Small Dense LDL Cholesterol Concentrations Predict Risk for Coronary Heart Disease: the Atherosclerosis Risk in Communities (ARIC) Study

Abstract

Objective

To investigate the relationship between plasma levels of small dense low-density lipoprotein cholesterol (sdLDL-C) and risk for incident coronary heart disease (CHD) in a prospective study among Atherosclerosis Risk in Communities (ARIC) study participants.

Approach and Results

Plasma sdLDL-C was measured in 11,419 men and women of the biracial ARIC study using a newly developed homogeneous assay. A proportional hazards model was used to examine the relationship between sdLDL-C, vascular risk factors, and risk for CHD events (n=1,158) over a period of ≈11 years. Plasma sdLDL-C levels were strongly correlated with an atherogenic lipid profile and were higher in diabetics than nondiabetics (49.6 vs. 42.3 mg/dL, p<0.0001). In a model that included established risk factors, sdLDL-C was associated with incident CHD with a hazard ratio (HR) of 1.51 (95%CI: 1.21–1.88) for the highest versus the lowest quartile, respectively. Even in individuals considered to be at low cardiovascular risk based on their LDL-C levels, sdLDL-C predicted risk for incident CHD (HR 1.61; 95% CI 1.04–2.49). Genome-wide association analyses identified genetic variants in 8 loci associated with sdLDL-C levels. These loci were in or close to genes previously associated with risk for CHD. We discovered 1 novel locus, PCSK7, for which genetic variation was significantly associated with sdLDL-C and other lipid factors.

Conclusions

sdLDL-C was associated with incident CHD in ARIC study participants. The novel association of genetic variants in PCSK7 with sdLDL-C and other lipid traits may provide new insights into the role of this gene in lipid metabolism.

Low-density lipoprotein cholesterol (LDL-C) is considered one of the most important risk factors for cardiovascular disease and remains the primary target for current cardiovascular risk reduction strategies.^1,2 However many individuals with LDL-C within the normal range still develop cardiovascular disease.^3,4 LDL particles are a heterogeneous population,⁵ and it has long been hypothesized that a subfraction of LDL, particularly small dense LDL (sdLDL), possesses increased atherogenic potential and thus contributes to this observation. A number of mechanisms have been proposed to explain the enhanced atherogenicity of sdLDL,^6–10 including 1) a lower affinity for the LDL receptor, 2) facilitated entry into the arterial wall, 3) greater arterial retention due to increased binding to proteoglycans, and 4) greater susceptibility to oxidation. Since sdLDL particles are smaller and contain less cholesterol, increased levels of sdLDL also represent an increased number of atherogenic particles, which may not be reflected by the levels of LDL-C.

Two of the earliest and most widely used methods for LDL classification involved density and size determinations based upon ultracentrifugal and/or nondenaturing gradient density gel electrophoresis procedures. These resulted in the division of LDL particles somewhat arbitrarily into two categories for clinical assessment: sdLDL and large buoyant LDL (lbLDL). These also led to the development of a two phenotype classification system, with phenotype A (or pattern A) characterized as individuals with a predominance of lbLDL particles and phenotype B (or pattern B) as individuals with a predominance of sdLDL particles.¹¹ This classification schema has been widely used, and pattern B has been recognized as a risk marker for cardiovascular disease. More recently, the efficacy of nuclear magnetic resonance methodology to determine both particle numbers and sizes of various lipoprotein fractions, including LDL, has been demonstrated.¹²

The distribution of LDL subfractions is determined by both genetic and environmental factors.^13,14 Furthermore, the concentration of sdLDL is highly correlated with triglyceride level and is increased in individuals with diabetes or the metabolic syndrome.¹⁵ Therefore, it is plausible that genetic variants that affect circulating levels of sdLDL-C may also influence other lipid traits (e.g. triglycerides and HDL-C) and may aid in the identification of genes involved in the causal pathway linking atherogenic dyslipidemia characterized by sdLDL-C and CHD.

sdLDL has been found to be associated with increased risk for cardiovascular disease in cross-sectional studies^16–18 as well as prospective observational studies.^19–21 However, in most of these studies, sdLDL did not remain an independent risk predictor when adjusted for other lipid risk factor traits. Until recently, the methods available for the measurement of sdLDL were generally limited to nonquantitative, laborious, and/or highly complex techniques and therefore not readily adaptable to a large number of samples in a routine clinical laboratory environment. Recently, Ito and coworkers developed a simple homogeneous assay adaptable to autoanalyzers for the quantification of sdLDL-C.²² Thus far, few epidemiological studies have examined whether the cholesterol content of sdLDL can predict future cardiovascular events. In the present study, we measured sdLDL-C in 11,419 men and women of the Atherosclerosis Risk in Communities (ARIC) study. These participants were followed over a period of 11 years during which the incidence of coronary heart disease (CHD) was measured. The purpose of this study was to evaluate whether sdLDL-C is a better predictor of risk for CHD than LDL-C and other traditional or nontraditional cardiovascular risk factors. To better understand the genetic determinants of sdLDL-C and lbLDL-C, we investigated the association of both sdLDL-C and lbLDL-C levels with genetic markers spanning the genome.

