Identification of lactate-related subgroups and prognostic model in triple-negative breast cancer

Abstract

Background

Triple-negative breast cancer (TNBC) is a highly aggressive subtype of breast cancer that exhibits elevated glycolytic capacity. Lactate, as a byproduct of glycolysis, is considered a major oncometabolite that plays an important role in oncogenesis and remodeling of the tumor microenvironment. However, the potential roles of lactate in TNBC are not yet fully understood. In this study, our goal was to identify prognosis-related lactate genes (PLGs) and construct a lactate-related prognostic model (LRPM) for TNBC.

Methods

First, we applied lactate-related genes to classify TNBC samples using a hierarchical clustering algorithm. Then, we performed the log-rank analysis and the least absolute shrinkage and selection operator analysis to screen PLGs and construct the LRPM. The biological functions of the identified PLGs in TNBC were investigated using CCK8 assay and clone formation assay. Finally, we constructed a nomogram based on the lactate-risk score and tumor clinical stage. We used the operating characteristic curve and decision curve analysis to evaluate the predictive capability of the nomogram.

Results

Our results showed that the TNBC samples could be classified into two subgroups with different survival probabilities. Three genes (NDUFAF3, CARS2 and FH), which can suppress TNBC cell proliferation, were identified as PLGs. Moreover, the LRPM and nomogram exhibited excellent predictive performance for TNBC patient prognosis.

Conclusion

We have developed a novel LRPM that enables risk stratification and identification of poor molecular subtypes in TNBC patients, showing great potential in clinical practice.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00432-023-05171-6.

Introduction

Breast cancer is a prevalent cause of cancer-related mortality in females on a global scale, ranking second among all other causes (Giaquinto et al. 2022). The mortality rate of breast cancer has been declining over the past few decades due to the widespread use of chemotherapy, endocrine therapy, and targeted therapy (Pondé et al. 2019). Nevertheless, triple-negative breast cancer (TNBC), which is characterized by the lack of expression of estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (Her2), remains the most challenging subtype to treat (Waks and Winer 2019). Neither endocrine therapy nor Her2-targeted therapy is effective for TNBC, and chemotherapy remains the primary treatment. Recently, immune therapy such as immune-checkpoint inhibitors (ICI) has shown promising results in advanced-stage TNBC (Bianchini et al. 2016). However, a subset of TNBC patients still suffers from chemotherapy resistance, low response and efficacy of immune therapy. Consequently, exploring potential biomarkers for treatment response judgment and prognostic prediction has significant clinical value in the individualized treatment of TNBC.

Upregulation of glycolysis is a nearly universal characteristic of primary and metastatic cancers (Gatenby and Gillies 2004). Lactate, a byproduct of glycolysis, has recently garnered attention due to its implication in immunity, and research on the interactions between lactate and the tumor microenvironment is rapidly growing. Robert A. Gatenby proposed that tumor cells can alter the local microenvironment in a manner that is harmless to itself but fatal to other competing cells through persistent aerobic glycolysis. As a result, these cells have a powerful growth advantage, which promotes tumor proliferation and invasion (Gatenby and Gillies 2004). Previous studies have provided evidence that the accumulation of lactate undermines tumor immune surveillance and leads to poor antitumor immunity by inhibiting the cytotoxic activity of NK cells (Husain et al. 2013), CD8⁺ T cells (Elia et al. 2022) and cytotoxic T cells (CTLs) (Mendler et al. 2012), or promoting apoptosis of naive T cells (Xia et al. 2017). Additionally, Chen et al. reported that tumor-derived lactate mediates the functional polarization of tumor-associated macrophages (Colegio et al. 2014) and promotes breast cancer metastasis (Chen et al. 2017). Meanwhile, several other studies have revealed that lactate is associated with an increased incidence of metastasis and poor outcome in various cancers, such as cervical cancer (Walenta et al. 2000), head and neck cancers (Brizel et al. 2001). Nonetheless, the exact contribution of lactate and lactate-associated genes (LRGs) in TNBC subtype identification and prognostic prediction is still unclear.

