A lactylation-related gene signature predicts metastasis and prognosis in breast cancer

Abstract

Breast cancer (BC) remains one of the most prevalent and deadly cancers among women worldwide. Nearly one-third of early-stage cases eventually develop metastases, dropping the five-year survival rate to just 24%. Lactylation plays an important role in various tumor metastases, but its role in breast cancer is not yet fully understood. Here, we developed a prognostic model based on lactylation-related genes using data from The Cancer Genome Atlas (TCGA), validated by three Gene Expression Omnibus (GEO) datasets. Functional enrichment analysis identified key biological processes, and further analyses explored associations with clinical characteristics, immune cell infiltration, drug sensitivity, and molecular docking affinities. Finally, we further verified the functional role of lactylation and prognostic genes through experimental verification. Our study demonstrates that lactylation-related genes serve as predictive biomarkers for BC metastasis and proposed novel therapeutic targets for further clinical intervention.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-26165-z.

Introduction

Breast cancer (BC) is the most commonly diagnosed cancer and the leading cause of death in women globally¹. While early detection and treatment have improved survival rates, metastasis is still the leading cause of mortality in BC patients. Nearly one-third of early-stage BC cases eventually develop metastasis, reducing their five-year survival to only 24%^2,3, with the greatest risk occurring within the first five years after treatment⁴. This not only significantly increases the complexity of treatment but also severely impacts the prognosis and quality of life for patients. This underscores the urgent need to better understand the mechanisms driving metastasis and identify new therapeutic targets.

BC metastasis is a multifactorial process, primarily involving the following aspects: epithelial-mesenchymal transition (EMT) of cancer cells, acquisition of cancer stem cell properties^5–9, extracellular matrix remodeling^10–12, and dynamic interactions within the tumor microenvironment^13–15, etc. These processes are regulated by post-translational modification (PTMs). Recent studies have emphasized the role of PTMs in BC metastasis. For instance, HOMER3 facilitated growth factor-mediated β-Catenin tyrosine phosphorylation and activation to promote metastasis in triple-negative BC¹⁶; TRIM28 promoted BC progression and metastasis by ubiquitinating and degrading BRD7¹⁷; and CBP-dependent acetylation stabilizes Slug to promote EMT and migration of BC cells¹⁸. Nonetheless, clinical interventions targeting the above-mentioned modifications (such as kinase inhibitors, deacetylase inhibitors) still couldn’t completely block the metastatic process, suggesting that other novel modifications may be involved in the regulation of metastasis through independent or synergistic mechanisms.

Lactylation, a newly discovered PTM, has been found to regulate multiple tumors. Lactylation‌ affects gene regulation and cellular function‌ by covalently coupling a lactyl group to lysine residues of proteins^19,20. While lactate serves as a rapid energy substrate under physiological conditions, its accumulation in tumors supports metabolic reprogramming and metastatic behavior^21,22. In BC cells, lactate-driven lactylation promotes migration and invasion by supplying both energy and biosynthetic precursors^23[,24[,25 alters the extracellular matrix to facilitate angiogenesis and lymphangiogenesis, and facilitates metastasis. In colorectal cancer, tumor-derived lactate degraded the extracellular matrix and helped tumor cells evade immune surveillance^26,27.

Moreover, lactylation modulates immune responses: lactate-induced lactylation of key regulators can drive immunosuppression and resistance to immunotherapy^28,29. Professor Fan Guangjian’s team found that lactylation of APOC2-K70 was identified as a key driver of tumor metastasis and resistance to immunotherapy³⁰. This indicated that lactylation may play a role in tumor metastasis by influencing immune function. Despite these findings, the specific role of lactylation in BC metastasis remains poorly defined.

Here, we report the first integrative analysis of lactylation in BC metastatic risk by combining large-scale bioinformatics with in vitro validation. We present a comprehensive analysis integrating TCGA and GEO datasets to explore lactylation-related genes in BC metastasis. We construct a prognostic model, validate it through bioinformatics and laboratory experiments, and assess its potential for guiding clinical interventions.

Materials and methods

Data procurement

Clinical and RNA sequencing data were obtained from TCGA. After processing, 906 non-metastatic (M0) and 22 metastatic (M1) BC cases were included. Additionally, we examined the GEO database. By using the keywords “breast cancer”, “metastatic”, “recurrence”, we selected three datasets: GSE11121, encompassing 200 patients (46 with distant metastasis and 154 without); GSE6532, which included 87 patients (28 with distant metastasis and 59 without); GSE19615, which included 115 patients (14 with distant recurrence and 101 without).

