Integrated bioinformatics analysis identifies cuproptosis and ferroptosis gene signatures via subgroup clustering: Prognostic and therapeutic implications in breast cancer

Abstract

Objective

This study aimed to investigate the prognostic value and biological implications of cuproptosis/ferroptosis-related genes in breast cancer, and to develop a robust molecular signature for risk stratification and treatment guidance.

Methods

Integrative bioinformatic analysis of TCGA and GEO datasets identified cuproptosis/ferroptosis-related genes. LASSO and Cox regression were used to refine a multi-gene signature. Unsupervised clustering stratified molecular subtypes. Genomic alterations, immune microenvironment composition, and drug sensitivity were assessed using bioinformatic tools. Experimental validation included immunohistochemistry and Western blot using clinical samples and cell lines (MCF-10A, MDA-MB-231, BT-549, SKBR-3, MCF-7).

Results

A four-gene signature (MTDH, PROM2, IFNG, SLC1A4) with independent prognostic value was established. Subtype-specific expression patterns emerged: IFNG associated with basal-like tumors and cytotoxic immunity, while SLC1A4 (HER2-enriched/luminal) and MTDH/PROM2 correlated with immunosuppressive microenvironments. Disease progression showed stage-dependent attenuation of IFNG/MTDH/PROM2 and progressive SLC1A4 upregulation. Genomic profiling revealed recurrent mutations and frequent copy-number amplifications, notably of MTDH (60.37 %) and loci on chromosomes 2/3. Functional assays demonstrated IFNG/MTDH/SLC1A4-mediated resistance to anthracyclines and microtubule inhibitors, contrasting with PROM2-associated chemosensitivity. Unsupervised clustering identified two molecular subtypes with divergent survival, characterized by distinct pathway activation. Prognostic nomograms achieved robust predictive accuracy (C-index: 0.755). Validation confirmed upregulation of all four genes in breast cancer.

Conclusion

The cuproptosis/ferroptosis-related genes-based signature serves as a reliable prognostic tool, reflecting immune landscape remodeling and genomic instability in BRCA. These findings provide insights into subtype-specific therapeutic vulnerabilities and suggest potential strategies for targeting cuproptosis/ferroptosis pathways in precision oncology.

Highlights

• •First molecular taxonomy linking cuproptosis/ferroptosis imbalance to immune evasion, therapy resistance, and tumor evolution in breast cancer.
• •A 4-gene signature (MTDH, PROM2, IFNG, SLC1A4) with independent prognostic value, integrated into clinically robust nomograms (C-index = 0.754) for precision prognosis.
• •Subtype-specific roles identified: IFNG associates with basal-like tumors and cytotoxic immunity, while SLC1A4/MTDH/PROM2 link to HER2/luminal subtypes and immunosuppression.
• •Functional validation reveals signature genes modulate chemoresistance (IFNG/MTDH/SLC1A4) or enhance sensitivity to anthracyclines/microtubule inhibitors (PROM2).
• •IHC-confirmed tumor-specific upregulation of signature genes bridges transcriptomic dysregulation to protein-level evidence, supporting actionable biomarkers for chemotherapy optimization.

1. Introduction

Breast cancer (BRCA) remains a leading threat to women's health globally, with high incidence and mortality rates necessitating improved prognostic tools and therapeutic strategies [1]. Distant metastasis, particularly to the brain and lymph nodes, is a major cause of treatment failure in BRCA patients [2,3], and recent multi-omics and single-cell studies have shed light on the complex tumor ecosystem and potential therapeutic targets in metastatic BRCA [2,4]. Current prediction models suffer from narrow perspectives and suboptimal accuracy, failing to meet clinical demands for precise risk stratification and personalized treatment. Recent advances in programmed cell death research have identified ferroptosis and cuproptosis as promising therapeutic targets, yet their integrative roles in BRCA pathogenesis-especially in the context of tumor microenvironment remodeling and metastasis-remain underexplored [5].

In ferroptosis research, while iron-dependent lipid peroxidation has been linked to BRCA progression, three critical limitations persist: (1) Most prognostic signatures focus on single cancer types, such as the 15-FRG signature developed for ovarian cancer [6], which may not generalize to BRCA's molecular heterogeneity, including subtype-specific differences in ferroptosis vulnerability; (2) Mechanistic studies predominantly concentrate on basic pathways (e.g., System Xc⁻/GSH/GPX4 axis) [7], with limited exploration of cross-talk with other cell death modalities or tumor microenvironment components like cancer-associated fibroblasts (CAFs) and immunosuppressive macrophages [2,8]; (3) Bibliometric analyses reveal that clinical translation lags behind mechanistic research, particularly regarding immunotherapy and tumor microenvironment interactions [9], despite evidence that ferroptosis regulation modulates responses to immune checkpoint inhibitors [10]. These gaps highlight the need for integrative approaches that bridge single-pathway findings to clinical applicability, as demonstrated by recent multi-omics studies linking ferroptosis to BRCA metastasis [4,11].

In cuproptosis research, this copper-triggered, TCA cycle-disrupting cell death mechanism shows substantial therapeutic potential in BRCA [12], yet critical limitations in current research hinder its clinical utility: (1) Subtype-specific biases predominate, as exemplified by studies focusing exclusively on HER2-positive cohorts [13], which fail to account for luminal or triple-negative breast cancer (TNBC) biology-especially the distinct metabolic and redox profiles of TNBC [3]; (2) Prognostic models rely heavily on non-coding RNA associations [14] rather than elucidating functional synergies with other cell death pathways or multi-omics-derived molecular signatures [4]; (3) Cross-talk with ferroptosis remains underexplored, despite evidence that copper-iron metabolic coupling regulates tumor redox homeostasis and that both pathways are dysregulated in BRCA lymph node metastasis [3]. Notably, even landmark cuproptosis studies in BRCA [13,14] have overlooked the potential of dual-pathway targeting, focusing instead on isolated CRGs or lncRNA networks, while recent cell death-focused multi-omics analyses have highlighted the value of integrating metal-dependent death pathways for prognostic modeling [5].

Furthermore, despite isolated studies on cuproptosis/ferroptosis-related genes (CRGs/FRGs), synergistic interactions between these metal-dependent death pathways in BRCA have not been systematically investigated-especially in the context of the immunosuppressive tumor microenvironment and metastatic progression [2,3]-representing a critical knowledge gap in the field. Leveraging TCGA database and integrating insights from recent BRCA multi-omics [4] and single-cell studies [2], this study employs comprehensive bioinformatics approaches to systematically characterize CRGs/FRGs expression profiles, mutation patterns, and clinicopathological correlations in BRCA, and to construct a prognostic model to unravel the synergistic mechanisms of CRGs/FRGs in tumor progressionand microenvironment remodeling.

This research addresses critical knowledge gaps by integrating two emerging cell death modalities, offering dual-targeted therapeutic potential. By establishing a robust prognostic assessment framework, the findings aim to improve prognostic precision for BRCA patients, identify novel therapeutic vulnerabilities through the synergistic targeting of copper and iron metabolism pathways, and lay the groundwork for personalized treatment paradigms that enhance patient outcomes. Unlike previous studies focusing solely on FRGs [15,16] or cuproptosis mechanism exploration [12,17], our integrated framework demonstrates improved prognostic accuracy (C-index = 0.755) and reveals novel cross-pathway regulatory nodes that could serve as dual therapeutic targets.