In this study, we collected transcriptomic data and corresponding clinical information of TNBC patients from The Cancer Genome Atlas Project (TCGA) cohort and the Fudan University Shanghai Cancer Center (FUSCC) cohort. Firstly, we systematically analyzed the expression profiles of LRGs and identified prognosis-related lactate genes (PLGs) in TNBC. Then, we constructed a robust lactate-related prognosis model (LRPM) and divided TNBC patients into high-risk and low-risk groups according to the median lactate-risk score (LRS). We also investigated the associations between LRS and immune signature, immune infiltration as well as drug sensitivity. Furthermore, we developed a nomogram that combined the LRPM and tumor clinical stage to quickly assess the prognosis of TNBC patients. Our study sheds light on the significant role of lactate in TNBC.

Materials and methods

Dataset source and preprocessing

RNA transcriptome sequencing data and the corresponding clinical information of breast cancers (BRCA) were acquired from The Cancer Genome Atlas (TCGA) via USCS Xena (https://xenabrowser.net/datapages/) in accordance with the website’s guidance. To select TNBC samples from BRCA, patients who were immunohistochemically negative for ER, PR and HER2 were included, resulting in 116 patients being enrolled in the TCGA-TNBC cohort. Additionally, RNA sequencing data, somatic mutation profile and clinical information of 298 TNBC patients treated at the Department of Breast Surgery at Fudan University Shanghai Cancer Center (FUSCC) designated the FUSCC-TNBC cohort, were downloaded from the National Omics Data Encyclopedia (NODE) (https://www.biosino.org/node/project/detail/OEP000155) (Gong et al. 2021). Table 1 displays relevant grouping information and clinicopathological characteristics. The METABRIC dataset was downloaded from the cBioPortal (http://www.cbioportal.org/) following the website’s guidance. The GSE76250 and GSE18864 cohorts were downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) according to the website’s guidance.

Table 1.

Clinical characteristics of TCGA cohort and FUSCC coho

Variables	TCGA	FUSCC
	Discovery cohort	Validation cohort
Age (years)
< 50	59	141
≥ 50	57	117
Gender
Female	116	258
Male	0	0
Clinical stage
I	20	0
II	72	41
II–III	0	20
III	20	174
IV	1	0
Unknown	3	23
OS status
Alive	97	228
Dead	19	30

Collection of LRGs

We obtained a total of 287 genes associated with lactate metabolism from the Molecular Signatures Database (MSigDB; https://www.gsea-msigdb.org/gsea/msigdb/). Nine human gene sets were included using the search keyword “lactic”, and after deleting duplicates, 287 lactate-related genes were identified.

Analysis of LRGs in TNBC

Principal component analysis (PCA) was conducted and visualized using the “FactoMineR” and “factoextra” package in R. Differences in LRG expression and lactate metabolic enzyme expression between TNBC and normal samples were analyzed using the “limma” package in R, and the results were included in Supplement Tables S3 and S4. The cut-off criterion was p < 0.05 and a fold change (FC) > 2. Somatic mutation profiles of TNBC patients (Supplement Tables S5 and S6) were analyzed and visualized using Hiplot online tools (https://hiplot.com.cn/home/index.html) according to the website’s guidance.

Clustering analysis

Hierarchical Clustering was performed using the “cluster” package in R. Survival analysis of distinct subtypes of patients was analyzed and visualized using the “survival” and “survminer” package in R. To better understand the biological functions related to distinct clusters, Gene Ontology (GO) annotation and Gene Set Enrichment Analysis (GSEA) were performed using an online tool (https://www.bioinformatics.com.cn/). The results were included in Supplement Tables S7–S11, and some of the important results in functional enrichment analysis were visualized via Hiplot online tools.

Construction and validation of the lactate-related prognostic model (LRPM)

Based on log-rank analysis (Supplement Table S12), Five LRGs (NDUFAF3, NDUFS7, NDUFB8, CARS2, and FH) were identified. Subsequently, using the “glmnet” package in R, three LRGs (NDUFAF3, CARS2 and FH) were selected to construct the LRPM via least absolute shrinkage and selection operator (LASSO) analysis. The lactate-risk score (LRS) was calculated for each patient using the following formula: LRS = mRNAgene1 × coefficientsgene1 + mRNAgene2 × coefficientsgene2 + mRNAgene3 × coefficientsgene3. Then, the patients were divided into two groups based on the median value of LRS, and survival analysis was used to evaluate the prognostic value of this model. The performance of the model was also assessed using receiver operating characteristic (ROC) curve analysis, which was performed using the “timeROC” package in R.

Plasmid construction and transfection

The coding regions of NDUFAF3, FH and CARS2 coding regions were individually tagged with FLAG and cloned into empty loading plasmids (pSin-EF2-puro) to obtain the overexpression plasmids pSin-EF2-puro-NDUFAF3-FLAG, pSin-EF2-puro-FH-FLAG and pSin-EF2-puro-CARS2-FLAG.

