Introduction
Esophageal carcinoma (ESCA) is the 7th most commonly diagnosed cancer worldwide [
1], and those with advanced ESCA still suffer from limited treatment options and poor survival rates [
2]. First-line chemotherapy regimens for advanced ESCA typically involve fluorouracil- or platinum-based treatments, which achieve a response rate of 40–60% [
3]. However, if first-line treatment fails, patients have a median overall survival (OS) of only 5–10 months, and there is currently no efficient standard second-line therapy available. Despite the comprehensive characterization of the human genome [
4,
5], targeted therapy for ESCA remains challenging. The tumor–node–metastasis (TNM) classification system is widely used in clinics to assess risk and make treatment decisions [
6]. However, even cases with similar clinicopathological characteristics can exhibit significantly different risks of death and recurrence due to substantial molecular heterogeneities [
7,
8]. Therefore, it is crucial to further investigate the molecular mechanisms associated with the genetic diversity of ESCA to enable precise diagnosis and the development of individualized treatments.
Whole-genome mRNA expression profiling refers to the comprehensive analysis of gene expression patterns in the entire genome of cancer cells, providing detailed information about the molecular characteristics and functional changes in tumors [
9]. By analyzing the expression levels of thousands of genes, researchers can identify gene signatures, pathways and molecular subtypes associated with specific tumor phenotypes or patient outcomes to improve our understanding of tumor biology and identify novel therapeutic targets [
10‐
12]. Thus, while the TNM staging system remains essential for clinical decision-making in esophageal carcinoma, whole-genome mRNA expression profiling research offers additional layers of molecular information that can potentially enhance personalized medicine approaches, improve prognostic accuracy, guide treatment selection, and facilitate the discovery of novel therapeutic targets [
13]. However, the relationship between these identified genes and the clinical characteristics of ESCA still requires comprehensive elucidation, especially considering the lack of validation of these identified targets in cell lines and patient samples.
Cancer has been proposed as a disease associated with metabolic disturbances [
14]. Various genes and mutations linked to cancer interfere with different metabolic processes that support the proliferation of cancer cells, including aerobic glycolysis, one-carbon metabolism, and glycogenolysis [
15,
16]. In ESCA, intratumoral metabolism is also influenced by the heterogeneity of gene mutations [
17,
18]. Glycolytic ESCA cells experience a state of relative oxidative stress due to increased levels of reactive oxygen species (ROS), which is counterbalanced by the activation of redox metabolic pathways [
19]. Additionally, metabolic changes such as obesity and elevated triglycerides (TG) observed in metabolic syndrome (MetS) are associated with the risk [
20,
21]. Furthermore, a retrospective study has confirmed that MetS, characterized by hypertriglyceridemia and impaired fasting glucose, predicts a higher risk of relapse in early ESCA cases [
22]. Numerous studies have suggested that immune function can undergo plasticity in different metabolic contexts [
23,
24]. Some research focuses on specific metabolic patterns to modulate immune polarization and function, opening possibilities for treating immune-related diseases like cancer [
25]. Previous studies have also demonstrated the impact of the tumor microenvironment (TME) on T cell metabolism, particularly in terms of tumor response, offering new insights for opportunities to enhance T cell response through metabolic manipulation, potentially improving the effectiveness of anticancer immunotherapy [
26]. Overall, these findings emphasize the importance of investigating the genetic landscape of ESCA at the metabolic level. Thus, accurate identification and characterization of metabolic subpopulations are crucial for a better understanding of ESCA and optimizing anticancer immunotherapy's effectiveness.
In this study, we aimed to investigate the importance of metabolism-immune interactions and contribute to the understanding of ESCA biology. For this, we classify ESCA patients into metabolic subtypes based on gene expression profiles from the GSE19417 and TCGA-ESCA datasets, investigate the relationship between these subtypes and immune landscapes, establish a metabolic index for accurate prognosis assessment, identify metabolic genes associated with the index, validate their expression in patient samples and cell lines, and assess their functional significance. Collectively, the findings provide important insights into ESCA heterogeneity, which could be used as a reference for strategizing personalized treatment interventions for ESCA patients.
Materials and methods
ESCA datasets
To investigate the characteristic changes of ESCA in metabolism-related pathways, we collected clinical, SNP, and RNA-seq data from 80 ESCA patients from the TCGA database (
https://cancergenome.nih.gov/) based on the following criteria: (a) availability of comprehensive follow-up information and (b) availability of comprehensive gene expression profiles in ESCA. Furthermore, we obtained the expression levels and clinical data from the GSE19417 dataset (
n = 70, samples with insufficient prognosis data were excluded) from the GEO database [
27].
