CD36 links fatty acid metabolism to tumor immune microenvironment remodeling in breast cancer

Highlights

• •A 12-gene fatty acid metabolism model predicts breast cancer survival.
• •High-risk tumors show immunosuppressive TIME and altered drug sensitivity.
• •CD36 suppresses malignancy and modulates lipid/immune molecule expression.
• •CD36-high tumors have lower FGA but higher immune marker scores in TCGA.
• •Germline cis-eQTL GWAS shows CD36 is tumor-context dependent, not inherited.

Abstract

Background

Fatty acid metabolism influences breast cancer prognosis and immune surveillance, but interactions with somatic mutations, the tumor immune microenvironment (TIME), and germline variation remain unclear.

Methods

Using TCGA and GEO cohorts, we constructed a fatty acid metabolism‑related prognostic model usingdifferential expression, univariate Cox, and least absolute shrinkage and selection operator (LASSO)regression. Performance, immune infiltration, drug sensitivity, and CD36 function were evaluated in public data and MDA‑MB‑231 cells. Mutation‑TIME links were examined in a cBioPortal‑processed TCGA‑BRCA cohort with matched mRNA, somatic mutations, TMB, FGA, driver alterations, and TIME marker scores. Germline cis‑eQTL SNPs for signature genes and CD36 were tested against BCAC GWAS summary statistics.

Results

The model stratified prognosis (high‑risk: OS HR = 2.15, 95%CI:1.68–2.75, P < 0.001; independent factor). High‑risk tumors showed immunosuppressive TIME and altered drug sensitivity. CD36 overexpression suppressed malignancy and modulated lipid/immune molecules. In 994 mutation‑annotated tumors, CD36‑high (n = 497) had lower fraction genome altered (FGA)than CD36‑low (median 0.221 vs. 0.328, FDR = 7.45 × 10⁻⁹), similar tumor mutation burden (TMB)(1.37 vs. 1.43, FDR = 0.294), and higher cytotoxic T‑cell, immune checkpoint, and suppressive myeloid/Treg scores. Recurrent PIK3CA/TP53 alterations and co‑mutation/mutual‑exclusivity patterns contextualized CD36‑TIME heterogeneity. Germline cis‑eQTLs for signature genes and CD36 showed no robust association with breast cancer risk, and CD36 was null across subtypes.

Conclusion

The fatty acid metabolism model predicts survival and immune status; CD36 links lipid metabolism to tumor‑immune regulation. CD36‑associated immune phenotypes are associated with genomic alteration context, supporting a mutation‑TIME framework, while germline findings indicate that common germline variants regulating CD36 do not measurably influence breast cancer risk, suggesting the CD36 axis operates primarily through tumor-context-dependent mechanisms.

Graphical abstract

This study establishes a fatty acid metabolism–related prognostic model in breast cancer that stratifies patients into high- and low-risk groups with distinct survival, immune infiltration, and drug sensitivity. CD36 is identified as a core gene linking lipid metabolism to immune regulation. CD36 overexpression suppresses malignant phenotypes in triple-negative breast cancer cells and modulates lipid-metabolic and inflammatory molecules. In a mutation-aware TCGA-BRCA extension, CD36-high tumors show lower fraction genome altered but higher cytotoxic T-cell, checkpoint, and suppressive myeloid/Treg marker scores, with frequent PIK3CA/TP53 alterations and co-mutation patterns contextualizing immune heterogeneity. Germline cis-eQTL analysis reveals that CD36 and signature genes are not associated with breast cancer risk via inherited variants, supporting a tumor-context rather than germline-driven mechanism. The findings propose a CD36-centered metabolic–immune–mutation framework for prognostic stratification and biomarker development.

Introduction

Breast cancer is one of the most common malignancies among women, with incidence and mortality rates ranking among the highest globally. According to the latest Cancer Statistics, 2025, breast cancer accounts for 32% of all female cancers, posing a serious threat to women’s health [1]. Despite significant advances in early diagnosis, surgical techniques, radiotherapy, and systemic therapies—including chemotherapy, endocrine therapy, targeted therapy, and immunotherapy—tumor heterogeneity leads to substantial variability in patient outcomes. Recurrence and distant metastasis remain the primary causes of treatment failure and mortality in breast cancer [2]. Therefore, an in‑depth understanding of the molecular mechanisms underlying breast cancer development and the identification of novel prognostic markers and therapeutic targets are of great clinical importance for achieving personalized medicine in the era of precision oncology.

Metabolic reprogramming is a hallmark of cancer, and enhanced lipid synthesis has been closely linked to chemotherapy failure in breast cancer. Lipid metabolism reprogramming is now recognized as an emerging hallmark of cancer [3]. Fatty acid metabolism, encompassing fatty acid uptake, de novo synthesis, β‑oxidation, and lipid storage, plays a critical role in the initiation, progression, and metastasis of breast tumors [4]. Fatty acids not only serve as essential structural components of cell membranes, providing necessary biosynthetic precursors for tumor cells, but also generate substantial adenosine triphosphate (ATP) through β‑oxidation to support the high energy demands of malignant cells. Moreover, specific lipid molecules and their metabolic intermediates act as signaling molecules involved in regulating cell proliferation, survival, invasion, and angiogenesis [5]. In recent years, numerous studies have confirmed that the expression of fatty acid metabolism‑related genes (FAMRGs) is frequently dysregulated in various cancers, including breast cancer, and such aberrations are associated with clinicopathological features and poor prognosis [[6], [7], [8], [9]]. For instance, fatty acid synthase (FASN) is overexpressed in multiple cancer types and is considered a marker of adverse prognosis [[10], [11], [12]]. However, fatty acid metabolism is a complex biological process involving numerous genes and intricate regulatory networks, and changes in individual genes often fail to fully capture the overall impact on tumors. Therefore, a systematic analysis of the expression patterns and interactions among multiple FAMRGs to construct a prognostic model integrating multi‑gene information may enable a more accurate assessment of patient risk and provide novel insights into tumor metabolic heterogeneity.

