Author + information
- aKing's British Heart Foundation Centre, King's College London, United Kingdom
- bDepartment of Neurology, Medical University Innsbruck, Innsbruck, Austria
- ↵∗Reprint requests and correspondence:
Prof. Manuel Mayr, King’s British Heart Foundation Centre, King’s College London, 125 Coldharbour Lane, London SE5 9NU, United Kingdom.
In this issue of the Journal, Fan et al. (1) present discovery metabolomics data from Chinese cohorts of patients with coronary artery disease. Patients were categorized into subjects with normal coronary arteries, nonobstructive coronary atherosclerosis, stable angina pectoris, unstable angina pectoris, and acute myocardial infarction. The authors provide evidence for improved classification on the basis of metabolomic profiles. Strengths of the study are the large number of samples analyzed and the fact that findings in the discovery cohort were replicated in independent cohorts. Multiple metabolites were combined and used in predictive models to support the authors’ conclusions.
Metabolomics is an emerging “-omics” technology that aspires to define the totality of metabolite concentrations. The method of choice for metabolomics is liquid chromatography (LC) tandem mass spectrometry (MS/MS, or combined as LC-MS/MS), which offers superior sensitivity and specificity compared with nuclear magnetic resonance spectroscopy. Nuclear magnetic resonance spectroscopy only measures a handful of cellular metabolites (usually <100). In comparison, most LC-MS/MS studies report hundreds of metabolites that are reproducibly detected in plasma. In principle, thousands of small-molecule metabolites can be detected, but LC-MS/MS has a bias toward high-abundant metabolites and cannot yet determine all low-abundant metabolites in plasma. In addition to endogenous metabolites, small-molecule analysis can identify exogenous metabolites, the main ones of which relate to diet and medication. Both endogenous and exogenous metabolites may constitute novel biomarkers: the former for disease processes, and the latter for dietary habits or therapy compliance.
Another important consideration in metabolomics is the distinction between “untargeted” and “targeted” metabolomic analyses. In “untargeted” metabolomics, samples are analyzed without authentic standards (Figure 1A). This approach offers the advantage of no a priori decisions on which metabolites to measure, but the lack of standards might compromise the quantitative accuracy and confidence in the metabolite identifications. In “targeted” metabolomics, a panel of metabolites is pre-selected for analysis (Figure 1B). These metabolites are then identified and quantified using authentic standards (2,3). This approach is robust and is used clinically for identifying drugs of abuse or monitoring drug metabolites (4).
Fan et al. (1) used untargeted metabolomics. The signals were assigned to putative metabolite identifications based on 2 criteria: the mass/charge ratio in the MS spectrum and the retention time on the LC system. The researchers reported that 89 metabolites were differentially expressed, one-half of which were further validated by using authentic standards, by confirming the retention time and mass/charge ratio of the peaks observed in the untargeted analysis. However, accurate mass and retention times alone are not sufficient for unambiguous metabolite identification. The putative identification of the analyte has to be further confirmed by fragmentation via MS/MS. Only if the MS/MS spectrum is consistent with the assigned metabolite is the metabolite identification confirmed (5).
Another major issue in untargeted metabolomics is the accuracy of quantitation. Fan et al. (1) spiked a single internal standard into the metabolite extracts to adjust for run-to-run variability. In targeted metabolomics, the plasma samples are spiked with authentic standards for every metabolite analyzed. Authentic standards are important to account for quantitation errors, such as those due to common phenomena like “ion suppression” in LC-MS/MS analysis (Figure 1C). The analytes eluting from the LC system must be ionized to allow their transition from the liquid phase into the gas phase before they can enter the orifice of the MS. Among other factors, such as the structural composition of the analyte, the efficiency of the ionization process is dependent on coeluting metabolites. High-abundant metabolites can suppress the ionization of low-abundant metabolites by picking up most of the available charges at the cone of the spray tip (6). Thus, the signal intensity is not solely dependent on the plasma concentration of the metabolite. The ionization efficiency, on the other hand, is in part determined by the “matrix” (7). To paraphrase a famous movie quote: “What is the matrix? The matrix is everywhere. It is the world of small molecules that has been pulled over your analyte of interest to blind you from the truth.” In brief, coeluting metabolites or contaminants will influence the quantitation accuracy for your analyte of interest. In the study by Fan et al. (1), plasma samples were analyzed within an LC gradient time of <15 min. Numerous metabolites will coelute at any 1 time. Without the use of authentic standards, matrix effects cannot be accounted for. An untargeted metabolomics approach may be a tool for hypothesis-free discovery science, but it requires further confirmation by targeted analysis.
A particular challenge in metabolomics is the interpretation of the large amount of data. This new study from Fan et al. (1) lacks enough detailed information on statistical procedures to permit replication; yet, this is a prerequisite, especially as such impressive predictive performances (>90% of samples classified correctly) were reported. The central process of identifying differentially expressed metabolites and determining significance levels involved a sequence of Student t tests, Bonferroni correction, orthogonal partial least squares discriminant analysis, and false discovery rate adjustment. It is not entirely clear in what order and to which metabolites each method was applied, although this would affect results. Methods both more established and more efficient should have been used for variable selection and control of multiplicity in the setting of high-data dimensionality. These concerns were mitigated, to some extent, by the demonstration of applicability of main results to 3 independent, albeit small, cohorts. Unfortunately, out-of-sample predictive performance was described mainly by the proportion of samples classified correctly, which as a performance measure is hampered by dependence on class prevalence and, here, by the need to introduce cut-offs on predicted risks thereby discarding information. A more complete and nuanced description of out-of-sample predictive performances of metabolites, including perhaps the established net reclassification improvement or integrated discrimination improvement and their variants, and confidence intervals, would have been advisable.
Over the last decade, metabolomics has been more widely used in cardiovascular research (8). An important branch of metabolomics is lipidomics, especially in the context of coronary artery disease (9). Technological advances in MS have facilitated the identification of ever more metabolites. However, a major limitation remains: the vast majority of metabolites are not tissue specific. Cardiac troponins are an example of successful biomarkers because of their exquisite tissue specificity. In the present study, metabolite performance was not compared against established biomarkers. Apart from the lack of tissue specificity, the interpretation of metabolic profiles was complicated by confounding effects of medication, such as statin therapy and heparin administration. Whereas statins lower lipid levels, heparin activates lipoprotein lipase. Administration of heparin will have profound effects on metabolic profiles. As expected, heparin was predominantly given to patients with acute myocardial infarction and unstable angina.
Thus, many issues remain to be addressed before claims can be made regarding the clinical utility of the presented findings (1), in particular the pre-analytical variation and confounding by medication, the reproducibility of the measurements, and most importantly, whether the additional information ultimately helps the clinician to manage patients and guide treatment decision by outperforming existing biomarkers (10).
↵∗ 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.
This research was funded/supported by the National Institute of Health Research (NIHR) Biomedical Research Centre based at Guy’s and St. Thomas’ NHS Foundation Trust and King’s College London in partnership with King’s College Hospital. Prof. Mayr is a Senior Fellow of the British Heart Foundation; and has filed a patent on lipid biomarkers. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- American College of Cardiology Foundation
- Fan Y.,
- Li Y.,
- Chen Y.,
- et al.
- Zhu Z.J.,
- Schultz A.W.,
- Wang J.,
- et al.
- Mayr M.
- Stegemann C.,
- Pechlaner R.,
- Willeit P.,
- et al.