Materials and methods

Materials and Methods are available in the online-only Data Supplement.

Results

Baseline Characteristics

Key baseline (visit 4) demographics of the 11,419 ARIC participants are described in Table 1. The mean age of the study cohort was 62.8 years; 78% of study participants were Caucasian, and 56% were female. Of the study participants, 58% had a history of smoking cigarettes; 16.9% had diabetes mellitus, and 44.6% were classified with metabolic syndrome according to Adult Treatment Panel III criteria.²³ Prevalent CHD was present in 972 individuals at the baseline visit, and these individuals were excluded from analyses for incident events. In the remaining 10,225 individuals (excluding 222 who had missing data on incident CHD), incident CHD developed in 1158 participants over an average of 11 years of follow-up. The mean plasma sdLDL-C level was 43.5 mg/dL, and the mean sdLDL-C/LDL-C ratio was 0.35. Supplemental Table I (available online at http://atvb.ahajournals.org) shows the race- and gender-specific plasma lipid levels. Mean sdLDL-C levels were higher in Caucasians than African Americans and in men than women, whereas LDL-C was slightly higher in women than men but was not different between races. The proportion of LDL-C which was sdLDL-C was higher in Caucasians than African Americans and in men than women. Total cholesterol and high-density lipoprotein cholesterol (HDL-C) levels were higher in women than men, while triglyceride levels were higher in Caucasians than African Americans.

Table 1.

Description of 11,419 ARIC participants at baseline (ARIC visit 4)

Characteristic	N=11,419
Age, years	62.83±5.67
Race
African American	2,515 (22)
Caucasian	8,904 (78)
Sex
Female	6,385 (56)
Male	5,034 (44)
Ever smoked cigarettes	6,635 (58)
Ever drank alcohol	8,989 (79)
Statin use	1,308 (11.5)
Prevalent CHD	972 (8.7)
Diabetes mellitus	1943 (16.9)
Metabolic syndrome	5127 (44.6)
sdLDL-C, mg/dL	43.48±20.76
lbLDL-C, mg/dL	79.22±28.52
sdLDL-C/LDL-C	0.35±0.1496

Data are presented as mean±SD or n (%). CHD: coronary heart disease; lbLDL-C: large buoyant low-density lipoprotein cholesterol; sdLDL-C: small dense low-density lipoprotein cholesterol; LDL-C: low-density lipoprotein cholesterol.

Association of sdLDL-C with Cardiovascular Risk Factors

Table 2 shows the means or proportions of baseline traditional risk factors and other characteristics by sdLDL-C quartiles. Generally, individuals with sdLDL-C levels in the highest quartile had proatherogenic lipid profiles, were more likely to have diabetes, hypertension, and metabolic syndrome, and had higher body mass index (BMI) and plasma high-sensitivity C-reactive protein (hs-CRP) levels. Statin use was higher in those individuals with sdLDL-C levels in the 3^rd and 4^th quartile. sdLDL-C levels were not associated with smoking status.

Table 2.

Adjusted means±SE or proportions±SE of cardiovascular risk factors by quartiles of sdLDL-C and p-value for trend across quartiles, adjusted for age, race, and gender.*

Characteristic	Q1	Q2	Q3	Q4	P-trend
Age, years*	62.77±0.106	62.95±0.106	62.80±0.106	62.79±0.106	0.8500
Male, %*	0.40±0.038	0.42±0.038	0.49±0.038	0.45±0.038	<.0001
African American, %*	0.30±0.041	0.23±0.044	0.19±0.048	0.14±0.055	<.0001
Total cholesterol, mg/dL	172.08±0.557	194.61±0.556	207.09±0.558	229.94±0.559	<.0001
Triglycerides, mg/dL	91.70±1.370	116.20±1.368	154.69±1.373	213.49±1.374	<.0001
LDL-C, mg/dL	97.55±0.537	119.30±0.537	129.36±0.542	145.64±0.551	<.0001
HDL-C, mg/dL	56.16±0.271	52.37±0.271	47.85±0.272	43.47±0.272	<.0001
BMI, kg/m2	27.80±0.103	28.46±0.103	29.20±0.103	29.79±0.103	<.0001
Smoking, %	59±4.0	62±4.0	57±3.9	58±4.0	0.1125
Diabetes, %	11.0±5.8	13.0±5.6	17.0±5.1	23.0±4.6	<.0001
Hypertensive, %	40.0±4.0	46.0±3.9	51.0±3.9	54.0±3.9	<.0001
Statin use, %	9.0±6.7	11.0±6.0	12.0±5.8	12.0±5.7	<.0001
Metabolic syndrome, %	23.0±4.5	33.0±4.0	50.0±3.8	73.0±4.3	<.0001
log hs-CRP	0.75±0.020	0.84±0.020	0.94±0.020	1.05±0.021	<.0001