The tumor immune microenvironment (TIME) is a critical determinant of tumor fate and therapeutic response. The type, density, and functional status of immune cell infiltration are closely associated with patient prognosis and the efficacy of immunotherapy [13]. Notably, there is a complex interplay between lipid metabolism and immune responses [14]. Aberrant lipid metabolism in tumor cells directly drives malignant progression. It also profoundly influences infiltrating immune cells by depleting essential nutrients, producing metabolites (e.g., lactate and ketone bodies), and expressing immunomodulatory molecules. These effects collectively shape an immunosuppressive microenvironment that facilitates immune evasion. For example, lipid accumulation in tumor cells can induce immunosuppressive functions in myeloid‑derived suppressor cells and tumor‑associated macrophages while contributing to the exhaustion of CD8⁺ T cells [15]. Thus, integrating fatty acid metabolism characteristics with immune landscape analysis may help elucidate how metabolic abnormalities shape the immunosuppressive microenvironment and offer new approaches for identifying patients likely to benefit from immunotherapy.

Against this background, this study aimed to systematically characterize the expression profile and clinical significance of FAMRGs in breast cancer. Using transcriptomic data and clinical information from public databases, we identified a set of core FAMRGs with prognostic value through differential expression analysis, univariate Cox regression, and LASSO regression. We then constructed a multi-gene prognostic risk score model based on these genes. We then comprehensively evaluated the model’s performance in predicting breast cancer prognosis and explored its associations with clinicopathological features, the tumor immune microenvironment, and drug sensitivity. Through in‑depth analysis of risk model‑driven immune differences, we identified CD36 as a key molecule linking lipid metabolism and immune regulation, and further validated its expression pattern, clinical significance, biological functions, and regulatory effects on lipid metabolism and immune‑related molecules in clinical samples and cell‑based assays. The findings of this study are expected to provide a novel molecular tool for prognostic stratification in breast cancer and offer theoretical support and potential intervention targets for combined therapeutic strategies targeting metabolism and immunity.

Beyond somatic alteration, germline genetic variation shapes cancer susceptibility and the composition of the tumor immune microenvironment. Genome-wide association studies (GWAS) have mapped many common single-nucleotide polymorphisms (SNPs) to breast cancer risk [16], and a large share of these germline variants act as expression quantitative trait loci (cis-eQTLs) that set gene expression in blood and in disease-relevant tissues [17,18]. Whether germline SNPs that regulate fatty acid metabolism genes and CD36 contribute to breast cancer risk has not been tested. We therefore integrated germline cis-eQTL data with breast cancer GWAS summary statistics, so that the prognostic and immune findings could be read against the inherited, variant-level genetic background [19]. Importantly, to bridge the gap between the prognostic model and the underlying genetic-alteration context, we further performed a mutation-aware analysis focusing on CD36—the core gene identified from the model—to investigate how recurrent driver mutations and genomic instability metrics (FGA, TMB) may contextualize the CD36-associated immune phenotypes observed in the risk stratification.

Materials and methods

Data acquisition and processing

Gene expression profiles and clinicopathological data were obtained from public databases, including The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO). RNA‑sequencing (RNA‑seq) data (fragments per kilobase of transcript per million mapped reads, FPKM) and complete clinical information (including age, stage, T/N/M classification, survival time, and status) for breast cancer were downloaded from TCGA. Breast cancer transcriptomic datasets with complete survival information were retrieved from GEO as external validation cohorts. Raw data were normalized, and gene symbols were uniformly mapped. Fatty acid metabolism‑related genes (FAMRGs) were systematically curated from the Molecular Signatures Database (MSigDB) and the literature.

Differential expression analysis and prognostic model construction

Differential expression analysis between breast cancer tissues and paired adjacent normal tissues in TCGA was performed using the limma package in R (version 4.1.0), with |log₂ fold change| > 1 and adjusted P < 0.05 considered significant. Differentially expressed FAMRGs were intersected with survival data, and a preliminary set of prognosis‑related genes was identified by univariate Cox regression (P < 0.05). LASSO Cox regression with 10‑fold cross‑validation was subsequently performed using the glmnet package to select the optimal λ and the corresponding set of genes for model construction. The risk score was calculated as Σ(expression of each gene × regression coefficient), and patients were divided into high‑ and low‑risk groups based on the median risk score.

Evaluation of prognostic model performance

Kaplan–Meier analysis was used to compare overall survival (OS) and progression‑free survival (PFS) between the high‑ and low‑risk groups, with differences assessed by the log‑rank test. Univariate and multivariate Cox regression analyses were performed to evaluate the independent prognostic value of the risk score after adjusting for covariates such as age and stage. Time‑dependent ROC curves were plotted, and the area under the curve (AUC) was calculated to assess predictive accuracy at 1, 3, and 5 years. The concordance index (C‑index) was used to compare the predictive performance of the model with that of traditional clinical variables. Associations between risk group and T/N stage, grade, and other clinicopathological features were analyzed using the χ² test or Fisher’s exact test.

Nomogram construction and validation

A nomogram integrating the risk score and independent clinical variables identified by multivariate Cox regression (e.g., age, tumor stage) was constructed using the rms package to predict individualized survival probabilities. Calibration curves were plotted to evaluate the agreement between predicted and observed survival at 1, 3, and 5 years. The predictive performance of the nomogram was compared with that of the risk score alone and individual clinical features using the ROC‑AUC.