2. Materials and methods

2.1. Computational environment

All analyses were performed using a standardized bioinformatics workflow in R (version 4.4.2; https://www.r-project.org/) and Perl (version 5.38.2.2;

https://www.perl.org/get.html). Software configurations and dependencies were managed via Bioconductor and CRAN repositories to ensure reproducibility.

2.2. Data acquisition and curation

2.2.1. Multi-omics data integration

Transcriptomic profiles (RNA-seq FPKM), clinical metadata, and somatic mutation data for BRCA were obtained from The Cancer Genome Atlas (TCGA), encompassing 1118 tumor and 113 normal adjacent tissue (NAT) samples. Independent validation datasets, including microarray expression data (GSE42568) and platform annotations (GPL570), were downloaded from the Gene Expression Omnibus (GEO). Genome-level copy number variation (CNV) data were acquired from the UCSC XENA repository (GDC TCGA Breast Cancer cohort).

2.2.2. Core gene set compilation

A curated set of 19 CRGs was assembled based on a systematic review of peer-reviewed literature [[18], [19], [20], [21], [22], [23]]. The ferroptosis-related gene set (n = 382) was retrieved from FerrDb (http://www.zhounan.org/ferrdb), a manually curated database of ferroptosis regulators.

2.3. Data preprocessing

2.3.1. Cohort harmonization

Raw TCGA transcriptomic data were processed using Perl to construct a unified tumor–NAT matrix (n = 1231). Clinical variables (age, TNM stage, survival status) were reformatted for subsequent survival analyses. GEO survival metadata were annotated using clinical supplements corresponding to each accession. Tumor mutational burden (TMB) was defined as the number of nonsynonymous mutations per megabase, derived from MuTect2-processed Mutation Annotation Format (MAF) files.

2.3.2. Batch effect correction

Technical variations between TCGA and GEO datasets were mitigated using the limma and sva packages for cross-platform normalization and batch effect adjustment.

2.4. Analytical framework

2.4.1. Core analytical modules

The analytical pipeline began with differential expression analysis of CRGs/FRGs between tumor tissues and NATs using moderated t-tests within the limma package (FDR-adjusted P < 0.05). Expression patterns were visualized via hierarchical clustering (pheatmap) and volcano plots (ggplot2). For survival analysis, univariate Cox proportional hazards regression was applied to identify genes associated with prognosis (P < 0.05). Optimal expression cutoffs were determined using maximally selected rank statistics. Patients were stratified into high- and low-risk groups based on these thresholds. Kaplan-Meier (K-M) survival curves were generated, and between-group differences were assessed using log-rank tests.

2.4.2. Multimodal network analysis

Regulatory networks were constructed by integrating protein-protein interaction (PPI) data from STRING (confidence score>0.7; https://cn.string-db.org/) and miRNA-mRNA interactions predicted via GeneMANIA (https://genemania.org/) and TargetScan (https://www.targetscan.org/vert_80/). Networks were visualized using Cytoscape.

A machine learning-based prognostic model was developed using LASSO-penalized Cox regression with 10-fold cross-validation and minimum mean squared error (MSE) criteria for feature selection. A multivariate Cox model was used to compute risk scores as a linear combination of gene expression values and their regression coefficients (Risk Score = $\sum \left({GeneExpression}_i\times {\beta }_i\right)$). Model performance was evaluated using time-dependent receiver operating characteristic (ROC) curves.

Genomic alteration profiling characterized somatic mutations via waterfall plots (maftools) and CNVs through Circos plots (RCircos).

2.4.3. Molecular subtyping

Molecular subtypes were identified via consensus clustering based on cuproptosis and ferroptosis gene expression profiles. Subsequent analyses included pathway enrichment (GSVA with Hallmark gene sets), immune infiltration estimation (CIBERSORTx), and functional annotation (GO and KEGG via clusterProfiler).

2.4.4. Translational validation

Clinical applicability was assessed using a prognostic nomogram that integrated risk scores and clinicopathological variables. Calibration curves and Harrell's concordance index (C-index) were used to evaluate nomogram accuracy. Additionally, expression levels of MTDH, PROM2, IFNG, and SLC1A4 in BRCA were validated using the Breast Cancer Gene-Expression Miner v5.2 (bc-GenExMiner v5.2). Immunohistochemical validation of candidate proteins was performed using the Human Protein Atlas (HPA), supplemented by Western blot analyses conducted in a panel of BRCA cell lines—including MCF-10A, MDA-MB-231, BT-549, SKBR-3, and MCF-7—which were preserved at the Guangxi Key Laboratory of Diagnosis and Treatment for Breast Cancer. Furthermore, paired tumor tissues and adjacent normal specimens were collected from the Department of Breast Surgery, Wuming Hospital Affiliated to Guangxi Medical University, with approval granted by the hospital's Institutional Ethics Committee.

2.5. Statistical design

The statistical approach followed a predefined hypothesis-driven protocol: univariate Cox regression identified candidate genes (P < 0.05), multivariate models were adjusted for age, stage, and molecular subtype, and the Benjamini–Hochberg method was applied to control the false discovery rate (FDR) for multiple testing. All analytical code and processed data have been archived to ensure reproducibility. Statistical significance was defined as two-tailed P < 0.05 unless otherwise indicated.

3. Results

3.1. Transcriptomic profiling and cohort stratification

Integration of TCGA transcriptomic data (1118 tumors vs. 113 NATs) with GEO survival metadata yielded a tumor-specific analytical matrix after batch effect correction using limma and sva (Table 1). Tumor mutational burden (TMB) calculations revealed distinct genomic instability patterns across BRCA molecular subtypes, with somatic mutation frequencies ranging from 0.8 to 2.1 mutations per megabase (see Fig. 1).

Table 1.

TCGA and GEO breast cancer patient characteristics.

Clinical characteristics		TCGA (Total 1097)a	GEO (Total 104)
Age		58 (26–90)	59 (31–90)
Gender	Female	1085	104
	Male	12	0
Stage	Ⅰ	183	NA
	Ⅱ	621	NA
	Ⅲ	249	NA
	Ⅳ	20	NA
	NA	24	NA
Grade	1	NA	11
	2	NA	40
	3	NA	53
	NA	NA	0
T-classification	T1	281	NA
	T2	635	NA
	T3	138	NA
	T4	40	NA
	NA	3	NA
M-classification	M0	912	NA
	M1	22	NA
	MX	163	NA
N-classification	N0	516	45
	N1	364	59
	N2	120	0
	N3	77	0
	NX	20	0
Status	Alive	948	69
	Death	149	35

DataexpressasMean(min–max).

^aThe remaining value after removing the sample with missing data from the total number of people.

Fig. 1.

Flowchart of the study design.

The study design flowchart details a multi-step approach to investigate cuproptosis and ferroptosis-related genes in BRCA. It integrates multiple datasets, performs a series of analyses including regression, clustering, and functional enrichment, and uses various techniques for validation and evaluation.

