Large Multiethnic Candidate Gene Study for C-Reactive Protein Levels: Identification of a Novel Association at CD36 in African Americans

Abstract

C-reactive protein (CRP) is a heritable biomarker of systemic inflammation and a predictor of cardiovascular disease (CVD). Large-scale genetic association studies for CRP have largely focused on individuals of European descent. We sought to uncover novel genetic variants for CRP in a multi-ethnic sample using the ITMAT Broad-CARe (IBC) array, a custom 50,000 SNP gene-centric array having dense coverage of over 2,000 candidate CVD genes. We performed analyses on 7570 African Americans (AA) from the Candidate gene Association Resource (CARe) study and race-combined meta-analyses that included 29,939 additional individuals of European descent from CARe, the Women’s Health Initiative (WHI) and KORA studies. We observed array-wide significance (p<2.2×10⁻⁶) for four loci in AA, three of which have been reported previously in individuals of European descent (IL6R, p=2.0×10⁻⁶; CRP, p=4.2×10⁻⁷¹; APOE, p=1.6×10⁻⁶). The fourth significant locus, CD36 (p=1.6×10⁻⁶), was observed at a functional variant (rs3211938) that is extremely rare in individuals of European descent. We replicated the CD36 finding (p=1.8×10⁻⁵) in an independent sample of 8041 AA women from WHI; a meta-analysis combining the CARe and WHI AA results at rs3211938 reached genome-wide significance (p=1.5×10⁻¹⁰). In the race-combined meta-analyses, 13 loci reached significance, including ten (CRP, TOMM40/APOE/APOC1, HNF1A, LEPR, GCKR, IL6R, IL1RN, NLRP3, HNF4A and BAZ1B/BCL7B) previously associated with CRP, and one (ARNTL) previously reported to be nominally associated with CRP. Two novel loci were also detected (RPS6KB1, p=2.0×10⁻⁶; CD36, p=1.4×10⁻⁶). These results highlight both shared and unique genetic risk factors for CRP in AA compared to populations of European descent.

INTRODUCTION

C-reactive protein (CRP) is a pentameric acute-phase protein that is a hallmark of low-grade systemic inflammation [1]. Vascular inflammation is thought to play a role in the development and progression of atherosclerosis, ultimately leading to plaque rupture and cardiovascular disease (CVD) events such as myocardial infarction [2]. Associations between CRP and CVD outcomes have been remarkably consistent despite varying study designs, target populations, and case classification methods [3]. Observational studies have shown increased levels of CRP to be present in individuals with factors such as older age [4], female sex [5], smoking [6], obesity [7], diabetes, atherosclerotic CVD [8,9], sleep curtailment [10] and sleep apnea[11]. Heritability estimates for CRP concentration range from 35–40% [12], supporting that genetic factors are likely to influence variation in CRP levels [3].

Multiple candidate gene and genome-wide association studies (GWAS) have been performed for CRP resulting in several reported associated loci. These associations include genes known to be involved in the regulation of inflammatory and metabolic pathways, some of which were not previously known to directly influence CRP levels. The Women’s Genome Health Study identified single nucleotide polymorphisms (SNPs) associated with CRP in the leptin-receptor gene (LEPR), glucokinase regulatory protein (GCKR), and hepatic transcription factor gene (HNF1A) [13]. Recently, a GWAS meta-analysis of participants of European ancestry confirmed association of previously identified loci with CRP and introduced 11 novel loci, including NLRP3, HNF4A, RORA, IRF1, and IL1F10 [14]. To our knowledge, only two published GWASs for CRP in individuals of African ancestry have been conducted [15,16], and the first was based on a relatively small cohort of individuals[14]. This earlier study identified several variants in the CRP gene that were associated with CRP, but no other loci were statistically significant [15]. The latter study, which included 8280 African American (AA) women from the Women’s Health Initiative (WHI) study, also identified a number of variants associated with CRP in the CRP gene as well as significant evidence for associations in or near IL1F10/IL1RN, TREM2, HNF1A and TOMM40/APOE [16].