Drug sensitivity analysis

Associations between the risk score and drug sensitivity were analyzed using the oncoPredict package, with half‑maximal inhibitory concentration (IC₅₀) data from the Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Therapeutics Response Portal (CTRP) databases. Differences in IC₅₀ values for drugs such as doxorubicin, paclitaxel, and olaparib between high‑ and low‑risk groups were compared using the Wilcoxon rank‑sum test.

Immune microenvironment analysis

The ESTIMATE algorithm was used to calculate stromal and immune scores and tumor purity. ssGSEA was employed to quantify the infiltration levels of 28 immune cell types, and CIBERSORT was used to analyze the proportions of immune cell subsets. Gene set variation analysis (GSVA) was performed to evaluate enrichment scores for immune‑related functional sets (e.g., antigen presentation, co‑stimulation, immune checkpoints). The TIDE algorithm was used to predict the potential response to immunotherapy. Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and GSEA pathway enrichment analyses were conducted on differentially expressed genes between high‑ and low‑risk groups.

Core gene screening and bioinformatics analysis

A PPI network of differentially expressed FAMRGs was constructed using STRING (version 11.5) and visualized with Cytoscape (version 3.9.1). The CytoHubba plugin (with the maximum clique centrality algorithm) was used to identify core modules, and CD36 was identified as a key target gene based on Kaplan–Meier analysis. Patients were divided into CD36‑high and CD36‑low expression groups according to the median expression level. Differential expression analysis and GO/KEGG/GSEA functional annotation were performed, and differences in ESTIMATE scores and ssGSEA immune infiltration were compared between the two groups.

Cell culture and transfection

The human triple‑negative breast cancer cell line MDA‑MB‑231 (obtained from the Chinese Academy of Sciences Cell Bank) was routinely cultured in Dulbecco’s modified Eagle’s medium supplemented with 10% fetal bovine serum and 1% penicillin‑streptomycin at 37 °C with 5% CO₂. CD36 overexpression plasmid and empty vector were synthesized by General Biosystems; CD36 short hairpin RNA (shRNA) and negative control were designed by Ribobio. Stable CD36‑overexpressing and CD36‑knockdown cell lines were established using the GeneCopoeia lentiviral system (LPP‑D0159‑Lv155‑400).

Western blotting

Cell pellets were lysed using radioimmunoprecipitation assay buffer containing protease inhibitors to extract total protein. Protein concentrations were determined using the bicinchoninic acid assay. Equal amounts of protein were separated by sodium dodecyl sulfate‑polyacrylamide gel electrophoresis and transferred to polyvinylidene difluoride membranes. Membranes were blocked with 5% non‑fat milk for 1 h at room temperature and then incubated overnight at 4 °C with primary antibodies against CD36 (Abcam, 1:1000) and glyceraldehyde‑3‑phosphate dehydrogenase (GAPDH; Cell Signaling Technology, 1:5000). After washing with Tris‑buffered saline containing Tween 20, membranes were incubated with horseradish peroxidase‑conjugated secondary antibodies for 1 h at room temperature, and protein bands were visualized using enhanced chemiluminescence.

Cell proliferation and migration assays

For proliferation assays, transfected cells were seeded in 96‑well plates at 3 × 10³ cells per well. At 0, 24, 48, and 72 h, CCK‑8 reagent was added and incubated for 2 h, and absorbance was measured at 450 nm using a microplate reader. For migration assays, cells were cultured until >90% confluency, and a sterile pipette tip was used to create a vertical scratch. Detached cells were washed away with phosphate‑buffered saline, and cells were cultured in serum‑free medium. Images were captured at 0 and 24 h, and the wound healing area was measured using ImageJ software to calculate the migration rate.

Quantitative real‑time PCR

Total RNA was extracted using TRIzol reagent and reverse‑transcribed into complementary DNA. qRT‑PCR was performed using SYBR Green Premix Pro TaqHS with the following cycling conditions: 95 °C for 30 s, followed by 40 cycles of 95 °C for 5 s and 60 °C for 30 s. GAPDH was used as an internal control, and relative expression levels were calculated using the 2⁻ΔΔCt method. The expression levels of lipid metabolism‑related enzymes (ACSL4, ACC, FASN, SCD) and inflammatory factors (CCL2, TNF‑α, IL‑6, IFN‑γ) were assessed.

Immunohistochemistry

Human breast cancer paraffin sections were deparaffinized, rehydrated, subjected to antigen retrieval in sodium citrate buffer, treated with 3% hydrogen peroxide to block endogenous peroxidase activity, and blocked with goat serum. Sections were incubated overnight at 4 °C with primary antibodies against CD36, Ki67, CD8, or PD‑L1. After washing with phosphate‑buffered saline, sections were incubated with horseradish peroxidase‑conjugated secondary antibodies for 1 h at room temperature, followed by 3,3′‑diaminobenzidine staining, hematoxylin counterstaining, dehydration, and mounting. Images were captured using a microscope, and the number of positive cells or staining intensity was analyzed using ImageJ software.

