediators provide a link between genetics and risk of disease. Inflammatory biomarker concentrations are heritable phenotypes.10,11 For instance, estimated C-reactive protein (CRP) concentration heritability is at least 20 per cent.12 Several single nucleotide polymorphisms (SNPs) are associated with adjusted CRP concentrations and the degree of chronic low-grade inflammation.13–15 However, the genetic contribution to systemic concentrations of most inflammatory markers remains incompletely understood.
We hypothesized that in a community-based cohort, enrolled irrespective of phenotype, SNPs in inflammatory gene regions are associated with concentrations of pro-inflammatory biomarkers. The well-characterized Framingham Heart Study provides a unique opportunity to examine the association of genetic and environmental factors with inflammatory biomarkers.
Participants were eligible if they attended the seventh examination cycle (1998–2001, n=5124) of the Framingham Heart Study Offspring, a white, community-based cohort of European ancestry enrolled in 1971.16,17 Reasons for exclusion from analyses were off-site visits (n=207), none of the 11 biomarkers available (n=10), and missing covariate data (n=17). Heritability and correlations were estimated in 1843 phenotyped individuals in 567 families, in addition to 1468 unrelated participants. To be resource effective and to maximize statistical power, we focused on unrelated individuals and hence genotyped 1565 randomly selected individuals on the standard Offspring unrelated plate set. This plate set is one of several standard Framingham plate sets that is publicly available to researchers.
According to protocol, all participants underwent routine medical history, physical examination, and laboratory testing at the Framingham Heart Study (see Supplement for details). The study was approved by Boston University Medical Center Institutional Review Board; participants signed informed consent.
Fasting biomarkers, selected to represent various phases and functions in the inflammation process (Supplement Table 1) included: CD40 ligand, CRP, fibrinogen, intercellular adhesion molecule-1 [ICAM1], interleukin-6, urinary isoprostanes indexed to urinary creatinine (isoprostanes), monocyte chemoattractant protein-1, myeloperoxidase, P-selectin, tumor necrosis factor receptor II, and tumor necrosis factor-alpha. Biomarkers methods have been detailed previously.18 The mean inflammatory biomarkers’ intra-assay coefficients of variation were <10%.
Genotyping (2942 SNPs in 233 candidate inflammation genes) was conducted by Perlegen Sciences, Inc. (Mountain View, CA) using high-density oligonucleotide, photolithographic microarrays (DNA chips). Common SNPs were chosen from a genome-wide compilation discovered by Perlegen Sciences and supplemented with others from the HapMap project (build 35), if they had >4% (or unknown) minor allele frequency in the HapMap CEU samples or were coding SNPs. To obtain maximum information, a binning procedure was used to identify tagging SNPs with a criterion of r2>0.8 to create bins; one or two SNPs were selected from each bin depending on bin size. Candidate gene selection details have been reported earlier.18 Because SNPs with low call rates showed excess departure from HWE equilibrium, we restricted our analyses to a subset of 1834 SNPs with call rate ≥98% and HWE p>0.01. Only SNPs with at least 5 minor allele carriers in the Framingham sample were evaluated for association. An additional 109 SNPs in 9 candidate inflammatory genes previously genotyped by the CardioGenomics project (http://cardiogenomics.med.harvard.edu/genes/gene-list) on the Sequenom MassARRAY platform, with call rate ≥90% and HWE p≥0.01 were included in the present report, for a total of 1943 SNPs in 233 genes.
Multiple regression analysis was performed on the log-transformed biomarker phenotypes to obtain residuals adjusted for age, sex, cohort (Omni), current smoking, systolic and diastolic blood pressure, hypertension treatment, body mass index, waist circumference, total/high-density lipoprotein cholesterol, triglycerides, lipid lowering medication, glucose, diabetes, aspirin (≥3 days per week), hormone replacement therapy and prevalent cardiovascular disease. For genetic analyses, we adjusted for the same covariates across markers for simplification. The residuals of the log-transformed biomarker phenotypes were rescaled to mean 0,SD 1.
The statistical methods for assessing heritability for biomarkers was described previously.10 Sequential Oligogenic Linkage Analysis (SOLAR, (www.sfbr.org/pages/genetics_projects.php?p=37)) was used to estimate residual log-biomarker concentration heritability for age- and sex-adjusted and multivariable-adjusted models and to calculate correlations. The correlation coefficient between any two covariate-adjusted natural log-transformed inflammatory biomarker concentrations was decomposed into genetic and environmental components.