BMI: body mass index; HDL-C: high-density lipoprotein cholesterol; hs-CRP: high-sensitivity C-reactive protein; LDL-C: low-density lipoprotein cholesterol.

The correlation between sdLDL-C levels and various traditional and nontraditional cardiovascular risk factors are shown in Table 3. Strong positive correlations with sdLDL-C (|r|>0.50) were found for the lipid risk factors non-HDL-C, apolipoprotein (apo) B, LDL-C, total cholesterol, and log triglycerides. Circulating levels of lipoprotein-associated phospholipase A₂ (Lp-PLA₂) activity and lactate showed moderate positive correlations and HDL-C showed a moderate negative correlation with sdLDL-C (0.25<|r|<0.50). Weaker correlations with sdLDL-C were found for fasting plasma glucose, apo AI, lbLDL-C, and log hs-CRP. High-sensitivity cardiac troponin T was not significantly correlated with sdLDL-C.

Table 3.

Correlation of sdLDL-C with traditional and novel cardiovascular risk factors

Risk Factor	n	Pearson R	P-value
Non-HDL-C	11,419	0.721	<.0001
ApoB	10,720	0.706	<.0001
Log triglycerides	11,419	0.641	<.0001
Total cholesterol	11,419	0.608	<.0001
LDL-C	11,234	0.543	<.0001
Lp-PLA2 activity	11,108	0.319	<.0001
HDL-C	11,419	−0.291	<.0001
Lactate	11,417	0.253	<.0001
Fasting plasma glucose	10,902	0.169	<.0001
ApoAI	10,720	−0.086	<.0001
lbLDL-C	11,234	−0.076	<.0001
Log hs-CRP	11,202	0.072	<.0001
hs-cTroponin T	11,130	−0.013	0.1549

apo: apolipoprotein; HDL-C: high-density lipoprotein cholesterol; hs-CRP: high-sensitivity C-reactive protein; hs-cTroponin T: high-sensitivity cardiac troponin T; lbLDL-C: large buoyant low-density lipoprotein cholesterol; LDL-C: low-density lipoprotein cholesterol; Lp-PLA₂ lipoprotein-associated phospholipase A₂.

sdLDL-C Levels and Incident CHD Events

The cumulative incidence curves for CHD risk by sdLDL-C and lbLDL-C quartiles, adjusted by age, race, and gender, are shown in Figure 1. The incidence of CHD events increased proportionately over the follow-up years for participants in each consecutive quartile of sdLDL-C. In contrast, lbLDL-C did not exhibit a concentration-dependent relationship with future incident CHD events. sdLDL-C and LDL-C levels were moderately correlated (Supplemental Figure IA; r=0.54; available online at http://atvb.ahajournals.org) but often discordant. Supplemental Figure IB (available online at http://atvb.ahajournals.org) displays the prevalence and magnitude of this discordance. We examined concordant and discordant subgroups separately using a similar analysis approach as previously described by Otvos et al.²⁴ We defined discordance as a difference of >24 percentile units (points outside the dashed lines in Supplemental Figure IB; available online at http://atvb.ahajournals.org). Supplemental Figure II (available online at http://atvb.ahajournals.org) shows the cumulative incidence curves for CHD risk for the subgroup with sdLDL-C>LDL-C discordance compared to the concordant and the discordant sdLDL-C<LDL-C subgroups. The sdLDL-C>LDL-C discordant subgroup showed the highest CHD risk compared to the concordant and discordant sdLDL-C<LDL-C subgroups.

Figure 1.

Cumulative incidence curves for risk of CHD by sdLDL-C and lbLDL-C quartiles, adjusted for age, race, and gender.