Mutation-aware TCGA-BRCA analysis of CD36, driver alterations, and TIME

To evaluate mutation-associated tumor immune microenvironment heterogeneity and to extend the prognostic model findings into the genetic-alteration context, we performed an additional public-cohort analysis using the cBioPortal-processed TCGA-BRCA PanCancer Atlas dataset. This analysis was designed as a mechanistic extension of the 12-gene prognostic model: because CD36 was identified as the core gene driving the immune differences between high- and low-risk groups, we sought to determine whether the CD36-associated immune phenotypes are further modified by the underlying somatic mutation landscape, including driver gene alterations, FGA, and TMB. Samples with matched mRNA expression, somatic mutation calls, clinical TMB fields, FGA values, recurrent breast cancer driver alteration status, and clinical annotations were integrated. CD36 expression was dichotomized into CD36-high and CD36-low groups by the cohort median. Bulk RNA-derived marker scores were calculated for cytotoxic T-cell activity, checkpoint/exhaustion, antigen presentation, IFNG/chemokine signaling, and suppressive myeloid/Treg activity using predefined marker-gene modules. Between-group differences were evaluated by Mann-Whitney tests with Benjamini-Hochberg FDR correction. Focused driver mutation frequencies, driver-pair co-mutation or mutual-exclusivity patterns, and CD36-high versus CD36-low immune/mutation metric differences within mutated driver strata were then summarized. Survival analyses for mutation-aware CD36/TIME strata were exploratory and were not treated as externally validated clinical prediction models.

Integration of germline cis-eQTL variants with breast cancer GWAS

To test whether inherited regulation of the fatty acid metabolism-related signature contributes to breast cancer susceptibility, we integrated germline expression quantitative trait locus (cis-eQTL) variants with breast cancer genome-wide association study (GWAS) summary statistics using the SMR framework and its heterogeneity in dependent instruments (HEIDI) test [19]. For each gene the lead cis-eQTL single-nucleotide polymorphism (SNP) served as the germline instrument, for example rs2007622 for HSPH1 and rs7755 for CD36. cis-eQTL SNPs for the 12 model genes and CD36 were obtained from whole-blood GWAS of gene expression in the eQTLGen Consortium (31,684 individuals) [17] and from breast mammary tissue in the Genotype-Tissue Expression project version 8 (396 donors) [18]. Breast cancer GWAS association statistics were taken from the Breast Cancer Association Consortium OncoArray meta-analysis of 122,977 cases and 105,974 controls, with estrogen-receptor-positive (69,501 cases) and estrogen-receptor-negative (21,468 cases) strata analyzed in parallel [16]. All datasets were aligned to the GRCh37 assembly, and 1000 Genomes European samples provided the linkage-disequilibrium reference. The cis-eQTL SNP instrument-selection threshold was 5 × 10⁻⁸ for the eQTLGen blood GWAS and 1 × 10⁻⁵ for the smaller breast tissue panel. Each estimate was expressed as an odds ratio for breast cancer per one standard deviation higher genetically predicted expression, and a HEIDI P value above 0.05 was read as evidence against horizontal pleiotropy or a linkage artefact [19,20]. GWAS associations were corrected for multiple testing within each subtype and tissue panel using the Benjamini-Hochberg procedure.

Statistical analysis

Data were analyzed using R software (version 4.1.0) or GraphPad Prism (version 8.0). Quantitative data are presented as mean ± standard deviation. Comparisons between two groups were performed using the t‑test or Wilcoxon rank‑sum test, and comparisons among multiple groups were performed using one‑way analysis of variance. Survival analysis was conducted using the Kaplan–Meier method with the log‑rank test. Cox regression was used to identify independent prognostic factors. All tests were two‑sided, and P < 0.05 was considered statistically significant.

Results

Identification of fatty acid metabolism‑related genes and construction of a prognostic model

Transcriptomic data from breast cancer tissues and paired adjacent normal tissues were obtained from public databases to systematically characterize the expression patterns and prognostic significance of FAMRGs. Differential expression analysis revealed a series of FAMRGs with significantly altered expression in tumor tissues (Fig. 1A, B). To identify genes with prognostic value, we integrated gene expression data with patient survival information and performed univariate Cox regression to preliminarily screen for prognosis‑related candidates. LASSO regression was subsequently applied to refine the selection and construct a robust prognostic signature, ultimately identifying a set of 12 core FAMRGs (Fig. 1C–E). The mutational landscape of these genes in the study cohort is presented as a waterfall plot (Fig. 1F). A risk score was calculated for each patient based on the expression levels and regression coefficients of the 12 genes, and patients were divided into high‑ and low‑risk groups according to the median risk score. Survival analysis demonstrated that patients in the high‑risk group had significantly shorter OS and PFS compared with those in the low‑risk group (P < 0.05; Fig. 1G, H), indicating favorable discriminatory performance of the prognostic model.

Fig. 1.

Expression of fatty acid metabolism-related genes and construction of the prognostic model. (A, B) Differential expression analysis of fatty acid metabolism-related genes in breast cancer tissues. (C, D, E) LASSO regression analysis and forest plot of the 12 prognostic FAMRGs. (F) Waterfall plot of mutation frequencies of the 12 prognostic FAMRGs. (G) Overall survival (OS) curves for high- vs. low-risk groups. (H) Progression-free survival (PFS) curves for high- vs. low-risk groups. TCGA-BRCA cohort. Log-rank test. *P &lt; 0.05, **P &lt; 0.01, ***P &lt; 0.001.

Independent prognostic value of the fatty acid metabolism‑related risk model in breast cancer

The prognostic model based on the 12 key FAMRGs demonstrated robust performance in predicting outcomes in breast cancer. Univariate and multivariate Cox regression analyses revealed that the risk score was an independent prognostic factor, remaining significant after adjustment for clinical covariates including age and tumor stage (Fig. 2A, B). Time‑dependent ROC curve analysis showed that the model maintained good discriminative ability for predicting 1‑, 3‑, and 5‑year OS, with AUC values >0.7 (Fig. 2C). The C‑index further confirmed that the predictive performance of the model was comparable to, or even better than, that of traditional clinical features (Fig. 2D). Association analyses with clinicopathological characteristics revealed significant differences in tumor stage, T stage, and N stage between the high‑ and low‑risk groups (Fig. 2E–J), supporting the clinical relevance of the risk stratification. To enable individualized quantitative prognostic assessment, a nomogram integrating the risk score and key clinical indicators was constructed (Fig. 2K). Calibration curves demonstrated good agreement between nomogram‑predicted and actual observed survival probabilities at 1, 3, and 5 years (Fig. 2L). ROC curve comparisons showed that the nomogram achieved a higher AUC than either the risk score alone or any individual clinical feature, indicating superior predictive performance of the integrated model (Fig. 2M).