We sought to extend what is known regarding the genetic underpinnings of CRP by performing multi-ethnic meta-analyses, including individuals of both African and European ancestry genotyped across a densely covered gene-based array. Participants for the primary analyses came from eight community-based cohorts from the Candidate Gene and Association Resource (CARe) consortium (AAs and European Americans [EAs]), WHI (EAs) and the Cooperative Health Research in the Region of Augsburg (KORA) study (Europeans). All participants had available genotype data from the ITMAT Broad-CARe (IBC) Chip, a custom 50,000 SNP gene-centric array having dense coverage of over 2000 candidate genes within CVD related pathways. An independent sample of AA participants from the WHI study with IBC chip data were used as a follow-up sample for interesting findings.

MATERIALS AND METHODS

Each study was reviewed by a local ethics board and all participants consented to genetic research. Genotype and phenotype data for all study participants, with the exception of KORA participants, are available through the NCBI dbGaP resource (www.ncbi.nlm.nih.gov/gap).

Study samples

CARe

The CARe (Candidate Gene Association Resource) consortium consists of nine studies. The purpose of the consortium was to bring together deeply-phenotyped prospective cohort studies to increase power for genetic association scans of CVD and other disorders [17]. Cohorts included in these analyses of CRP levels are: Atherosclerosis Risk in Communities (ARIC) (n=7572 EA; n=1983 AA), Coronary Artery Risk in Young Adults (CARDIA) (n=1318 EA; n=1118 AA), Cleveland Family Study (CFS) (n=281 EA; n=369 AA), the Cardiovascular Health Study (CHS) (n=3919 EA; n=736 AA), Framingham Heart Study (FHS) (n=7543 EA), Jackson Heart Study (JHS) (n=2026 AA), and Multi-Ethnic Study of Atherosclerosis (MESA) (n=2051 EA; n=1338 AA).

WHI

The Women’s Health Initiative (WHI) is one of the largest (n=161,808) studies of women's health ever undertaken in the U.S. [18]. A diverse population was recruited from 1993–1998 at 40 clinical centers across the U.S. A total of n=4389 EA WHI subjects with CRP measures were included in the current study.

KORA

The MONItoring of trends and determinants in Cardiovascular disease/ Cooperative Health Research in the Region of Augsburg (MONICA/KORA) study is a series of population-based surveys conducted in the region of Augsburg in Southern Germany [19]. The sample used in the current study consisted of n=2866 EA subjects with CRP measures selected from 1075 participants for KORA S12 and 1800 participants for KORA F3.

Further details of the participating CARe, WHI and KORA studies are reported in the Supplemental Materials.

IBC genotype array

The IBC SNP array is described in detail in Keating et al. [20]. The IBC SNP array includes 49,320 SNPs selected across ~2000 candidate loci for CVD. The array includes SNPs that capture patterns of genetic variation in both European- and African-descent populations. Genotyping for the CARe cohorts was performed at the Broad Institute (Cambridge, MA).

Quality control of genetic data

Criteria for DNA sample exclusion based on genotype data included sex mismatch, discordance among duplicate samples, or sample call rate <90%. Approximately 2.5% of CARe subjects initially included in genotyping efforts did not pass the required call rate. For each set of duplicates or monozygotic twins, data from the sample with the highest genotyping call rate were retained. SNPs were excluded when monomorphic, the call rate was <95%, or when significant departures from expected Hardy-Weinberg equilibrium (HWE) genotype proportions were observed (p<10⁻⁵ in EAs). Given the genetic admixture in AAs, there was no HWE filter used for these samples. After these exclusions were applied, data remained on 47,539 SNPs.