3.2. Differential gene expression and survival analysis

Comparative transcriptomic profiling between BRCA tumors and paired NATs identified 110 significantly dysregulated genes associated with cuproptosis and ferroptosis (|log₂FC|>1, FDR-adjusted P < 0.05; Fig. 2A–B). K-M survival analysis revealed divergent prognostic associations: elevated expression of 19 genes (G6PD, TFRC, CHAC1 et al.) correlated with reduced overall survival (OS), whereas increased expression of 8 genes (ATP7B, FLT3, IFNG et al.) predicted favorable clinical outcomes (log-rank test, P < 0.05; Fig. 2C–N, Fig. 3). Univariate Cox proportional hazards models further identified 27 CRGs/FRGs with significant prognostic relevance in BRCA (Table 2). Notably, three FRGs-G6PD, SLC7A5, and MTDH-were identified as independent risk factors for disease progression, while the CRG ATP7B functioned as a protective prognostic factor (Fig. 4A).

Fig. 2.

Transcriptomic profiling and survival significance of cuproptosis- and ferroptosis-associated genes in BRCA.

(A) Heatmap visualization of differentially expressed genes (DEGs) linked to cuproptosis and ferroptosis across BRCA cohorts.

(B) Volcano plot of cuproptosis/ferroptosis-associated DEGs, highlighting significantly upregulated (red; FDR-adjusted P < 0.05, log2FC > 1) and downregulated (green; FDR-adjusted P < 0.05, log2FC < −1) genes.

(C–N) Kaplan-Meier survival curves comparing overall survival between high- and low-expression groups of cuproptosis/ferroptosis-associated genes.

Fig. 3.

Prognostic impact of cuproptosis- and ferroptosis-associated gene expression on survival outcomes in BRCA.

(A–O) Kaplan-Meier survival analyses comparing overall survival between high- and low-expression groups of cuproptosis/ferroptosis-associated genes.

Table 2.

Univariable Cox proportional hazards regression analysis of prognostic biomarkers in BRCA.

ID	HR (95 % CI)	P-value
MTDH	1.4167 (1.1701–1.7152)	0.0004
G6PD	1.3478 (1.1434–1.5887)	0.0004
SLC7A5	1.1770 (1.0731–1.2909)	0.0005
PROM2	1.3326 (1.1230–1.5813)	0.0010
NGB	1.3611 (1.1166–1.6591)	0.0023
TP63	0.8544 (0.7665–0.9524)	0.0045
LONP1	1.4067 (1.1062–1.7889)	0.0054
IFNG	0.7438 (0.6013–0.9199)	0.0063
HILPDA	1.2537 (1.0643–1.4767)	0.0068
HSF1	1.3528 (1.0863–1.6846)	0.0069
CHAC1	1.2043 (1.0520–1.3786)	0.0071
TRIB3	1.2588 (1.0643–1.4889)	0.0072
AURKA	1.1865 (1.0460–1.3459)	0.0078
SLC1A4	0.8362 (0.7329–0.9542)	0.0079
FLT3	0.8652 (0.7748–0.9663)	0.0102
IL33	0.8841 (0.8044–0.9717)	0.0106
PRDX1	1.3417 (1.0703–1.6819)	0.0108
VEGFA	1.2238 (1.0464–1.4313)	0.0115
ATP7B	0.8561 (0.7563–0.9689)	0.0139
CA9	1.0833 (1.0132–1.1583)	0.0191
SLC2A1	1.1848 (1.0255–1.3689)	0.0214
TFRC	1.1573 (1.0108–1.3250)	0.0344
SLC3A2	1.2933 (1.0187–1.6418)	0.0347
FANCD2	1.2091 (1.0078–1.4507)	0.0410
CXCL2	0.8929 (0.7989–0.9980)	0.0460
PLIN4	0.9278 (0.8613–0.9995)	0.0484
RRM2	1.1293 (1.0007–1.2745)	0.0487

Fig. 4.

Molecular network architecture and prognostic modeling of cuproptosis- and ferroptosis-associated genes in BRCA.

(A) Prognostic co-expression network of cuproptosis/ferroptosis-associated DEGs, highlighting hub genes and their regulatory interactions.

(B) PPI network of prognostic DEGs constructed via the STRING database.

(C) The length of the column reflects the number of neighboring interactors.

(D) LASSO regression coefficient profiles for 13 cuproptosis/ferroptosis-associated prognostic genes, illustrating feature selection under regularization.

(E) Ten-fold cross-validation curve for optimal penalty parameter (λ) selection in LASSO regression, with λ.min corresponding to minimal deviance.

(F) Forest plot of multivariable Cox regression analysis identifying four independent prognostic genes (MTDH, IFNG, SLC1A4, PROM2).

(G–J) K-M survival curves comparing overall survival between high- and low-expression groups of MTDH (G), IFNG (H), SLC1A4 (I), and PROM2 (J). Log-rank P-values and HRs are annotated.

(K) K-M survival analysis stratified by high-versus low-risk groups defined by the cuproptosis/ferroptosis prognostic signature, demonstrating significant survival divergence.

To delineate functional interactions among these genes, PPI networks were constructed using the STRING database, retaining high-confidence interactions (confidence score>0.7). Topological analysis identified SLC2A1, SLC3A2, IFNG, and SLC7A5 as central network hubs with the highest node connectivity, suggesting pivotal roles in mediating cuproptosis-ferroptosis crosstalk (Fig. 4B–C). Feature selection via LASSO regression with 10-fold cross-validation narrowed the candidate predictors to 13 genes, with the optimal regularization parameter λ selected by minimizing the mean squared error (Table 3, Fig. 4D–E). Multivariate Cox regression further refined the prognostic model to four core biomarkers: MTDH and PROM2 as risk factors (hazard ratio [HR]>1), and IFNG and SLC1A4 as protective factors (HR < 1) (Table 4, Fig. 4F–J). Risk stratification based on optimal expression cutoffs demonstrated significantly worse OS in high-risk patients compared to low-risk patients (P < 0.0001; Fig. 4K).

Table 3.

LASSO regression analysis of prognostic gene selection in BRCA.

Gene Symbol	Coefficient Estimate
MTDH	0.338
IFNG	−0.267
SLC1A4	−0.213
PROM2	0.204
NGB	0.172
G6PD	0.150
LONP1	0.136
ATP7B	−0.101
TP63	−0.067
CXCL2	−0.048
FLT3	−0.028
TRIB3	0.025
SLC7A5	0.009

Table 4.

Multivariable Cox proportional hazards regression analysis of independent prognostic factors in Breast Cancer.

Gene Symbol	HR (95 % CI)	P-value
MTDH	1.5134 (1.2250–1.8697)	0.0001
IFNG	0.7074 (0.5680–0.8810)	0.0020
SLC1A4	0.7631 (0.6485–0.8978)	0.0011
PROM2	1.2887 (1.0831–1.5333)	0.0042
NGB	1.2139 (0.9748–1.5117)	0.0833
G6PD	1.1855 (0.9742–1.4427)	0.0893
LONP1	1.2150 (0.9402–1.5702)	0.1367
ATP7B	0.8712 (0.7474–1.0155)	0.0779
TP63	0.9298 (0.8212–1.0528)	0.2511
CXCL2	0.9236 (0.8160–1.0453)	0.2081
FLT3	0.9523 (0.8477–1.0696)	0.4094
TRIB3	1.0486 (0.8692–1.2651)	0.6200
SLC7A5	1.0080 (0.8934–1.1372)	0.8973

3.3. Clinicogenomic correlations and molecular characterization

Integrative analysis of the four-gene signature revealed distinct clinicopathological associations and subtype-specific expression patterns in BRCA. The protective factors IFNG and SLC1A4 showed divergent expression profiles: IFNG was overexpressed predominantly in basal-like tumors, while SLC1A4 was enriched in HER2-enriched, Luminal A, and Luminal B subtypes. In contrast, the risk factors MTDH and PROM2 were preferentially upregulated in basal-like and HER2-enriched carcinomas, respectively (Fig. 5A–D). Longitudinal assessment across pathological stages indicated progressive downregulation of IFNG, MTDH, and PROM2 with advancing disease, whereas SLC1A4 expression increased with stage (Fig. 5E), collectively implicating these CRGs/FRGs in disease progression.