Fig. 2.

Association of the model with clinical characteristics and functional analysis. (A, B) Univariate and multivariate Cox regression analyses of the risk score. (C) Time-dependent ROC curves for 1-, 3-, and 5-year survival. (D) Concordance index (C-index) of the model and clinical features. (E–J) Risk score associations with clinicopathological characteristics. (K, L) Nomogram and calibration curves for survival prediction. (M) ROC curve comparison of nomogram and clinical variables. TCGA-BRCA cohort. Cox regression, log-rank test. **P &lt; 0.01, ***P &lt; 0.001.

Anticancer drug sensitivity analysis based on fatty acid metabolism risk stratification

To investigate differences in treatment response between risk groups, we systematically analyzed associations between the risk model and drug sensitivity using the oncoPredict R package. Screening against public pharmacogenomic databases containing IC₅₀ data identified candidate drugs significantly correlated with the risk score, and we focused on comparing sensitivity differences for 12 commonly used breast cancer chemotherapeutic and targeted agents. Notably, the low‑risk group exhibited lower IC₅₀ values—indicative of higher sensitivity—for the majority of tested drugs, including doxorubicin, paclitaxel, and olaparib, compared with the high‑risk group (Fig. 3A–L). These findings suggest that the fatty acid metabolism‑based risk signature has potential utility in guiding individualized drug selection and highlight the heterogeneity in drug responses between different risk strata.

Fig. 3.

Differences in sensitivity to 12 breast cancer drugs between risk groups defined by the model. (A–L) IC₅₀ comparisons for 12 agents between high- and low-risk groups. GDSC/CTRP databases. Wilcoxon rank-sum test. **P &lt; 0.01.

Immune landscape analysis reveals significant differences between high‑ and low‑risk groups and identification of CD36 as a core gene

To explore the relationship between the risk score model and the tumor immune microenvironment, we performed a systematic immune characterization of patients in the high‑ and low‑risk groups. Functional enrichment analysis revealed that differentially expressed genes between the two groups were predominantly enriched in immune‑related pathways, including cytokine–cytokine receptor interaction, chemokine signaling, and T‑cell receptor signaling (Fig. 4A–C). Immune function scores indicated that the high‑risk group had significantly lower scores for antigen presentation, co‑stimulatory molecule expression, and immune checkpoint‑related functions compared with the low‑risk group (Fig. 4D). ssGSEA immune cell infiltration analysis showed that the low‑risk group had higher infiltration levels of antitumor immune cells, including activated CD8⁺ T cells, activated CD4⁺ T cells, macrophages, and dendritic cells (Fig. 4E). TIDE scores suggested a potentially weaker response to immune checkpoint blockade in the high‑risk group (Fig. 4F). To identify core molecules driving the immune differences between the two groups, we constructed a PPI network of differentially expressed genes, performed topological analysis to identify core modules, and integrated prognostic validation, ultimately identifying CD36 as a key target gene for subsequent functional validation (Fig. 4G, H). Collectively, these results indicate that the risk model effectively stratifies patients with distinct immune phenotypes, and the core gene CD36 may play a critical role in regulating the tumor immune microenvironment.

Fig. 4.

Immune landscape analysis associated with the risk score model and screening of core genes. (A–C) GO, KEGG, and GSEA enrichment analyses between groups. (D) Immune-related function scores in high- and low-risk groups. (E) ssGSEA immune cell infiltration in high- and low-risk groups. (F) TIDE scores in high- and low-risk groups. (G, H) PPI network and core gene identification (CD36). TCGA-BRCA cohort. ssGSEA, TIDE. **P &lt; 0.01, ***P &lt; 0.001.

Association of CD36 expression with clinical features, molecular pathways, and the tumor microenvironment in breast cancer

To investigate the biological functions of CD36 in breast cancer, we systematically analyzed its expression in relation to clinicopathological features, molecular pathways, and the tumor microenvironment. CD36 expression was significantly downregulated in breast cancer tissues compared with adjacent normal tissues, and low CD36 expression was significantly associated with lymph node metastasis (Fig. 5A–C), suggesting that CD36 downregulation may contribute to breast cancer initiation and progression. When patients were divided into CD36‑high and CD36‑low expression groups based on the median expression level, differential expression analysis revealed distinct expression profiles (Fig. 5D). Functional enrichment analysis indicated that CD36‑associated differentially expressed genes were primarily enriched in biological processes such as fatty acid metabolism, peroxisome proliferator‑activated receptor (PPAR) signaling, lipid transport, and inflammatory responses (Fig. 5E, F). GSEA further confirmed that the CD36‑high expression group was enriched in fatty acid metabolism, adipocytokine signaling, and PPAR signaling pathways (Fig. 5G). Analysis of the tumor microenvironment revealed that the CD36‑high group had significantly higher overall infiltration of stromal and immune cells compared with the CD36‑low group (Fig. 5H). Quantitative analysis of immune cell subsets showed that the CD36‑high group exhibited higher infiltration of antitumor immune cells, including activated CD8⁺ T cells, macrophages, dendritic cells, and neutrophils (Fig. 5I). These findings suggest that high CD36 expression is associated with favorable clinical phenotypes in breast cancer and may reshape the tumor immune microenvironment by modulating lipid metabolism pathways.