Differential expression analysis

Differentially expressed genes (DEGs) were identified using the “Limma” R package (p < 0.05, |FC| > 1.2). Heatmaps and volcano plots were generated using “pheatmap” and “ggplot2”.

Acquisition of prognostic lactylation genes

The lactylation gene set in this study was obtained from previously published data^31,32, including 1438 genes (Supplementary Table 1).

Identification of key genes

Intersection genes between DEGs and lactylation-related genes were identified via Venn diagrams.

Enrichment analysis

Functional enrichment was performed using the R package “clusterProfiler”. This analysis included both Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) assessments^33–35.

Mutational analysis

Somatic mutation data from TCGA was analyzed with the “maftools” R package. We constructed waterfall plots and mutation spectrum plots, respectively. Additional analysis and protein domain information were sourced from the database cBioPortal (http://www.cbioportal.org/) to evaluate the possible effect of lactylation modification.

Establishment and verification of the risk model

A LASSO-Cox regression model was developed using “glmnet” to select prognostic genes. The risk score was computed using the following formula: Risk score = exp (β1 × 1 + β2 × 2 + … + βnXn). Based on the optimal cut-off value, patients were classified into two groups: the high-risk group and the low-risk group. To analyze and assess the prognostic differences between the two groups, we applied the “survfit” function and the log-rank test to examine both the differences and their statistical significance. Additionally, we used the R package “pROC” to calculate the area under the curve (AUC) for evaluating the prognostic model’s accuracy.

Nomogram construction and evaluation

A prognostic nomogram integrating clinical features and risk scores was built with the “rms” package. Calibration plots assessed its predictive accuracy.

Survival analysis and the relationship between the prognostic genes and clinicopathological features

To explore the diagnostic values of the risk genes, we utilized the “survminer” package in R for Kaplan-Meier survival analysis, including overall survival (OS), distant recurrence free survival (DRFS), and distant metastasis free survival (DMFS), and the log-rank test was applied to group comparisons.

Immune infiltration analysis and single-cell sequencing analysis

The degree of immune cell infiltration and the presence of immune checkpoints were compared across different risk-score subgroups. To ensure the stability of the results, the “EPIC”, “quanTIseq”, “CIBERSORT”, “ESTIMATE”, “MCPCounter”, and “IPS” tools were employed. Data on breast cancer single cells were obtained from the GEO dataset GSE176078³⁶. We analyzed the data on the online website (https://singlecell.broadinstitute.org/).

Drug sensitivity analysis

To identify drugs that target the hub genes, we searched CTRP databases and GSCA databases. By searching these databases, we aimed to identify potential drug candidates that could target the key genes, providing valuable insights for new therapeutic strategies.

Exploration of the expression of target genes in different BC cell lines

In this study, the online Expression Atlas database (https://www.ebi.ac.uk/gxa/home) was selected. Maintained by the European Bioinformatics Institute (EBI), this database integrates a vast amount of gene expression data from public resources.

Molecular docking

Target protein structures were obtained from the Protein Data Bank (PDB, https://www.rcsb.org/), and potential ligands from the database PubChem (https://pubchem.ncbi.nlm.nih.gov/). Docking was performed via the online docking platform CB-Dock (http://clab.labshare.cn/cb-dock/). After completion, import the docking results into the Protein-Ligand Interaction Profiler (PLIP, https://plip-tool.biotec.tu-dresden.de/plip-web/plip/index). Analyze interactions such as hydrogen bonds and hydrophobic interactions to comprehensively understand the binding characteristics, providing a basis for studying their biological functions.

Cell lines and culture

MCF7 and MDA-MB-231 cells were cultured for assays. These cell models were cultured and maintained in DMEM. Cell cultures were incubated under standard conditions (37 °C, 5% CO₂, 95% humidity) throughout all experimental procedures. Regular quality control assessments confirmed the absence of Mycoplasma contamination in all cellular systems before experimental use.

Clinical tissue samples

From January 2021 to August 2024, breast cancer patients who underwent surgical resection at the Tumor Hospital Affiliated with Nantong University were selected. 23 pairs of breast cancer specimens and metastatic tissues were collected. Comprehensive clinical records were obtained with the informed consent of all participants. This study was ethically approved by the Institutional Review Board of the Tumor Hospital Affiliated with Nantong University (No. 2021-L081).