For transient transfection, the human TNBC cell lines HCC1806 were cultured in RPMI-1640 (Invitrogen) medium supplemented with 8% foetal bovine serum (FBS, Gibico). HCC1806 cells were transfected with the overexpression plasmids using Neofect (Cat#TF20121201) and harvested for RT-qPCR to determine the efficiency of gene expression.

RNA extraction and RT-qPCR

Total RNA was extracted from TNBC cells using RNA-Quick Purification Kit (ESscience). Reverse transcription was performed using the Reverse Transcriptase Kit (Promega). Quantitative PCR reaction was conducted on a LightCycler 480 System (Roche) with SYBR qPCR Master mix (Vazyme). Relative gene expression was normalized to GAPDH expression. The primer sequences were as follows:

• NDUFAF3 forward: 5′-GCTTTTCCCTCTTCTGGTTGCTG-3′

• NDUFAF3 reverse: 5′-GTTGAAGGTGGCACAGGCATTG-3′

• FH forward: 5′-CCGCTGAAGTAAACCAGGATTATG-3′

• FH reverse: 5′-ATCCAGTCTGCCATACCACGAG-3′

• CARS2 forward: 5′-CGAGAAGTCCTGCTGTGTTTGG-3′

• CARS2 reverse: 5′-ACCACACCATGCAAGGTAGCCT-3′

Cell proliferation assays

For CCK8 assays, HCC1806 cells were seeded into 96-wells plates at a density of 1000 cells per well. A volume of 10ul of Cell Counting Kit-8 (CCK8, TargetMol) was added to each well on the indicated days and incubation at 37 °C for 2 h. Then, the absorbance at 450 nm was detected using a spectrophotometer. For colony formation assay, single-cell suspensions were seeded into 6-well plates (800 cells per well) and then incubation 14 days. The plates were fixed with methanol for 30 min and then stained with crystal violet for 1 h. Colonies containing more than 50 cells were counted using ImageJ.

Establishing and assessing a nomogram

To predict the 1-, 3-, and 5-year survival rate of TNBC patients, we constructed a nomogram based on LRS and tumour clinical stage using the “regplot” package in R. The calibration curve was used to estimate the consistency between the predicted survival and actual survival, using the “calibrate” package in R. To evaluate the specificity and sensitivity of the nomogram, the time-dependent ROC curves and the decision cure analysis (DCA) were performed, using “timeROC”, and “ggDCA” package in R respectively.

Tumor microenvironment analysis

We investigated the immune infiltration of distinct breast cancer subtypes using the EPIC and TIMER algorithms. The xCell algorithms were applied to discriminate immune cell phenotypes in two different risk groups, and the results were listed in Supplement Tables S13–S15. Additionally, we calculated the ESTIMATE score, which can indicate stromal cell and immune cell infiltration profiles, using the “estimate” package (Supplement Table S16) and visualized the results using the “ggplot” package. Furthermore, we calculate TIED score, MSI score, Dysfunction score and Exclusion score in Tumor Immune Dysfunction and Exclusion (TIDE) (http://tide.dfci.harvard.edu/login/) to evaluate the potential response to immunotherapy, and the results were listed in Supplement Tables S17 and 18. Finally, we conducted a correlation analysis among the expression of immune-related genes, LRSs and the expression of three PLGs, using the cor R package. The results were listed in Supplement Table S19.

Drug sensitivity analysis

We predicted the semi-inhibitory concentration (IC50) values of 198 drugs in the Genomics of Drug Sensitivity in Cancer (GDSC) database using the “OncoPredict” package in R (Supplement Table S20). We conducted correlation analysis to clarify the relationship between drug sensitivity and LRS, as well as comparing the differences in IC50 values in two different lactate-risk groups.

Statistical analysis

All statistical analyses were conducted using R v4.2.2 software. Comparisons between the two groups were calculated using the unpaired Student’s t test with two-sided p values less than 0.05 considered statistically significant. The p value in multiple analyses was corrected using the False discovery rate (FDR). Correlation analyses were conducted by Pearson correlation test.