We used proportional hazards regression analyses to investigate the association of incident CHD with baseline levels of sdLDL-C and LDL-C modeled in quartiles, using quartile 1 as the referent group (Table 4). In the basic model adjusted for age, race, and gender (model 1), individuals with sdLDL-C levels in the highest quartile had a 2-fold higher risk for incident CHD compared to those in the lowest quartile (hazard ratio [HR] 2.00; 95% confidence interval [CI] 1.69–2.37). After additional adjustment for smoking, BMI, hypertension, HDL-C, log triglycerides, lipid-lowering medications, diabetes, diabetes medications, and log hs-CRP (model 2), risk for incident CHD was attenuated but remained significant (HR 1.51; 95% CI 1.21–1.88). sdLDL-C was not significantly associated with risk for incident CHD following further adjustment for other lipid risk factors, such as LDL-C, apo B, and total cholesterol. This may be due in part to over adjustment of the multivariable model and is not surprising given the strong correlations of sdLDL-C with these lipid risk factors. In comparison, individuals with LDL-C levels in the highest quartile had a 56% and 68% higher risk for incident CHD (HR 1.56; 95% CI 1.32–1.83 and HR 1.68; 95% CI 1.42–1.99) in the basic model (model 1) and more fully adjusted model (model 2), respectively.

Table 4.

Hazard ratio (95% confidence interval) for incident CHD by sdLDL-C and LDL-C quartiles

	Quartile of sdLDL-C*
	2	3	4	P†
Model 1‡	1.19 (0.99–1.43)	1.44 (1.21–1.72)	2.00 (1.69–2.37)	<0.0001
Model 2§	1.10 (0.90–1.33)	1.21 (0.99–1.48)	1.51 (1.21–1.88)	0.0008
	Quartile of LDL-C*
	2	3	4	P†
Model 1‡	1.01 (0.85–1.21)	1.15 (0.97–1.37)	1.56 (1.32–1.83)	<0.0001
Model 2§	1.08 (0.90–1.30)	1.23 (1.03–1.47)	1.68 (1.42–1.99)	<0.0001

sdLDL-C: small dense low-density lipoprotein cholesterol; LDL-C: low-density lipoprotein cholesterol.

^*Lowest quartile (1) is reference.

^†P-values (Pr > chiSq) for linear hypothesis testing results of sdLDL-C quartiles.

^‡Adjusted for age, sex, and race.

^§Adjusted for model 1 variables + smoking, body mass index, hypertension, high-density lipoprotein cholesterol, log(triglycerides), lipid-lowering medications, diabetes, diabetes medications, and log(hs-CRP).

We further investigated the association of sdLDL-C with risk for incident CHD in participants stratified by LDL-C risk categories (i.e., LDL-C <100 mg/dL and LDL-C ≥100 mg/dL). In these analyses, we used a multivariable model (adjusting for age, gender, race, ever smoking, BMI, hypertension, diabetes, diabetes medication, and log hs-CRP), with quartile 1 for sdLDL-C in the LDL-C <100 mg/dL risk category as the referent group. Even in individuals with LDL-C levels <100 mg/dL, sdLDL-C was predictive of CHD risk across increasing sdLDL-C quartiles (Figure 2). Participants with LDL-C <100 mg/dL and sdLDL-C levels in the 4^th quartile had a 61% increase in risk for incident CHD (HR 1.61; 95% CI 1.04–2.49) compared with individuals with sdLDL-C levels in the 1^st quartile. In comparison, participants with LDL-C ≥100 mg/dL and sdLDL-C levels in the 4^th quartile had an 86% increase in risk for incident CHD (HR 1.86; 95% CI 1.48–2.33) compared with the same referent group.

Figure 2.

Adjusted hazard ratios for incident CHD by sdLDL-C quartiles stratified by LDL-C risk categories, adjusted for age, sex, and race, smoking, BMI, hypertension, diabetes, diabetes medications, and log hs-CRP.

In additional analyses, we examined the effects on CHD risk of sdLDL-C discordance among ARIC participants with low LDL-C (<100 mg/dL; <25^th percentile) or equivalently low sdLDL-C (<27.8 mg/dL; <25^th percentile) (Figure 3). The cumulative incidence of CHD events was higher among individuals with low LDL-C but discordantly higher sdLDL-C (10.9%) compared to individuals with low sdLDL-C but discordantly higher LDL-C (7.9%). Not surprisingly, the cumulative incidence of CHD events was highest among individuals with concordantly higher levels of LDL-C and sdLDL-C (12.7%) and lowest among individuals with concordantly lower levels of LDL-C and sdLDL-C (7.6%).

Figure 3.

Cumulative incidence of cardiovascular events in sub-groups with LDL-C <100 mg/dL (<25^th percentile) and/or sdLDL-C <27.8 mg/dL (<25^th percentile), from proportional hazards models adjusted for age, gender, and race.

Genome-Wide Association Study of sdLDL-C

We performed genome-wide association study (GWAS) analyses of sdLDL-C and Table 5 summarizes our primary findings. In total, 127 single-nucleotide polymorphisms (SNPs) were significantly associated with sdLDL-C (p<5×10⁻⁸). These SNPs clustered at 8 different loci on chromosomes 1, 2, 7, 8, 11, and 19 and were located within 14 different genes (or gene clusters). With the exception of PCSK7, genetic variants within all of these genes have previously been found to be related to pathways involved in lipid metabolism and vascular inflammation (www.genome.gov).

