Author + information
- aDuke Center for Applied Genomics & Precision Medicine, Duke University, Durham, North Carolina
- bDuke Molecular Physiology Institute, Duke University, Durham, North Carolina
- cDivision of Cardiology, Department of Medicine, Duke University, Durham, North Carolina
- ↵∗Reprint requests and correspondence:
Dr. Svati Shah, Duke Molecular Physiology Institute, Duke University Medical Center, 300 North Duke Street, Durham, North Carolina 27701.
Statins are powerful medications to prevent cardiovascular disease (CVD). It is well known that statins’ primary mechanism of action is inhibiting HMGCR (3-hydroxy-3-methylglutaryl- coenzyme A reductase), the enzyme that catalyzes the rate-limiting step of cholesterol biosynthesis. However, statin therapy is associated with pleiotropic effects such as improvements in endothelial function, inflammation, and oxidation that may additionally explain their ability to prevent CVD beyond low-density lipoprotein cholesterol (LDL) lowering. The mechanism and extent to which the effects of statins are due to on-target effects related to HMGCR inhibition, downstream effects of LDL reduction, or off-target effects are not well described. In this issue of the Journal, Würtz et al. (1) perform an elegant study marrying metabolomic profiling with genetics to conduct a “natural” clinical trial evaluating the on-target effects of statins.
Hardly unique to statins, the importance of on- versus off-target effects is critical to most medications in use and in development. Similarly, late drug failures (i.e., in clinical trial phase III or IV) are not uncommon and contribute to skyrocketing drug costs. One approach to help prioritize drug targets in early-phase testing is to develop improved human models of drug response. The recent explosion in “omics” technologies provides novel opportunities to explore human biology in a more comprehensive, high-throughput, systems-based fashion. By identifying unknown on- or off-target drug effects, such technologies enable better models of drug responses in human systems and, thus, have the potential to accelerate drug development.
An emerging “omics” platform with roots in the science of biochemistry, metabolomics aims to measure the byproducts of metabolism on a broad scale. Modern metabolomics platforms can report on hundreds of small molecules per sample in a standardized fashion in large numbers of samples at low cost. Metabolomics studies can perform “unbiased” mechanistic investigations, identify novel biomarkers, and when incorporated with genetic data, gain deeper insights into the basis of disease (2). Pharmacometabolomics is an application of metabolomics to drug responses. By identifying metabolic pathways associated with drug response, pharmacometabolomics is one approach to understanding on- versus off-target drug effects.
The gold standard study design for identifying drug effects is a placebo-controlled clinical trial. Although robust, this approach is prohibitively expensive at large scales and is limited to the drug of interest. An alternative and more convenient approach is to use observational data from patients who are prescribed medications as part of their medical care. However, because the use of a drug is based on a clinical decision and not randomized, confounders plague the observational approach, limiting the ability to attribute an association with drug use to causation by the drug. The study by Würtz et al. (1) uses an observational study design to study the effects of statins, and, in a clever approach, they incorporate genetics to overcome the limitations of this design.
The concept of Mendelian randomization in genetics research can be thought of as “nature’s clinical trial,” capitalizing on the random allocation of genetic variants at meiosis. Mendelian randomization studies have been used to identify novel drug targets. For example, the development of proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitors was motivated by the observation that patients carrying loss-of-function PCSK9 mutations have lower LDL and rates of CVD compared with those without the mutation, thus mimicking a randomized trial of PCSK9 inhibition. Conversely, Mendelian randomization has also been used to refute a causal link of potential candidate drug targets (e.g., C-reactive protein) (3). By linking genomic and electronic health record data, phenome-wide association studies or “pheWAS” can use Mendelian randomization to identify novel diseases for drugs with known mechanisms of action (4) and potentially rescue failed drugs by identifying novel indications. Lastly, these studies can dissect on- and off-target drug effects to better understand variation in drug response and adverse drug effects (5).