Immunohistochemistry (IHC)

The tumor specimens were collected from the Tumor Hospital Affiliated with Nantong University. Subsequently, the sections were heated to 95 °C in a 0.01 M citric acid buffer solution with a pH value of 6.0. Next, these tissue sections were incubated with the Anti-L-Lactyl Lysine Rabbit mAb (Catalog Number: PTM-1401RM, Jingjie, China). After that, an appropriate detection system was employed for further treatment. High-resolution images of the stained sections were then captured by means of a scanning microscope (Nikon, Japan). Finally, the assessment of lactylation staining was carried out by two independent pathologists who had no access to the relevant clinical details.

Western blot assays (WB)

To detect the lactylation levels in cells with different metastatic abilities, MDA-MB-23, MCF-7 and clinical specimen were selected for WB experiments. The cellular samples underwent two cycles of PBS rinsing followed by lysis using a cocktail of phenylmethylsulfonyl fluoride (PMSF, 1:100 dilution; Beyotime Biotechnology, Shanghai) and standard lysis buffer. Protein separation was achieved through SDS-PAGE electrophoresis with subsequent transfer onto polyvinylidene fluoride membranes (Invitrogen, CA). Membranes were blocked with 5% non-fat dry milk for 2 h at room temperature. Then we cut a whole membrane into two pieces according to the maker in the middle and incubated them respectively with Anti-L-Lactyl Lysine Rabbit mAb (Catalog Number: PTM-1401RM, Jingjie, China) and GAPDH Monoclonal antibody (Catalog Number: 60004-1-Ig, Proteintech, China) overnight at 4 °C. Following TBST washing, membranes were probed with horseradish peroxidase-conjugated secondary antibodies (1:1000; Proteintech Group) for 1 h at room temperature. Protein bands were visualized using enhanced chemiluminescent substrate, with densitometric analysis performed using ImageJ software after normalization to GAPDH expression levels.

Quantitative real-time polymerase chain reaction (qPCR)

MDA-MB-231, MCF-7 cells and clinical specimen were collected for qPCR. Total RNA isolation was conducted with TRIzol reagent (Thermo Fisher Scientific, Waltham, MA) following the manufacturer’s protocols. For cDNA synthesis, we used the EvoM-MLV Reverse Transcriptase system (Accurate Biology, Hunan, China). Gene-specific primers designed for target sequences were commercially sourced from Sangon Biotech (Shanghai, China), with detailed oligonucleotide information presented in Table 1.

Table 1.

The primer sequence of 3 genes.

Wound healing assay

MDA-MB-231 and MCF7 cells were cultured in a 6-well plate and scratched with a 200µL pipette tip. Next, the cells were cultured, and the images taken under a microscope were analyzed at different time points. Statistical analysis was performed using GraphPad Prism 10.1.

Transwell assay

The Matrigel (Corning, China) was diluted at a 1:8 ratio and uniformly applied to the upper surface of the transwell invasion. Cells were resuspended in serum-free medium at a density of 5 × 10⁴ cells/mL and carefully seeded onto the pre-coated matrix layer. The lower chamber was filled with culture medium supplemented with 20% fetal bovine serum (FBS) to create a chemoattractant gradient. Following a 48-hour incubation period under standard cell culture conditions, adherent cells were fixed using 4% paraformaldehyde solution and stained with crystal violet (Beyotime, Beijing, China). Three randomly selected fields of view per sample were imaged using a light microscope to quantify cell migration/invasion activity.

Chemicals and reagents

Lactate, 2-Deoxy-D-glucose, and Oxamic acid sodium were purchased from MedChemExpress (MCE, Shanghai, China).

Cell counting kit-8

MCF-7 and MDA-MB-231 cells were inoculated into 96-well plates. Treat with drugs (Lactate, 2-Deoxy-D-glucose, and Oxamic acid sodium) at concentrations of 0, 1, 2, 5, 10, and 20 mM for 24 h. The samples were incubated at 37 °C for 1.5 h with CCK-8 Cell Counting Kit (Vazyme, Nanjing, China), and the absorbance at 450 nm was measured.

Detection of lactate content

Cells in a 6-well plate were cultured to 70%–80% confluency. Then, added 0, 1, 2, 5, 10, and 20 mM of Lactate, 2-Deoxy-D-glucose, and Oxamic acid sodium for 24 h. Further detection of the lactate content was carried out using the L-lactic acid detection kit (WST-8 method) (Beyotime, Shanghai, China).

Statistical analysis

All statistical analyses were conducted using R version 4.4.3 and relevant software packages. The t-test and Wilcoxon rank-sum test were chosen based on whether the data met the assumption of normal distribution. Log-rank test or Cox regression analysis was used in Kaplan–Meier analysis. A p-value less than 0.05 was considered statistically significant.

