Identification of molecular subtypes and a prognostic signature based on machine learning and purine metabolism-related genes in breast cancer

Abstract

Breast cancer (BC), one of the most prevalent malignant tumors worldwide, lacks efficacious diagnostic biomarkers and therapeutic targets. This study harnesses multi-omics data to identify novel purine metabolism-related genes (PMRG) as potential biomarkers and risk signatures. Univariate Cox regression was employed to assess the correlation between PMRGs and BC patient prognosis, while a Lasso Cox model was constructed to derive a prognostic signature. Gene set enrichment analysis was utilized to investigate functional differences. Kaplan-Meier survival curves were plotted to examine overall survival disparities between these 2 risk groups, with further exploration into the relationship between the prognostic signature, immune landscape, and drug sensitivity. Ultimately, a predictive nomogram was developed based on these findings. BC patients were stratified into 2 distinct molecular subtypes with significantly different prognoses using the identified PMRG-based signature, which comprised 17 PMRGs. This signature emerged as an independent prognosticator for BC and was integrated into a nomogram along with age, chemotherapy/radiotherapy treatment history, and clinical staging to accurately predict patient outcomes. Moreover, the signature showed associations with the tumor immune microenvironment and drug responsiveness, where lower-risk patients exhibited increased chemotherapeutic sensitivity, immune scores, and decreased tumor purity. Gene set enrichment analysis highlighted significant activation in pathways such as the complement and coagulation cascades, ribosome biogenesis, MAPK signaling, cAMP signaling, and drug metabolism pathways in the low-risk group. The derived PMRG-based signature holds promise for predicting the prognosis of BC patients and guiding their clinical management, including immunotherapy interventions.

Keywords: breast cancer; immune landscape; immunotherapy; nomogram; purine metabolism.

Figure 1.

Molecular subtyping of breast cancer based on PMRGs associated with prognosis. (A) Somatic mutation landscape of PMRGs. (B) Volcano plot illustrating differential expression of PMRGs in breast cancer. (C) Hazard ratios of PMRGs relevant to prognosis. (D–F) Consensus clustering of breast cancer based on the selected PMRGs with prognostic significance. (G) Kaplan–Meier survival analysis comparing the prognostic differences among the PMRG-derived molecular subtypes. PMRG = purine metabolism-related genes.

Figure 2.

Construction and evaluation of the PMRG-derived breast cancer risk signature. (A and B) Lasso Cox regression analysis for screening and dimensionality reduction of PMRGs associated with prognosis. (C) Coefficients of the 17 PMRGs in the riskscore. (D–F) Stratification of the TCGA–BRCA cohort into high- and low-risk groups based on the riskscore followed by Kaplan–Meier survival analysis and receiver operating characteristic (ROC) curve analysis. (G–I) Similar stratification, survival analysis, and ROC curve analysis for the GSE96058 cohort using the riskscore. BRCA = breast invasive carcinoma, PMRG = purine metabolism-related genes, TCGA = The Cancer Genome Atlas.

Figure 3.

Association of the PMRG-derived risk signature with clinical, mutational, and treatment features. (A) Heatmap displaying the expression of PMRGs composing the risk signature alongside annotated clinical-pathological information. (B) Comparison of the riskscore between dead and alive patient groups. ****P < .0001. (C) Comparison of the riskscore between patients receiving chemotherapy and those who did not. *P < .05. (D) Oncoplot showing somatic mutations across high and low-risk groups. ****P < .0001. (E) Comparison of tumor mutation burden (TMB) between high and low-risk groups. (F) Scatterplot illustrating the correlation between the riskscore and TMB. PMRG = purine metabolism-related genes.

Figure 4.

Relationship between the PMRG-derived risk signature and drug sensitivity. (A) Comparison of drug sensitivity between high and low-risk groups. *P < .05, **P < .01, ***P < .001, ****P < .0001. (B) Heatmap representing correlations between the riskscore and PMRG gene expression with drug sensitivity. PMRG = purine metabolism-related genes.

Figure 5.

Analysis of the relationship between PMRGs and the tumor immune microenvironment. (A) Comparison of infiltration levels of 22 immune cell types between high and low-risk groups. ns: not significant, *P < .05, **P < .01, ***P < .001, ****P < .0001. (B) Heatmap depicting correlations between the riskscore and its constituent genes with immune cell infiltration. (C) Comparison of stromalscore, immunescore, estimatescore, and tumor purity between high and low-risk groups. ****P < .0001. (D) Comparison of infiltrating progenitor score (IPS) between high and low-risk groups. ****P < .0001. IPS = immunophenoscore, PMRG = purine metabolism-related genes.

Figure 6.

Gene expression profiling distinguishing high and low-risk groups. (A) Gene ontology enrichment analysis of genes differentially expressed between high and low-risk groups. (B) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of genes differentially expressed between high and low-risk groups.

Figure 7.

Development and evaluation of a breast cancer nomogram. (A) Univariate Cox regression analysis identifying prognostic factors for breast cancer. (B) Multivariate Cox regression analysis identifying independent prognostic factors for breast cancer. (C) Nomogram constructed for predicting 1-, 3-, and 5-year overall survival in breast cancer patients. (D) Calibration curve analysis of the nomogram. (E) Decision curve analysis demonstrating superior prognostic performance of the nomogram compared to other prognostic factors. (F) Time-dependent ROC analysis for the nomogram’s predictive accuracy of 1-, 3-, and 5-year overall survival rates. ROC = receiver operating characteristic.