Table 5.

Association of the top SNPs with sdLDL-C

sdLDL-C
SNP	Location	Chromosome	# Sign SNPs	Coded Allele	Allele Frequency	N	Beta	SEBeta	P	Gene
rs964184	11q23.3	11	12	C	0.8588	6979	−5.70008	0.511384	7.46×10−29	APOA5/A4/C3/A1
rs4420638	19q13.32	19	6	A	0.8264	6979	−5.45295	0.490271	9.77×10−29	APOE/C1/C4/C2
rs2075650	19q13.32	19	3	A	0.8611	6979	−7.15912	0.748784	1.17×10−21	TOMM40
rs660240	1p13.3	1	10	C	0.7873	6979	3.45306	0.432418	1.40×10−15	PSRC1/CELSR2/SORT1
rs6589564	11q23.3	11	18	C	0.0723	6979	5.53037	0.700672	2.95×10−15	BUD13
rs2075290	11q23.3	11	5	C	0.0716	6979	5.48955	0.701319	4.98×10−15	ZNF259
rs2980853	8q24.13	8	38	A	0.5315	6979	2.63962	0.353669	8.42×10−14	TRIB1
rs1260326	2p23.3	2	3	C	0.5856	6979	−2.60009	0.366214	1.25×10−12	GCKR
rs7254892	19q13.32	19	1	A	0.0534	6979	−10.9757	1.78507	7.82×10−10	PVRL2
rs562338	2p23–2p24	2	24	A	0.1811	6979	−2.65678	0.464554	1.07×10−08	APOB
rs508487	11q23-q24	11	1	C	0.9402	6979	−4.95643	0.868883	1.17×10−08	PCSK7
rs4803760	19p13.2	19	1	C	0.7692	6979	2.73614	0.49648	3.57×10−08	BCAM
rs1178977	7q11.23	7	3	A	0.8000	6979	2.55226	0.466442	4.46×10−08	MLXIPL
rs6976930	7q11.23	7	2	A	0.2001	6979	−2.54050	0.464764	4.60×10−08	BAZ1B

Association of PCSK7 SNP rs508487 Genotype with Circulating Lipids and CHD

A novel finding from the current GWAS analysis was the significant association of sdLDL-C with SNP rs508487 (at locus 11q23-q24) located in the PCSK7 gene. We investigated the effect of rs508487 genotype on circulating lipid levels (Table 6). Each copy of the minor allele at this SNP raised sdLDL-C by ~4 mg/dL and triglycerides by ~20 mg/dL. In contrast, each copy of the minor allele lowered lbLDL-C by ~3 mg/dL; rs508487 genotype had no significant effect on circulating LDL-C levels. Interestingly, 2 copies of the minor allele increased HDL-C by ~5 mg/dL and total cholesterol by ~14 mg/dL.

Table 6.

Association of PCSK7 SNP rs508487 genotype with circulating lipids

	Genotype
Lipid, mg/dL (mean±SE)	CC	CT	TT	P
sdLDL-C	44.3±20.6	48.5±21.4	51.5±23.4	<0.0001
lbLDL-C	78.7±27.5	75.5±29.7	72.7±23.6	0.02
sdLDL-C/LDL-C	0.36±0.15	0.40±0.16	0.41±0.13	<0.0001
LDL-C	123.0±32.8	124.0±32.6	124.2±33.8	0.75
HDL-C	50.2±16.4	48.6±16.1	54.9±13.4	0.03
Total cholesterol	201.5±35.6	205.0±35.0	215.3±40.4	0.02
Triglycerides	141.5±67.5	162.1±77.5	181.4±88.5	<0.0001

HDL-C: high-density lipoprotein cholesterol; lbLDL-C: large buoyant low-density lipoprotein cholesterol; sdLDL-C: small dense low-density lipoprotein cholesterol; LDL-C: low-density lipoprotein cholesterol.

Given that the PCSK7 variant (rs508487) is located in the chromosome 11 region harboring the APOA5-APOA4-APOC3-APOA1 gene cluster, we investigated whether the novel association result with PCSK7 is due to linkage disequilibrium with previously reported variants in this cluster, especially previously reported functional variants in APOA5. One variant, rs662799 (APOA5 T-1131C) was in significant linkage disequilibrium with rs508487 (R²=0.53). We repeated the association analysis between PCSK7 rs508487 and sdLDL-C in 6,069 individuals homozygous for the wild type allele at APOA5 rs662799, and the results were attenuated a bit but still nominally statistically significant (p=0.0027).