The study by Würtz et al. (1) integrates Mendelian randomization concepts with population-based pharmacometabolomics to uncover a more granular picture of the on-target effects of statins. In their approach, the authors use a nuclear magnetic resonance (NMR)-based platform to construct a “signature” of the response to statin therapy using serially collected plasma samples, in which a subset of patients was initiated on statin therapy. As expected, they found that statins were associated with lowering of very-low-density lipoprotein (VLDL), LDL subclasses, intermediate-density lipoprotein (IDL), triglycerides, remnant cholesterol, and several fatty acids: saturated, unsaturated, and omega-6 fatty acids. Interestingly, the greater magnitude of IDL and VLDL lowering, compared with the more modest lowering of triglycerides, suggests that statins may be more effective in reducing remnant cholesterol than previously appreciated. Although the NMR platform used in this study focused on lipids, several nonlipid molecules were included that are likely related to the known pleiotropic effects of statins. For example, the authors noted a reduction in GlycA (an inflammatory biomarker) and acetate with statin therapy.
Importantly, to overcome the inherent limitations that accompanied their observational study design—confounding by disease, concomitant changes in diet, lifestyle, and medications—the authors then turned their attention to existing, population cohorts with available genetic and NMR metabolomics data. They focused on a specific genetic variant (rs12916) in the HMGCR gene where a higher number of T alleles is associated with lower HMGCR gene expression. The effect of the genetic variant is small compared with the effect of statins: 1.65 SD decrease in LDL for statins versus a 0.096 SD decrease for each rs12916-T allele. Despite the small genetic effect on HMGCR function, because of the cohort’s large size, the authors were able to demonstrate striking concordance between the effects of statins and the genetic variant on most lipids. The correlation between the effect sizes for statin therapy versus the genetic effect was 0.94, suggesting that the effects of statins and the rs12916 HMGCR genetic variant were highly similar with respect to changes in lipid profiles. Further, the large decrease in IDL and small VLDL with statin therapy was concordant with the genetic studies, suggesting additional effects of statins beyond LDL lowering in reducing the risk of atherosclerosis. This finding is consistent with the greater importance of non–high-density lipoprotein cholesterol versus LDL in predicting CVD risk (6). Therefore, HMGCR inhibition (through statins or genetic effects) led, not only to lowering of LDL, but also additional atherogenic particles.
These exciting findings will no doubt lead to future investigations that will further unravel the remaining uncertainties related to the pleiotropic effects of statins and refining the residual lipid risk in patients treated with statins. NMR metabolomics is powerful for subclassification of lipids; extension to other genetic variants across the genome should identify off-target lipid effects of statins not mediated by HMGCR. More comprehensive and sensitive platforms profiling a larger number of metabolites, including metabolites at lower abundance not detected with NMR, will likely identify additional nonlipid molecular pathways. This likelihood is suggested by the nonlipid metabolites in this study: GlycA was lower in the genetic study than would be predicted by the changes in the statin therapy group. Similarly, a branched chain amino acid isoleucine and the aromatic amino acids decreased with statin therapy but were unchanged in the genetic study.
Lastly, whereas the changes observed can be attributed to inhibiting or lowering HMGCR, it is unknown to what extent the changes require LDL lowering or some alternative function of HMGCR on lipid metabolism. Analogous studies using ezetimibe and NPC1L1 genetic variants or PCSK9 inhibitors and PCSK9 genetic variants will facilitate teasing out whether the effects identified in this study can be generalized to alternative methods of LDL lowering. We anticipate that future work by this and other groups will help fill in these knowledge gaps.
In summary, the work by Würtz et al. (1) represents an important proof-of-concept study that observational data can be used to characterize drug responses using pharmacometabolomics by confirming these results with a “natural” clinical trial using Mendelian randomization concepts. With the increasing availability of large, population-based cohorts with banked biological samples and associated medical record data, we can envision future studies that characterize metabolomic differences based on drug, disease, or environment. Using an analogous approach to that of Würtz et al. (1), by selecting genetic variants with predictable effects on gene function, one can begin to identify novel genes that mimic (or antagonize) the observed effects that may help accelerate drugs in development, motivate novel drug development for new targets, and potentially repurpose existing medications for new indications.
↵∗ Editorials published in the Journal of the American College of Cardiology reflect the views of the authors and do not necessarily represent the views of JACC or the American College of Cardiology.
Both authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- 2016 American College of Cardiology Foundation
- Würtz P.,
- Wang Q.,
- Soininen P.,
- et al.
- Shah S.H.,
- Newgard C.B.
- C Reactive Protein Coronary Heart Disease Genetics Collaboration (CCGC),
- Wensley F.,
- Gao P.,
- Burgess S.,
- et al.
- Sofat R.,
- Hingorani A.D.,
- Smeeth L.,
- et al.