Genetic Variation at the Phospholipid Transfer Protein Locus Affects Its Activity and High-Density Lipoprotein Size and Is a Novel Marker of Cardiovascular Disease Susceptibility

All statistical analyses were performed with SPSS version 16.0.2. Two-sided P values of <0.05 were considered statistically significant. Effects of single SNPs or of PLTP gene score on continuous variables were examined by ANOVA; in case of significant differences, per-allele effects and corresponding 95% confidence intervals (CIs) were estimated with linear regression analysis. To estimate the relative risk of CVD in EPIC-Norfolk, conditional logistic regression was used to calculate odds ratios (ORs) and 95% CIs per allele change in PLTP gene score, with the lowest PLTP gene score as the reference category. Conditional logistic regression took into account the matching for sex, age, and enrollment time and was adjusted for Framingham Risk Score.³⁰ In the Premature Atherosclerosis Patients and Sanquin Blood Bank Controls (PAS/SQ), CHAOS-SEARCH, Rotterdam, and SHEEP studies, differences between cases and controls were analyzed by standard contingency table analysis with 2-tailed χ² test probabilities; linearity of this relationship was assessed with logistic regression analysis. To estimate the relative risk of CVD in all studies combined, logistic regression was used. Heterogeneity of the relationship between PLTP gene score and CVD risk across different studies was assessed by the use of a linear mixed model with study random effects for the estimated study-specific log OR. LD plots were created with Haploview, version 4.1.³¹

Relationships between tagSNPs and PLTP activity are shown in Tables 1 and 2. Only carriership of rs378114 G alleles and rs6065904 T alleles was related to lower PLTP activity in both the Groningen study (−3.1 arbitrary units [AU] [95% CI, −0.1 to −6.1] per rs378114 G allele, P=4.5×10⁻²; −7.4 AU [95% CI, −4.5 to −10.4] per rs6065904 T allele, P=1.0×10⁻⁶) and the DALI study (−7.7 AU [95% CI, −3.0 to −12.4] per rs378114 G allele, P=1.5×10⁻³; −7.6 AU [95% CI, −3.9 to −11.4] per rs6065904 T allele, P=8.7×10⁻⁵). We constructed a PLTP gene score based on rs378114 (presence of 0, 1, or 2 G alleles) and rs6065904 (presence of 0, 1, or 2 T alleles) on the basis of their effects on PLTP activity in these 2 studies (177 participants in DALI and 145 in the Groningen Study had complete gene score data). This gene score represents the number of PLTP activity–decreasing alleles, ranging from 0 to 4. A higher PLTP gene score was associated with lower PLTP concentration (−0.69 mg/L [95% CI, −0.24 to −1.14] per allele; P=3.0×10−³) and activity (−6.0 AU [95% CI, −3.4 to −8.6] per allele; P=1.2×10⁻⁵) in the DALI study and lower PLTP activity (−3.9 AU [95% CI, −2.1 to −5.7] per allele; P=3.5×10⁻⁵) in the Groningen study (Figure 1A through 1C). In the 2 studies combined, the per-allele effect of PLTP gene score on PLTP activity was −5.0 AU (95% CI, −3.4 to −6.7; P=4.4×10⁻⁹). To exclude an important contribution of type 2 diabetes, we reassessed the relationship, selecting only the Groningen study participants without diabetes mellitus, and found similar associations between PLTP gene score and PLTP activity (−4.7 AU [95% CI, −2.4 to −7.1] per allele; P=1.3×10⁻⁴). In addition, in pooled data from both studies, PLTP gene score was significantly associated with PLTP activity after controlling for age, sex, and diabetes status (−5.1 AU [95% CI, −3.5 to −6.7] per allele; P=4.1×10⁻⁹). Finally, to validate the PLTP gene score, we evaluated its association with PLTP RNA expression in the expanded HLC and found that it was linearly associated with less efficient PLTP transcription (−0.47 AU per allele [95% CI, −0.058 to −0.037]; P=3.2×10⁻¹⁸; Figure 1D).