We investigated the relationship between PCSK7 SNP rs508487 genotype and CHD in the ARIC study and did not observe a significant association. However, the power for this particular analysis was limited due to the fact that the number of ARIC CHD cases with one or two minor alleles at this SNP was low. Therefore, we examined the association of rs508487 with 40,260 CHD cases from the CARDIoGRAM study and found that rs508487 was significantly associated with CHD (Odds Ratio=1.13; 95% confidence interval, 1.06–1.21, P=0.00017).

We next analyzed rare variants on the Illumina exome chip designated as nonsynonymous, splicing, or stop gain in the PCSK7 gene for association with sdLDL-C. We found a total of 7 nonsynonymous PCSK7 variants among Caucasian ARIC participants, resulting in amino acid substitutions in the wild type PCSK7 protein. Since these variants had minor allele frequencies below 1%, we analyzed them collectively for their effect on sdLDL-C. Individuals who were carriers of any of the rare PCSK7 variants had a significant increase in circulating levels of sdLDL-C (~ 7.5 mg/dL; P=0.012) and triglycerides (ln(TG) ~0.145; P=0.043) compared to non-carriers (Table 7).

Table 7.

Change in circulating lipids among Caucasian carriers of rare nonsynonymous PCSK7 variants

Lipid	Change (β)	SE	P
sdLDL-C (mg/dL)	7.53	3.00	0.012
Ln (triglycerides)	0.144	0.071	0.043
HDL-C (mg/dL)	−3.58	2.12	0.091

sdLDL-C: small dense low-density lipoprotein cholesterol; HDL-C: high-density lipoprotein cholesterol.

Discussion

In the current study, we investigated the relationship between plasma levels of sdLDL-C and risk of incident CHD in the predominantly biracial ARIC study cohort using a newly developed automated homogeneous sdLDL-C assay. Elevated plasma sdLDL-C levels were associated with increased risk of incident CHD in a multivariable model (hazard ratio (HR) 1.51; 95%CI: 1.21–1.88) and even in individuals considered to be at low cardiovascular risk based on their LDL-C levels, sd-LDL-C predicted risk for incident CHD (HR 1.61; 95% CI 1.04–2.49). Using genome-wide association (GWA) analyses, we discovered one novel locus, PCSK7, for which genetic variation was significantly associated with sdLDL-C levels and other lipid traits. Subsequent examination in the CARDIoGRAM Study showed a significant association of the PCSK7 SNP rs508487 with CHD.

sdLDL-C and Risk for Incident CHD

Among ARIC participants, the mean baseline plasma sdLDL-C level was 43.5 mg/dL, which represented on average 35% of the total LDL-C concentration. sdLDL-C and sdLDL-C/LDL-C ratio were higher in Caucasians than African -Americans and higher in men than women. Our findings related to circulating sdLDL-C levels seem in general agreement with a report from the Framingham Offspring Study, which showed that men had higher sdLDL-C levels and a higher percentage of LDL-C as sdLDL-C compared to women.²⁵ Although we found higher mean sdLDL-C and sdLDL-C/LDL-C ratio overall than those reported in the Framingham Offspring Study, these differences may be in part due to differences in study populations (e.g., the ARIC cohort has a higher prevalence of obesity and metabolic syndrome than the Framingham Offspring Study) and sdLDL-C assay methodologies.

Plasma levels of sdLDL-C were adversely associated with cardiovascular lipid risk factors, a finding consistent with previous reports showing a correlation of sdLDL with an atherogenic lipid profile. We also found significant correlations between sdLDL-C and nonlipid risk factors, such as fasting glucose and lactate levels. Even though sdLDL-C was associated with diabetes and metabolic syndrome in previous studies, we report a remarkable increase in prevalence of metabolic syndrome among individuals with sdLDL-C levels in the highest quartile (73%) compared with those in the lowest quartile (23%). sdLDL-C was also correlated with inflammatory markers such as Lp-PLA₂ activity and hs-CRP.

