Author + information
- aDepartment of Medicine, Duke University School of Medicine, Durham, North Carolina
- bDivision of Cardiology, Department of Medicine, Duke University School of Medicine, Durham, North Carolina
- cDuke Clinical Research Institute, Durham, North Carolina
- dDuke Molecular Physiology Institute, Durham, North Carolina
- ↵∗Address for correspondence:
Dr. Svati H. Shah, Duke University School of Medicine, 300 North Duke Street, Duke University, Durham, North Carolina 27701.
Heart failure (HF) is a heterogeneous clinical syndrome affecting more than 5 million people in North America alone. In the past decade, we have witnessed significant advances in neurohormonal modulation, electrophysiologic therapies, and mechanical circulation; however, patients with HF still suffer from excess morbidity and mortality on par with many malignancies. How might we stem the tide of the swelling HF epidemic in the 21st century? One way we could address this challenge is by identifying and targeting alternative biological pathways involved in HF pathophysiology. In this issue of JACC: Heart Failure Lanfear et al. (1) present data further supporting metabolism as one such pathophysiologic axis on which we should focus in the coming years.
Using targeted metabolomic profiling of plasma in a large cohort of patients with a history of HF and reduced ejection fraction, Lanfear et al. (1) identify significant differences in circulating metabolites among key HF subgroups: diabetics versus nondiabetics, ischemic versus nonischemic HF etiology, African Americans versus other races, and male versus female. Further, plasma metabolite profiles differ significantly by age, as well as by New York Heart Association functional class and 6-min walk distance. Most important, they identify a set of 13 metabolites independently associated with survival. In alignment with prior studies, these metabolites were involved in the tricarboxylic acid cycle (e.g., succinate, α-ketoglutarate, and fumarate), fatty acid oxidation (e.g., long-chain acylcarnitines), and nitric oxide synthesis (e.g., arginine). Overall, these data contribute to an expanding body of literature suggesting that perturbations of these and other key energy-producing pathways mediate HF pathogenesis and progression, and contribute to residual risk beyond neurohormonal, hemodynamic, and electrophysiologic abnormalities (2).
As the authors acknowledge, however, these data do not identify a new HF biomarker, or clarify mechanisms of HF-associated metabolic impairments. Rather, the strength of this investigation is broader in scope: It constitutes further proof of concept of the potential of metabolism, and metabolomics, as a primary investigative tool for refining phenotypes, improving risk stratification, and identifying novel therapeutic targets in HF. But, how do we begin to realize this great potential? The steps will be many, and the journey will require focus and collaboration from investigators in clinical, translational, and basic arenas. The present article highlights several key principles that can maximize the usefulness of metabolomics investigations in HF. These principles, or signposts, will help to guide us on the path to developing clinically impactful applications of metabolism and metabolomics in the management of HF (Figure 1).
Performing Dense Clinical Phenotyping
The first key principle for metabolomics investigations, and signpost on the journey to clinical usefulness, is dense clinical phenotyping. This is well-exemplified in the study by Lanfear et al. (1); they collected all of the variables included in the validated Meta-Analysis Global Group in Chronic Heart Failure (MAGGIC) risk score (3). These variables included age, sex, smoking, presence of diabetes, HF duration, systolic blood pressure, left ventricular ejection fraction, body mass index, New York Heart Association functional class, and use of angiotensin-converting enzyme inhibitors/angiotensin receptor blockers or beta-blockers. Additionally, they collected natriuretic peptide measurements on all subjects. Dense clinical phenotyping, as demonstrated in this study, is critical because it improves signal-to-noise ratio, and thereby facilitates detection of key metabolite associations with outcomes or subgroups of interest. Additionally, it enables thorough adjustment in multivariate modeling, and reduces confounding. In fact, circulating metabolite profiles are significantly impacted by a number of clinical measures and their associated comorbidities, such as coronary artery disease (CAD), diabetes mellitus, chronic kidney disease, and chronic obstructive pulmonary disease (2).
Importantly, Lanfear et al. (1) show that, although attenuated, the 13 metabolite profile remained predictive of mortality even after accounting for the important clinical covariates included in the MAGGIC score. However, the presence or severity of CAD is not adjusted for in this analysis. This omission is critical because several metabolites included in the 13 metabolite panel have been associated with CAD severity in previous studies published by our group, and discriminate ischemic vs. nonischemic etiology of HF in the present study (4). This lack of adjustment for CAD thus complicates interpretation of the mortality results because some of the associations could be primarily related to CAD as opposed to HF per se; nevertheless, the authors should be lauded for their attempts to minimize confounding and show independent association. Finally, although not pursued in this study, another important benchmark in metabolomics studies is demonstrating not only independent associations with mortality, but also incremental risk predictive capabilities beyond established risk factors, through use of the area under the curve and/or calculation of reclassification indices.