Results

Dysregulation of LRGs in TNBC

We first extracted 287 LRGs from the MSigDB and then explored the correlation between LRGs mRNA expression levels and lactate metabolomic abundance in the FUSCC-TNBC cohort (Supplementary Table S1). The results showed that three LRGs had positive correlations (correlation coefficient > 0.2) with lactate metabolomic abundance, while three LRGs had negative correlations (correlation coefficient < − 0.2), reflecting the complexity of the lactate metabolism (Fig. 1A, Supplementary Table S2). On the other hand, the PCA demonstrated the significant regulatory effects of LRGs and lactate metabolomic abundance in TNBC. It was feasible to discriminate TNBC from normal samples in the TCGA cohort based on the mRNA expression of LRGs, as well as in the FUSCC cohort based on the lactate metabolomic abundance (Fig. 1B). We further explored the LRGs expression profiles in the TCGA-TNBC cohort. There were 48 LRGs that showed significant differential expression (p < 0.05 and |log₂FC|≥ 1), with 35 being upregulated and 13 being downregulated (Fig. 1C). In addition, we determined whether the mRNA expression levels of the lactate metabolic enzymes were dysregulated in TNBC. The results from three individual datasets (TCGA, METABRIC, GSE76250) showed that most lactate metabolic enzymes were upregulated in TNBC patients (Fig. 1D). Furthermore, somatic mutation profiles of TNBC patients in the TCGA and FUSCC cohorts were analyzed and visualized. The results showed that both TNBC cohorts had low mutation rates (Fig. 1E, F).

Fig. 1.

The expression profiles of lactate-related genes in TNBC: A the correlation between mRNA expression of lactate-related genes (LRGs) and lactate metabolomic abundance in FUSCC-TNBC cohort. B Principal component analysis of TNBC and normal samples in the TCGA-TNBC and FUSCC-TNBC cohorts. C Expression differences of LRGs between TNBC and normal samples in the TCGA cohort. The y-axis represents the logarithm of fold changes of LRGs. D Diagram summarizing the metabolic genes involved in glycolysis. Bottom colors of each gene were depicted as the ratio of mRNA expression in TNBC samples of three datasets (TCGA, METABOLIC, GSE76250) compared to the normal samples. Red indicated upregulated genes and blue indicated downregulated genes. E, F Mutation frequency of LRGs in the TCGA cohort and FUSCC cohort

Identification of TNBC subtypes and their biological functions

To better clarify the lactate metabolism characteristics in TNBC, we divided a total of 258 TNBC patients from the FUSCC-TNBC cohort into subgroup A (n = 189) and subgroup B (n = 69) using the hierarchical clustering algorithm based on the expression of LRGs. The LRGs expression profiles of the two subgroups were illustrated in a heatmap which showed significant differences (Fig. 2A). Survival analysis revealed that patients in subgroup A had a worse prognosis in the previous 5 years than those in subgroup B (Fig. 2B). To explore the heterogeneity of two subgroups, we performed functional enrichment analysis after identifying the differentially expressed genes (DEGs). We found that the cluster-related DEGs were mainly enriched in biological processes linked to immune response processes (Fig. 2C). Moreover, GSEA results suggested that subgroup A was significantly associated with several cancer-related pathways, which could potentially account for the varied clinical outcomes among the different lactate subgroups (Fig. 2D). We further used the TIMER and EPIC algorithms to calculate the immune cell infiltration score. We found that the two clusters had distinct immune phenotypes, with subgroup A exhibiting a significantly higher abundance of most immune cells compared to subgroup B (Fig. 2E, F). As a result, subgroup B, which had a higher expression of LRGs, could be considered an immune-desert phenotype. Building on these findings, we suggested that lactate may have a crucial role to play in the immune microenvironment remodeling process in TNBC.

Fig. 2.

Identification of Lactate-Related Subgroups. A The FUCC-TNBC cohort was divided into two subtypes (subgroup A and subgroup B) based on expression levels of LRGs. The consensus matrix heatmap showed different expression profile of LRGs in the two subgroups, with samples displayed in columns. B Survival analysis of the two TNBC subtypes, revealed that patients in subgroup A had a lower 5-years survival probability (p = 0.035). C GO pathway enrichment analysis in the two subgroups. The x-axis represents the number of genes in the GO terms and the y-axis represents the enriched GO terms including Biological Process (BP), Molecular Function (MF) and Cell Component (CC). The p value is < 0.05. D GSEA analysis in the two subgroups. The GLYCOLYSIS pathway was enriched in subgroup B, while several cancer-related pathways were enriched in subgroup A. The p value is < 0.05. E, F Boxplot showed the different abundance of infiltrating immune cell types in the two subgroups

