Metabolomic profiling for predicting breast cancer treatment toxicities

Breast cancer is one of the most frequently diagnosed cancers and the second leading cause of cancer death among women. With over 4 million breast cancer survivors in the United States (over 7.8 million worldwide), there is growing interest in addressing quality-of-life issues associated with breast cancer treatment (1). Treatment-induced normal tissue toxicities are common side effects that often persist after treatment completion significantly impacting survivors’ quality of life (2). Thus, there is an increasing interest in exploring predictive models and targeted precision interventions to mitigate these side effects.

Metabolomics, a growing field of comprehensive evaluation of metabolites, may provide insight into biological pathways associated with cancer treatment-induced neurologic and metabolic toxicities. Metabolomics reflects overall changes within biological systems; metabolites, the end products of genomic, transcriptomic, and proteomic processes, are closely connected to the functions and characteristics of cells and tissues. Metabolomics may be utilized to predict breast cancer treatment-induced toxicities (3). Further, with the high-dimensional data produced by metabolomics, there is an increasing interest in leveraging machine learning algorithms to produce more robust models (4). Machine learning algorithms including the Least Absolute Shrinkage and Selection Operator (LASSO) and adaptive LASSO, can address the multicollinearity and non-independent data often associated with metabolomics (5).

The study by Piffoux et al. explored the application of metabolomic profiling in predicting breast cancer treatment-related neurologic and metabolic toxicities (6). Using untargeted high-resolution metabolomic profiles from 992 patients with estrogen receptor-positive (ER⁺)/human epidermal growth factor receptor 2-negative (HER2⁻) breast cancer, the study evaluated the effectiveness of machine learning models, specifically the adaptive LASSO, in predicting these toxicities. Results showed that adaptive LASSO performed well with minimal bias and identified key metabolites for future research. Including low-frequency and nonannotated metabolites improved prediction accuracy. Overall, the study demonstrated that metabolomics, when combined with clinical data, provides enhanced predictive power for assessing treatment-related toxicities in breast cancer patients. This is valuable for preventing breast cancer treatment-induced neurologic and metabolic toxicities where the interplay of host metabolism, tumor biology, and external factors may contribute to treatment outcomes (7). Thus, findings from the study may be used in developing robust prediction models for treatment-related toxicities and targets for precision intervention.

Using metabolomic profiling, rigorous predictive modeling, and biological interpretation, Piffoux et al. identified key pathways including amino acid, caffeine, nitrogen, and glyoxylate metabolism as potential contributors to breast cancer treatment-related toxicities (6). Pathways including D-glutamine and D-glutamate metabolism, aminoacyl-tRNA biosynthesis, and arginine and proline metabolism were associated with metabolic toxicities, and disruptions to pathways involving glutamate, arginine, and aspartate metabolism were implicated in neurotoxicity models. Intriguingly, our research also showed that alanine, aspartate, and glutamate metabolic pathway had the lowest false discovery rate (FDR)-adjusted P value and the highest impact value for radiotherapy-induced skin toxicities (8). Further, nitrogen metabolism emerged as an essential contributor to systemic stress and toxicity, with alterations in nitrogen cycling linked to neurologic toxicities, including sensory and motor neuropathies. Disruptions in nitrogen metabolism may promote inflammation and reactive nitrogen species, leading to oxidative damage and neuroinflammation implicated in neuropathy (9,10).