Data analysis

Participants with CRP measurement over 100 mg/L were excluded from analysis as these observations would be highly influential and potentially a result of acute infection. We natural log-transformed CRP level in order to generate an approximately normal distribution of model residuals, conditional on the covariates, to meet linear model assumptions. We assumed an additive genetic model in all tests of association. We used the linear regression model implemented in PLINK [21] for studies with unrelated individuals and the linear mixed effects model implemented in the program GWAF for cohorts with related individuals [22] to test for association between log-CRP and genotype at each SNP, adjusting for covariates. All models were stratified by cohort and race. Covariate adjustment was applied for age, sex, current smoking, body mass index and the first 10 principal components calculated using the program EIGENSTRAT [23] to control for potential population substructure. After obtaining cohort- and race-specific results, we performed a fixed-effects, inverse-variance-weighted meta-analysis using the METAL software [24]. Meta-analysis was performed separately by race and race-combined. Heterogeneity of effects across studies was tested in METAL using Cochran’s Q statistic.

Based on a simulation analysis performed, the effective number of independent tests was estimated to be 26,482 for the AAs and 20,544 for samples of primarily European ancestry after accounting for the linkage disequilibrium between markers on the IBC array [25]. To maintain an approximate overall type 1 error rate of 5%, a uniform statistical threshold of α = 2.2×10⁻⁶ (0.05/25,000) was recommended to declare array-wide (experiment-wide) significance [25].

Genotyping and imputation in WHI AA follow-up sample

Genome-wide genotyping for WHI AA participants was performed at Affymetrix using the Affymetrix 6.0 array. A total of 8421 AAs had genotype data that passed quality control. A reference sample of 761 AA NHLBI Exome Sequencing Project (ESP) participants was used for imputation of CD36 rs3211938 into 8041 AA WHI individuals with GWAS data using the programs MaCH 1.0.18 [26] and minimac [27]. Additional details on the genotype imputation are given in Auer et al. [28].

RESULTS

Characteristics of the 7570 AA and 29,939 European-ancestry study participants from CARe, WHI and KORA are reported in Table 1.

Table 1.

Study sample characteristics (% or mean±SD)

	ARIC		CARDIA		CFS		CHS		FHS	JHS	MESA		WHI	KORA
	EA	AA	EA	AA	EA	AA	EA	AA	EA	AA	EA	AA	EA	EU
N	7572	1983	1318	1118	281	369	3919	736	7543	2026	2051	1338	4389	2866
Female, %	53.9	64.1	53.2	59.2	53.0	58.0	56.3	62.5	53.8	60.6	52.0	53.7	100.0	48.9
Current Smoker, %	20.9	26.1	23.0	29.3	24.6	18.3	11.1	16.1	50.0	15.1	10.7	17.9	7.6	21.8
Diabetes, %	7.6	16.5	0.53	0.85	11.1	17.6	14.7	24.6	10.9	15.4	5.8	16.2	7.3	4.5
Age, y	54.1±5.7	52.9±5.7	25.7±3.3	24.4±3.8	44.3±19.3	40.4±18.6	72.8±5.6	73.0±5.7	48.8±13.7	50.0±12.0	62.2±10.1	61.7±9.8	68.2± 6.5	52.1±10.7
BMI, kg/m2	26.9±4.7	29.6±5.9	23.7±4.0	25.6±5.7	31.7±9.2	33.1±9.8	26.4±4.5	28.5±5.6	27.4±5.5	32.2±7.8	27.7±5.1	30.1±5.7	27.6 ±6.8	27.2±4.1
CRP, mg/L	4.0±5.6	5.8±7.1	2.4±3.7	4.3±6.3	3.8±5.2	4.6±6.2	4.3±6.7	6.1±8.2	3.3±6.2	5.1±7.7	3.4±5.1	4.7±6.9	4.2 ±6.1	2.8±5.2