Result

Identification of lactylation as an important biological process in BC metastasis

The study’s procedural overview is depicted in Fig. 1. First, we identified 324 up-regulated and 907 down-regulated DEGs by comparing metastatic and non-metastatic groups using TCGA BC data. These DEGs were subsequently visualized through volcano plots and heatmaps (Fig. 2A,B). The GO analysis of the up-regulated DEGs revealed significant enrichment in ATP binding, transport activity, and glutamate secretion (Fig. 2C), while the GO analysis of the down-regulated DEGs identified significant enrichment in immune response, protein processing, fatty acid metabolism, and mucopolysaccharide binding processes, as shown in Fig. 2D. KEGG pathway analysis revealed that up-regulated DEGs were significantly enriched in fructose and mannose metabolism and ABC transporters (Fig. 2E). In contrast, down-regulated DEGs were significantly enriched in the MAPK signaling pathway, TNF signaling pathway, and pathways associated with immune checkpoint function (including PD-1 and PD-L1) (Fig. 2F). These results indicated that lactylation was an important biological process in BC metastasis. Next, we employed the maftools visualization module to analyze the mutational characteristics of all DEGs. Somatic mutation analysis using the “maftools” package revealed a predominance of missense mutations (Fig. 2G). Transition/transversion (Ti/Tv) analysis further characterized the mutation spectrum, identifying six principal base substitution types across samples (Fig. 2H).

Fig. 1.

The flowchart and graphic abstract of this study.

Fig. 2.

Identification of metastasis-related differential genes. (A) Volcano map of differential gene expression; (B) Heat Map of differential gene expression; (C) GO enrichment analysis of the up DEGs; (D) GO enrichment analysis of the down DEGs; (E) KEGG enrichment analysis of the up DEGs^33–35; (F) KEGG enrichment analysis of the down DEGs^33–35; (G) The mutation of the DEGs; (H) Transition/Transversion ratio of DEGs.

To investigate this further, we performed an intersection analysis between the differentially expressed genes and a curated list of lactylation-related genes (see supplementary Table 1), identifying nine overlapping genes defined as lactylation-related metastatic differentially expressed genes (LRMDEGs) (Fig. 3A). Protein-protein interaction (PPI) analysis using GeneMANIA demonstrated significant interactions among these genes, notably between RGS1 and GBP2 (Fig. 3B,C). These genes were found to be associated with apoptosis, DNA damage repair, and EMT (Fig. 3D) and exhibited differential expression across various clinical stages of breast cancer (Fig. 3E,F). Pan-cancer analysis revealed expression alterations of these genes in multiple cancer types (Fig. 3G), supporting their relevance in oncogenesis and metastasis.

Fig. 3.

Lactylation-related metastatically differentially expressed genes. (A) 9 genes were obtained by taking the intersections of the DEGs and lactylation related genes; (B) PPI network analysis of the 9 genes were involved; (C) Correlation analysis among 9 molecules; (D) Links between nine molecules and common biological processes and pathways in cancer; (E, F) Relationship between 9 molecules and different clinic stages of BC; (G) Expression of 9 genes in 10 different tumors.

Development and validation of lactylation-related metastatic risk model

To assess the prognostic utility of the identified LRMDEGs, we conducted a LASSO Cox regression analysis. Eight genes were selected to construct the risk model (Fig. 4A,B), with the risk score defined as:

Fig. 4.

Risk model based on lactylation-related metastatically differentially expressed genes. (A, B) LASSO coefficient curves of prognosis related genes; (C) ROC analysis of the risk score in patients with BC in 1, 3, 5 and 10 years; (D) Univariate Cox regression analysis to screen 8 prognosis related genes; (E) The relationship between different risk scores and OS events, and changes in the expression of individual genes; (F) KM analysis of the risk score in patients.

Risk Score = (− 0.0179) × GBP2 + (− 0.0323) × NR4A2 + (− 0.0084) × RGS1 + 0.0624 × SLC5A8 + (− 0.0223) × JCHAIN + 0.1050 × RIMS1 + (− 0.1250) × CLEC2D + 0.1444 × STAT4

