Breast cancer patients who later suffer heart damage from one of the world’s most widely used chemotherapy drugs already carry a distinct chemical fingerprint in their blood before a single treatment dose, according to a new study that combined untargeted metabolomics with machine learning to hunt for early warning signs of chemotherapy-induced heart failure. The research, led by a team at the University of Arkansas for Medical Sciences (UAMS), suggests that a simple blood test drawn before chemotherapy could one day identify which women are most vulnerable to the heart-weakening side effects of doxorubicin, allowing doctors to intervene before irreversible damage occurs.

Doxorubicin, an anthracycline antibiotic discovered more than half a century ago, remains one of the most effective chemotherapy agents ever developed. It works by inhibiting topoisomerase II, an enzyme essential for DNA replication, thereby blocking the uncontrolled cell division that drives tumor growth. But its success comes with a devastating trade-off. Clinical analyses have reported occurrence rates of roughly 6 percent for clinically overt cardiotoxicity and 18 percent for subclinical cardiotoxicity among patients treated with the drug, and other studies have found cardiac dysfunction ranging from 30 percent in adult survivors to as much as 60 percent in children. Doxorubicin-induced congestive heart failure is unpredictable, carries a poor prognosis independent of cancer outcomes, and currently cannot be predicted with any validated clinical biomarker.

The clinical challenge lies in the drug’s insidious mechanism of harm. Cardiac toxicity typically begins as subclinical myocardial injury, followed by an early and often asymptomatic decline in the left ventricular ejection fraction, or LVEF, a measure of how much blood the heart pumps with each contraction. By the time symptoms of heart failure appear, damage may already be irreversible. Echocardiography and radionuclide angiography can monitor ventricular function, but the former is operator-dependent while the latter exposes patients to radiation. Cardiac troponins, the standard blood markers of heart muscle injury in emergency settings, have not been validated for detecting chemotherapy-related cardiotoxicity in clinical studies.

The UAMS team enrolled 27 patients with early-stage breast cancer who were scheduled to receive a predefined regimen of doxorubicin (60 mg/m²) combined with cyclophosphamide (600 mg/m²) for four cycles every two weeks at the Winthrop P. Rockefeller Cancer Institute. Blood was drawn at two points: before chemotherapy began, designated T0, and after the first cycle, designated T1. Cardiac function was assessed with multigated acquisition (MUGA) scans at baseline and after the fourth cycle. Patients whose LVEF declined more than 10 percentage points from baseline or fell below 50 percent were classified as the abnormal cardiotoxicity group, while those with a decline of 10 points or less and a post-treatment LVEF of at least 50 percent formed the normal group. Eight patients ultimately fell into the abnormal group and 19 into the normal group.

Plasma samples were processed under strict protocols and shipped frozen on dry ice to Metabolon, Inc., in Durham, North Carolina, for untargeted metabolomic profiling. The platform used a Waters ACQUITY ultra-performance liquid chromatography system coupled to a Thermo Scientific Q-Exactive high-resolution mass spectrometer with a heated electrospray ionization source and Orbitrap mass analyzer. Metabolites were identified by matching ion features against a library of 3,300 purified standard compounds using three criteria: retention index, accurate mass within 10 parts per million, and MS/MS spectral comparison. The raw assay yielded 1,285 metabolites, which were filtered for missing values and low variability, log-transformed, and quality-controlled to produce a final dataset of 54 samples and 1,124 metabolites, of which 913 were chemically annotated.

Statistical analysis of baseline samples, before any chemotherapy had been administered, revealed a striking separation in plasma metabolomic profiles between the two groups, confirmed by principal component analysis. A univariate screen identified 100 metabolites significantly different between the groups at baseline, of which 78 were chemically annotated. Patients who would go on to develop cardiotoxicity showed elevated levels of dicarboxylic fatty acids such as sebacate (a ten-carbon dicarboxylate, with a log2 fold change of 1.33 and a p-value of 0.005) and suberate (an eight-carbon dicarboxylate), along with taurine, inosine, sphingosine, and oxindolylalanine. At the same time, they showed depressed levels of key antioxidant vitamins, including retinol (vitamin A, log2 fold change of −0.54) and alpha-tocopherol (vitamin E, log2 fold change of −1.90), along with reduced pyruvate, a central metabolite of energy metabolism, and diminished phospholipids involved in membrane structure. The researchers interpreted these patterns as evidence of pre-existing impairment in mitochondrial fatty acid oxidation and compromised antioxidant defenses in patients destined for heart injury.

A second analytical strategy tracked how metabolite levels changed dynamically during treatment. Using linear mixed-effects models adjusted for age, race, and body mass index, the team identified 78 metabolites whose trajectories over the first chemotherapy cycle differed significantly between groups. Ten metabolites overlapped between the baseline and longitudinal analyses, including 3-phosphoglycerate, inosine, taurine, suberate, sebacate, and sphingadienine, forming a focused candidate biomarker panel. Pathway enrichment analysis of these longitudinal changes pointed to significant disruptions in three interconnected metabolic networks: galactose metabolism, involving glycerol and sorbitol; purine metabolism, involving xanthine, adenosine, and inosine; and beta-alanine metabolism, involving spermidine and ureidopionic acid. These findings suggest that chemotherapy shifts carbohydrate, nucleotide, and amino acid metabolism in ways that may amplify cardiac vulnerability.