Choosing Sample Source and Metabolomics Platform to Address Study Question
Although the clinical aspects of study design and phenotyping are vital, additional signposts that guide our efforts involve laboratory-related considerations. To better characterize HF-associated metabolic impairments and maximize the usefulness of metabolomics investigations, investigators must carefully choose the source or type of biospecimen to use for analysis. Similar to the present study, the vast majority of metabolomics investigations in HF have analyzed metabolites in peripheral blood (5). The advantages of this analysis include ease of access, lower cost, and potentially greater clinical usefulness for identified biomarkers. However, because the circulating metabolome is an integrated marker of the metabolic activity of multiple organ systems, we cannot easily discern the contributions of specific organs to a given abnormality. It will be interesting to evaluate whether the results highlighted in the present study are byproducts of myocardial, skeletal muscle, vascular, or other central sources using studies that combine plasma with tissue samples from these organ systems and/or paired afferent and efferent blood samples (e.g., central venous and coronary sinus blood samples).
Further, the choice of metabolomics platform is a critical design question that should fit the study purpose and hypothesis. Nontargeted, or unbiased, metabolomics affords a more comprehensive survey of the small molecules present in a sample, but by nature does not permit absolute quantification. Whereas this bottom-up approach is ideally suited to generating hypotheses, targeted metabolomic profiling as exemplified by Lanfear et al. (1) is better suited to hypothesis testing. With this top-down approach, a smaller number of metabolites of interest are pre-specified (typically 50 to 100), and quantified in an absolute manner. The targeted platform used in the present analysis assayed a variety of amino acids, organic acids, and acylcarnitines; this collection of metabolites generally reports on carbohydrate metabolism, urea cycle, and fatty acid oxidation. Although this constitutes a significant number of metabolites of interest to metabolic derangements relevant to HF, key metabolites with diagnostic and prognostic value in HF were not assessed (e.g., phospholipids, sphingolipids, spermidine, dimethylarginine, and trimethylamine N-oxide) (5). As the evidence base grows on the circulating metabolome and its associated impairments in HF, and as metabolomics technology continues to improve, more comprehensive and accurate snapshots of the metabolome incorporating known and novel pathways will likely facilitate refinement of predictive models and disease mechanisms.
Optimizing Design to Reduce Type I Error
One of the challenges of metabolomics investigations is dealing with the complex, multidimensional datasets thereby generated. When attempting to identify metabolite associations with clinical phenotypes, tens or hundreds of statistical tests may be performed. Accordingly, another signpost on our journey to clinical usefulness involves the issue of multiple comparisons. Although Bonferroni adjustment is sometimes used to reduce type I error, it is likely too conservative given the correlation between metabolites in shared biological pathways. Multidimensional data reduction methods are thus being increasingly used in the molecular “big data” space, including the lasso techniques performed in the present study. Such methods effectively manage the burden of multiple comparisons and reduce type I error without being overly conservative.
Another design principle that enhances robustness and reduces false discovery rate is the use of validation cohorts. Lanfear et al. (1) accomplished this by splitting a single cohort into derivation and validation cohorts with robust cross-validation techniques. However, for eventual clinical translation, the novel associations found herein will need to be validated in external cohorts from other centers. Accordingly, to maximize the impact of a given metabolomics investigation, investigators should carefully consider how they may build in a replication or validation cohort to enhance the robustness of their study and reduce the likelihood of finding a significant but nongeneralizable signal.
Dating back to the 1930s, the notion of metabolic dysfunction in HF is nothing new. Technological advancements, especially in metabolomics, has allowed us to better characterize HF-associated metabolic impairments. Once conceptualized as “an engine running out of fuel,” we now understand HF-associated metabolic dysfunction to be a complex system of derangements involving multiple biochemical pathways and cross-talk between major organ systems. The study by Lanfear et al. (1) further demonstrates the promise of metabolic impairments, and metabolomics, to refine HF phenotypes, enhance risk stratification, and identify therapeutic targets. Nevertheless, we still have a significant distance to traverse on the path to realizing this promise. By following the signposts identified herein, we can maximize the usefulness of metabolomics investigations, and continue progressing toward the goal of developing novel metabolic diagnostics and therapeutics to improve the lives of HF patients in the coming years.
↵∗ Editorials published in JACC: Heart Failure reflect the views of the authors and do not necessarily represent the views of JACC: Heart Failure or the American College of Cardiology.
Dr. Shah has received funding support through a sponsored research agreement between BMS and Duke University; has a patent on a metabolomic finding; and a family member has received consultant fees or honoraria from Biosense Webster, Boston Scientific, CardioNet, and St. Jude. Dr. Hunter has reported that he has no relationships relevant to the contents of this paper to disclose.
- 2017 American College of Cardiology Foundation
- Lanfear D.E.,
- Gibbs J.J.,
- Li J.,
- et al.
- Doehner W.,
- Frenneaux M.,
- Anker S.D.
- Pocock S.J.,
- Ariti C.A.,
- McMurray J.J.V.,
- et al.
- Hunter W.G.,
- Kelly J.P.,
- McGarrah R.W.,
- Kraus W.E.,
- Shah S.H.