Identification of LRGs and their biological functions

As patients with different lactate metabolic profiles showed different immune statuses and prognoses, we believed it was necessary to establish an LRPM to assist in determining TNBC clinical treatment strategies. We identified five PLGs using log-rank analysis and performed LASSO to establish an LRPM composed of three PLGs (Fig. 3A). Survival analyses indicated that patients with low expressions of the above three PLGs had poor prognosis (Fig. 3B). To evaluate the three PLGs protein expression levels, we utilized a public database called the Human protein Atlas. The results showed that NDUFAF3 and FH were downregulated in breast cancers compared to normal tissues, while CARS2 showed no significant difference (Supplementary Fig. 1A). The protein expression profile in TNBC needs further exploration.

Fig. 3.

Identified PLGs in TNBC: A Coefficient profiles of five PLGs (NDUFAF3, NDUFS7, NDUFB8, CARS2 and FH) in the LASSO analysis and identification of the best parameter (lambda). B Survival analysis based on the expression of three PLGs (NDUFAF3, FH and CARS2) included in lactate-related prognosis model (LPRM). The low expression of these PLGs is correlated with worse survival probability. C CCK8-assay of HCC1806 cells transfected with pSin-EF2-NDUFAF3, pSin-EF2-FH, pSin-EF2-CARS2 or empty vector. D Representative images (left) and quantification (right) of the colony formation assay in HCC1806 cells transfected with pSin-EF2-NDUFAF3, pSin-EF2-FH, pSin-EF2-CARS2 or empty vector. The above experiments were independently repeated at least three times

Since the three PLGs were downregulated in TNBC, we investigated whether they could inhibit TNBC cell proliferation. Overexpression of NDUFAF3, FH, and CARS2 in HCC1806 substantially decreased cell proliferation and colony formation (Fig. 3C, D, Supplementary Fig. 1B). These results suggested that the downregulation of NDUFAF3, FH, and CARS2 played an important role in TNBC.

Establishment of LRPM in TCGA cohort and validation in FUSCC cohort

Subsequently, we calculated the LRS for each patient based on the expression of the three PLGs (NDUFAF3, FH, CARS), and the patients were categorized into the high-risk group and the low-risk group by utilizing the median LRS as the dividing point. As expected, the high-risk group showed higher lactate abundance (Supplementary Fig. 2). As shown in Fig. 4A, patients with high LRS scores had shorter overall survival (OS) times than those with low LRS scores. Meanwhile, as shown in Fig. 4B, the LRS plot and survival status plot demonstrated that the number of TNBC patients with dead status gradually increased with increasing LRS, and the heatmap revealed that the three PLGs were significantly downregulated in the high-risk group. Additionally, the time-dependent ROC analyses showed that the areas under the curve (AUCs) value were 0.668, 0.707, 0.777, 0.647 for 1-, 3-, 5- and 10-year survival, respectively (Fig. 4C).

Fig. 4.

Construction and validation of LRPM: A, D overall survival (OS) analysis of TNBC patients in the high-risk and low-risk groups. B, E Ranked dot plot showing the distribution of survival status and LRS of each patient, and a heatmap showing the expressions of three PLGs. Upper panel: the TNBC patients were divided according to the median LRS. Blue represents the low-risk group and red represents the high-risk group. Middle panel: the y-axis represents the overall survival time of the patients divided into two groups at the Upper panel. Lower panel: the expression profile of NDUFAF3, FH and CARS2 in the different risk groups. C, F The time-dependent ROC analysis of LPRM in predicting 1-, 3-, 5- and 10-years OS. The AUC value which represents the area under the curve represents the precision of LPRM

To further validate the performance of the LRPM in different populations, we designated the FUSCC-TNBC cohort as the validation cohort. We calculated the LRS of each patient using the same formula and separated TNBC patients into the high-risk group and the low-risk group using the median LRS as the cut-off point. Similarly, TNBC patients with high-risk scores had poorer OS than those with low-risk scores (Fig. 4D). The LRS, survival status and expression levels of the three PLGs of every TNBC patient were displayed in Fig. 4E. Moreover, the AUCs were 0.979, 0.613, 0.583, 0.797 for 1-, 3-, 5- and 10-year survival, respectively (Fig. 4F). The above results indicate an excellent predictive performance of the LPRM in TNBC.