Over the 11-year follow-up period of this study, 1158 (11.3%) participants developed CHD. The cumulative incidence curves clearly illustrate the direct relation between sdLDL-C levels and CHD risk, whereas we did not find a similar relation between lbLDL-C and CHD. These results suggest that the sdLDL subfraction is a major contributor to the risk for incident CHD that is associated with LDL-C. Circulating levels of sdLDL-C were significantly associated with increased risk for CHD in a model adjusted for age, sex, and race and in a more fully adjusted model that also included smoking, BMI, hypertension, HDL-C, triglycerides, lipid-lowering medications, diabetes, diabetes medications, and hs-CRP. However, sdLDL-C was not an independent predictor of incident CHD when we further adjusted for other lipid risk factors, such as LDL-C, apo B, and total cholesterol, which is not surprising given the strong correlations of sdLDL-C with these other lipid risk factors, and our results are in agreement with previous studies reporting that sdLDL was not an independent predictor of cardiovascular disease.^19–21 Interestingly, sdLDL-C showed predictive power for CHD risk even in individuals with optimal LDL-C levels as defined in the current guidelines (<100 mg/dL).²³ Several investigators have emphasized that the number of particles as measured by nuclear magnetic resonance is more important for assessment of cardiovascular risk than LDL subclass, LDL particle size, or LDL-C concentration.^26,27 Since sdLDL particles contain less cholesterol than lbLDL particles, there are more sdLDL particles than lbLDL particles at a given LDL-C concentration. Whether the total number of particles or the cholesterol payload per particle is more important to cardiovascular risk remains a topic of discussion. However, it is important to note that the particle number theory does not take into account the accumulating evidence pointing to different atherogenic properties of LDL subclasses. A limitation of this study is the fact that we did not have particle number information available and thus, we were not able to address this question specifically.

Genetics of sdLDL-C

GWAS analysis identified a large number of SNPs, clustered at 8 different loci on chromosomes 1, 2, 7, 8, 11 and 19 that were significantly associated with sdLDL-C. With the exception of PCSK7, genetic variants located in all the genes associated with sdLDL-C levels have been reported previously in GWAS studies of blood lipid levels.²⁸ Our GWAS findings are in general agreement with an earlier report by Chasman and coworkers who employed a comprehensive GWAS analysis to identify largely similar loci that affect the nuclear magnetic resonance–based measures of concentration and size of LDL, HDL and VLDL in women.²⁹ Although the study by Chasman et al did not find a significant association with LDL particle size or concentration at the PCSK7 locus, this apparent discrepancy may be due to notable differences between the two studies, such as differences in methodologies to measure lipoprotein fractions, genotyping methods and study populations. Indeed, a recent report from the Multi-Ethnic Study of Atherosclerosis (MESA) compared the identical automated assay of sdLDL-C as was used in our study to nuclear magnetic resonance-derived small LDL concentrations with regards to risk prediction for incident CHD.³⁰ The authors showed that the new automated assay of sdLDL-C identified risk of CHD whereas the nuclear magnetic resonance-derived small LDL concentrations did not convey a significant risk of CHD in the MESA cohort. Therefore, if these two different methodologies show different associations with cardiovascular risk it is plausible that they may also lead to different GWAS findings.

Genetic variants within a number of the genes associated with sdLDL-C in our study have previously been found to be associated with increased risk for cardiovascular disease in meta-analyses of GWA studies.^31,32 SNP rs4420638, which is located in the APOE-APOC1-APOC4-APOC2 gene cluster, was also associated with Lp-PLA₂ activity and CHD in a meta-analysis of GWA studies from 5 community-based studies.³³ In addition, we have previously shown an association of the SNP rs780094 in GCKR with metabolic syndrome prevalence and incident diabetes in the ARIC study.³⁴ rs780094 was also significantly associated with sdLDL-C (p=4.08×10⁻¹²) in our current study.

Our findings have important implications in light of recent observations from mendelian randomization studies investigating genetic determinants of HDL-C levels and risk for incident CHD. Unlike data from human mendelian diseases, which support a causal role for LDL-C in risk for CHD,^35,36, evidence for a causal role of HDL-C from mendelian randomization studies is inconsistent and complicated by the fact that most SNPs associated with HDL-C levels affect multiple lipid traits. Voight and coworkers recently showed that a genetic risk score consisting of 14 SNPs exclusively associated with HDL-C was not associated with risk for myocardial infarction, in contrast to a genetic risk score for LDL-C.³⁷ These investigators had previously shown that a number of SNPs associated with HDL-C were also associated with other lipid traits such as triglycerides and LDL-C.³⁸ The SNPs associated with HDL-C that were most strongly associated with increased risk for myocardial infarction and that influenced other lipid traits were located in or near the APOA5-APOA4-APOC3-APOA1, TRIB1, and LPL genes or gene clusters. In the current study, we showed that these genes are also associated with sdLDL-C or sdLDL-C/LDL-C ratio (see Supplemental Table II [available online at http://atvb.ahajournals.org]).

Association of PCSK7 SNP-rs508487 Genotype with Circulating Lipids

A novel finding from the current GWAS analysis is the significant association of sdLDL-C with SNP rs508487 (at locus 11q23-q24), located in the PCSK7 gene. Investigation of the effect of rs508487 genotype on circulating lipid levels showed that each copy of the minor allele at this SNP raised sdLDL-C by ~4 mg/dL. It is important to note that genetic variation at the PCSK7 locus was not associated with LDL-C levels in our study. Since LDL-C is a more commonly used lipid measurement, it is plausible that genetic variations in the PCSK7 gene have not been associated with circulating lipids in previous GWAS studies.