EA=European American; AA=African American; EU=European

African Americans

Four loci reached IBC array-wide significance (p<2.2×10⁻⁶) in AAs (Table 2), including three loci reported in previous GWAS for CRP in individuals of European descent (CRP, IL6R, APOE). The fourth significant result, at rs3211938 in the gene encoding the cluster differentiation 36-membrane protein (CD36; p= 1.4×10⁻⁶), has not been reported previously for association with CRP. A locus zoom plot of the CD36 region is presented in Supplementary Figure 1, demonstrating the focused association at rs3211938. A Forest plot illustrating cohort specific results is presented in Supplementary Figure 2. We further tested this SNP using imputed genotype data in an independent sample of 8041 African Americans women from WHI. The association was confirmed, with the minor allele of CD36 rs3211938 associated with 0.128 (+/− 0.030) lower CRP levels (p=1.8×10⁻⁵). Together, in a combined meta-analysis of CARe and WHI AA participants, results at CD36 rs3211938 reached genome-wide significance (p =1.5×10⁻¹⁰). Of note, CD36 rs3211938 was not studied in the recent WHI AA GWAS [16]. There was no evidence for any heterogeneity in the results across the AA cohorts for any of the four significant loci (Table 2).

Table 2.

Loci associated with CRP in the African American samples (N=7570)

Chr.	No. SNPs P<2.2×10−6	Most Significant	Position	A1/A2	Freq.*	Beta(SE)*	p-value Het**	p-value	Gene
1q23	13	rs3093058	157951939	T/A	0.17	0.39(0.022)	0.69	4.2×10−71	CRP
1q21	2	rs8192284	152693594	A/C	0.14	−0.12(0.025)	0.60	2.0×10−6	IL6R
7q21	1	rs3211938	80138385	T/G	0.08	−0.14(0.030)	0.22	1.6×10−6	CD36
19q13	2	rs769450	50102284	G/A	0.37	0.083(0.017)	0.16	1.6×10−6	TOMM40

^*Allele frequency and beta estimate is presented for second allele (A2).

^**P-value from Cochran’s Q-test for heterogeneity of effects across cohorts.

Out of the five loci previously reported to be associated with CRP in AA women [16], we had good proxies for SNPs at CRP and TOMM40 but no available satisfactory proxies for the remaining gene regions (HNF1A, ILF10/IL1RN, and TREM2). Our top result occurred at CRP rs3093058 (p=4.2×10⁻⁷¹) (Table 2), which is in strong linkage disequilibrium (LD) with, and our best available proxy for (based on 1000 Genomes data in individuals of African ancestry (YRI)), rs16827466 (r²=0.89), the top variant reported by Reiner et al.[16]. Our significant result at APOE was for a SNP, rs769450 (p=2.0×10⁻⁶), which is in modest LD (r²=0.28) with the top SNP, rs1160985 in nearby TOMM40, reported by Reiner et al. Our best proxy SNP for rs1160985 (rs405509, r²=0.66) demonstrated nominal evidence for association with CRP (p=1.2×10⁻⁴). We had poor proxies available for the top SNPs reported for the other three significant loci in Reiner et al. Our best proxy for IL1F10/IL1RN rs6734238, rs17042795 (r²=0.28) demonstrated a trend towards association (p=0.068), as did our best proxy for HNF1A rs7979473 (rs1169293, r²=0.33, p=0.052). We did not find any evidence for association at our best proxy (rs6933067, r²=0.39, p=0.14) for the reported novel TREM2 rs7748513 association. When our analyses were restricted to AA women, rs3093058 was significant (p=9.4×10⁻⁴⁴), while rs769450 (p=0.0011), rs1169293 (p=0.027) and rs6933067 (p=0.029) were nominally significant. No evidence for association was observed for rs405509 (p=0.24) or rs17042795 (p=0.38).