Fig. 5.

Association of CD36 expression with clinical features, molecular pathways, and the tumor microenvironment in breast cancer. (A–C) CD36 expression vs. clinicopathological features. (D) Heatmap of DEGs between CD36-high and -low groups. (E, F) GO/KEGG enrichment analysis of CD36-associated DEGs (bar plot and bubble plot). (G) GSEA results. (H) Stromal and immune scores (ESTIMATE) in CD36-high vs. -low groups. (I) ssGSEA immune cell infiltration in CD36-high vs. -low groups. TCGA-BRCA cohort. CD36-high/low by median. Wilcoxon test. **P &lt; 0.01, ***P &lt; 0.001.

Association of CD36 expression with drug sensitivity in breast cancer

To explore the relationship between CD36 expression levels and drug sensitivity in breast cancer, we compared IC₅₀ values for six commonly used therapeutic agents between CD36‑high and CD36‑low expression groups. CD36 expression exhibited divergent correlation patterns with sensitivity to different drug classes. Specifically, IC₅₀ values for the antimetabolite methotrexate were positively correlated with CD36 expression, indicating reduced sensitivity in the CD36‑high group (Fig. 6D). In contrast, IC₅₀ values for five tyrosine kinase inhibitors, including lapatinib and gefitinib, were negatively correlated with CD36 expression, suggesting higher sensitivity in the CD36‑high group (Fig. 6A–C, E, F). These results indicate that high CD36 expression is associated with reduced response to antimetabolites but enhanced sensitivity to tyrosine kinase inhibitors, suggesting its potential utility as a biomarker for individualized treatment selection in breast cancer.

Fig. 6.

Differences in sensitivity to six breast cancer drugs based on CD36 expression. (A–F) Changes in drug sensitivity to six breast cancer agents according to CD36 expression levels.

CD36 regulates malignant phenotypes and the expression of lipid metabolism‑ and immunity‑related molecules in breast cancer cells

To functionally validate the role of CD36 in breast cancer, in vitro experiments were performed using MDA‑MB‑231 cells. Western blotting confirmed the successful establishment of CD36‑knockdown and CD36‑overexpressing stable cell lines (Fig. 7A, B). Functional assays demonstrated that CD36 overexpression significantly suppressed the proliferation and migration of MDA‑MB‑231 cells, whereas CD36 knockdown promoted these malignant phenotypes (Fig. 7C–F), indicating that CD36 inhibits malignant progression in breast cancer. Mechanistically, qRT‑PCR revealed that CD36 overexpression significantly downregulated the expression of the lipid metabolism‑related enzymes FASN and SCD while upregulating ACSL4 and ACC (Fig. 7G). In addition, CD36 overexpression reduced the levels of the pro‑inflammatory factors CCL2 and IL‑6 and increased the expression of the antitumor immunity‑related factors IFN‑γ and TNF‑α (Fig. 7H). Immunohistochemical analysis of clinical tissue samples further demonstrated that the CD36‑high expression group exhibited a significantly lower Ki67 proliferation index, increased CD8⁺ T‑cell infiltration density, and reduced PD‑L1 expression (Fig. 7I–K). Collectively, these findings indicate that CD36 suppresses breast cancer cell proliferation and migration and modulates lipid metabolism and inflammatory factor secretion, thereby contributing to an improved tumor immune microenvironment.

Fig. 7.

Functional validation of CD36 in breast cancer cells and its regulation of lipid metabolism- and immunity-related molecules. (A, B) Western blot validation of CD36 knockdown and overexpression in MDA-MB-231 cells. (C) CCK-8 proliferation assay. (D) Glutaraldehyde level detection. (E, F) Wound healing migration assay. (G) qRT-PCR analysis of lipid metabolism enzymes (ACSL4, ACACA, FASN, SCD). (H) qRT-PCR analysis of inflammatory factors (CCL2, TNF, IL6, IFNG). (I) Immunohistochemical staining of CD8 and PD-L1 in clinical breast cancer specimens. (J) Ki67 proliferation index quantification. (K) CD8-positive cell density quantification. MDA-MB-231 cells, n = 3 replicates. Mean ± SD. Student's t-test. *P &lt; 0.05, **P &lt; 0.01, ***P &lt; 0.001; ns, not significant.

Mutation-aware CD36/TIME analysis reveals genomic alteration contexts linked to immune heterogeneity

Having established that CD36 is the core gene linking the prognostic risk model to immune microenvironment differences, we next sought to investigate whether this CD36-associated immune phenotype is associated with the underlying somatic mutation landscape. To this end, we extended the analysis to a cBioPortal-processed TCGA-BRCA PanCancer Atlas cohort containing 994 tumors with matched mRNA, mutation, TMB, FGA, driver alteration, and clinical annotation fields. This mutation-aware analysis was not intended to replace the prognostic model but rather to provide a genetic-alteration framework for understanding the heterogeneity of the CD36-TIME axis identified by the risk model. CD36-high and CD36-low groups each contained 497 tumors. In this public cohort, CD36-high tumors showed a lower FGA than CD36-low tumors (median 0.221 versus 0.328; FDR = 7.45 × 10⁻⁹), whereas nonsynonymous TMB did not differ strongly between groups (median 1.37 versus 1.43; FDR = 0.294).