Fig. 5.

Molecular subtype-specific expression dynamics, functional networks, and immune correlations of cuproptosis/ferroptosis-associated genes in BRCA.

(A–D) Subtype-dependent expression patterns of IFNG (A), SLC1A4 (B), MTDH (C), and PROM2 (D) across BRCA molecular subtypes (basal-like, HER2-enriched, Luminal A/B). Boxplots denote median expression with significance thresholds.

(E) Temporal expression dynamics of IFNG, MTDH, PROM2, and SLC1A4 across pathological stages (I-IV), demonstrating progressive attenuation (IFNG/MTDH/PROM2) versus induction (SLC1A4) with disease progression.

(F) Functional interactome of the four-gene signature generated via GeneMANIA, highlighting co-expressed/co-functional partners (20 genes) involved in ferroptosis/cuproptosis regulation. Edge thickness reflects interaction confidence scores.

(G) Immune infiltration correlations of IFNG (positive) and SLC1A4/MTDH/PROM2 (negative) with cytotoxic lymphocytes (CD8⁺ T cells, NK cells) and immunosuppressive subsets (Tregs, neutrophils) quantified.

(H) Epigenetic-immune interplay: Inverse correlations between IFNG promoter methylation and immune infiltration versus positive associations of SLC1A4/PROM2 hypermethylation with immunosuppressive niches.

Functional annotation via GeneMANIA identified 20 co-expressed and functionally interactive partners (Fig. 5F). Immune landscape analysis revealed dichotomous immunoregulatory roles: the immunosuppressive triad (SLC1A4, PROM2, MTDH) correlated negatively with cytotoxic effectors (NK cells, CD8⁺ T cells) but positively with regulatory T cell subsets (nTregs, Tr1). In contrast, the immunostimulatory hub IFNG correlated positively with broad immune infiltration, cytotoxic lineages (Th1 cells, exhausted CD8⁺ T cells), and central memory cells, while correlating inversely with immunosuppressive neutrophils (Fig. 5G). Epigenetic analyses further supported these observations, showing an inverse correlation between IFNG promoter methylation and immune infiltration, whereas hypermethylation of PROM2 and SLC1A4 was associated with immunosuppressive microenvironments (Fig. 5H).

Genomic profiling identified recurrent somatic alterations among prognostic genes. FANCD2, ATP7B, PLIN4, CA9, and PROM2 exhibited the highest mutation frequencies (incidence≥1 %; C > T transitions predominant), whereas HILPDA, VEGFA, and NGB showed no mutations (Fig. 6A). CNV analysis revealed widespread amplification events (100 % penetrance), with chromosomes 2 and 3 identified as genomic vulnerability hotspots. Key loci on chromosome 2 included RRM2 (2p25.1; involved in ribonucleotide reduction), SLC1A4 (2p16.1; glutamate transporter), and PROM2 (2q11.2; regulates membrane curvature). Chromosome 3 harbored functionally pivotal genes such as FANCD2 (3p25.3; DNA repair), TP63 (3q28; epithelial differentiation), and TFRC (3q29; iron acquisition), collectively displaying the highest CNV density among prognosis-associated genes (Fig. 6B–C). Focal CNV assessment of core prognostic markers showed MTDH as the most frequently amplified (60.37 % amplifications vs. 2.69 % deletions), followed by IFNG (24.26 % amplifications vs. 10.09 % deletions; Table 5,Fig. 6D).

Fig. 6.

Genomic and epigenetic landscapes of cuproptosis/ferroptosis-associated genes in BRCA: mutational profiles, therapeutic correlations, and regulatory networks.

(A) Mutational landscape of cuproptosis/ferroptosis-associated genes, visualized as a waterfall plot. C > T transitions dominate somatic alterations.

(B) CNV frequencies of cuproptosis/ferroptosis genes.

(C) Circos plot mapping chromosomal loci of recurrent CNVs in cuproptosis/ferroptosis genes, with chromosomes 2 (RRM2, SLC1A4, PROM2) and 3 (FANCD2, TP63, TFRC) identified as genomic instability hotspots.

(D) Proportional distribution of CNV subtypes (heterozygous/homozygous amplifications/deletions) for the four-gene prognostic signature (MTDH, IFNG, SLC1A4, PROM2).

(E) Pan-cancer Spearman correlation analysis of gene expression with drug sensitivity.

(F) Pathway modulation signatures mediated by the four-gene hub.

(G) Experimentally validated miRNA-mRNA regulatory network upstream of the four-gene signature, reconstructed using Cytoscape.

Table 5.

Copy number alterations of four CRGs/FRGs in BRCA.

Gene symbol	IFNG	MTDH	PROM2	SLC1A4
Total amp. (%)	24.26	60.37	13.15	16.76
Total dele. (%)	10.09	2.69	17.04	15.37
Hete amp. (%)	21.57	45.56	12.50	15.56
Hete dele. (%)	10.09	2.69	17.04	15.37
Homo amp. (%)	2.69	14.81	0.65	1.20
Homo dele. (%)	0.00	0.00	0.00	0.00

Amplification,Amp., Deletion, Dele., Heterozygous, Hete., Homozygou, Homo.

Total amp.(%): the percentage of samples with copy number amplification, including heterozygous and homozygous amplification.

Total dele.(%): the percentage of samples with copy number deletion, including heterozygous and homozygous deletion.

Hete amp.(%): the percentage of samples with copy number heterozygous amplification.

Hete dele.(%): the percentage of samples with copy number heterozygous deletion.

Homo amp.(%): the percentage of samples with copy number homozygous amplification.

Homo dele.(%): the percentage of samples with copy number homozygous deletion.

Chemosensitivity profiling indicated that IFNG, MTDH, and SLC1A4 were associated with resistance to anthracyclines (e.g., doxorubicin) and microtubule-targeting agents (e.g., vincristine), while PROM2 overexpression correlated with increased drug sensitivity (Fig. 6E). Pathway activation mapping revealed mechanism-specific profiles: IFNG activated apoptosis, DNA damage response, and mTOR signaling but suppressed hormone receptor pathways; MTDH promoted cell cycle progression and RTK signaling while inhibiting RAS-MAPK; PROM2 activated RTK signaling alongside cell cycle arrest; and SLC1A4 activated steroid receptors but suppressed PI3K-Akt signaling (Fig. 6F). Reconstruction of upstream regulatory networks identified functionally validated miRNA–mRNA interactions, including miR-493-5p targeting MTDH and miR-371-5p regulating IFNG, visualized using integrated Cytoscape networks (Fig. 6G).

3.4. Molecular subtyping and functional annotation

Unsupervised consensus clustering of BRCA samples based on CRGs/FRGs expression profiles stratified patients into two molecularly distinct subtypes (A and B). Principal component analysis revealed clear separation between subtypes, with pronounced spatial segregation of subtypes A and B in multivariate projection plots (Fig. 7A–C). Survival outcomes differed significantly across subtypes (P < 0.001), with subtype A exhibiting the most favorable median OS (Fig. 7D). Transcriptomic heatmap visualization revealed polarized expression patterns: CRGs/FRGs were broadly upregulated in subtype A but markedly suppressed in subtype B, suggesting divergent susceptibilities to regulated cell death (Fig. 7E).