In addition, glyoxylate and dicarboxylate metabolism were pinpointed as critical pathways in treatment-induced metabolic toxicities. These pathways are involved in energy homeostasis and detoxifying reactive metabolites, often dysregulated during cancer treatments (11). Disruptions in glyoxylate metabolism illustrate the complex interplay between endogenous processes and environmental exposures, with prior research showing that glyoxylate metabolism promotes systemic changes in energy balance and metabolic stress, hallmark features of treatment-related toxicities (12,13). Additionally, neurologic toxicities including paresthesia, neurosensory, and neuromotor impairments were frequently observed in Piffoux et al.’s study and were primarily linked to amino acid and nitrogen metabolism, pathways impacting neural health and dysfunction. Moreover, many of the metabolites associated with neurologic toxicities were exogenous in origin, pointing to the potential influence of environmental factors, diet, and microbiome in modulating treatment responses (14,15). On the other hand, metabolic toxicities contribute to cancer treatment-related weight gain (16). Other factors, such as treatment-related fatigue and depression may also contribute to weight gain. These findings highlight the importance of integrating both endogenous and exogenous factors into prediction models to better understand treatment toxicities and inform personalized interventions for breast cancer survivors.

Considering the growing burden of treatment-related toxicities, metabolomics presents a promising approach for identifying predictive biomarkers. However, despite its significant contributions, the study may have a few limitations in translational applicability. First, the study reported moderate or suboptimal performance of the models with area under the curve (AUC) values (about 0.55–0.6) and not all models were validated even though they have a significantly better predictive power than clinical variables alone. These moderate AUC values highlight the need for further refinement in modeling and integration of additional variables and genetic factors to improve predictive accuracy. Second, while the study included some baseline clinical variables, such as diabetes and body mass index (BMI), it did not fully address many potential confounders or effect modifiers, which could influence metabolomic profiles and treatment-related toxicities. Third, as pointed out by the investigators, the study relies on a single baseline metabolomic sample, which may limit its ability to evaluate cancer treatment-induced metabolomic changes that may be critical in predicting normal tissue toxicities. Integrating longitudinal sampling, particularly immediately after cancer treatments may provide a better estimate of the dynamic changes, thereby enhancing the predictive values of metabolomics. Fourth, this study focused on the ER⁺/HER2⁻ subtype breast cancer patients from the prospective CANTO (CANcer TOxicities) cohort. More diverse study populations to include different tumor subtypes may enhance the generalizability of the study findings. Metabolomic profiles are shaped by genetics, diet, microbiota, and environmental exposures all of which vary across diverse breast cancer subtype populations. Fifth, the Piffoux et al.’s study may provide an overall model to explore whether baseline metabolomic profiles could predict the risk of developing neurologic toxicity due to all breast cancer treatments. Adjusting for taxane exposure in their models may contribute to suboptimal predictive values. To refine their data analysis, subgroup analysis stratified by taxane, or chemotherapy may present a better chemotherapy-specific predictive model. Stratification by chemotherapy can lead to more balanced treatment groups, reducing potential bias from confounding and increasing the precision of treatment effect estimates. Lastly, while the study employed a discovery-validation framework, external validations in independent cohorts are critical to validate the study findings.

Deep learning methods did not perform very well compared to “simpler methods” like LASSO in this study. However, this is probably due to limitations in the dataset, including a small sample size, a limited number of events, and data analysis not chemotherapy-specific, rather than issues with deep learning methods themselves. To improve metabolomic-based toxicity prediction models of breast cancer treatments, future research should include longitudinal metabolomic sampling to capture changes induced by cancer treatment and refine prediction models. Integrating metabolomics with other omics data, such as genomics, proteomics, and transcriptomics may provide a more comprehensive understanding of the molecular mechanisms to enhance predictive model accuracy. Internal and external validation with more diverse study cohorts with different tumor subtypes may help to develop more generalizable prediction models. External validation studies are essential to confirm their findings and further support clinical implementation. Applying their prediction modeling framework to other cancer treatments and diseases may further establish metabolomics as a key tool in predicting treatment-induced normal tissue toxicities and targeted precision intervention.

Acknowledgments

None.

Footnote

Provenance and Peer Review: This article was commissioned by the Editorial Office, Translational Cancer Research. The article has undergone external peer review.

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-261/prf

Funding: This study was supported by the National Institutes of Health and the Florida Breast Cancer Foundation [Grant Nos. R01CA135288, R03CA195643, R21CA234880, and R21CA234880-S1 (to J.J.H.)].