Analysis of variance (ANOVA) was performed to compare means of log-biomarker residuals (model1: age and sex; model2: age, sex and multiple variables) among inflammatory SNP genotypes using a general genetic model (2 degrees of freedom). ANOVA is not robust for SNPs with low MAF; we used the nonparametric Kruskal-Wallis test instead for SNPs with fewer than 10 individuals in the lowest frequency genotype category. Within each biomarker phenotype, the q-value method,19 a variation of the false discovery rate method, to adjust for multiple testing. We used a threshold of q<0.25 to identify potentially important findings, meaning that the expected proportion of false positive tests among the tests we report within each phenotype is 25%.
We assessed potential interactions of 10 SNPs with the smallest p-value for each biomarker with sex, age, smoking status, and body mass index using linear regression. The full set of covariates were included in the model, as well as the SNP (coded with 2 degrees of freedom), and a 2 parameter SNP by covariate interaction term.
We searched PubMed for English-language literature that reported SNP-biomarker associations with the inflammatory phenotypes characterized in our sample in studies comprising at least 500 individuals that reached statistical significance level of p≤0.05 and provided the direction of association. We identified SNPs reported in the publications or allowed for proxies with an LD r2 of ≥0.5 in our database and provided the association p-value. We omitted CRP and P-selectin associations in cis-acting SELP and CRP genes because we have previously reported on both.20,21 In the online supplement we provide the comprehensive results on SNP-circulating biomarker association studies that were available for the search terms inflammation, inflammatory biomarkers, and single nucleotide polymorphisms, or genetics (August 2008). Phenotype residuals were created using SAS version 8.1 (Cary, NC, http://www.sas.com/presscenter/guidelines.html). Genetic analyses were performed with R (www.r-project.org). All authors had full access to the data, take responsibility for its integrity, and have read and agree to the manuscript as written.
External replication was attempted in the previously described AtheroGene cohort22 in up to 1752 patients with documented coronary artery disease and 430 controls free of manifest cardiovascular disease. We confined replication to top findings in the current study to limit the number of tests performed. SNPs were selected if they had not been reported in the literature in comparable studies, were in trans-acting genes and were in low linkage disequilibrium (r2<0.5) with other top SNPs. Residuals were created using age, sex, case-control status, smoking status, body mass index, total/HDL cholesterol, triglycerides, serum glucose, diabetes, hypertension, and lipid treatment.
The clinical and laboratory characteristics of the study cohort have been reported before.20 The heritability and genotype samples’ characteristics are outlined in Supplement Table 2. Briefly, the mean age of the genotyped sample was 62±9 years, 51% were women, and the cardiovascular disease prevalence was 13%. Clinical variables explained between 2.4 (CD40 ligand) to 28.5% (CRP) of inflammatory biomarker variability (Supplement Table 3).
All inflammatory traits were heritable (p<0.05; Table 1, second/third column); estimated multivariable-adjusted heritability values ranged from 10.9% (isoprostanes) to 44.8% (P-selectin); age- and sex-adjusted results were generally slightly higher. The Pearson correlation coefficients and portion of correlation due to genetic factors also are displayed. Significant environmental correlations were observed for 24 biomarker combinations. Strongest overall pairwise correlations were observed for CRP with fibrinogen (ρ=0.48), and interleukin-6 (ρ=0.39). Six genetic correlations were seen with highest correlation coefficients for fibrinogen and CRP (0.14), and for interleukin-6 and P-selectin (0.12).
To account for multiple testing we computed false discovery rates.23 A total of 45 associations were significant at a cutoff q-value<0.25. Lowest p-values for trans-acting (not involving the protein-coding gene) SNPs were observed for APCS (rs1374486, p=1.01*10−7, and rs6695377, 5’ near gene, p=1.85*10−7) with MCP-1 concentrations, IL6R (rs8192284, Ala/Asp missense, p=3.36*10−5) with interleukin-6 concentrations, and MPO in relation to myeloperoxidase (rs28730837, Val/Ala missense , p=1.9×10−5). SNPs with a q-value<0.25 across phenotypes not previously reported in the Framingham Study (SNPs in the CRP, CCL2 and SELP genes) are tabulated (Table 2). The top cis-acting associations for SNPs not previously reported by our group (SELP SNPs- P-selectin concentrations were previously reported20) were observed in the ICAM1 gene in relation to ICAM-1 concentrations (rs1799969, Arg/Gly missense, p=1.32*10−8). Results for the top SNPs (q-value<0.25) presented in Table 2 were consistent with age- and sex-adjusted models (Supplement Table 10).
Accounting for multiple testing, there was no evidence for strong interactions between the SNPs most highly associated with each phenotype and sex, age, smoking status, and body mass index (Supplement Table 4).