The area under the curve (AUC) values for predicting 1-, 3-, 5-, and 10-year survival were 0.58, 0.56, 0.61, and 0.62, respectively, indicating moderate prognostic accuracy (Fig. 4C). Multivariate regression analysis identified SLC5A8, STAT4, and CLEC2D as significant prognostic factors (Fig. 4D). Stratifying patients by the optimal cutoff into high- and low-risk groups revealed that high-risk individuals had significantly lower survival (Fig. 4E,F). Validation in three independent GEO datasets (GSE11121, GSE6532 and GSE19615) confirmed the model’s prognostic performance. Patients in the high-risk group exhibited poorer distant recurrence free survival and metastasis free survival (Fig. 5A–I). AUC values for GSE11121 were 0.78, 0.67, and 0.62 for 1-, 3-, and 5-year predictions (Fig. 5C), while AUCs for GSE6532 were 0.79, 0.72, 0.61 for 3-, 5-, and 10-year predictions, respectively (Fig. 5F), AUCs for GSE19615 were 0.79, 0.63, 0.66 for 1-, 3-, and 5-year predictions(Fig. 5I). These results indicate that the risk model possesses predictive and broadly applicable features for BC recurrence and metastasis.

Fig. 5.

Verification model based on lactylation-related metastatic differentially expressed genes. (A, D, G) The relationship between different risk scores and final events, and changes in the expression of individual genes; (B, E, H) KM analysis of the risk score in patients; (C, F, I) ROC analysis of the risk score in patients with BC in different years.

Integration of risk model via nomogram construction and clinical characteristics

KM analysis revealed that elevated expression of CLEC2D, NR4A2, JCHAIN, GBP2, and SLC5A8 was significantly associated with poorer overall survival (Fig. 6B-D,F,H). Conversely, STAT4, RIMS1, and RGS1 exhibited no significant association with survival outcomes (Fig. 6A,E,G). Different genes were significantly different between high- and low-risk groups (Fig. 7A). Multivariate analysis demonstrated that the risk score remained an independent predictor of prognosis when adjusted for clinical variables such as T stage, N stage, M stage, age, and overall clinical stage (Fig. 7B). A prognostic nomogram integrating the risk score with clinical parameters was constructed (Fig. 7C), and calibration plots indicated good concordance between predicted and observed survival outcomes (Fig. 7D). Stratification by T stage, M stage, and clinical stage further confirmed that the risk score was significantly associated with disease progression, independent of nodal status (N stage) (Fig. 7E–H).

Fig. 6.

Survival curve analysis of 8 genes. (A-H) KM analysis of STAT4, CLEC2D, NR4A2, JCHAIN, RIMS1, GBP2, RGS1, SLC5A8.

Fig. 7.

Prognosis of genes and clinicopathological features between different risk groups. (A) The expression of 8 genes in different risk groups; (B) Univariate Cox regression analysis to screen risk score and clinic features; (C) Nomogram integrating risk score and clinical features; (D) Calibration of the nomogram at 3 and 5 years; (E-H) The classification of different risk score in T stage, M stage, N stage and clinic stages and their connection.

Immunological characteristics of high- and low-risk groups

GO and KEGG analyses were conducted on DEGs between high- and low-risk cohorts. These genes were enriched in immune-related processes, including immune response and immune system process (Fig. 8A,B). Immune cell infiltration analysis via the CIBERSORT algorithm revealed distinct immune profiles between the two groups (Fig. 8C). Consistent results were obtained using alternative methods, including ESTIMATE, quanTIseq, MCPCounter, EPIC, and IPS (Fig. 8D–G). The low-risk group exhibited significantly higher immune and stromal scores, suggesting a more active tumor immune microenvironment. IPS scores were also elevated in the low-risk group, indicating greater potential responsiveness to immune checkpoint blockade, particularly anti-PD-1 therapy (Fig. 8H). Correlation analysis showed that higher risk scores were positively associated with increased infiltration of resting natural killer cells, neutrophils, and M0 macrophages (Fig. 9A−C). Associations between prognostic genes (SLC5A8, STAT4, CLEC2D; see Fig. 4D) and immune subsets further supported the immunomodulatory relevance of the model (Fig. 9D–F).

Fig. 8.

Immune infiltration analysis. (A) GO enrichment analysis of the differential genes in different risk assessment groups; (B) KEGG enrichment analysis of the differential genes in different risk assessment groups; (C) The comparison of CIBERSORT scores derived from 22 different immune cells; (D) The ESTIMATE algorithm between high and low risk subgroups; (E) The quanTIseq algorithm between high and low risk subgroups; (F) The IPS algorithm between high and low risk subgroups; (G) The MCPCounter algorithm between high and low risk subgroups; (H) The EPIC algorithm between high and low risk subgroups.

Fig. 9.