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-261/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Heidary Z, Ghaemi M, Hossein Rashidi B, et al. Quality of Life in Breast Cancer Patients: A Systematic Review of the Qualitative Studies. Cancer Control 2023;30:10732748231168318. [Crossref] [PubMed]
2. Vaz-Luis I, Cottu P, Mesleard C, et al. UNICANCER: French prospective cohort study of treatment-related chronic toxicity in women with localised breast cancer (CANTO). ESMO Open 2019;4:e000562. [Crossref] [PubMed]
3. Qiu S, Cai Y, Yao H, et al. Small molecule metabolites: discovery of biomarkers and therapeutic targets. Signal Transduct Target Ther 2023;8:132. [Crossref] [PubMed]
4. Coler EA, Chen W, Melnik AV, et al. Metabolomics in the era of artificial intelligence. Microbiota Host 2024; [Crossref]
5. Lee AM, Hu J, Xu Y, et al. Using Machine Learning to Identify Metabolomic Signatures of Pediatric Chronic Kidney Disease Etiology. J Am Soc Nephrol 2022;33:375-86. [Crossref] [PubMed]
6. Piffoux M, Jacquemin J, Pétéra M, et al. Metabolomic Prediction of Breast Cancer Treatment-Induced Neurologic and Metabolic Toxicities. Clin Cancer Res 2024;30:4654-66. [Crossref] [PubMed]
7. Plaza-Diaz J, Álvarez-Mercado AI. The Interplay between Microbiota and Chemotherapy-Derived Metabolites in Breast Cancer. Metabolites 2023;13:703. [Crossref] [PubMed]
8. McMahon AN, Lee E, Takita C, et al. Metabolomics in Radiotherapy-Induced Early Adverse Skin Reactions of Breast Cancer Patients. Breast Cancer (Dove Med Press) 2024;16:369-77. [Crossref] [PubMed]
9. Lana JV, Rios A, Takeyama R, et al. Nebulized Glutathione as a Key Antioxidant for the Treatment of Oxidative Stress in Neurodegenerative Conditions. Nutrients 2024;16:2476. [Crossref] [PubMed]
10. Ruiter-Lopez L, Khan MAS, Wang X, et al. Roles of Oxidative Stress and Autophagy in Alcohol-Mediated Brain Damage. Antioxidants (Basel) 2025;14:302. [Crossref] [PubMed]
11. Zhang Y, Wang J, Dai N, et al. Alteration of plasma metabolites associated with chemoradiosensitivity in esophageal squamous cell carcinoma via untargeted metabolomics approach. BMC Cancer 2020;20:835. [Crossref] [PubMed]
12. Tufail M, Jiang CH, Li N. Altered metabolism in cancer: insights into energy pathways and therapeutic targets. Mol Cancer 2024;23:203. [Crossref] [PubMed]
13. Ramsay S, Hyvärinen E, González-Arriagada W, et al. Radiation-induced changes in salivary metabolite profile and pathways in head and neck cancer patients. Clin Oral Investig 2025;29:145. [Crossref] [PubMed]
14. Chen L, Zhernakova DV, Kurilshikov A, et al. Influence of the microbiome, diet and genetics on inter-individual variation in the human plasma metabolome. Nat Med 2022;28:2333-43. [Crossref] [PubMed]
15. Xia S, Xu Z, Cheng C, et al. Blood metabolites, protein regulatory networks and their roles in pan-cancer: a mendelian randomisation study. Discov Oncol 2025;16:721. [Crossref] [PubMed]
16. Goodman W, Lally P, Fisher A, et al. Weight Management Experiences Among People Affected by Overweight and Obesity Who Are Living With and Beyond Colorectal, Breast or Prostate Cancer: A Cross-Sectional Survey. Cancer Med 2025;14:e70885. [Crossref] [PubMed]