Characteristics of study participants across PLTP gene score categories in EPIC-Norfolk are shown in Table I in the online-only Data Supplement. Conventional risk factors did not differ between PLTP gene score categories. Standard lipid profiles were similar between gene scores, although HDL cholesterol levels showed a nonsignificant tendency to be lower with higher gene scores. There were no differences in apolipoprotein A-I (a structural component of HDL) or apolipoprotein B (a structural component of very-low-density lipoprotein) levels.

We determined whether the PLTP gene score was associated with parameters related to HDL remodeling. The number of HDL particles increased with a rising PLTP gene score (0.37 nmol/L [95% CI, 0.16 to 0.57] per allele; P=6.2×10⁻⁴; Figure 2A). This was due predominantly to a higher concentration of small HDL particles (0.79 nmol/L [95% CI, 0.61 to 0.97] per allele; P=3.4×10⁻¹⁷; Figure 2b), whereas the number of large HDL particles actually decreased with higher PLTP gene scores (−0.37 nmol/L [95% CI, −0.50 to −0.24] per allele; P=3.9×10⁻⁸; Figure 2C). Consequently, people with a higher PLTP gene score exhibited a lower HDL size as measured by nuclear magnetic resonance spectroscopy (−50 pm [95% CI, −32 to −67] per allele; P=2.8×10⁻⁸; Figure 2D). HDL size measured by gradient gel electrophoresis yielded a similar result (−47 pm [95% CI, −31 to −63] per allele; P=9.0×10⁻⁹).

Next, we determined whether PLTP gene score was associated with an altered risk of CVD. Combining data from all 5 CVD studies, we found an inverse association between PLTP gene score and risk of CVD (OR, 0.94 per allele increase in score; 95% CI, 0.90 to 0.98; P=1.2×10⁻³; Figure 3). This association is consistent with the proposed biological role of PLTP. Although the pattern of association between gene score and risk of CVD appears to differ between studies, we found no formal statistical evidence of heterogeneity between studies (P=0.16). However, in light of evidence for heterogeneity in genetic structure between Swedish and other European populations,^32–35 we also present combined analyses after exclusion of the SHEEP study. In only British and Dutch individuals, the association between PLTP gene score and CVD risk was more pronounced (OR, 0.91 per allele; 95% CI, 0.87 to 0.95; P=2.3×10⁻⁵; Figure 3). Comparing groups with the highest exposure differential, reflecting the greatest difference in PLTP activity, we found that, in all studies combined, the OR for CVD for the highest versus the lowest gene score was 0.69 (95% CI, 0.55 to 0.86; P=1.0×10⁻³; Table 3). In only British and Dutch individuals, the OR for CVD for the highest versus the lowest gene score was 0.58 (95% CI, 0.45 to 0.75; P=3.2×10⁻⁵).

An understanding of correlations among nearby gene variants (LD structure) permits the selection of tagSNPs (ie, SNPs that most efficiently represent other SNPs that are not genotyped) in a genomic region with high LD.²⁶ We found that alleles of 2 PLTP tagSNPs were linearly associated with lower plasma PLTP activity in 2 independent studies and with PLTP transcription efficiency in a third study. Although it remains unknown which genetic variant is causally responsible for these associations, these data provide solid support for the hypothesis that the PTLP gene score is a valid genetic proxy for PLTP activity.

In vitro, PLTP remodels HDL into large and small particles and mediates the dissociation of lipid-poor apolipoprotein A-I.^6,36–39 Evidence indicates that the actions of PLTP initially lead to the formation of a fusion product of 2 HDL particles, which subsequently rearranges into 3 small conversion products or alternatively into 1 large conversion product in a process that is accompanied by the shedding of apolipoprotein A-I.⁵ Our findings suggest that the net effect of these actions in vivo is a reduction in the number of small HDL particles and an increase in the number of large HDL particles. This is in line with observations in patients with type 1 diabetes mellitus in whom elevated PLTP activity was associated with more large and fewer small HDL particles.⁴⁰ In combination with the absence of a difference in apolipoprotein A-I and HDL cholesterol levels, this suggests that, in humans, across this range of PLTP activity modulation, PLTP does not materially affect apolipoprotein production or catabolism but repackages HDL cholesterol into fewer and larger particles in vivo. This contrasts with an article by Engler et al,¹⁴ which revealed a small association signal for a PLTP SNP with HDL cholesterol levels in 107 hyperalphalipoproteinemic patients. Unfortunately, in this study, plasma PLTP activity was not analyzed.