To translate the statistical findings into predictive power, the researchers turned to machine learning. They first narrowed the metabolite lists using stepwise logistic regression guided by the Akaike Information Criterion, then trained Random Forest classifiers with leave-one-out cross-validation to make maximum use of the small sample. At baseline, two metabolites emerged as the strongest predictors: sebacate (importance score 100) and 2-hydroxyhippurate, a xenobiotic metabolite also known as salicylurate (importance score 62.17). The baseline model achieved an accuracy of 81.5 percent and a receiver operating characteristic area under the curve of 0.855, with high specificity of 89.5 percent but moderate sensitivity of 62.5 percent. A second model built on longitudinal changes from the first chemotherapy cycle identified orotate (importance 100), picolinate (importance 55.48), and suberate (importance 44.12) as key predictors, achieving 80 percent accuracy and an AUC of 0.798, again with high specificity and lower sensitivity. The pattern of high specificity with limited sensitivity suggests these markers are more effective at confirming low risk than at catching every case of developing cardiotoxicity, a limitation the authors attribute to the small cohort and class imbalance.

The biological significance of these candidates is rooted in well-characterized cardiac physiology. Dicarboxylic acids such as sebacate and suberate are products of omega-oxidation, an alternate fatty acid breakdown pathway in the endoplasmic reticulum that activates when mitochondrial beta-oxidation is impaired. Elevated levels suggest fatty acid overflow and insufficient carnitine availability, processes that strain mitochondria, impair contractility, and promote cardiac cell death. Elevated orotate, an intermediate in pyrimidine biosynthesis, has previously been linked to endothelial dysfunction through disruption of insulin- and metformin-induced nitric oxide production in blood vessel cells. Picolinate, a metabolite of the tryptophan-kynurenine pathway, has been associated with coronary heart disease and belongs to a family of metabolites implicated in both neuroprotection and neurotoxicity, hinting at systemic connections between chemotherapy’s cardiac and cognitive side effects.

The authors caution that the study is exploratory and has important limitations. With 27 participants, statistical power is limited and the risk of overfitting is real. Screening thresholds were deliberately lenient, and false discovery rate correction was not applied, making the findings hypothesis-generating rather than definitive. Concomitant medications, including beta-blockers and metformin, may have influenced metabolomic profiles, and plasma metabolomics reflects systemic metabolism rather than direct evidence of cardiomyocyte-specific injury. Nevertheless, the research, funded by the National Institute of General Medical Sciences and published in the journal Metabolomics, demonstrates that the blood of patients at risk of chemotherapy heart damage tells a chemically distinct story before treatment even begins. If validated in larger cohorts, such metabolomic signatures could enable clinicians to personalize doxorubicin dosing, deploy cardioprotective strategies preemptively, and preserve the anticancer power of one of oncology’s most valuable weapons without sacrificing the hearts of the patients it saves.

Subject of Research: People

Subject of Research: Biology

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

Article References: Singh, A., Jun, S.-R., Wallis, K., S. Lan, R., Todorova, V., Joseph Su, L., Makhoul, S., & Hsu, P.-C. (2026). Metabolic phenotypes of doxorubicin-induced cardiotoxicity among patients with breast cancer. Metabolomics, 22(4), Article 119. https://doi.org/10.1007/s11306-026-02469-7

Image Credits: AI Generated

DOI: 10.1007/s11306-026-02469-7

Keywords: breast cancer, doxorubicin, cardiotoxicity, untargeted metabolomics, biomarkers, machine learning, LVEF, fatty acid oxidation, Random Forest

Cite Scienmag News

APA
MLA
Chicago

Nathaniel Bowman. (September 9, 2026). Metabolic signatures reveal doxorubicin heart toxicity risk in breast cancer patients. Scienmag. https://scienmag.com/metabolic-signatures-reveal-doxorubicin-heart-toxicity-risk-in-breast-cancer-patients/

Copy citation
Download RIS

Tags: anthracycline chemotherapy side effectsanthracycline-induced heart damageblood-based biomarkers for cardiotoxicityblood-based early detection of cardiotoxicitybreast cancer chemotherapy side effectsbreast cancer treatment toxicity assessmentcardiac health monitoring in cancer therapydoxorubicin heart toxicity riskdoxorubicin-induced heart failure biomarkersearly detection of chemotherapy-related cardiotoxicityidentifying vulnerable patients before chemotherapymachine learning for chemotherapy risk predictionmachine learning for early toxicity detectionmetabolic fingerprinting in cancer patientsmetabolic signaturesmetabolic signatures of chemotherapy toxicitymetabolomic profiling of cancer patientspersonalized cancer therapy risk assessmentpre-treatment blood tests for cardiac riskpredictive blood tests for chemotherapy adverse effectsprevention of chemotherapy-induced heart failureprevention of chemotherapy-related heart damageuntargeted metabolomics in breast cancer treatmentuntargeted metabolomics in cancer treatment