PCSK7 encodes subtilisin-like/kexin proprotein convertase type 7 (PCSK7), a calcium-dependent serine endoprotease. PCSK7 has previously been implicated as a mediator of adipogenesis³⁹ and in the processing of VEGF-D, a critical step for binding of the angiogenic receptor VEGFR-2.⁴⁰ Furthermore, recent data show that internalization of PCSK7 from the plasma membrane is mediated by clathrin-coated vesicles,⁴¹ which are also implicated in the internalization of other cellular receptors such as the LDL receptor and various scavenger receptors. Although the physiologic role of PCSK7 is not clearly understood, it is plausible that PCSK7 could be involved in the processing of LDL and/or scavenger receptors, thereby modulating circulating lipid levels. Alternatively, recent studies suggest a potential role of protein convertases, including PCSK7, in lipid metabolism through proteolytic activation of angiopoietins and proteolytic inactivation of lipases.^42,43 In contrast to PCSK9, another member of the proprotein convertase family, to our knowledge PCSK7 has not been previously associated with cardiovascular lipid risk factors in other GWA studies. However, we should be cautious in the interpretation of our findings regarding the association of PCSK7 genotype with circulating lipids. A limitation of our study is that we did not measure protein or mRNA levels of PCSK7. Furthermore, the PCSK7 SNP rs508487 is in close proximity to the APOA5-APOA4-APOC3-APOA1 gene cluster, which has also been shown to affect circulating triglyceride levels. However, our exome chip data show that rare variants in the PCSK7 gene, which lead to amino acid substitutions in the PCSK7 protein, are associated with sdLDL-C and other lipid traits. Additional in vitro or animal studies using transgenic or PCSK7-knockout mouse models are needed to investigate the potential role of PCSK7 in lipid metabolism. Our findings also highlight the important issue of pleiotropy as PCSK7 was associated with circulating levels of sdLDL-C, triglycerides and HDL-C.

Conclusions

In summary, our results showed that sdLDL-C was highly correlated with an atherogenic lipid profile and, in contrast to lbLDL-C, predicted future CHD events in ARIC participants. Furthermore, sdLDL-C predicted risk for incident CHD even in individuals who would be considered at low cardiovascular risk based on their LDL-C level. This new homogenous sdLDL-C assay could be readily implemented in most routine clinical chemistry laboratories, provided that its clinical value can be confirmed in future studies. GWAS analysis identified significant associations of sdLDL-C with genetic variants in 14 different genes, all but 1 of which have been previously linked to cardiovascular disease risk. Our GWAS findings, together with findings from previous studies showing genetic variants in the same genes associated with other lipid traits, highlight the importance of pleiotropy in the development of cardiovascular disease. The novel finding of a significant association of sdLDL-C with genetic variants in PCSK7, a member of the subtilisin-like/kexin proprotein convertase family, provide new insights into the role of this gene in lipid metabolism.

Significance.

Low-density lipoprotein cholesterol (LDL-C) is considered one of the most important risk factors for cardiovascular disease and remains the primary target for current cardiovascular risk reduction strategies. LDL particles are a heterogeneous population, and it has long been hypothesized that a subfraction of LDL, small dense LDL (sdLDL), possesses atherogenic potential. In the current study, we investigated the relationship between plasma levels of sdLDL-C and risk of incident CHD in the biracial ARIC study cohort. Elevated plasma sdLDL-C levels were associated with increased risk of incident CHD, even in individuals considered to be at low cardiovascular risk based on their LDL-C levels. Using genome-wide association (GWA) analyses, we discovered one novel locus, PCSK7, for which genetic variation was significantly associated with sdLDL-C levels and other lipid traits. Together these findings provide new insights into the role of the PCSK7 gene in lipid metabolism and risk of cardiovascular disease.

Abbreviations

apo

apolipoprotein

ARIC

Atherosclerosis Risk in Communities

BMI

body mass index

CHD

coronary heart disease

CI

confidence interval

GWAS

genome-wide association study

HDL-C

high-density lipoprotein cholesterol

HR

hazard ratio

hs-CRP

high-sensitivity C-reactive protein

lbLDL

large buoyant low-density lipoprotein

lbLDL-C

large buoyant low-density lipoprotein cholesterol

LDL-C

low-density lipoprotein cholesterol

Lp-PLA2

lipoprotein-associated phospholipase A2

sdLDL

small dense low-density lipoprotein

sdLDL-C

small dense low-density lipoprotein cholesterol

SNP

single-nucleotide polymorphism