Lipoprotein signatures of cholesteryl ester transfer protein and HMG-CoA reductase inhibition

Cholesteryl ester transfer protein (CETP) inhibition reduces vascular event risk, but confusion surrounds its effects on low-density lipoprotein (LDL) cholesterol. Here, we clarify associations of genetic inhibition of CETP on detailed lipoprotein measures and compare those to genetic inhibition of 3-hydroxy-3-methylglutaryl-coenzyme A reductase (HMGCR). We used an allele associated with lower CETP expression (rs247617) to mimic CETP inhibition and an allele associated with lower HMGCR expression (rs12916) to mimic the well-known effects of statins for comparison. The study consists of 65,427 participants of European ancestries with detailed lipoprotein subclass profiling from nuclear magnetic resonance spectroscopy. Genetic associations were scaled to 10% reduction in relative risk of coronary heart disease (CHD). We also examined observational associations of the lipoprotein subclass measures with risk of incident CHD in 3 population-based cohorts totalling 616 incident cases and 13,564 controls during 8-year follow-up. Genetic inhibition of CETP and HMGCR resulted in near-identical associations with LDL cholesterol concentration estimated by the Friedewald equation. Inhibition of HMGCR had relatively consistent associations on lower cholesterol concentrations across all apolipoprotein B-containing lipoproteins. In contrast, the associations of the inhibition of CETP were stronger on lower remnant and very-low-density lipoprotein (VLDL) cholesterol, but there were no associations on cholesterol concentrations in LDL defined by particle size (diameter 18–26 nm) (−0.02 SD LDL defined by particle size; 95% CI: −0.10 to 0.05 for CETP versus −0.24 SD, 95% CI −0.30 to −0.18 for HMGCR). Inhibition of CETP was strongly associated with lower proportion of triglycerides in all high-density lipoprotein (HDL) particles. In observational analyses, a higher triglyceride composition within HDL subclasses was associated with higher risk of CHD, independently of total cholesterol and triglycerides (strongest hazard ratio per 1 SD higher triglyceride composition in very large HDL 1.35; 95% CI: 1.18–1.54). In conclusion, CETP inhibition does not appear to affect size-specific LDL cholesterol but is likely to lower CHD risk by lowering concentrations of other atherogenic, apolipoprotein B-containing lipoproteins (such as remnant and VLDLs). Inhibition of CETP also lowers triglyceride composition in HDL particles, a phenomenon reflecting combined effects of circulating HDL, triglycerides, and apolipoprotein B-containing particles and is associated with a lower CHD risk in observational analyses. Our results reveal that conventional composite lipid assays may mask heterogeneous effects of emerging lipid-altering therapies.


Introduction
Definitive evidence on the causal role of low-density lipoproteins (LDL) in cardiovascular disease comes from trials of LDL cholesterol lowering compounds, 1 which have shown beneficial effects on risk of coronary heart disease (CHD) and stroke. Consistent effects have been seen for drugs acting on related pathways, such as 3-hydroxy-3-methylglutaryl-coenzyme A reductase (HMGCR) inhibitors, i.e., statins, and proprotein convertase subtilisin-kexin type 9 (PCSK9) inhibitors, 2 both of which upregulate hepatic LDL receptor expression, and for drugs acting on other pathways, such as ezetimibe, which inhibits intestinal absorption of cholesterol. 3 However, trials of drugs primarily designed to alter concentrations of lipids other than LDL cholesterol have had mixed results. 4 One such example is the class of drugs designed to inhibit cholesteryl ester transfer protein (CETP), a lipid transport protein responsible for the exchange of triglycerides and cholesteryl esters between apolipoprotein B-containing particles and highdensity lipoprotein (HDL) particles. CETP inhibitors were developed initially on the basis of their HDL cholesterol raising effects. While accumulating genetic evidence suggests that HDL cholesterol concentration is unlikely to be causally related to CHD, 5 there were two strong reasons to believe that CETP inhibition may still reduce vascular risk: (i) genetic studies of CETP variants have shown associations with CHD 6 and (ii) some CETP inhibitors not only increase HDL cholesterol but also appear to lower LDL cholesterol as measured by conventional assays. 7 The recent findings from the phase III REVEAL trial showed that treatment with the CETP inhibitor anacetrapib led to a reduction in risk of coronary events that was proportional to the reduction in non-HDL cholesterol. 8 Interestingly, anacetrapib appeared to have discrepant effects based on the assay used to quantify LDL cholesterol (using beta-quant, direct or Friedewald estimation). 7 This discrepant effect was also identified in a genetic study that approximated a factorial clinical trial of CETP inhibition and statin therapy. 9 Thus, while both CETP inhibitors and statins lower Friedewald-estimated LDL cholesterol, which also includes cholesterol carried by other lipoprotein particles, it is possible that the drugs have differential effects on the concentration and content of lipids in different apolipoprotein-B containing lipoproteins.
