Preliminary metabolomics-based predictive models for myelosuppression in breast cancer patients with different anthracycline chemotherapy

Abstract

Abstract Background: Myelosuppression is a common adverse effect in breast cancer patients receiving anthracyclines combined with cyclophosphamide chemotherapy. Screening potential biomarkers and building predictive models have implications for clinical management of myelosuppression. Methodology: This study collected 103 breast cancer patients in Shenzhen, China, from September 2020 to January 2022, including two different chemotherapy (Epirubicin or Doxorubicin). The plasma samples were collected 48 hours after chemotherapy. Plasma metabolomics were measured using Ultra-High-Performance Liquid Chromatography-Tandem Q-Exactive. After identification of metabolites and screening of potential differential metabolites, we mapped the key pathways. Further, we established a LASSO-logistic predictive model for myelosuppression. Results: In the group of patients treated with doxorubicin, we identified 19 differential metabolites. These metabolites were primarily associated with 6 metabolic pathways, including sphingolipid metabolic pathway, glycine, serine and threonine metabolic pathway, glycerol phospholipid metabolic pathway, tryptophan metabolic pathway, primary bile acid biosynthesis pathway and purine metabolic pathway. From these, a final logistic regression model was developed, incorporating seven variables. This model exhibited an accuracy of 84.91% in predicting chemotherapy-related myelosuppression, with an impressive area under the ROC curve of 0.9571. Similarly, in patients who received epirubicin, we uncovered 20 differential metabolites that were linked to 14 metabolic pathways, encompassing taurine and hypotaurine metabolism, D-glutamine and D-glutamic acid metabolism, and more. A final logistic regression model was constructed, encompassing 12 variables. This model demonstrated an accuracy of 82.35% in predicting the occurrence of chemotherapy-related myelosuppression, with an area under the ROC curve reaching 0.9198. Conclusion: In this study, potential biomarkers and metabolic pathways related to myelosuppression were screened by non-targeted metabolomics. The prediction models of myelosuppression in different chemotherapy populations was constructed, and the models demonstrated robust discriminatory and predictive capabilities. Our novel findings suggested that metabolomics provided unique insights into formulating individualized medication strategies for anthracyclines combined with cyclophosphamide in breast cancer patients.