The link between individual genes and immune cells. (A) Correlation between risk score and NK cells resting; (B) Correlation between risk score and neutrophil; (C) Correlation between risk score and M0 macrophage; (D–F) Link between SLC5A8, STAT4, CLEC2D, and immune cells.

Structural and mutational analysis of key genes

To systematically investigate lactylation-mediated regulation, we analyzed cBioPortal data to delineate protein-coding domains and genomic alterations in the eight LRMDEGs (Fig. 10A–I).

Fig. 10.

Signature gene structure and mutation distribution. (A) The aggregate mutation conditions of the 8 genes; (B−I) The structural domains of each gene.

Drug sensitivity profiling

Drug response data from the GDSC and CTRP databases were analyzed to examine correlations between the expression of key LRMDEGs and chemotherapeutic sensitivity. Notably, significant associations were observed with responsiveness to commonly used agents such as doxorubicin (Fig. 11A,B). These findings offer a potential basis for personalized therapeutic strategies.

Fig. 11.

Drug sensitivity analysis. (A) Drug sensitivity analysis of GDSC databases; (B) Drug sensitivity analysis of CTRP databases.

Experimental validation and molecular docking

Analysis of single-cell RNA-seq data (Fig. 12A) revealed distinct expression patterns among the eight LRMDEGs. NR4A2, RGS1, and CLEC2D were primarily enriched in T and B lymphocyte subsets. STAT4 expression was largely confined to T cells, while GBP2 demonstrated near-exclusive localization to T cells and perivascular lymphocytes (PVL). In contrast, SLC5A8 and RIMS1 were predominantly expressed in epithelial cells, while JCHAIN was characteristic of plasmablast populations. These expression patterns collectively delineate an intricate interplay network between immune cells and tumor cells within the tumor microenvironment. Focusing on the three independently prognostic genes: SLC5A8, STAT4, and CLEC2D (Fig. 4D), we evaluated their expression in BC cell lines (Fig. 12B). Comparison between the low-metastatic MCF-7 and the highly metastatic MDA-MB-231^17,37 cell lines revealed significantly reduced CLEC2D expression in MDA-MB-231 cells, consistent with its protective role (HR < 1).

Fig. 12.

Single-cell sequencing and molecular docking. (A) Result of Single cell sequencing; (B) Expression of three genes in different BC cell lines; (C) The structure diagram of CLEC2D; (D) Molecular docking.

Next, experiments were conducted to verify lactylation and the expression of hub LRMDEGs in the risk model. Assays, including wound healing and Transwell, demonstrated markedly enhanced migratory and invasive capabilities in MDA-MB-231 cells relative to MCF7 cells (Fig. 13A–D), consistent with the known metastatic potential of these cell lines. To further confirmed the three independently prognostic genes, we further examined their expression levels using the TCGA database and QPCR. The research results from TCGA indicate that the expressions of SLC5A8 and CLEC2D have significantly decreased, while the expression of STAT4 shows no significant difference. The results of QPCR were consistent with those of TCGA (Fig. 13E,F). IHC staining analysis of clinical specimens confirmed elevated lactylation in metastatic lesions compared with primary tumors (Fig. 14A,B). Western blot experiments further validated increased lactylation levels in MDA-MB-231 cells and metastasis patients (Fig. 14C,D).

Fig. 13.

Experimental verification. (A, B) Wound healing assay to compare the migratory capacity of BC cells; (C, D) Transwell assay to compare the invasion ability of BC cells; (E) The expression of SLC5A8, STAT4, CLEC2D of TCGA database; (F) q-PCR detected the expression of SLC5A8, STAT4, CLEC2D in MCF-7, MDA-MB-231 and clinical specimen.

Fig. 14.

Clinical validation of lactylation. (A, B) IHC staining of lactylation in BC patients; (C, D) WB tested the situation of lactylation in BC cells and clinical specimens (the uncut image can be found in Supplementary Figure-1).

To explore therapeutic potential, molecular docking analysis was performed on CLEC2D (Fig. 12C). Deoxyribose 5-phosphate was identified as a candidate ligand with high binding affinity. The docking simulations revealed multiple interaction modes, including hydrogen bonds, π-π stacking, and hydrophobic contacts (Fig. 12D), suggesting its viability as a molecular modulator of CLEC2D activity.