Construction and assessment of a clinical nomogram

We then performed multivariate COX analysis combining the LRS with clinical characteristics, including age and tumor clinical stage. The results showed that LRS acted as a prognostic risk factor for TNBC independent of clinical characteristics (Supplementary Fig. 3A). Additionally, we explored the relationships between the LRS and clinical characteristics of TNBC patients and noted that patients over 55 years old and those with advanced tumor stages had a higher probability of being categorized as the high-risk group (Supplementary Fig. 3B, C). Furthermore, our analysis revealed that the high-risk group patients were more likely to suffer from breast cancer with advanced N stage, indicating more lymph node metastases (Supplementary Fig. 3D). However, no significant differences were observed between patients stratified by the T stage and M stage (data not shown). Moreover, we discovered that LRS was linked to the response to neoadjuvant chemotherapy of TNBC patients in GSE18864 cohort (Supplementary Fig. 3E), patients in the high-risk group were more likely to have chemotherapy no-response, as indicated by the miller-payne response grade of 0–1. Overall, these results demonstrate that LRS acts as an independent risk factor in TNBC and suggest that the LRS can be used to construct a clinical nomogram to help predict the prognosis of TNBC patients.

For improve the clinical application of LRS, we developed a nomogram to predict 1-year, 3-year and 5-year OS by integrating LRS and tumor clinical stage in the TCGA cohort (Fig. 5A). To internally validate the nomogram, we conducted a calibration plot. The results demonstrated outstanding consistency between the predicted values generated by the nomogram and the actual 1-, 3-, and 5-year OS (Fig. 5B). In addition, we evaluated the predictive efficiency of the nomogram by calculating the AUCs for 1-, 3- and 5-year OS. The result showed that the nomogram had higher efficiency than other clinical factors such as age, tumor clinical stage and LRS, with AUCs of 0.83, 0.88 and 0.75, respectively (Fig. 5C). Moreover, we conducted decision curve analysis (DCA) to evaluate the predicted performance of age, tumor clinical stage, LRS and nomogram. As shown in Fig. 5D, the nomogram exhibited a better clinical benefit. Overall, these results indicate that the nomogram has high accuracy for predicting the prognosis of TNBC patients and may aid clinical management.

Fig. 5.

Construction and assessment of a clinical nomogram: A the nomogram based on LRS and the clinical stage was used for the prediction of the overall survival of TNBC patients. The “score” represents a scoring scale for each factor and the “Total score” represents the sum of the scoring scale. The overall survival rate of 1-, 3- and 5-years was inferred according to the “Total score” and shown in the lower panel. B Calibration curves were used to assess the accuracy of prognostic prediction. The x-axis represents the OS predicted by the above nomogram and the y-axis represents the true OS. C The Time-dependent ROC analysis of different predictors, including age, the above nomogram, LRS and clinical stage, to predict 1-, 3- and 5-year OS. The AUC value which represents the area under the curve represents the precision of each predictor. D The decision curve analysis (DCA) of the above nomogram. The y-axis represents the net benefit at different thresholds of different predictors including age, the above nomogram, LRS and clinical stage

Associations between lactate-risk score and immune infiltration

Previous studies have highlighted the significant role of lactate in the remodelling of the tumor immune microenvironment. To explore the relationship between LRS and immune infiltration, we performed GSEA analysis using the LRS-correlated genes and calculated the immune cell infiltrate score by utilizing the xCell algorithm. Our findings revealed that signaling pathways involved in the tumor immunosuppressive microenvironment and inflammatory response were significantly enriched in the high-risk group (Fig. 6A). In addition, the infiltration of CD4⁺ T cells notably decreased in the TNBC of the high-risk group (Fig. 6C). Notably, our results also indicated that LRS was significantly correlated with the modulation of the tumor microenvironment modulation, such as angiogenesis, myogenesis and epithelial-mesenchymal transition (Fig. 6B). Correspondingly, the infiltration of cancer-associated fibroblasts and hematopoietic stem cells showed a meaningful increase in the high-risk group (Fig. 6D). In addition, it was observed that the high-risk group displayed a greater degree of stromal score and ESTIMATE score, but a reduced level of tumor purity in comparison to the low-risk group (Fig. 6E). Our finding suggested that lactate is a crucial factor in remodeling the immune microenvironment within the tumor.

Fig. 6.