We were able to replicate two previously reported ICAM1 SNPs in our database (Supplement Table 6); rs1799969 was congruent with our top ICAM1 finding. SNPs in IL6, CD14 and NOS3 genes in relation to interleukin-6 concentrations were not replicated. We could confirm rs8192284 in the IL6R gene in relation to interleukin-6 concentrations, as well as three SNPs in the CCL2 gene in association with MCP-1.
In the AtheroGene cohort, predominantly consisting of coronary artery disease patients (n=895–1752), only rs3732764 in the P2RY12 gene, reached borderline significance, p=0.05. None of the other top findings could be replicated (Supplement Table 7).
We report heritability and genetic associations for a broad panel of carefully selected inflammatory biomarkers and SNPs in a moderately-sized community-based sample. We observed significant heritability for 11 inflammatory biomarker traits, with heritability estimates ranging from 10.9% to 44.8%, including estimates for isoprostanes, myeloperoxidase and tumor necrosis factor receptor II that have not been published before. We detected substantial environmental correlations between many systemic biomarkers, and some pairwise genetic correlations between biomarker traits. Top findings of the broad candidate gene approach, comprising 1943 SNPs, confirmed recent results from the literature for cis-acting SNPs in ICAM1 and CRP genes and trans-acting IL6R association with interleukin-6 concentrations. In addition, novel associations we report were significant cis-acting MPO SNPs with myeloperoxidase concentrations, and trans-acting APCS SNPs in relation to monocyte chemoattractant protein-1. We were not able to replicate our results in a cohort of patients with prevalent coronary artery disease. We present age- and sex-adjusted, as well as multivariable-adjusted phenotype-genotype associations online, so that investigators can download the data and conduct their own analyses. Furthermore, to place our results into perspective, we include comprehensive reviews of the inflammatory biomarker heritability and candidate gene literature in the Supplement.
The examination of bivariate biomarker trait correlations, partitioned into shared genetic and environmental components,24 clearly revealed that environmental factors contributed a larger extent to observed correlations of circulating biomarker concentrations compared to additive genetic effects. The strength of the genetic and environmental correlations we observed was lower than reported in recent twin studies.25 Only CRP in relation to fibrinogen, and ICAM1 and interleukin-6 showed moderate genetic, as well as environmental correlations. Compared to the prior literature, we provide correlations for a large inflammatory marker panel.
Heritability for inflammatory biomarkers has been reported by Framingham and other researchers for extensively investigated traits like CRP, interleukin-6, ICAM1 and monocyte chemoattractant protein-1 (Supplement Table 5).10–12,26,27 The present cohort convincingly demonstrated a modest to moderate proportion of variability explained by descent in a large panel of distinct inflammatory biomarkers, even biomarkers with known higher intra-individual variability and measurement coefficients of variation like isoprostanes.28
As coding genes have the highest likelihood of association with encoded proteins, the majority of biomarker candidate gene association analyses have been performed for cis-acting genes. Not surprisingly, our strongest association finding was in the SELP gene in relation to P-selectin concentrations, which has previously been reported in Framingham20 and independent studies.29,30 For several tagging SNPs we and others were able to show moderate associations between common genetic variation in the respective coding genes for CRP14,21,31 and ICAM132,33 concentrations after accounting for known covariates (for additional replication please see Supplement Table 6).
We further hypothesized that inflammatory genes are related to circulating biomarkers not coded for by the gene (trans-acting genes). We extended current knowledge by examining a broad panel of 233 inflammatory candidate genes. The aim was to capture trans-acting genotypes that might contribute at the genetic level to the known strong interrelations of inflammatory pathways at the biomarker level. We confirmed the strong association of SNP rs8192284 in IL6R with interleukin-6 phenotype.34 The observed relation has a functional explanation since the amino acid exchange leads to an alteration at the receptor cleavage site, which increases soluble interleukin-6 receptor concentrations and thus affects, circulating interleukin-6.34,35 The independent evidence in ethnically different groups and a plausible pathophysiological explanation have turned this SNP into a very promising candidate polymorphism.
