Characteristic of Metabolic Status in Heart Failure and Its Impact in Outcome Perspective

Abstract

Metabolic alterations have been documented in peripheral tissues in heart failure (HF). Outcomes might be improved by early identification of risk. However, the prognostic information offered is still far from enough. We hypothesized that plasma metabolic profiling potentially provides risk stratification for HF patients. Of 61 patients hospitalized due to acute decompensated HF, 31 developed HF-related events in one year after discharge (Event group), and the other 30 patients did not (Non-event group). The plasma collected during hospital admission was analyzed by an ultra-high performance liquid chromatography time-of-flight mass spectrometry (UPLC-TOFMS)-based metabolomic approach. The orthogonal projection to latent structure discriminant analysis (OPLS-DA) reveals that the metabolomics profile is able to distinguish between events in HF. Levels of 19 metabolites including acylcarnitines, lysophospholipids, dimethylxanthine, dimethyluric acid, tryptophan, phenylacetylglutamine, and hypoxanthine are significantly different between patients with and without event (p < 0.05). Established risk prediction models of event patients by using receiver operating characteristics analysis reveal that the combination of tetradecenoylcarnitine, dimethylxanthine, phenylacetylglutamine, and hypoxanthine has better discrimination than B-type natriuretic peptide (BNP) (AUC 0.871 and 0.602, respectively). These findings suggest that metabolomics-derived metabolic profiling have the potential of identifying patients with high risk of HF-related events and provide insights related to HF outcome.

Keywords: BNP; dimethylxanthine; heart failure; metabolomics; phenylacetylglutamine.

Figure 1

Changes in metabolome of heart failure patients. The OPLS-DA score plot (A) show considerable separation between 30 non-event patients (blue) and 31 heart failure (HF) event patients (green). The ellipse shown in the model represents the Hotelling T2 with 95% confidence. Each point represents one patient, and the cohort to which they belong is color-coded as shown in the legend box. The loading plot (B) of metabolite profiles, each point in represents one metabolite. (C) Score plots of orthogonal projection to latent structure discriminant analysis (OPLS-DA) prediction model is used to determine the success of the models for classifying the event and non-event groups. All of the sample were applied to construct the model, and the Y axis-predicted scatter plots assigned samples to either event or non-event group using a priori cut-off at 0. (D) The top 50 metabolites were ranked according to variable importance in the projection (VIP) score of OPLS-DA model. The heat map shows the ratio of abundance (log2) between event and non-event group. The un-identified metabolites are showed with metabolite ID.

Figure 2

Identification of metabolites between event and non-event groups. (A) Pathway analysis reveals that several pathways were changed in HF patients with event group. Metabolic disturbances are mapped to pathways involved in amino acid metabolism, caffeine metabolism, and purine metabolism. (B) Scheme of metabolic disturbance associated with poor heart failure-related outcomes. Red metabolites: those significantly increasing in event group; blue metabolite: those significantly decreasing in event group; dark metabolites: not measured or no significant difference between event and non-event groups. (C) Venn diagram shows the number of metabolites with VIP greater than 1 and significantly (p < 0.05) altered between event and non-event groups. (D) Significantly different metabolites were correlated with level of BNP by Pearson’s correlation coefficient analysis. * p < 0.05, ** p < 0.01, *** p < 0.001.

Figure 3

ROC AUC analysis plots. Comparison between the AUC with (A) ROC curves for the discrimination of the event patients and non-event patients using BNP, four metabolites (tetradecenoylcarnitine + dimethylxanthine + phenylacetylglutamine + hypoxanthine (TDPH)). (B) ROC curves for the discrimination of the event patients (death) and non-event patients using BNP, four metabolites (tetradecenoylcarnitine + dimethylxanthine + phenylacetylglutamine + hypoxanthine (TDPH)). The AUC values for respective ROCs in event versus non-event (A,C) and event (death) versus non-event (B,D) models are shown. Statistically significant differences between two AUCs with DeLong’s test are indicated *** p < 0.001. Significant increases in ability of prediction as assessed by integrated discrimination improvement (IDI) and the net reclassification improvement (NDI) are indicated # and †, respectively. ### or ††† p < 0.001.