Fig. 7.

Molecular subtyping and functional annotation of cuproptosis/ferroptosis-associated genes in BRCA: clustering, survival, and pathway Dynamics

(A) Consensus clustering heatmap of BRCA samples based on cuproptosis/ferroptosis-associated gene expression, revealing two molecular subtypes (A and B).

(B) Consensus clustering index plot illustrating sample assignment stability across clustering iterations.

(C) Cumulative distribution function (CDF) curve analysis for optimal cluster number determination, with delta area calculations.

(D) K-M survival curves comparing overall survival across molecular subtypes.

(E) Hierarchical clustering heatmap of 1309 DEGs across subtypes.

(F) Immune cell infiltration disparities among subtypes

(G) GSVA of KEGG pathways across subtype comparisons.

(H) PCA plot demonstrating robust spatial segregation of subtypes in transcriptomic space.

(I) Venn diagram of co-expressed DEGs across subtype pairs.

(J) GO enrichment analysis of DEGs.

(K) KEGG pathway enrichment map highlighting PI3K-Akt signaling pathway as the central mechanistic hub modulated by subtype-specific DEGs.

Subtype A demonstrated significantly higher infiltration levels of anti-tumor effector cells—including activated CD8⁺ T cells, Th1 cells, CD56^diᵐ NK cells, and activated dendritic cells-compared to subtype B. In contrast, immunosuppressive cells such as Tregs and MDSCs were less abundant in subtype A than in subtype B, indicating an activated immune phenotype with strong anti-tumor potential in subtype A. Conversely, subtype B showed reduced infiltration of effector immune cells and elevated immunosuppressive cell populations, suggesting a tumor-promoting immune environment conducive to immune escape and treatment resistance (Fig. 7F). Gene Set Variation Analysis (GSVA) highlighted subtype-specific pathway activities. Subtype A showed relative inhibition, while subtype B exhibited activation across several key pathways (Table 6, Fig. 7G). Cross-subtype differential expression analysis identified 1309 core genes distinguishing the two subtypes (Fig. 7H–I). Functional annotation via GO analysis indicated significant enrichment in terms related to the collagen-containing extracellular matrix (Table 7, Fig. 7J). KEGG pathway analysis further implicated the PI3K-Akt signaling pathway as a central mechanism influenced by subtype-specific gene expression patterns (Table 8, Fig. 7K).

Table 6.

GSEA of top 10 differentially activated pathways across BRCA molecular subtype comparisons.

ID	P	adj.P
KEGG_Cell_Cycle	6.38E-107	1.19E-104
KEGG_DNA_Replication	8.57E-93	7.97E-91
KEGG_Pentose_Phosphate_Pathway	4.99E-91	3.09E-89
KEGG_Proteasome	1.03E-88	4.80E-87
KEGG_Pyrimidine_Metabolism	3.39E-78	1.26E-76
KEGG_Homologous_Recombination	5.64E-78	1.75E-76
KEGG_Cysteine_and_Methionine_Metabolism	4.68E-72	1.24E-70
KEGG_Mismatch_Repair	1.91E-71	4.44E-70
KEGG_Oocyte_Meiosis	2.42E-69	5.01E-68
KEGG_Fructose_and_Mannose_Metbolism	2.61E-62	4.86E-61

Table 7.

GO functional enrichment analysis of top 5 significantly enriched terms in cuproptosis- and ferroptosis-associated molecular subtypes of BRCA.

GO category	Enriched term description	P	adj.P	Count
BP	Mitotic nuclear division	6.69E-21	3.69E-17	66
	Mitotic sister chromatid segregation	2.29E-20	6.30E-17	53
	Nuclear division	1.32E-19	2.42E-16	84
	Organelle fission	2.75E-19	3.79E-16	89
	Sister chromatid segregation	4.32E-19	4.76E-16	57
CC	Collagen-containing extracellular matrix	4.01E-34	2.04E-31	104
	Condensed chromosome	8.10E-11	2.06E-08	49
	Outer kinetochore	2.55E-10	4.33E-08	12
	Chromosome, centromeric region	7.22E-10	9.18E-08	45
	Condensed chromosome, centromeric region	9.72E-10	9.89E-08	36
MF	Extracellular matrix structural constituent	6.91E-19	7.59E-16	48
	Heparin binding	1.81E-15	9.97E-13	44
	Glycosaminoglycan binding	1.26E-14	4.61E-12	52
	Sulfur compound binding	1.97E-13	5.41E-11	54
	Integrin binding	1.18E-08	2.59E-06	32

Table 8.

KEGG pathway enrichment analysis of top 5 significantly enriched pathways in cuproptosis- and ferroptosis-associated differentially expressed genes across molecular subtypes of BRCA.

Pathway	P	adj.P	Count
Cell cycle	4.88E-15	1.57E-12	42
PPAR signaling pathway	9.21E-08	1.48E-05	20
ECM-receptor interaction	1.45E-06	1.41E-04	20
Oocyte meiosis	1.76E-06	1.41E-04	26
PI3K-Akt signaling pathway	1.07E-05	6.85E-04	47

3.5. Molecular taxonomy and prognostic stratification

Unsupervised consensus clustering based on cuproptosis/ferroptosis-related gene expression defined three molecular subtypes (A, B, and C) with high cluster stability (Fig. 8A–C). Subtype A was associated with significantly better clinical outcomes compared to subtypes B and C (Fig. 8D). Hierarchical clustering of differentially expressed genes revealed a polarized regulatory pattern: cuproptosis/ferroptosis regulators were upregulated in subtypes A and B but downregulated in subtype C (Fig. 8E). Among the 27 prognostic CRGs/FRGs (e.g., ATP7B), most were upregulated in subtype B and downregulated in subtype A, illustrating pronounced subtype-specific regulation (Fig. 8F). The significant concordance between CFRGcluster (A, B) and genecluster (A, B, C) supported the reliability of the classification and gene expression patterns (Fig. 8G–H). PCA based on cuproptosis/ferroptosis signatures stratified patients into high- (PC1>1) and low-score (PC1≤1) groups using maximally selected rank statistics. Patients in the high-score group showed significantly prolonged median overall survival (11.7 vs. 9.8 years, P < 0.001; Fig. 8I), a finding validated through Kaplan-Meier analysis (Fig. 8J–K). Sankey diagram visualization illustrated dynamic transitions between subtypes (Fig. 8L).

Fig. 8.

Molecular taxonomy, prognostic stratification, and inter-subtype dynamics of cuproptosis/ferroptosis-associated transcriptomic profiles in BRCA.

(A) Consensus clustering heatmap of BRCA samples based on DEGs linked to cuproptosis/ferroptosis, stratifying patients into three molecularly distinct subgroups.

(B) Consensus clustering index plot evaluating sample assignment stability across iterative clustering runs (k = 2–10), with k = 3 demonstrating optimal cluster robustness.

(C) Cumulative distribution function (CDF) delta area plot for determining optimal cluster number, highlighting maximal relative change at k = 3.

(D) K-M survival analysis of molecular subgroups, revealing significant survival disparities.

(E) Transcriptomic heatmap of DEGs across subgroups.

