Metabolic phenotypes of doxorubicin-induced cardiotoxicity among patients with breast cancer

Figures

• 

  Fig. 1. Study design and analytical approaches

• 

  Fig. 2. Analysis workflow for the untargeted metabolomic data

• 

  Fig. 3. Principal component analysis of untargeted metabolomic profiles at baseline (T0). PCA plots show sample clustering by cardiotoxicity group (normal vs. abnormal ejection fraction) at baseline with 32.52% variance after data filtering out missing values < 50%, metabolites with standard deviation < 0.15, and after log transformation

• 

  Fig. 4. Differential plasma metabolites between NL vs. ABN at baseline and pre-post differences after the first cycle of DOX-containing chemotherapy depicted by volcano plots and Venn diagram. (A) Volcano plot of baseline metabolites, with the x-axis representing |Log2FC| ≥ 0.5 and the y-axis showing -log10 (P-value). Significant metabolites (P-value < 0.05) were highlighted as colored crosses, with downregulated metabolites in ABN group indicated by blue crosses and upregulated metabolites in ABN group indicated by red crosses. (B) Volcano plot of pre-post metabolite differences after the 1 st cycle of DOX-containing chemotherapy, with effect size on the x-axis and –log₁₀(P-value) on the y-axis. Significant metabolites (P-value < 0.05) were highlighted as colored crosses, with negative effect size metabolites represented by blue crosses (increased in the ABN group at T1) and positive effect size metabolites represented by red crosses (decreased in the ABN group at T1). (C) Venn diagram analysis between baseline and pre-post metabolite differences after the first cycle of DOX-containing chemotherapy

• 

  Fig. 5. Multi-marker analysis with RandomForest modeling at baseline. (A) Metabolites with |Log2FC| ≥ 0.5 and P-value < 0.1 in ANOVA test of comparison at baseline were used as input. Two metabolites -2-hydroxyhippurate (salicylurate) and sebacate (C10-DC) were identified in the final Random Forest model with importance score of 100 and 62.17, respectively. (B) ROC curve from LOOCV-based Random Forest analysis showing moderate discrimination (AUC = 0.855)

• 

  Fig. 6. Multi-marker Random Forest Analysis of pre-post difference after 1 st DOX-containing chemotherapy (T0-T1). (A) Metabolites with P-value < 0.05 from an interaction of cardiotoxicity and time variables in the linear mixed model of pre-post design were used as an input. Three metabolites-orotate, picolinate, and suberate (C8-DC) were identified in the final Random Forest model with importance 100, 55.48, and 44.12. (B) ROC curve from LOOCV-based Random Forest analysis showing moderate discrimination (AUC = 0.798)

• 

  Fig. 7. Pathway enrichment analysis of metabolic perturbations linked to DOX-induced cardiotoxicity in patients with breast cancer

• 

  Fig. 8. Schematic representation of circulating metabolite perturbations associated with doxorubicin treatment and LVEF decline in patients with breast cancer