In mice, systemic expression of PLTP has been shown to be proatherogenic using PLTP knockout⁴¹ and human^42,43 and murine⁴⁴ PLTP overexpression models. Most studies in humans accordingly show a positive association between plasma PLTP activity and atherosclerosis.⁹ Schlitt et al¹² reported that PLTP activity is elevated in CAD patients compared with healthy control subjects. In another study, PLTP activity was independently and positively related to carotid intima-media thickness, a measure of subclinical atherosclerosis, in patients with type 2 diabetes mellitus.¹⁰ Furthermore, it has been reported that high plasma PLTP activity was related to fatal and nonfatal cardiovascular events in CAD patients treated with statins.¹¹ Only 1 study has demonstrated a lower plasma PLTP activity in subjects with peripheral vascular disease.¹³ To date, no studies have examined the relationship between genetic variants in PLTP and the risk of CVD.

We found consistent relationships between PLTP gene score and CVD risk in 4 of the 5 studies. There are several potential explanations for the absence of association in the SHEEP study. It may be the result of sampling variation or of differences in LD structure between study populations, reflecting the fact that we are not directly assessing the causal variant(s) underlying these associations.^32–35 However, genotype frequencies for the 2 SNPs and for the gene score were similar between all studies (Tables II and III in the online-only Data Supplement). Alternatively, other genetic and environmental factors may specifically affect the relation between PLTP activity and CVD risk in the Swedish population. These considerations notwithstanding, our findings using the combined data of all 5 CVD studies comprising >16 000 individuals indicate that a genetic predisposition toward lower PLTP activity is associated with a lower risk of CVD. The relationship between gene score and CVD risk was linear, suggesting a “gene-dose” effect. Furthermore, in all studies combined, people with the highest PLTP gene score exhibited a 31% reduction in CVD risk compared with the lowest PLTP gene score.

Under the assumptions of mendelian randomization,⁴⁵ genetic association analyses are expected to be less confounded than direct assessments of plasma parameters and disease risk. Because variants in the PLTP locus show a very good level of specificity to PLTP transcription, plasma PLTP concentration and activity, and HDL particle number and size, we believe the association between these genetic variants and CVD risk allows us to make more robust statements about the relationship between this biological risk marker and disease susceptibility. Taken together, our results strengthen the hypothesis that PLTP has proatherogenic potential.

The PLTP gene score was not associated with any conventional risk factor for CVD but was strongly associated with HDL particle distribution. It is therefore tempting to speculate that the increased numbers of (small) HDL particles explain the lower CVD risk in people with a high PLTP gene score. Theoretically, such an increase may augment plasma capacity to accept cholesterol from cellular cholesterol efflux pathways, the most important putative antiatherogenic property of HDL.⁴⁶ This would be in line with observations in mice that elevation of systemic PLTP was found to impair excretion of macrophage cholesterol into the feces.⁴⁷ We attempted to further substantiate this possibility with our own data but found that statistical adjustment for HDL particle parameters in EPIC-Norfolk did not materially affect the relationship between PLTP gene score and CAD risk (data not shown). However, even if an HDL particle parameter would be the mediator of the observed risk reduction, random measurement error in this variable can be expected to limit its ability to abolish the relationship between PLTP gene score and CVD risk in a statistical model; therefore, this potential mechanism cannot be excluded.

In recent years, the focus of HDL research has gradually shifted from HDL cholesterol levels to HDL functionality, reflecting the fact that HDL cholesterol levels may be a poor measure of the efficiency of these particles to perform their antiatherogenic functions in vivo.⁴⁶ Our finding of a reproducible association between cardiovascular risk and genetic variants with a clear impact on HDL particle distribution, but not on HDL cholesterol levels, is in general agreement with this concept.

We demonstrate for the first time in humans that genetic variation at the PLTP locus is associated with reduced PLTP messenger RNA transcription efficiency, a lower plasma PLTP activity, a higher concentration of HDL particles of smaller size, and a reduced risk of CVD, suggesting that PLTP has proatherogenic potential. Our findings are consistent with the notion that modulation of specific elements of HDL metabolism may offer cardiovascular benefit.