(F) Differential expression analysis of cuproptosis/ferroptosis-associated genes between molecular subgroups.

(G) Boxplots of cuproptosis/ferroptosis gene expression across molecular subtypes (A, B, C).

(H) Genotype-phenotype associations: Boxplots comparing gene expression distributions.

(I) Survival analysis stratified by PCA-derived risk scores.

(J–K) Scatterplots correlating PCA scores with survival status (alive/deceased) and time-to-event.

(L) Sankey diagram visualizing dynamic inter-subtype transitions and clinicopathological parameter associations.

Multivariate Cox regression incorporating clinicopathological covariates (age>65, gender, TNM stage) identified the PCA-derived CFRG score as an independent prognostic factor. The high-CFRG-score group exhibited significantly better survival across most clinical stages, suggesting its broad prognostic relevance. Specifically, the score maintained significant prognostic value across age and gender subgroups (Fig. 9A–C). In nodal staging (N), high CFRG scores were associated with improved survival in N0, N2, and N3 stages, though not in N1, indicating that combined lymph node and CFRG scoring enhances prognostic accuracy (Fig. 9D–G). Similarly, among patients with T2-stage tumors, a high CFRG score strongly predicted favorable outcomes (Fig. 9H–K). Immune correlation analysis suggested that the survival advantage in the high-CFRG group may be linked to increased infiltration of immune effector cells (e.g., activated T cells and NK cells), which enhance anti-tumor immunity. In contrast, type 1/type 2 helper T cells correlated negatively with CFRG score, indicating reduced infiltration in high-scoring patients (Fig. 9L). Immune checkpoint analysis revealed significant downregulation of BTLA and eight other checkpoint genes in the high-CFRG group, pointing to robust associations and potential therapeutic relevance for immune checkpoint inhibitors (Fig. 10A–H).

Fig. 9.

Clinical stratification and immune correlates of prognosis in BRCA: survival and microenvironmental associations.

(A–B) K-M survival analysis stratified by age (≤65 vs. >65 years).

(C) Survival comparison by gender.

(D–G) Nodal stage (N0/N1/N2/N3)-dependent survival disparities.

(H–K) Tumor stage (T1/T2/T3/T4)-specific survival outcomes.

(L) Immune cell correlation analysis of PCA-derived prognostic scores.

Fig. 10.

Prognostic nomogram validation and immune checkpoint associations in BRCA: predictive modeling and therapeutic implications.

(A–H) Differential expression analysis of immune checkpoint molecules (BTLA, CD274 [PD-L1], CTLA4, HAVCR2 [TIM-3], LAG3, PDCD1 [PD-1], PDCD1LG2 [PD-L2], SIRPA) between high- and low-PCA risk groups.

(I) Prognostic nomogram integrating clinicopathological parameters (age, TNM stage) and ferroptosis/cuproptosis PCA risk scores to predict 1-, 3-, and 5-year survival probabilities. Total points correlate with escalating mortality risk.

(J) Calibration plot demonstrating concordance between nomogram-predicted and observed survival outcomes (C-index = 0.755, 95 % CI: 0.712–0.798). Proximity to the diagonal confirms model accuracy.

(K) Time-dependent receiver operating characteristic (ROC) analysis of the nomogram, yielding area under the curve (AUC) values of 0.826 (1-year), 0.768 (3-year), and 0.757 (5-year).

(L) Comparative ROC analysis validating the nomogram's superior predictive performance over standalone clinical variables.

3.6. Prognostic nomogram construction and validation

By integrating clinicopathological parameters with cuproptosis/ferroptosis-related molecular signatures, we constructed a prognostic nomogram using multivariate Cox regression. The model identified the CFRGscore (derived from PCA) as a protective factor, in contrast to other risk variables. The nomogram incorporated five key predictors to estimate 1-, 3-, and 5-year survival probabilities, which were 0.99, 0.941, and 0.887, respectively, with survival probability decreasing over time. The length of the “Points” axis visually reflected the weight of each variable. The CFRGscore exhibited the widest point axis span, indicating its central role in the model and underscoring the independent prognostic value of the molecular signature (Fig. 10I). Calibration curves demonstrated high agreement between predicted and observed outcomes, with a concordance index (C-index) exceeding 0.75, reflecting strong discriminatory power. The narrow 95 % confidence interval indicated result reliability, and the close fit of the calibration curve further confirmed the accuracy of survival predictions (Fig. 10J). Time-dependent ROC analysis yielded area under the curve (AUC) values of 0.826, 0.768, and 0.757 for 1-, 3-, and 5-year survival, respectively (Fig. 10K). The integration of CFRGscore with clinical features significantly improved predictive performance compared to any single indicator, supporting the utility of multidimensional prognostic modeling (Fig. 10L).

3.7. Protein-level validation via immunohistochemistry

Analysis using bc-GenExMiner revealed progressively elevated expression of the four genes-MTDH, PROM2, SLC1A4, and IFNG-in healthy individuals, adjacent normal tissues, and tumor tissues, respectively (Fig. 11A–D). Immunohistochemistry results from the Human Protein Atlas (HPA) database confirmed significantly higher protein expression of these markers in BRCA tissues compared to normal adjacent samples (Fig. 11E). Western blot analysis further validated these findings, showing upregulated expression of all four proteins across multiple BRCA cell lines relative to normal breast epithelial cells. Additionally, we performed Western blot analysis on clinical samples comprising tumor tissues and paired adjacent normal tissues from BRCA patients. The results consistently demonstrated elevated protein expression in tumor samples, corroborating the transcriptomic findings (Fig. 11F). This concordance across transcriptional and translational levels reinforces the potential involvement of these genes in BRCA tumorigenesis and progression. Their marked overexpression highlights them as promising candidates for future therapeutic targeting.

Fig. 11.

Validation of prognostic signature protein overexpression.

The results of the bc-GenExMiner data show that the expression of four genes, MTDH (A), PROM2 (B), SLC1A4 (C), and IFNG (D), is sequentially upregulated in healthy individuals, adjacent tumor tissues, and tumor tissues.

(E) The IHC results from the HPA database indicate that the protein expression levels in BRCA tissues are significantly higher than those in adjacent normal tissues.

(F) Western blot results show that compared to normal breast epithelial cells and adjacent non-cancerous tissues, the expression levels of the aforementioned four genes are increased to varying degrees in different BRCA cell lines and BRCA tissues. Scale bars: 200 μm.

4. Discussion

Breast cancer remains a leading cause of morbidity and mortality among women worldwide [24], with distant metastasis (especially brain and lymph node metastasis) being the primary driver of treatment failure [2,3]. Single-cell RNA sequencing studies have revealed the complex cellular landscape of metastatic breast cancer, highlighting the critical role of tumor microenvironment (TME) remodeling and cell-cell communication in progression [2]. Current prognostic models are often limited in perspective and accuracy, underscoring the need for more effective therapeutic strategies and precise assessment tools. Recent advances have highlighted regulated cell death mechanisms as promising avenues for novel anticancer therapies, with multi-omics analyses providing robust frameworks to unravel their synergistic roles [4,5].