In this study, we used the established approach of exploiting genetic variants near the proteincoding genes of drug targets to investigate detailed lipid and lipoprotein subclass signatures of CETP inhibition. We compared the association of variants in CETP with HMGCR 10 (to proxy statin treatment) to gauge insight into how these two therapies alter the lipoprotein milieu. We also present findings that the triglyceride composition, in contrast to circulating concentrations, of HDL particles is associated with CHD and may relate to a new mechanism by which CETP inhibition reduces risk of CHD.

Prospective and cross-sectional studies and lipoprotein quantification
We used genetic and lipoprotein data from four population-based Finnish cohorts and one crosssectional study in the UK (cohort characteristics are presented in Online Table 1 14 and a tri-ethnic UK community-based cohort SABRE (n = 4,857 individuals aged 40-69 y). 15 A nuclear magnetic resonance (NMR)-based methodology was used to quantify lipoprotein lipids and subclasses. Details of this platform have been published previously, 16,17 and it has been widely applied in genetic and epidemiological studies. 10,18 Where possible, we excluded individuals receiving lipid lowering medication, pregnant women and those who had a high proportion (>30%) of values missing across the lipid traits. All measures were first adjusted for sex, age (if applicable), genotyping batch (if applicable) and ten first principal components from genomic data and the resulting residuals were transformed to normal distribution by inverse rank-based normal transformation. Details of study-specific genotyping are provided in Online Table 2 in the Supplementary Appendix.

SNP analysis
We selected variants as genetic proxies of CETP and HMGCR inhibition on the basis of robust associations with circulating lipids in GWAS consortia 19,20 and target gene expression. The P=5.4x10 -94 ) associates with lower gene expression across several tissues, verified through Genotype To Expression (https://gtexportal.org) project data. 22 We used an additive model for each cohort separately (see Online Table 1 for details of analysis software). In order to make the lipoprotein and lipid estimates comparable, the estimates for CETP rs247617 and HMGCR rs12916 were scaled to the same CHD association as reported by the CARDIoGRAMplusC4D GWAS Consortium. 23 The per-allele log odds (logOR) for CHD was 0.0358 for HMGCR rs12916 and 0.0309 for CETP rs247617; subsequently the summary statistics of each individual cohort and each metabolite were scaled to -0.105 logOR of CHD (equivalent to an odds ratio [OR] of CHD of 0.90) to align the estimates to a 10% lower risk of CHD. The cohort specific association results of lipoprotein and lipid measures with both variants were then combined using an inverse variance weighted fixed effect meta-analysis.
Our focus for this study was to evaluate the impact of variants in CETP and HMGCR on the entire cascade of apolipoprotein B-containing lipoproteins and HDL subclasses. Therefore, we decided a priori to examine 191 lipoprotein and lipid traits available from the NMR platform. 18,24 Focusing on these 191 traits, we estimated that 28 principal components explain 99% of their variation in the Finnish cohorts and therefore we used a P-value threshold of 0.05/28=0.002 to denote evidence in favor of an association. Abbreviations and full descriptions of the lipoprotein measures studied are listed in Online Table 3.