Combined Race Meta-Analysis

We observed significant signals (p<2.2×10⁻⁶) at 13 loci, including seven loci widely reported to be associated with circulating CRP levels (CRP, TOMM40/APOE/APOC1, HNF1A, LEPR, GCKR, IL6R, and IL1RN) and three (BAZ1B/ BCL7B, NLRP3 and HNF4A) recently reported as significant by the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) consortium (Table 3)[14]. We also observed array-wide significance for ARNTL, which was reported by CHARGE to be “suggestive” (i.e. nominally associated with p<1.0×10⁻⁵ with CRP levels), and two novel loci, RPS6KB1 and CD36.

Table 3.

Loci associated with CRP in a meta-analysis of African Americans (AA) and participants of European (Eur) descent (n=37,509).

Chr.	Most Significant	A1/ A2	Freq. (Eur)*	Beta(SE) (Eur)*	p-value (Eur)	p-value Het (Eur)	Freq. (AA)*	Beta(SE) (AA)*	p-value (AA)	p-value Het(AA)	p-value (Meta)	p-value Het(Meta)	Gene
1p31	rs1805096	G/A	0.38	−0.098(0.009)	3.3×10−30	0.013	0.45	−0.061(0.017)	2.5×10−4	0.98	5.6×10−32	0.044	LEPR
1q21	rs4129267	C/T	0.40	−0.079(0.008)	5.2×10−21	0.32	0.13	−0.12(0.025)	5.7×10−7	0.52	8.0×10−26	0.28	IL6R
1q23	rs3091244	A/G	0.62	−0.17(0.009)	3.5×10−91	0.34	0.45	−0.24(0.017)	5.1×10−45	0.11	7.8×10−132	0.011	CRP
1q44	rs12239046	C/T	0.39	−0.039(0.009)	5.3×10−6	0.93	0.51	−0.064(0.017)	1.4×10−4	0.98	7.3×10−9	0.98	NLRP3
2p23	rs1260326	C/T	0.43	0.094(0.008)	5.3×10−29	0.57	0.14	0.056(0.024)	0.019	0.63	1.0×10−29	0.57	GCKR
2q13	rs4251961	T/C	0.38	0.060(0.009)	1.4×10−12	0.37	0.18	0.081(0.022)	2.0×10−4	0.055	1.8×10−15	0.12	IL1RN
7q11	rs714052	A/G	0.12	−0.065(0.013)	3.9×10−7	0.75	0.042	−0.056(0.044)	0.20	0.80	1.7×10−7	0.91	BAZ1B
7q21	rs3211938	T/G	<0.001	−0.42(0.54)	0.44	0.96	0.085	−0.14(0.030)	1.6×10−6	0.22	1.4×10−6	0.40	CD36
11p15	rs6486121	T/C	0.37	−0.043(0.009)	6.4×10−7	0.91	0.58	−0.037(0.017)	0.026	0.098	1.3×10−7	0.57	ARNTL
12q24	rs2244608	A/G	0.34	−0.11(0.009)	4.6×10−39	0.51	0.14	−0.076(0.024)	0.0016	0.65	9.5×10−41	0.54	HNF1A
17q23	rs1292034	G/A	0.45	−0.037(0.008)	6.0×10−6	0.043	0.82	−0.039(0.02)	0.079	0.53	1.2×10−6	0.13	RPS6KB1
19q13	rs2075650	A/G	0.13	−0.18(0.012)	2.2×10−47	0.59	0.13	0.039(0.02)	0.11	0.75	1.9×10−34	1.7×10−9	TOMM40
20q13	rs1800961	C/T	0.030	−0.13(0.024)	2.0×10−7	0.63	0.0070	−0.14(0.10)	0.18	0.82	7.8×10−8	0.85	HNF4A

^*Allele frequencies and beta estimates are presented for second allele (A2).