Ferroptosis, an iron-dependent form of cell death driven by lipid peroxidation, plays a critical role in BRCA progression [25]. Dysregulation of FRGs affects key processes such as proliferation and apoptosis, influencing patient outcomes [[26], [27], [28]]. However, the relationship between FRGs expression and prognosis remains incompletely understood, hindering prognostic and therapeutic optimization. More recently, cuproptosis has emerged as a novel cell death pathway, initiated by copper binding to lipoylated TCA cycle proteins, resulting in protein aggregation and cell death [20]. Like ferroptosis, cuproptosis shows significant therapeutic potential, with copper concentrations modulating CRGs expression and influencing tumor survival across cancers [29,30], and its dysregulation has been linked to breast cancer lymph node metastasis [3].

While isolated studies on CRGs or FRGs exist, integrated analyses of both mechanisms in breast cancer are lacking. Their cooperative role in carcinogenesis and prognosis remains poorly elucidated. Here, we performed a comprehensive multi-omics analysis of CRGs/FRGs in BRCA, aligning with recent trends in pan-cancer multi-omics research that emphasize integrative molecular characterization [4]. Integrating data from TCGA and GEO, we developed a prognostic signature-CFRGscore-based on four genes (MTDH, PROM2, IFNG, SLC1A4), which served as an independent prognostic factor enabling effective risk stratification. Unsupervised clustering further revealed three novel molecular subtypes with distinct clinical outcomes, immune microenvironments, and pathway activities. Genomic analyses indicated that CNVs and mutations may drive dysregulation of these genes. Finally, protein-level validation via bc-GenExMiner, immunohistochemistry, and Western blot confirmed differential expression of core genes, strengthening the clinical relevance of our findings.

Firstly, transcriptomic profiling of BRCA tissues identified 110 differentially expressed genes (DEGs) at the intersection of cuproptosis and ferroptosis pathways, with 27 genes demonstrating significant prognostic value. Notably, SLC7A11 and ATP7B were confirmed as key regulators of redox homeostasis, consistent with their known roles in glutathione synthesis and copper export, respectively [31,32]. K-M survival analysis indicated that elevated expression of G6PD and TFRC was associated with poorer outcomes, while IFNG and ATP7B expression correlated with improved survival. Protein-protein interaction network analysis highlighted SLC2A1, the SLC7A5-SLC3A2 heterodimer, and IFNG as central nodes, suggesting their roles in a metabolic-immune network influencing tumor progression. Functional characterization revealed two complementary mechanisms: (1) SLC2A1 (a glucose transporter) and SLC7A5-SLC3A2 (a leucine transporter) coordinate nutrient uptake under metabolic stress to maintain cystine-glutamate exchange [33], and (2) FNG enhances immune activation through PD-L1 downregulation and sensitizes cells to ferroptosis via xCT inhibition [34], positioning it as a potential therapeutic target capable of simultaneously modulating immune and metabolic pathways-consistent with findings that ferroptosis regulation modulates immunotherapy response [10].

Using LASSO-Cox regression with 10-fold cross-validation, we constructed a multivariate prognostic model based on two risk genes (MTDH, PROM2) and two protective genes (IFNG, SLC1A4). The oncogenic roles of MTDH and PROM2 are linked to PI3K-Akt signaling and Notch-mediated stemness, respectively [35,36], potentially conferring resistance to metal ion-mediated cell death. In contrast, IFNG promotes antitumor immunity through T cell and NK cell activation [37,38], while SLC1A4, a neutral amino acid transporter, may modulate ferroptosis susceptibility via glutamine metabolism and redox regulation [39,40]. Genomic alterations, including frequent amplification of MTDH and hypomethylation of the IFNG promoter, further underscore the interplay between genetic and epigenetic regulation in BRCA [41]. These findings suggest several translational approaches: dual inhibition of SLC2A1 and SLC7A5 to disrupt redox balance; combination therapy involving IFNG stimulation, ferroptosis inducers, and immune checkpoint blockade; and use of DNMT inhibitors to target SLC1A4 hypermethylation.

Subtype-specific expression patterns emphasized significant tumor heterogeneity. High IFNG expression in basal-like tumors aligns with their immunogenic phenotype and may explain their increased responsiveness to immunotherapy, as observed in recent trials [42]. SLC1A4 was upregulated in HER2-positive and Luminal subtypes, possibly supporting amino acid metabolism in these proliferative tumors [43,44]. MTDH enrichment in basal-like tumors reinforces their aggressive nature [45,46], whereas PROM2 overexpression in HER2-positive tumors may contribute to stemness and therapy resistance [47]-a finding consistent with multi-omics studies linking stemness pathways to drug resistance in HER2-positive breast cancer [3]. CRGs/FRGs-based consensus clustering identified three molecular subtypes (A, B, C), with Subtypes B and C exhibiting poorer prognosis, genomic instability, and immunosuppressive traits, suggesting more aggressive disease forms. Expression dynamics across disease stages revealed declining levels of IFNG, MTDH, and PROM2 in advanced tumors, while SLC1A4 increased, supporting their potential as stage-specific biomarkers. Drug sensitivity analysis indicated that high expression of IFNG, MTDH, and SLC1A4 was associated with resistance to conventional agents (e.g., doxorubicin, etoposide, vincristine), whereas PROM2 overexpression correlated with increased chemosensitivity. These results support a strategy for therapy personalization: patients with high IFNG/MTDH/SLC1A4 expression may benefit from alternative regimens, while those with PROM2-high tumors could respond favorably to conventional chemotherapy.

Epigenetic profiling identified IFNG promoter hypomethylation as a biomarker of immunogenic microenvironments, while SLC1A4 hypermethylation was linked to immunosuppressive niches, suggesting potential utility for DNA methyltransferase inhibitors to enhance IFNG-mediated immune activation [48]. Recurrent copy number variations were observed at chromosomal hotspots 2p25.1 (RRM2) and 3q28 (TP63), indicating universal amplification events (100 % penetrance) implicating dysregulated dNTP synthesis and epithelial plasticity in therapeutic resistance [49,50]. PROM2 mutations (1 % incidence) and its apical membrane localization suggest a potential role in membrane curvature-mediated drug efflux, consistent with its association with chemosensitivity [51]. Among the prognostic gene set, FANCD2, ATP7B, PLIN4, CA9, and PROM2 exhibited the highest mutation frequencies (≥1 %). FANCD2 mutations are associated with impaired DNA damage repair and genomic instability [[52], [53], [54]], while ATP7B alterations may disrupt copper homeostasis and mitochondrial function [55,56]. PLIN4 and CA9 mutations likely promote metabolic adaptation and hypoxia survival [57,58], and PROM2 variants may enhance stemness and metastatic potential [59]. These high-frequency mutations collectively contribute to BRCA pathogenesis through synergistic multi-pathway interactions. A predominance of C-to-T substitutions suggests involvement of DNA methylation-related mutagenesis [60]. The absence of mutations in HILPDA, VEGFA, and NGB implies essential, conserved roles in tumor biology [61,62].

Besides, CNV analysis revealed widespread copy number gains across all prognostic genes, with chromosomes 2 and 3 showing particularly high densities of CRG/FRG alterations. These gains likely drive pro-tumorigenic expression and may rewire cell death regulatory networks to support survival. Elevated CNVs in IFNG and MTDH further link genomic instability to chemoresistance and progression [63,64], highlighting these regions as candidates for mechanistic investigation and therapeutic targeting. Consensus clustering delineated three molecular subtypes (A, B, C), with subtype A associated with superior survival, extracellular matrix remodeling, and cytotoxic immune infiltration. Subtype B showed upregulation of cuproptosis/ferroptosis regulators and oxidative phosphorylation, potentially conferring resistance to glycolysis inhibitors, whereas subtype A exhibited enhanced adipocytokine signaling, which may sensitize cells to ferroptosis. These patterns support subtype-specific therapeutic strategies [65,66]. KEGG enrichment revealed distinct pathway activities: subtype A demonstrated low metabolic and proliferative activity with a quiescent phenotype, while subtype B was characterized by high metabolic reprogramming, proliferation, and DNA repair-an active phenotype consistent with its poorer prognosis. These metabolic and functional distinctions provide a rationale for tailored interventions.