Less evidence is available that would help to explain the top finding of two SNPs in the APCS gene in relation to serum monocyte chemoattractant protein-1 concentrations. The APCS SNPs, all located in the 5’ gene region without relevant LD to SNPs with known function, belong to a gene with multiple polyadenylation sites encoding a highly conserved glycoprotein of the pentraxin family, serum amyloid P component.36 Serum amyloid component shares considerable sequence homology with CRP resulting from gene duplication during evolution. Serum amyloid P opsonizes apoptotic cells, an important step in their clearance.37 Amyloid P is found in atherosclerotic plaques38 and circulating concentrations have been related to clinical cardiovascular disease in an elderly, multiethnic community-based cohort.39 Genetic data in humans are scant. We reported a linkage peak for MCP-1 on chromosome 1 which extends over the APCS gene locus and provides additional evidence for a potential association.10 Ongoing genome-wide association studies will help to identify the true variants. None of the polymorphisms in the CCL2 gene reported in the Framingham Heart Study cohort reached experiment-wide significance, but showed nominally significant associations with the same directionality.40
Myeloperoxidase, has been recognized for its role in non-infectious inflammatory diseases, and as an important modulator of vasomotor function in vascular inflammation.41 Two non-HapMap SNPs have been described in association with myeloperoxidase activity (rs28365049, rs34097845). The functional promotor polymorphism (−463G/A) containing an Alu element is related to myeloperoxidase expression.42 It has been linked to inflammatory diseases like Alzheimer’s disease,43 and atherosclerotic disease.44 In the current cohort we provide evidence on the significance of a new SNP, rs28730837, a Val/Ala missense variation, with regard to myeloperoxidase concentration.
Compared to the large number of published studies, only few met our inclusion criteria of sample size ≥500 participants for an in-silico replication attempt. We were able to replicate mostly cis-acting SNPs from the literature for ICAM1 and CCL2 genes and one prominent trans-acting association of the recently reported SNP rs8192284 in IL6R gene in relation to interleukin-6. Associations for SNPs in CD14 and NOS3 genes with interleukin-6, previously seen in patients with coronary artery disease could not be confirmed and may indicate spurious or disease-specific findings.45,46
The Framingham Study constitutes a single center family-based community cohort with limited referral bias, stringent biomarker quality control, well-documented, and routinely ascertained environmental confounders, which facilitate multivariable models and heritability analyses. The choice of multiple biomarkers from scientifically sound candidate pathways based on basic and human studies further increases the current study’s comprehensiveness. The broad range of SNPs chosen for association reduces bias observed in candidate gene approaches and may uncover both cis and trans regulators.47 Whereas our study underscores the problems of multiple testing, in contrast to many recent publications, a q-value method was applied with conservative thresholds, to minimize false positive findings without instituting overly strict Bonferroni corrections. In addition, we provide a downloadable excel file at our website of all inflammatory marker-SNP associations tested that will facilitate replication by external investigators.a Furthermore, we provide a comprehensive review of most prior publications examining heritability and the associations between SNPs and circulating inflammatory biomarkers.
Some limitations meriting consideration are that the significant results are currently restricted to one study group. We were unable to replicate our findings in an independent cohort with coronary artery disease. Non-replication may be due to a relatively low number of genotyped individuals and their pre-existing coronary disease, which is known to elevate biomarker concentrations. As noted by Chanock and colleagues, phenotype and participant heterogeneity will compromise the likelihood of replication.48 The selected nature of AtheroGene, is corroborated by the observation, that the repeatedly validated association of IL6R rs819228434 was replicated in the FHS cohort but not in the AtheroGene cohort. We used an older HapMap build (build 35) and thus may have missed important variants.
Intermediate cardiovascular phenotypes (i.e. hypertension) are strongly correlated with inflammatory activity; multivariable-adjustment may limit our ability to detect pleiotropic environmental and genetic effects related to inflammation. To reduce the high multiple testing burden, we specified à priori that our primary analyses would be multivariable-adjusted models.
Generalizability of the results is limited by the ethnically homogenous cohort; biomarker concentrations vary by ethnicity.49 For other ethnicities, a slightly different set of informative SNPs would have been chosen.50 On the other hand, the potential for population stratification was reduced by racial homogeneity.51 We acknowledge that our cohort had only moderate power to detect modest effects; a potential for false negative findings is evident. Inherent to single-cohort genetic association studies, our results should be viewed as hypothesis generating; replication in independent samples is necessary.
Supported by NIH/NHLBI contract N01-HC-25195 and NIH grants HL064753 & HL076784 AG028321 (E.J.B.), HL70139 (R.S.V). NIH Research career award HL04334 (R.S.V.), and K24 DK080140 (J.B.M.). German Research Foundation Research Fellowship SCHN 1149/1-1 (RS); TNF-alpha via American Diabetes Association Career Development Award and NCRR GCRC M01-RR-01066 (JBM).
aData file included as reviewers’ electronic supplement will be posted upon publication.
The authors report no conflicts of interest.