The direction and size of the effects were largely consistent between AAs and individuals of European descent for most of these 13 loci (Table 3). In some cases, the frequency of the minor allele was considerably different between the populations, which could largely explain the absence of significant evidence for association in one population or the other. One SNP, rs2075650 in TOMM40, had highly discrepant results between AAs and individuals of European descent (see Supplementary Figure 3). There was no evidence of association for the variant in AAs; in fact, there was a near trend for an effect in the opposite direction as was observed in participants of European descent. Both AAs and individuals of European descent had evidence for association at nearby SNP rs769450 (AAs see Table 2; Europeans, beta(SE) =0.033(0.009), p=1.0×10⁻⁴). The two variants are in weak LD (estimated r² in YRI = 0.088, r² in CEU = 0.13) in populations of African and European ancestry.

Analyses of European Americans not included in CHARGE consortium

A subset of the current study samples was also included in the CHARGE report, which included ~66,000 participants of European descent with CRP and genotype measurements [14]. Genotype data used in the CHARGE report were obtained across a variety of genome-wide marker platforms; genotype imputation was used to probabilistically infer missing genotype data for SNPs in the HapMap database. The current study, which includes 29,939 samples of European descent, has 21,900 overlapping European-ancestry samples with the CHARGE report (subjects of European descent from ARIC, CHS, FHS and KORA). Notably, no AA participants were included in the CHARGE report.

Focusing on the CHARGE non-overlap EA sample (EA subjects from CARDIA, CFS, MESA and WHI; n=8039), we identified significant evidence for six loci (CRP, TOMM40/APOE/APOC1, HNF1A, LEPR, GCKR, IL6R) widely recognized to be associated with CRP (Table 4). We observed evidence supporting two genes newly reported to be associated with CRP in the CHARGE report (HNF4A rs1800961 p=5.4×10⁻⁴; NLRP3 rs12239046 p=0.011), providing, to our knowledge, the first reported confirmation for these findings. We also observed indirect evidence, through a proxy SNP, supporting another novel CHARGE finding, rs13233571 in the BAZ1B/BCL7B gene cluster. We observed nominal evidence at BAZ1B rs714052 (p=7.6×10⁻³), which is in strong LD with the reported CHARGE BCL7B SNP (r²=0.93 in CEU HapMap samples). Additionally, we observed indirect evidence, through proxy SNP rs6486121 (p=0.0025), supporting an association at rs6486122 in ARNTL reported to be nominally significant in CHARGE. Interestingly, in this non-overlap EA sample, there was no evidence of association at RPS6KB1 rs1292034 (p=0.41), which we identified as a novel locus in our combined meta-analysis (see Supplementary Figure 4 for cohort specific results). Similar to AA results, there was no evidence for any heterogeneity in the results across the EA cohorts for any of the four significant loci in this non-overlap EA sample (Table 4).

Table 4.

Loci associated with CRP in the independent European American (n=8039) replication sample (CARDIA, CFS, MESA, WHI)

Chr.	No. SNPs P<2.2×10−6	Most Significant	Position	A1/A2	Freq.*	Beta(SE)*	het p-value	p-value	Gene
1p31	19	rs1805096	65874845	G/A	0.38	−0.14(0.017)	0.61	2.5×10−18	LEPR
1q21	5	rs4129267	152692888	C/T	0.40	−0.10(0.016)	0.69	6.1×10−10	IL6R
1q23	9	rs3091244	157951289	G/A	0.38	0.18(0.016)	0.61	2.9×10−28	CRP
2p23	3	rs1260326	27584444	C/T	0.43	0.085(0.016)	0.35	1.5×10−7	GCKR
12q24	5	rs2244608	119919810	A/G	0.31	−0.12(0.017)	0.38	1.2×10−14	HNF1A
19q13	6	rs12721046	50087459	G/A	0.13	0.18(0.023)	0.36	7.3×10−15	APOC1

^*Allele frequency and beta estimate is presented for second allele (A2).

