Development and validation of a Hypoxia and Lactate Metabolism Prognostic Score (HLMPS) for breast cancer using machine learning.

Background: Previous studies often overlooked the roles of hypoxia and lactate metabolism in the breast cancer (BRCA) microenvironment. This study developed and validated a novel prognostic model for BRCA based on hypoxia-related genes (HRGs) and lactate metabolism-related genes (LMRGs) using machine learning approaches. The aim was to identify molecular subtypes capable of predicting patient prognosis and treatment response, thereby facilitating precision medicine strategies for BRCA.

Methods: This study utilized bulk RNA-sequencing data from The Cancer Genome Atlas (TCGA) BRCA cohort (1,079 tumor samples; 99 normal samples) as the training set, with five independent validation cohorts (GSE19615, GSE20685, GSE20711, GSE42568, GSE58812) retrieved from the Gene Expression Omnibus (GEO) database. HRGs and LMRGs were identified from the Molecular Signatures Database (MSigDB). A machine learning-based integrative approach was employed to construct the Hypoxia and Lactate Metabolism Prognostic Score (HLMPS) via 10-fold cross-validation and multiple algorithm combinations. Model robustness was rigorously assessed through Kaplan-Meier survival analysis, time-dependent receiver operating characteristic (ROC) curves, and calibration plots with Brier score quantification.

Results: The HLMPS model demonstrated robust prognostic discrimination, with high-risk patients exhibiting significantly inferior overall survival compared to low-risk counterparts [training set areas under the curve (AUCs): 0.76, 0.77, 0.74 at 1/3/5 years; validation sets AUCs: 0.61, 0.65, 0.67 at 1/3/5 years]. Functional enrichment analysis revealed that patients with a high HLMPS tended to have dysregulation of cell cycle and neurodevelopmental pathways, while those with a low HLMPS exhibited activation of immune pathways, including T-cell receptor (TCR) signaling and antigen presentation. An Immune infiltration analysis showed that patients with a low HLMPS had higher levels of immune cell infiltration and better responsiveness to immunotherapy. Meanwhile, patients with a low HLMPS showed greater sensitivity to drugs such as irinotecan and palbociclib, while patients with a high HLMPS were more sensitive to drugs such as lapatinib and sorafenib.

Conclusions: The HLMPS model represents a novel and clinically actionable tool for prognosticating outcomes and therapeutic responses in BRCA patients. This study highlights the potential of precision medicine strategies that integrate HRGs and LMRGs based on tumor microenvironment (TME) features. Future work should focus on validating the HLMPS model in larger, multicenter cohorts and determining its clinical applicability in guiding personalized treatment decisions for patients with BRCA.