Results
Data from 62,400 individuals with extensive lipoprotein subclass profiling and genotypes were available. We combined data from five adult cohorts (mean age range from 31 to 52 years) and one cohort of adolescents (mean age 16 years) for the genetic analyses where 51% of participants of all six studies were female. Study specific and pooled estimates from meta-analyses of genetic and observational analyses for all 191 traits are presented in Online Figures 1-15. When examining triglycerides in apolipoprotein B-containing particles, CETP associated with lower circulating triglyceride concentrations in VLDL and IDL subclasses, while HMGCR had weaker effects on these measures, except in LDL subclasses (Fig. 2). CETP had a very strong association with higher HDL cholesterol (0.84; 95%CI: 0.76, 0.92) but HMGCR did not (0.04; 95%CI: -0.02, 0.10) (Fig. 3). Similarly, CETP was inversely associated with the total quantity of triglycerides in HDL particles (-0.23; 95%CI: -0.31,-0.15) but HMGCR was not (-0.03; 95%CI: -0.09, 0.02).
The lipoprotein particle structure is biophysically constrained, generating strong correlations between lipid measures within individual lipoprotein subclasses. [25][26][27][28] Notable differences in lipid concentrations in subclass particles would therefore suggest changes in the compositional proportions of these lipids. For genetic inhibition of CETP, the effects on circulating triglyceride concentrations in all HDL subclasses were weaker (XL-HDL and L-HDL) or even in the opposite direction (M-HDL and S-HDL) than the effects on cholesterol concentration in these subclasses (Fig. 3). Examining the genetic associations with the particle lipid compositions, the relative amount of triglycerides (in relation to all lipid molecules in the particles) was remarkably diminished in all HDL subclass particles by genetic inhibition of CETP (Fig. 4). Genetic inhibition of HMGCR did not associate with the triglyceride concentration or composition of any HDL subclass. These associations are in line with the known physiological roles of CETP and HMGCR and their inhibition. 29,30 In addition, as expected, CETP associated with higher compositions of triglycerides in most VLDL subclass particles and HMGCR showed directionally similar, albeit weaker associations.
To understand the clinical relevance of these HDL-related compositional changes arising from CETP inhibition, beyond reductions in cholesterol concentrations of apolipoprotein B-containing lipoprotein particles, we studied the observational associations of lipoprotein subclass lipid concentrations and compositions with CHD in three prospective population cohorts. The triglyceride concentration of HDL was associated with incident CHD when adjusted for nonlipid cardiovascular risk factors (Fig. 5)., However, when serum cholesterol and serum triglycerides were added to the model, as expected, the associations attenuated. In contrast, the triglyceride compositions of all the HDL subclass particles were positively associated with CHD, independent of circulating concentrations of cholesterol and triglycerides, with hazard ratios around 1.3 for all HDL subclasses (Fig. 5). In addition, compositional enrichment of cholesteryl esters in the largest VLDL particles (XXL-VLDL and XL-VLDL) was associated with risk of CHD (Online Fig. 11); genetic inhibition of CETP also impacted on these traits (Online Fig. 2).

Discussion
We used genetic variants in CETP and HMGCR to gain insight into the expected effects of therapeutic inhibition of CETP and HMG-coA reductase on circulating lipoproteins and lipids.
Our data show that while CETP and HMGCR have near identical effects on Friedewald-  35 This non-specificity of commonly-used "LDL" cholesterol assays is under-recognized and underlies the prevailing opinion that inhibitors of HMGCR and CETP both alter LDL cholesterol. However, our data show this not to be the case: using NMR spectroscopybased lipoprotein particle quantification, which defines individual lipoprotein subclasses based on particle size, 18,25,27 our findings demonstrate that CETP has negligible effect on cholesterol in size-specific LDL particles. In this way, the use of a composite lipid measure can obscure differential associations of a therapy or gene 26 with individual constituents of the composite, and can have clinical ramifications. For example, if a trial is powered to a given reduction in Friedewald LDL cholesterol, under the naïve assumption that the drug uniformly alters all the subcomponents, then the trial may not have the expected result if the drug has differential effects on these subcomponents. This is exemplified in the recent phase III ACCELERATE trial of evacetrapib, which was terminated for futility, and was powered to a difference in LDL cholesterol based on a composite assay. 36 The differential effects of CETP inhibition on composite markers such as Friedewald and directly-quantified LDL cholesterol compared to apolipoprotein B concentrations identified in the subsequent phase III REVEAL trial of anacetrapib 7 suggest that had ACCELERATE used an alternative measure of pro-atherogenic lipoproteins (e.g. apolipoprotein B or non-HDL-C 8 ) to gauge the expected vascular effect, the trial may have been more appropriately powered.
