Exploration of serum biomarkers in heart failure patients with preserved and reduced ejection fractions through analysis of heterogeneity

Fig. 1. Workflow of biomarker analysis in HFpEF and HFrEF (Graphical Abstract).

Fig. 2. OPLS-DA score plots and permutation test results of metabolic profiles among different groups and classification of identified metabolites. A–D: OPLS-DA score plots comparing the metabolic profiles of A: H_A_T vs. N, B: H_A_T vs. HFpEF, C: H_A_T vs. HFrEF, and D: HFpEF vs. HFrEF. Each ellipse represents the 95 % confidence interval for each group, showing separation between the different groups. E–H: Permutation test results for the OPLS-DA models corresponding to E: H_A_T vs. N, F: H_A_T vs. HFpEF, G: H_A_T vs. HFrEF, and H: HFpEF vs. HFrEF, demonstrating model robustness (R2) and predictive power (Q2). I: Distribution of identified metabolites across different confidence levels (Levels 1–4). J: Superclass classification of the identified metabolites, showing the proportions of various metabolite classes.

Fig. 3. Volcano plots and superclass distributions of significantly altered metabolites across comparisons. A–D: Volcano plots showing the log2-fold change (log2FC) vs. -log10 p-value for each metabolite. Red dots represent significantly upregulated metabolites, and blue dots represent significantly downregulated metabolites (p < 0.05). A: H_A_T vs. N, B: HFpEF vs. H_A_T, C: HFrEF vs. H_A_T, D: HFpEF vs. HFrEF. E–H: Superclass distribution of significantly altered metabolites, color-coded by upregulation (red) or downregulation (blue). Metabolites are grouped into superclasses, including benzenoids, lipids, organic acids, and others. E: H_A_T vs. N, F: HFpEF vs. H_A_T, G: HFrEF vs. H_A_T, H: HFpEF vs. HFrEF.

Fig. 4. KEGG pathway enrichment analysis across groups.

Fig. 5. Weighted gene coexpression network analysis (WGCNA) in the N–H_A_T-HFpEF and N–H_A_T-HFrEF groups. A & E: Soft-threshold power selection for network construction in the N–H_A_T-HFpEF (A) and N–H_A_T-HFrEF (E) groups. A power of 7 was chosen on the basis of the scale-free topology model fit (left) and mean connectivity (right). B & F: Cluster dendrogram of metabolites showing distinct coexpression modules via dynamic tree cutting. C & G: Module‒trait relationship heatmaps. D & H: Trend analysis across the N, H_A_T, HFpEF and HFrEF groups.

Fig. 6. ROC curves and boxplots of key metabolites across groups. A-E & N–O: ROC curves showing the diagnostic performance of DG (13:0/20:3(8Z,11Z,14Z)/0:0) [iso2] (A), sphingosine (B), glutarylcarnitine (C), Homo-l-arginine (D), (R)-3-hydroxybutyric acid (E), 5-hydroxy-l-tryptophan (N), and l-kynurenine (O). F: Combined ROC analysis of A-E metabolites for enhancing HFpEF diagnosis. P: Combined ROC analysis of N and O. G–M: Boxplots comparing metabolite levels across groups: normal (N), hypertension with diabetes (H_A_T), HFpEF, and HFrEF. Statistical significance is indicated (∗p < 0.05, ∗∗p < 0.01, ns = nonsignificant).