DISCUSSION

We performed a dense candidate gene-based scan of approximately 50,000 SNPs covering approximately 2,000 gene regions in a bi-racial sample of 37,509 individuals, including 21,900 EA subjects previously studied in the CHARGE GWAS [14]. We observed significant evidence for seven loci widely reported to be associated with CRP, which include: CRP, TOMM40/APOE/APOC1, HNF1A, LEPR, IL6R, GCKR, and IL1RN. We also observed IBC array-wide significant evidence for association at HNF4A, NLRP3, and BAZ1B/BCL7B, loci that were reported to be significantly associated with CRP in the CHARGE GWAS [14]. Analyses based on an independent subset of European decent samples not included in the CHARGE report provide supporting evidence for these associations. We also observed significant evidence for association at ARNTL, a locus reported as suggestive in the CHARGE report. Finally, we identified two novel loci, CD36 and RPS6KB1, which have not been previously reported to be significantly associated with CRP. The CD36 association at rs3211938 is specific to AAs. We replicated this finding in an independent AA sample (n=8041) from WHI. A meta-analysis at rs3211938 including results from the CARe AA cohorts and WHI resulted in a genome-wide significant association. No evidence for the RPS6KB1 rs1292034 association was observed in the non-overlap CHARGE sample of European descent (beta= −0.013, p=0.41); the evidence for this association was considerably stronger in the subset of CHARGE European descent samples (beta= −0.047; p=1.7×10⁻⁶). The direction and size of effect in AA was similar to the CHARGE EA subset for rs1292034, but the association was not significant (beta= −0.039; p=0.079). Given that rs1292034 was not reported to be significantly associated with CRP in the complete CHARGE study, future studies will be necessary to replicate our association between rs1292034 and CRP.

The CD36 gene encodes a cellular receptor that facilitates fatty acid (FA) uptake and the utilization of key metabolic tissues [29,30]. Individuals with mutations in CD36 have a defective FA uptake, which could result in a poor metabolic profile and elevated serum lipid levels [30,31]. Patients with elevated lipid levels (i.e. LDL cholesterol) are more likely to develop atherosclerosis, which most broadly defines cardiovascular disease risk. Various scavenger receptors have been recognized for their role in mediating the uptake of oxidized LDL (ox-LDL) leading to the formation of foam cells that is a precursor step in the development of atherosclerotic lesions [32]. CD36 is a scavenger receptor involved in the uptake of these oxidized lipids and thus plays a role in the formation of atheroma [32]. Because the uptake of ox-LDL occurs on multiple cells types, (endothelial cells, macrophages, vascular smooth muscle cells, etc), the role of CD36 in atherogenesis is complex. A study using hyperlipidemic mice showed that the deletion of CD36 or SR-A (a type A scavenger receptor) decreases lesion formation but does not abrogate foam cell formation suggesting that scavenger receptors may not be absolutely necessary for this process [32,33]. Interestingly, an in vitro experiment showed that the addition of CRP to ox-LDL in a cell culture system stimulated foam cell formation suggesting that CRP may have an active role in enhancing foam cell formation [34]. Together, these findings suggest that CRP and CD36 may have a cooperative role in atherogenesis.

The influence of natural selection, linked to malaria susceptibility, has resulted in the high genetic variation of CD36 in populations of African descent [35,36]; as a result, CD36 mutations are ten times more frequent in African populations versus individuals of primarily European descent [37]. The link to malaria susceptibility results in the gene’s role as a receptor for Plasmodium falciparum infected erythrocytes, which is found in malaria patients. Variants in this gene have also been reported be associated with metabolic syndrome, [37] high density lipoprotein (HDL) cholesterol levels [17,38], and abnormal serum FA [39–41]. A study using AA participants from HyperGEN demonstrated that CD36 variants account for ~3.4% of inter-individual HDL variability in the study population [42]. The variant allele at rs3211938 creates a premature stop codon and thus a truncated form of the CD36 protein. This variant has only been identified in populations of African descent, having thought to arise as a result of positive selective pressure, and thus its predictive impact on CD36 expression would not apply to other populations [43,44]. In a recent study by Love-Gregory, et al., the minor allele (G) for rs3211938 was associated with increased HDL and reduced CD36 protein expression on monocytes [38]. The same coding allele (G) for rs3211938 was associated with lower CRP levels in our analysis, suggesting that CD36 SNPs may elucidate a biological link between CRP and HDL levels. A recent GWAS study on platelet count and mean platelet volume using 16,388 AA individuals from the Continental Origins and Genetic Epidemiology Network (COGENT) identified novel associations at two intronic SNPs at the CD36 gene [45]. Previous evidence has demonstrated that one of the variants (rs17154155) is associated with platelet function as well as CD36 expression on platelets [46,47]; rs17154155 is in LD with the rs3211938 (r²= 0.27). Further analysis on less common CD36 variants in AAs from the National Heart Lung and Blood Institute Exome Sequence Project found rs3211938 to be associated with lower platelet count in this population [48]. Taken together these results highlight the potential for CD36 variants to be used as a predictive tool for CVD risk in AAs [38].