Crucial function of the lactylation mechanism

Next, to gain a deeper understanding of the impact of lactic acid and lactylation on breast cancer metastasis, we conducted a series of functional tests. First, we treated two types of cells with different concentrations of lactic acid inhibitors and enhancers, and then detected the cell survival ability and L-lactic acid levels. We determined the more appropriate concentrations for the subsequent experiments. For MCF-7, the concentrations were 2-Deoxy-D-glucose (2DG), 5 mM; Oxamic acid sodium (Oxamate), 20 mM; Lactate, 10 mM (Fig. 15C,D). For 231, the concentration of 2DG was 2 mM; Oxamate, 5 mM; Lactate, 10mM (Fig. 15A,B). Subsequently, we also examined the lactylation levels of cells under different lactate conditions and found that after increasing lactate in the metastatic cell line 231, the lactylation level significantly increased (Fig. 15E,F), indicating that lactate and lactylation play an indispensable role in breast cancer metastasis.

Fig. 15.

Cell survival and L-lactic acid levels. (A, B) CCK8 assay for cell viability; (C, D) Measure the level of L-lactic acid; (E, F) WB tested the situation of lactylation in BC cells with different lactate levels (the uncut image can be found in Supplementary Figure-2).

Next, we further conducted functional experiments on the two cells under different lactate levels and found that after reducing lactate in 231 cells, both migration and invasion abilities significantly decreased, while after increasing lactate, the functions were enhanced (Fig. 16A,C,D,F). After increasing lactate, the invasion ability of MCF-7 slightly improved, while the migration ability increased with the increase of lactate and decreased with the decrease of lactate (Fig. 16B,C,E,F). These experimental validations support the biological relevance of lactylation and its associated genes in metastatic breast cancer and provide a foundation for targeted therapeutic strategies.

Fig. 16.

Functional tests on cells were conducted under different lactate levels. (A, D) Wound healing assay of 231 with different lactate levels; (B, E) Wound healing assay of MCF-7 with different lactate levels; (C, F) Invasion experiment.

Discussion

Breast cancer is the most common type of cancer diagnosed in women. Various factors, such as fatigue³⁸ and hormonal levels³⁹, can contribute to it. Several potential prognostic biomarkers⁴⁰ have been reported in numerous studies. For instance, Guo et al.^41–44 have conducted analyses from various biological perspectives, providing new insights into the diagnosis and prognosis significance of BC. Current treatments mainly include surgery, chemotherapy, radiotherapy, endocrine therapy, targeted therapy, immunotherapy, gene therapy, and other innovative treatments⁴⁵. However, metastasis remains the leading cause of BC mortality.

Various modifications, including ubiquitination⁴⁶, phosphorylation⁴⁷, and methylation^41,42,48, have also been found to play an indispensable role in the occurrence and development of breast cancer. And current markers (such as pathological grade) are insufficient⁴⁹. There is therefore an urgent need to develop novel markers. In liver cancer⁵⁰, colorectal cancer⁵¹, and head and neck squamous cell carcinoma⁵², studies have shown that changes in lactate levels and lactate modification have an indispensable impact on metastasis. In breast cancer⁵³, they also play an important role, but the specific mechanism is still unclear. This study identifies lactylation-related genes as independent prognostic markers, enhancing our understanding of metastasis mechanisms. More importantly, this discovery provides a new perspective for the clinical evaluation of metastasis risk and treatment selection.

Differential gene expression analyses from TCGA revealed enrichment of pathways associated with glutamate secretion and fatty acid metabolism, both previously implicated in lactate signaling and metabolic reprogramming⁵⁴. Our experimental data demonstrated that lactylation levels were elevated in aggressive breast cancer cell lines and patient-derived metastatic tissues, supporting a functional link between lactylation and disease progression.

Currently, in many tumor studies, the lactylation-related prognosis model also plays a significant role^31,55–57. We developed a prognostic risk model based on eight lactylation-related genes. The model showed consistent performance across TCGA and three external GEO datasets (GSE11121, GSE6532 and GSE19615), validating its generalizability. Notably, patients in the high-risk group exhibited poorer survival outcomes, which is in line with some of the existing studies on lactylation’s role in cancer. The risk score demonstrated strong correlations with T stage, M stage, and overall clinical stage. We further integrated the risk score with clinicopathological variables to construct a nomogram that outperformed individual parameters, underscoring its translational potential.