A pivotal finding of this study is the robust association between the CFRGscore and tumor immune microenvironment (TIME) composition. Elevated risk scores correlated strongly with reduced infiltration of cytotoxic immune cells-including CD8⁺ T cells and NK cells-and increased abundance of immunosuppressive populations such as Tregs and MDSCs, indicative of an immune-excluded or desertified phenotype [67]. In contrast, low risk scores were characterized by an immunologically “hot” tumor microenvironment. We propose that CRGs/FRGs expression influences tumor cell fate and modulates immune activity through the release of signaling molecules (e.g., DAMPs, cytokines), thereby shaping TIME and clinical outcomes. This is consistent with reported roles of MTDH in suppressing antiviral responses and immune infiltration [68]. Notably, subtype A exhibited significantly higher levels of activated CD8⁺ T cells, Th1 cells, and dendritic cells, alongside lower Treg and MDSC infiltration, suggesting robust antitumor immunity. In contrast, subtype B displayed an immunosuppressive phenotype conducive to immune escape and treatment resistance. Specifically, CD8⁺ T cells are the “core weapon” for direct killing of tumor cells, which can recognize and eliminate cells expressing tumor antigens [69]. Th1 cells secrete IFN-γ, TNF-α and other proinflammatory cytokines, which enhance the killing activity of CD8⁺ T cells and the phagocytic function of macrophages [70,71]. T. ell (Treg cells) infiltration level in group A was significantly lower than that in group B. Treg cells inhibit the proliferation and function of activated CD4/CD8 T cells by secreting anti-inflammatory factors such as IL-10 and TGF-β, which are the core cells of immunosuppression [72]. In addition, CD56dim NK cells are the main killer subset of NK cells, and they highly express killer receptors, which can directly lyse tumor cells [73], indicating that group A has stronger innate anti-tumor killing ability, and can clear tumor cells early before T cell activation, forming a synergy with adaptive immunity.

PCA-based stratification further reinforced the clinical relevance of CRGs/FRGs expression patterns. High PCA scores correlated with improved survival and enhanced immune infiltration, possibly reflecting active antitumor immunity, whereas low scores were linked to immune evasion and poor prognosis. Sankey diagram visualization illustrated patient flow across subtypes, enhancing interpretability of cluster stability. Immune checkpoint analysis revealed elevated BTLA and reduced CTLA4/LAG3 expression in high-PCA-score groups, implying a more permissive immune context and potential sensitivity to immune checkpoint inhibitors [74,75]. The downregulation of certain checkpoints in high-risk patients may not indicate T cell exhaustion but rather inadequate priming, suggesting retained capacity for immune reactivation. It suggests that immune escape in high-risk tumors is not due to irrevocable T-cell exhaustion but rather a lack of effective T-cell activation and recruitment. This implies that high-risk patients could potentially be more responsive to immune checkpoint inhibitors (ICIs), as their underlying immune effector cells retain significant potential for reactivation. The CFRGscore warrants investigation as a novel biomarker for predicting ICIs response in future prospective clinical trials. We developed a prognostic nomogram integrating clinical variables (age, gender, TNM stage) and molecular features (PCA score), which demonstrated strong predictive accuracy for 1-, 3-, and 5-year survival (AUCs: 0.834, 0.763, 0.75; C-index = 0.75). The nomogram outperformed individual clinical parameters, providing a practical tool for personalized risk assessment and therapeutic decision-making.

This study aligns with growing interest in non-apoptotic cell death pathways in oncology but advances the field by integrating cuproptosis and ferroptosis within a unified prognostic and immune context framework. While ATP7B's association with favorable prognosis corroborates existing models of cuproptosis hyperactivation [76], our work is the first to implicate PROM2 and SLC1A4 as core immunometabolic regulators in a combined signature. The model's strength lies in its mechanistic grounding and multidimensional validation across genomic, transcriptomic, and proteomic levels. Compared to other published BRCA prognostic signatures, the unique strength of our model lies in its biology-driven rationale (focusing on emerging cell death pathways) and its deep integration with immune context and therapeutic response. Key strengths include multi-omics integration, comprehensive mechanistic exploration, and orthogonal validation. Limitations comprise the retrospective design, absence of environmental and lifestyle variables, and unvalidated causal mechanisms. Future studies should: (1) prospectively validate the CFRGscore for prognosis and immunotherapy guidance; (2) elucidate functional roles of SLC1A4 and PROM2 using experimental models; (3) explore therapeutic targeting of these pathways; and (4) integrate clinical and molecular variables to refine personalized treatment strategies for BRCA.

5. Conclusions

In conclusion, we developed and validated a novel prognostic signature based on CRGs/FRGs, which serves as an independent prognostic factor in BRCA. This signature effectively bridges the molecular characteristics of tumors (cell death susceptibility), the state of the immune microenvironment, and clinical outcomes, providing a new perspective for understanding BRCA pathogenesis.

Fundings

We gratefully acknowledge financial support from Innovation Project of Guangxi Graduate Education (No. YCSW2024242). We also gratefully acknowledge financial support from Scientific Research and Technology Development Program of Guangxi (RZ2400001773).

CRediT authorship contribution statement

Yuqin Wei: Conceptualization, Methodology, Resources, Validation, Visualization, Writing – original draft. Xingwen Wei: Methodology, Resources, Validation. Wei Zhao: Investigation, Supervision, Writing – review & editing.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Yuqin Wei reports financial support was provided by Department of Education of Guangxi Zhuang Autonomous Region. Wei Zhao reports financial support was provided by Department of Science and Technology of Guangxi Zhuang Autonomous Region. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Abbreviations

AUC, area under the curve; BRCA, breast cancer; CNVs, copy-number variations; CRGs, cuproptosis-related genes; DAMPs, damage-associated molecular patterns; DEGs, differentially expressed genes; DNMT, DNA methyltransferase; FC, fold change; FDR, false discovery rate; FRGs, ferroptosis-related genes; GEO, Gene Expression Omnibus; GO, Gene Ontology; GSVA, Gene set variation analysis; HPA, Human Protein Atlas; ICIs, immune checkpoint inhibitors; IHC, immunohistochemistry; K-M survival, Kaplan-Meier survival; KEGG, Kyoto Encyclopedia of Genes and Genomes; LASSO, least absolute shrinkage and selection operator; MAF, Mutation Annotation Format; MDSCs, myeloid-derived suppressor cells; NAT, normal adjacent tissue; NK cells, natural killer cells; OS, overall survival; PCA, principal component analysis; PPI, protein-protein interaction; ROC, receiver operating characteristic; TCA, tricarboxylic acid; TCGA, The Cancer Genome Atlas; TIME, tumor immune microenvironment; TMB, tumor mutational burden; Tregs, regulatory T cells.

Appendix A. Supplementary data

The following is the Supplementary data to this article:

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.