Immune cell infiltration in two risk groups: A, B GSEA analysis of high-risk group and low-risk group. C, D Horizontal bean plots showed the different abundance of infiltrating immune cell types in high-risk and low-risk groups. E Violin plots were used to show the difference in the immune score, stromal score, ESTIMATE score and tumor purity between high-risk and low-risk groups

To further investigate the relationship between LRS and ICI response, we calculated the TIDE score online and found that the high-risk group in the FUSCC-TNBC cohort showed a raised TIED score, Dysfunction score and Exclusion score, but no significant difference in microsatellite instability (MSI) score (Supplementary Fig. 4A). Similarly, the high-risk group in the TCGA-TNBC cohort displayed a greater degree of Dysfunction scores, but lower microsatellite instability (MSI) scores (Supplementary Fig. 4B). These differences imply that individuals in the high-risk group may not respond well to ICI therapy and that LRS could be utilized as a predictive marker of immunotherapy responsiveness. Additionally, we examined the relationship between immune-related genes (Zheng et al. 2022) expression and LRS, as well as three PLGs expression (Supplementary Fig. 4C). We observed no significant correlation between LRS/PLGs and classic checkpoint molecules, except for FH which was negatively correlated with some immune-inhibitory molecules such as VEGFE and TGFB1. This implies that FH may act as a hub gene for predicting immunotherapy response in TNBC.

Associations between lactate-risk score and drug susceptibility

To investigate whether there was a correlation between LRS and drug sensitivity, we utilized the “OncoPredict” package to predict the IC50 values of 198 drugs in the Drug Sensitivity in Cancer (GDSC) database. We observed a significant correlation between the LRS and several common anticancer drugs, including Fluorouracil, Alpelisib, Fulvestrant and Olaparib (Fig. 7A, B). Furthermore, we compared the IC50 values between the two risk groups. The results showed that the low-isk group exhibited enhanced sensitivity to several common chemotherapeutic drugs for TNBC treatment, such as Cisplatin, Docetaxel, Gemcitabine and Paclitaxel (Fig. 7C). These findings suggest that LRS can help identify suitable patients for appropriate therapy and provide a practical tool for TNBC treatment decision-marking.

Fig. 7.

The correlation between LRS and drug susceptibility: A, B the correlations between LRS and IC50 values. C Beanplots showing the high-risk group had higher IC50 values of common antitumor drugs in TNBC

Discussion

Breast cancer is generally considered a poorly immunogenic malignancy (Kandoth et al. 2013), but this is no always the case. Considerable evidence shows that compared to other subtypes of breast cancer, TNBC has a higher tumor mutation burden (TMB), frequent copy number changes (CNS) and more genetic instability (Bareche et al. 2018). As a result, TNBC should be a leading candidate for potential breast cancer immunotherapies (Zhu et al. 2020). A growing number of clinical trials aims to explore the role of immunotherapy in TNBC, but inconsistent results have been observed following different therapy strategies and tumor-stage patients (Keenan and Tolaney 2020). The results of IMpassion130 study shows that atezolizumab combined with paclitaxel prolongs progression-free survival (PFS) in PD-L1-positive metastatic TNBC patients (Schmid et al. 2018). Additionally, the KEYNOTE-355 study in metastatic TNBC patients demonstrates a meaningful improvement in PFS with pembrolizumab combined chemotherapy versus placebo combined chemotherapy (Cortes et al. 2020). However, the KEYNOTE-119 study reveals that the pembrolizumab does not significantly improve overall survival (OS) in metastatic TNBC patients (Winer et al. 2021). Since the efficacy of single-agent immune therapy in TNBC is low and varies among individuals (Keenan and Tolaney 2020), identifying novel biomarkers that can predict immunotherapy response in TNBC is a high priority for clinical development.

Lactate plays an important role in various diseases including tumor (Li et al. 2022a). The main feature of metabolic reprogramming is aerobic glycolysis and accumulation of lactate in tumor microenvironment, which plays an essential role in the oncogenesis and immunosuppression (Chen et al. 2022). The LRGs which abnormally expressed in TNBC were able to discriminate TNBC from normal samples, and the lactate metabolic enzymes (such as SLC2A1, SLC16A1 and GAPDH) were upregulated in TNBC compared to normal samples. This indicates the existence of lactate metabolic reprogramming in TNBC. We then divided TNBC patients into two subtypes based on the expression of LRGs. The glycolysis signaling pathway was significantly enriched in cluster B, which had a lower proportion of B cells, CD4⁺ T cells and CD8⁺ T cells. These findings suggest the link between lactate and immunity signature, and LRGs might serve as a novel biomarker for predicting response to immunotherapy in TNBC. In addition, previous reports have demonstrated that the gene signature based on LRGs has robust and effective predictability in Kidney Renal Clear Cell Carcinoma (Sun et al. 2022), Hepatocellular Carcinoma (Li et al. 2022b) and Lung Adenocarcinoma (Zhao et al. 2022). Here, we constructed the first LRPM in TNBC.