Metastasis is shaped not only by genetic alterations but also by immune dynamics^58–60. The tumor microenvironment is key for metastasis and impacts targeted therapies^13,61–63. Given the model’s association with immunomodulatory pathways, we compared immune infiltration between risk groups and observed a decrease in tumor-infiltrating lymphocytes in high-risk patients. Consistent with our findings, a recent study investigating lactylation-related genes in breast cancer prognosis³² demonstrated that enhanced tumor immune infiltration correlates with favorable clinical outcomes, highlighting the potential of lactylation dynamics as a biomarker for guiding immunotherapy strategies. Next, we explored three hub genes (SLC5A, STAT4, CLEC2D) that played an indispensable role. SLC5A8, a sodium-coupled monocarboxylate transporter, was downregulated in highly metastatic cells. Its known role in lactate reabsorption suggests that loss of SLC5A8 may disrupt lactate homeostasis and facilitate metastasis^64,65. STAT4, a member of the STAT (Signal Transducer and Activator of Transcription) family. It was a prognostic marker for triple negative breast cancer (TNBC), playing an important predictive role in BC metastasis⁶⁶. Although its relationship with lactylation remained unclear, activation of STAT5 in the same family promoted lactate accumulation and immunosuppression⁶⁷. CLEC2D can recognize glycosylated antigens, and lactic acid may affect its ligand binding. It was upregulated in clear cell renal cell carcinoma, predicting a poor prognosis⁶⁸. Its expression was significantly reduced in highly metastatic cell lines in our study. Our molecular docking studies revealed that CLEC2D may be targeted by small-molecule ligands, including deoxyribose 5-phosphate, offering a novel therapeutic avenue.

In conclusion, we have identified a novel panel of lactylation-related genes with significant prognostic relevance in metastatic breast cancer. These findings offer a basis for risk stratification, immunotherapy optimization, and the development of targeted interventions. Its primary strength lies in stratifying patients based on their metastatic potential and highlight the new findings regarding its significant association with metastasis free survival. Our work contributes to a growing body of evidence supporting the clinical utility of metabolic and epigenetic biomarkers in precision oncology. The study also explored these genes in the context of immunotherapy response, influencing patient selection for immunotherapy and the development of targeted therapies, thus promoting precision medicine in BC treatment. However, the research had limitations. Lactylation modification was dynamic and reversible, while bioinformatics analysis, relying on data at a specific time, struggles to capture its changes and comprehensively reflect impacts on gene function and prognosis. We anticipated overcoming these limitations in the future.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

BC

Breast cancer

TCGA

The cancer genome atlas

GEO

Gene expression omnibus public database

DEGs

Differentially expressed genes

ROC

Receiver operating characteristic curve

LRDEGs

Lactylation-related metastatic differentially expressed genes

OS

Overall survival

DRFS

Distant recurrence free survival

DMFS

Distant metastasis free survival

GO

Gene ontology

KEGG

Kyoto encyclopedia of genes and genomes

PTMs

Post translational modification of proteins

EMT

Epithelial mesenchymal transition

AUC

Area under the curve

PPI

Protein-protein interactions

MAF

Mutation annotation format

qPCR

Quantitative real-time polymerase chain reaction

EBI

European bioinformatics institute

IHC

Immunohistochemistry

PVL

Perivascular lymphocytes

WB

Western blot assays

FBS

Fetal bovine serum

MCE

MedChemExpress

Ti/Tv

Transition and transversion

PMSF

Phenylmethylsulfonyl fluoride

STAT

Signal transducer and activator of transcription

PDB

Protein data bank

PLIP

Protein-ligand interaction profiler

TNBC

Triple negative breast cancer

STAT

Signal transducer and activator of transcription

2DG

2-deoxy-D-glucose

Oxamate

Oxamic acid sodium

Author contributions

SZ: Writing—original draft, Experiment, Methodology. YL: Writing—original draft, Experiment. WT and MJ: Data curation, Formal analysis. CL: Financial support; SA: Writing—review & editing, Validation, Financial support. QH: Validation, Supervision, Financial support.

Funding

This research was supported by the Natural Science Foundation of Jiangsu Province (BK20221275), by Health Commission key project of Jiangsu Province (ZD2022039), by Health Committee of Nantong (MS2023060, QN2023027), and by the Nantong Medical Young Talents grant (Science and Education no.(2023)19).

Data availability

This study analyzed publicly available datasets including TCGA-BRCA (https://portal.gdc.cancer.gov/) and GEO databases including GSE11121, GSE6532 and GSE19615 (https://www.ncbi.nlm.nih.gov/geo/) to investigate the association between lactylation-related genes and breast cancer metastasis.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

The human tissue samples used in this study were collected from the Tumor Hospital Affiliated with Nantong University with the informed consent of all patients. Meanwhile, this study was approved by the Ethics Board of the Tumor Hospital Affiliated with Nantong University (No.2021-L081). In addition, the study was conducted in full accordance with the relevant guidelines and provisions of the Declaration of Helsinki and NIH human research guidelines.