A recent study from the CARe consortium, using the same AA participants included in this study, also found strong evidence for an association between HDL levels and CD36 rs3211938 [17]. HDL and CRP levels are modestly correlated across these cohorts (e.g. Spearman’s correlation for ARIC= −0.050 and JHS= −0.037). Inclusion of HDL as a covariate in the linear models for the CARe samples modestly impacted the association between CRP and rs3211938 (p = 1.4×10⁻⁶ for the model unadjusted for HDL vs. p =5.4×10⁻⁶ for the HDL-adjusted model), suggesting that the CD36 association with CRP levels is largely independent of HDL.

We also found evidence at ARNTL, also termed BMAL1, which is a core component of the circadian clock and a vital element of the central circadian pacemaker [49]. This locus had suggestive evidence for association in the CHARGE report. Molecular circadian clocks exist in peripheral tissues and coordinate gene transcription involved in a wide range of metabolic processes including gluconeogenesis, lipolysis, adipogenesis, and mitochondrial oxidative phosporylation to achieve an appropriate internal alignment of metabolic signaling as well as external alignment of cellular processes [50]. It has been well documented that pathologic events display circadian rhythms with an increase in incidences, such as myocardial infarction and ischemia, from dawn to noon [51,52]. Furthermore, experimental circadian misalignment is associated with abnormalities in blood pressure, glucose, and insulin levels [53] while shift work, a real-world model for circadian misalignment, is associated with diabetes and CVD [54,55]. Polymorphisms of ARNTL have also been associated with age at menarche [56–58]. Inactivation of ARNTL results in altered regulation of blood pressure, lipid metabolism, and glucose homeostasis; these changes have been observed in hypertensive mouse models [59–63]. The association between ARNTL and CRP suggests one possible mechanism of correlation between CRP and metabolic dysregulation, thereby increasing CVD risk.

Strengths of this study include the large AA and European-ancestry samples, with measured CRP levels, that have been genotyped on a large commercial candidate-gene-based genotyping panel. The generalizability to other populations remains to be determined. SNPs identified may not be causally linked to variation in CRP, rather they may be in linkage disequilibrium with the causal variants. In addition, the functional mechanisms linking genetic variants to CRP concentrations remain to be determined. Though we did not observe much evidence for heterogeneity of effect estimates across studies among our top findings, as with any meta-analyses, systemic differences between studies, including different technologies used to measure CRP levels, can lead to reduced power and biased aggregate effect size estimates. Finally, we had limited power to detect low frequency variants and SNPs with low effect size, especially in AAs. Additional genomic loci may be uncovered in larger samples and with broader coverage of genetic variation across the human genome.

Our findings provide increased insight about genetic variants influencing variation in CRP concentrations, including the identification of a newly reported African-specific CD36 variant associated with CRP level and the confirmation of three CRP loci identified in a recent GWAS. Overall, these findings are consistent with the role of metabolism and inflammatory pathway in the regulation of circulating CRP levels.