Three PLGs, including NDUFAF3, CARS2 and FH, were identified and used to construct the LPRM. NDUFAF3 (NADH: ubiquinone oxidoreductase complex assembly factor 3) is a mitochondrial respiratory chain complex I assembly protein. Mitochondrial complex I deficiency is the most common disorder of the oxidative phosphorylation system, and the deficiency of the NDUFAF3 has been identified as a cause of several mitochondrial disease (Saada et al. 2009; Baertling et al. 2017). Similarly, CARS2 (Cysteinyl-tRNA Synthetase 2), a mitochondrial aminoacyl-tRNA synthetase and novel cysteine persulfide synthase, has been reported to be correlated with defects in mitochondrial function and mitochondrial translation, such as mitochondrial epileptic encephalopathy (Fujii et al. 2019; Coughlin et al. 2015). The function and specific roles of NDUFAF3 and CARS2 in cancer remain unclear. However, our study revealed that NDUFAF3 and CARS2 might act as protective factors in TNBC. FH (Fumarate Hydratase) is one of the Krebs cycle genes, and its loss-of-function mutation results in epigenetic alterations and contributes to tumorigenesis (Xiao et al. 2012). Transcriptional downregulation of FH has also been found in colorectal cancer and clear cell carcinoma (Schmidt et al. 2020). Clinically, FH-deficient renal cell carcinoma is characterised by early metastasis and poor outcome (Sun et al. 2021). Consistent with previous studies, our study demonstrated that FH loss was correlated with poor outcome in TNBC patients.

We found that TNBC patients with different LRS had distinct immune components. The high-risk group was characterized by a reduced ratio of antitumor immune cells but a higher proportion of TME components and ESTIMATE score. Previous studies have shown that tumor-derived lactate is a key mediator in the tumour microenvironment. It has been reported that lactate can enter endothelial cells and stimulate the NF-kB/IL-8 pathway, promoting tumor angiogenesis (Certo et al. 2021). Furthermore, cancer-associated fibroblasts can secrete lactate to modulate acquired drug resistance in a NF-kB-dependent manner (Pucino et al. 2019). Notably, we observed a significant enrichment of inflammatory-response-related signaling pathways within the high-risk group. This observation is intriguing since a recent study has indicated that lactate has the capacity to stimulate a chronic inflammatory process by triggering a series of intracellular signals in inflammatory disease (Végran et al. 2011). Additionally, it has been reported that lactate, mediated by the upregulated lactate transporter SLC5A12, inhibits CD4⁺ T cell motility in inflamed tissues, and blockade of SLC5A12 diminishes the disease severity in an arthritis model (Jin et al. 2021).

Using the TIED algorithm, we estimated the potential response of ICI therapy in two different LRS groups and found that the low LRS might exhibit sensitivity to ICI therapy. Further exploration is needed to determine the correlation between LRS and other immunotherapy in TNBC. Furthermore, our research revealed that patients with a high-risk score displayed elevated IC50 values of common chemotherapeutic drugs, including Cisplatin, Docetaxel, Gemcitabine and Paclitaxel. These findings suggest that LRS could potentially serve as a marker for predicting chemotherapy and immunotherapy response.

In summarize, we identified three independent PLGs and constructed a novel LRPM for TNBC patients. The LRPM in our study shows promise in improving prognostic prediction accuracy and the ability to assess drug therapy response.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization, XX and YL; data curation, SH and LW; formal analysis, SH and LW; funding acquistion, XX; methodology, SH and YQ; project administration YX and HW; supervision, XX and YL; visualization, SH; writing-orginal draft, SH and LW; writing-review and editing, YL and XX; all authors have read and agreed to the published version of the manuscript.

Funding

This research was supported by the National Natural Science Foundation of China (no. 81974444, Xinhua Xie).

Data availability

All data can be obtained from the public database.

Declarations

Conflict of interest

The authors declare no conflict of interest.

Ethics approval and consent to participate

Not applicable.