Background: Breast cancer (BC) is the most commonly diagnosed cancer. Currently, mammography and breast ultrasonography are the main clinical screening methods for BC. Our study aimed to reveal the specific metabolic profiles of BC patients and explore the specific metabolic signatures in human plasma for BC diagnosis.
Methods: This study enrolled 216 participants, including BC patients, benign patients, and healthy controls (HC) and formed two cohorts, one training cohort and one testing cohort. Plasma samples were collected from each participant and subjected to perform nontargeted metabolomics and proteomics. The metabolic signatures for BC diagnosis were identified through machine learning.
Results: Metabolomics analysis revealed that BC patients showed a significant change of metabolic profiles compared to HC individuals. The alanine, aspartate and glutamate pathways, glutamine and glutamate metabolic pathways, and arginine biosynthesis pathways were the critical biological metabolic pathways in BC. Proteomics identified 29 upregulated and 2 downregulated proteins in BC. Our integrative analysis found that aspartate aminotransferase (GOT1), L-lactate dehydrogenase B chain (LDHB), glutathione synthetase (GSS), and glutathione peroxidase 3 (GPX3) were closely involved in these metabolic pathways. Support vector machine (SVM) demonstrated a predictive model with 47 metabolites, and this model achieved a high accuracy in BC prediction (AUC = 1). Besides, this panel of metabolites also showed a fairly high predictive power in the testing cohort between BC vs HC (AUC = 0.794), and benign vs HC (AUC = 0.879).
Conclusions: This study uncovered specific changes in the metabolic and proteomic profiling of breast cancer patients and identified a panel of 47 plasma metabolites, including sphingomyelins, glutamate, and cysteine could be potential diagnostic biomarkers for breast cancer.
Keywords: Breast neoplasms; Machine learning; Metabolomics; Plasma; Proteomics.
© 2022. The Author(s).