Mutation-aware TIME analysis further showed that CD36-high tumors had higher cytotoxic T-cell marker scores (median −0.263 versus −0.372; FDR = 3.87 × 10⁻⁵), checkpoint/exhaustion marker scores (median −0.220 versus −0.310; FDR = 0.0106), and suppressive myeloid/Treg marker scores (median −0.0446 versus −0.217; FDR = 2.58 × 10⁻⁸). The most frequent queried driver alterations were PIK3CA (34.7%), TP53 (34.6%), CDH1 (12.9%), GATA3 (12.4%), KMT2C (9.8%), MAP3K1 (9.0%), PTEN (5.6%), and NCOR1 (4.9%). Focused pairwise driver analysis identified strong mutual-exclusivity or co-mutation patterns, including TP53+CDH1 (odds ratio = 0.12; FDR = 2.36 × 10⁻¹²), TP53+GATA3 (odds ratio = 0.159; FDR = 6.13 × 10⁻¹⁰), and PIK3CA+TP53 (odds ratio = 0.45; FDR = 4.29 × 10⁻⁶).

Driver-stratified CD36/TIME testing indicated that the CD36-associated immune phenotype was not uniform across all driver-defined contexts. For example, the TP53-mutated stratum did not show a multiple-testing-supported CD36-high versus CD36-low cytotoxic-score shift (FDR = 0.872), whereas the broader driver/TIME matrix highlighted gene- and metric-specific heterogeneity. These data support Fig. 8 as a mutation-aware extension of the fatty acid metabolism-CD36-TIME axis, while remaining association-level and requiring independent validation.

Fig. 8.

Mutation-aware CD36/TIME context in TCGA-BRCA. CD36-high (n = 497) and CD36-low (n = 497) groups were defined by the cohort median expression. (A) FGA distribution showing significantly lower genomic instability in CD36-high versus CD36-low tumors. (B) Cytotoxic T-cell activity score (percentile) distribution across CD36 strata. (C) Heatmap of CD36-stratified immune phenotype composition, including cytotoxic T-cell, checkpoint/exhaustion, antigen presentation, IFNG/chemokine, and suppressive myeloid/Treg marker scores. (D) Delta matrix (CD36-high minus CD36-low) for immune metrics, FGA, and TMB across major driver mutation strata. Red indicates higher values in CD36-high; blue indicates lower values. (E) Pairwise co-mutation (orange) and mutual-exclusivity (blue) odds ratios among recurrent breast cancer driver genes (PIK3CA, TP53, CDH1, GATA3, KMT2C, MAP3K1, PTEN, NCOR1). All immune scores were derived from bulk RNA-seq marker-score proxies. Association-level analysis; requires independent validation.

Germline cis-eQTL variants and breast cancer GWAS for the fatty acid metabolism signature and CD36

Across the overall, estrogen-receptor-positive and estrogen-receptor-negative GWAS, germline SNPs that regulate the 12 model genes and CD36 showed little association with breast cancer risk (Fig. 9). No gene survived Benjamini-Hochberg correction in any subtype or tissue panel, and CD36 was consistently null, with its lead cis-eQTL SNP rs7755 carrying a breast cancer GWAS P = 0.70 and odds ratios between 0.99 and 1.02 per standard deviation of predicted expression. HSPH1 gave the only signal that recurred across analyses, where its lead cis-eQTL SNP rs2007622 was associated with higher risk in the overall GWAS (per-allele P = 0.019, odds ratio 1.08 per standard deviation, 95% confidence interval 1.01 to 1.16) and more strongly in the estrogen-receptor-positive GWAS (per-allele P = 0.0049, odds ratio 1.12, 95% confidence interval 1.03 to 1.21), with a HEIDI test consistent with a single shared germline variant. The PLA2G4A SNP rs12072973 in the estrogen-receptor-negative GWAS (odds ratio 1.21, 95% confidence interval 1.03 to 1.42) and the ALOX15B SNP rs150460215 in the estrogen-receptor-positive GWAS reached nominal significance without recurring elsewhere. Breast mammary tissue cis-eQTL SNPs were available for only two of the thirteen genes and were concordantly null. Therefore, common inherited cis-eQTL variants in the expression of the fatty acid metabolism signature and CD36 do not appear to account for their prognostic and immune associations in breast cancer in these GWAS datasets, although we cannot exclude the possibility that rare variants or non-regulatory region variants may play a role.

Fig. 9.

Germline cis-eQTL variants and breast cancer GWAS for the fatty acid metabolism signature and CD36. Odds ratios per 1-SD higher genetically predicted expression (95% CI). eQTLGen whole-blood cis-eQTL (n = 31,684) integrated with BCAC GWAS (overall, ER⁺, ER⁻). Dashed line = OR of 1. Filled points = nominal P &lt; 0.05; all passed HEIDI test. No gene survived Benjamini-Hochberg correction. CD36 and HSPH1 highlighted. Breast mammary tissue cis-eQTL available for two genes; both null. SMR/HEIDI test.

Discussion

This study establishes a fatty acid metabolism-related risk model that links breast cancer prognosis with immune microenvironmental variation. It also identifies CD36 as a focal gene that connects lipid metabolism, immune regulation, and tumor behavior. These findings are consistent with the broader literature that places metabolic reprogramming and immune escape among central features of cancer progression [13,14], but extend that framework by nominating a CD36-centered fatty acid metabolism axis in breast cancer rather than treating lipid metabolism and tumor immunity as separate processes.

The prognostic model stratified patients by survival and immune landscape, with the high-risk group showing an immune-suppressed phenotype and altered drug-sensitivity patterns. Prior studies have shown that immune cell composition, antigen presentation, and checkpoint activity influence tumor fate and therapy response [13,21,22]. Our results agree with that literature in showing that risk stratification is accompanied by immune-context differences, but the present evidence remains based on retrospective public cohorts and computational immune scoring rather than prospective immunotherapy response.