This highlights the need to understand, in detail, the consequences of lipid modifying therapies on lipoproteins and lipids in order to be able to gauge whether a composite measure (such as Friedewald LDL cholesterol) can be reliably used as an indicator of the likely beneficial effect of a therapy. This is unlikely to be limited to assays for LDL cholesterol. For example, assays that quantify triglycerides, measure the summation of triglycerides across multiple lipoprotein particle categories. Drugs currently under development that target triglycerides (such as apolipoprotein C-III inhibitors 37 ) have differential effects on triglycerides in lipoprotein subclass particles as demonstrated in a recent genetic study. 38 If triglycerides within different lipoprotein subclasses have heterogeneous effects on vascular disease, a clinical trial powered to the overall concentration of circulating triglycerides may give an inaccurate portrayal of the cardiovascular consequences arising from apolipoprotein C-III inhibition.
Another key finding is that the lipid compositions of lipoprotein particles can associate with disease risk independently of total lipid concentrations. While genetic inhibition of CETP increased circulating concentrations of cholesterol in all HDL subclasses, the triglyceride composition, i.e. the percentage of triglyceride molecules of all the lipid molecules in the particle, was markedly lower in all HDL particles. Intriguingly, our observational analyses, the first to explore lipoprotein particle lipid composition with CHD outcomes, revealed that triglyceride enrichment of HDL particles associates with higher risk for future CHD, independently of total circulating cholesterol and triglycerides. The largest hazard ratio for the triglyceride enrichment in medium HDL subclass particles was of a similar magnitude (~1.3) as that for LDL cholesterol and apolipoprotein B. 39 These findings suggest that lipoprotein particle compositions, independent of circulating lipid concentrations, could have a role in the development of CHD. While the causal role of these lipoprotein compositions remains unclear, these findings advocate the importance of moving from simple composite lipid measures towards more detailed molecular phenotyping of lipoprotein metabolism.
Key strengths of our analyses include the availability of detailed measurements of blood lipoprotein subclass concentrations and compositions from general population studies with incident CHD events, together with the availability of genome-wide genotyping. While we used CETP and HMGCR variants as genetic proxies for therapeutic inhibition, we note that the CETP genetic variant recapitulated the effects of CETP enzyme activity in relation to the role the enzyme has in shuttling esterified cholesterol from HDL to apolipoprotein B-containing particles in exchange for triglycerides. 29 Furthermore, prospective population-based data of patients taking statins with blood sampling before and after the commencement of therapy showed that genetic variants in HMGCR robustly recapitulated the effects of statin therapy on lipoprotein subclasses and lipids. 10 In conclusion, we have shown that, in contrast to genetic inhibition of HMG-CoA (proxying statin therapy), genetic inhibition of CETP does not alter circulating size-specific LDL cholesterol concentrations. This is masked by using conventional, non-specific assays for LDL cholesterol and may be problematic for ongoing and future clinical trials of lipid lowering therapies, especially when a non-specific marker of lipids is used to derive an expected effect of a drug with risk of disease. Our findings suggest potential additional mechanisms by which CETP inhibition could prevent CHD through reductions in the triglyceride composition of HDL particles. Our findings also call attention to the need for metabolic precision in measurements of lipoprotein lipids and in assessing the role of lipoprotein metabolism in cardiovascular disease in relation to ongoing treatment trials of novel lipid-altering therapies.