CD36 emerged from the model and core-gene analysis as a biologically plausible link between fatty acid uptake, lipid metabolism, and immune regulation [[23], [24], [25]]. The in vitro experiments support an inhibitory effect of CD36 overexpression on malignant phenotypes in MDA-MB-231 cells and suggest regulation of lipid metabolism- and immunity-related molecules. This experimental layer strengthens the computational nomination of CD36, although it does not by itself prove how CD36 shapes immune-cell recruitment or function within intact tumors.

The mutation-aware extension adds a genetic-alteration context to the CD36/TIME story and serves as a correlative complement to the 12-gene prognostic model. Whereas the model stratifies patients by survival and global immune features, the mutation-aware analysis asks whether the CD36-centric immune phenotype identified by the model is associated with specific driver mutations and genomic instability metrics. CD36-high tumors had lower FGA but similar nonsynonymous TMB compared with CD36-low tumors, and they showed higher cytotoxic, checkpoint/exhaustion, and suppressive myeloid/Treg marker scores. This pattern is compatible with current concepts that genetic alteration and immune pressure co-evolve during tumor progression [13], but it also suggests that CD36-associated immune states may be more closely linked to copy-number/genomic-instability context than to nonsynonymous mutation burden alone in bulk TCGA-BRCA data. Thus, the mutation-aware analysis does not replace the risk model but rather refines it by showing that the CD36-TIME association is not uniform across all mutational backgrounds—a finding that may inform future stratification strategies combining risk scores with mutation status.

Driver-level analysis further contextualized the CD36-associated TIME pattern. Frequent PIK3CA and TP53 alterations, together with focused co-mutation and mutual-exclusivity patterns, indicated that the immune-metabolic state captured by CD36 was embedded in a heterogeneous mutational landscape. However, CD36-by-driver interaction tests did not provide strong evidence for a single dominant driver mutation explaining the CD36/TIME association. Thus, the mutation-aware analysis should be interpreted as a hypothesis-generating map for biomarker refinement rather than as evidence of a causal mutational mechanism.

Germline SNPs that regulate expression of the 12 model genes and CD36 in blood and breast tissue showed no robust association with breast cancer risk across our GWAS analyses. CD36 was consistently null across overall and estrogen-receptor-defined disease [26]. Integrating cis-eQTL data with GWAS summary statistics asks whether a trait and gene expression share the same germline variant rather than reflecting linkage [19], and large breast cancer GWAS have mapped most common-variant risk to regulatory SNPs near genes other than these metabolic effectors [16]. The CD36 program described here therefore behaves as a tumor-context phenomenon rather than a constitutional one in the context of common germline variation, which is consistent with its emergence from tumor-versus-normal differential expression and from the mutation-aware microenvironment analysis rather than from heritable, GWAS-detectable expression variation. However, it is important to note that this interpretation is limited to common cis-eQTL variants captured by current GWAS; the potential contributions of rare variants or non-regulatory region variants to CD36 regulation and breast cancer susceptibility remain unexplored and cannot be excluded. HSPH1 gave a modest estrogen-receptor-positive-predominant GWAS signal at its cis-eQTL SNP rs2007622 that passed the HEIDI test but did not survive multiple-testing correction, so it remains a candidate germline variant for targeted replication rather than an established risk locus [17,19]. Read together, the germline variant layer and the somatic layer support a model in which lipid-metabolic remodeling of the breast tumor immune microenvironment is governed more by the somatic and transcriptional state of the tumor than by germline, GWAS-encoded expression set points.

Several limitations should be acknowledged. The prognostic model was constructed from retrospective public datasets and requires prospective, multicenter validation. The mutation-aware analysis used cBioPortal-processed TCGA-BRCA data and bulk RNA marker-score proxies. It did not incorporate raw variant calling, single-cell measurements, spatial immune mapping, or immunotherapy-treated breast cancer cohorts. The survival contrasts within mutation-aware CD36/TIME strata were exploratory and not externally validated. Future work should combine independent clinical cohorts, spatial or single-cell immune profiling, and targeted perturbation experiments to test whether CD36 directly links lipid metabolic remodeling to mutation-conditioned immune evasion.

Ethics approval

All animal experiments and clinical trials involved were conducted in accordance with the ethical policies and procedures approved by the Ethics Committee of the First Affiliated Hospital of Army Medical University (Approval No (A) KY2023068).

Consent for publication

No conflict of interest exits in the submission of this manuscript, and manuscript is approved by all authors for publication.

Funding

The present study was funded by General Program of the Chongqing Natural Science Foundation (CSTB2023NSCQ-MSX1110), the National Science Foundation for Young Scientists of China (Grant no. 81802665), Chongqing Graduate Research and Innovation Project (CYB23293), Chongqing Major Medical Research Program (Joint Program of Chongqing Municipal Health Commission and Science and Technology Bureau) (Grant no. 2024DBXM001), Chongqing Clinical Diagnosis and Treatment Center of Breast Cancer (Grant no. 425Z2a1), Military Key Clinical Specialty (Grant no. 51561Z23612).

CRediT authorship contribution statement

Xiujuan Wu: Writing – original draft, Funding acquisition, Data curation. Lin Ren: Methodology, Formal analysis, Data curation. Li Peng: Software, Methodology, Formal analysis. Xuanni Tan: Software, Resources. Wenting Yan: Data curation. Xiaowei Qi: Writing – review & editing, Conceptualization. Yan Wang: Writing – review & editing, Writing – original draft, Data curation. Yi Zhang: Writing – review & editing, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Availability of data and material

The public transcriptomic, clinical, drug-sensitivity, and mutation-context datasets analyzed in this study were obtained from TCGA, GEO, GDSC/CTRP, and cBioPortal-processed TCGA-BRCA PanCancer Atlas resources. Derived analysis outputs supporting the mutation-aware CD36/TIME extension are included in the project result files and figure audit package.