Data preprocessing
The clinical data of the ESCA samples obtained from TCGA were successfully matched with the corresponding RNA-seq data. Additionally, the ENSG identifiers were mapped to GeneSymbol, allowing us to obtain expression profiles for 21,652 genes. Similarly, for the samples from the GSE19417 dataset, the clinical data were consistent with the RNA-seq data, and expression profiles for 20,101 genes were obtained by mapping chip probes to gene names using probe annotation files.
In this study, a total of 2,923 metabolic genes enriched in 117 metabolic pathways were obtained from the Molecular Signatures Database (MSigDB) [
28]. These genes were then used to select metabolic genes associated with survival based on a threshold of log-rank
P < 0.05. Among these, expression data from 163 metabolic genes associated with survival were used for k-means unsupervised clustering. The clustering process was performed using the "ConsensusClusterPlus" package [
29] with 500 iterations. To determine the optimal cluster number, curves were plotted based on the consensus score and cumulative distribution function (CDF). Subsequently, SigClust analysis was conducted to compare the two subtypes and assess the significance of the clustering results.
Prediction of activity of immune-related pathway, immune infiltration, as well as response to immunotherapy
In this study, the ESTIMATE algorithm was employed to calculate the ESTIMATE/immune/stromal scores and tumor purity for all cancer samples [
30]. The xCell algorithm [
31] was used to predict the relative levels of different types of human immune cells within the tumor microenvironment (TME). To validate the results, established approaches such as EPIC [
32] and MCPcounter [
33] were combined. The activities of immune pathways in various subtypes were predicted using Single-sample GSEA (ssGSEA) with the GSVA R package [
34]. Enrichment scores were used to indicate the extent of up- or downregulation of coordinated genes in individual samples. The response to immunotherapy was assessed using the tumor immune dysfunction and exclusion (TIDE) algorithm (
http://tide.dfci.harvard.edu/). A higher TIDE score suggests a higher likelihood of immune escape, indicating that patients are less likely to benefit from immunotherapy.
Establishment of the typical metabolic index for ESCA cases
To establish a typical index for classifying metabolic subtypes, linear discriminant analysis (LDA) was employed. A total of 163 metabolic genes associated with prognosis were utilized, and each segment was subjected to Z-transformation. Fisher's LDA optimization criterion was used to ensure maximal dispersion of the centroids for each group. As a result, a linear combination, denoted as A, was determined to maximize the inter-class variance relative to the intra-class variance. The model indicated that the initial two features were sufficient to distinctly differentiate the different subtypes.
Weighted gene co-expression network analysis (WGCNA) and clustering analysis
Gene transcription data were analyzed using Weighted Gene Co-expression Network Analysis (WGCNA) [
35] to identify metabolic genes associated with co-expression modules. The gene expression data were collected from the TCGA database, and specific parameters were set as follows: a cluster threshold of 5, a median absolute deviation greater than 5%, and a β value of 50%. The expression matrix was then transformed into a topological matrix. Modules were obtained using average linkage with the following parameters: a height of 0.3, a minimum module size of 30, and a depth split of 1.
Sample collection and preparation
Twenty-seven patients diagnosed with esophageal squamous cell carcinoma (ESCA) were included in this study. The patient inclusion criteria were: (1) treatment naïve patients, (2) underwent surgical resection, (3) pathologically confirmed as ESCA, and (4) had complete baseline data required for data analysis for this study. Those who did not consent to the anonymous use of their tissue samples, underwent previous cancer treatment and were synchronously diagnosed with other cancer were excluded. The patients received treatment at the Department of Gastrointestinal Surgery of Fujian Medical University Second Affiliated Hospital. Prior to their participation, written informed consent was obtained from all patients, which was approved by the Institutional Review Board of Fujian Medical University Second Affiliated Hospital (2021 − 414). The detailed clinical characteristics of the patients are presented in Supplementary Table
1.
During surgery, fresh esophageal carcinoma and adjacent normal tissue samples were cryopreserved and stored in liquid nitrogen. Before the experiment, the specimens were thawed at room temperature.
Cell culture
The human normal esophageal epithelial Het-1A cells and esophageal squamous cell carcinoma cells (TE-1) were purchased from Shanghai Cell Resource Center (Shanghai, China). The cells were cultured in RPMI-1640 (Wako, Osaka, Japan) containing 10% fetal bovine serum (FBS) and antibiotics (100 U/ml penicillin and 100 µg/ml streptomycin) in a humidified 5% CO2 incubator at 37°C.
Small interfering RNA (siRNA) experiment
TE-1 cells were selected to investigate the functional role of MTHFD2. The experiment included two groups: the siNC Group and the si-MTHFD2 Group. The siRNAs for MTHFD2 (si-MTHFD sense 5’-GAGCAGUUGAAGAAACAUATT-3’, si-MTHFD antisense 5’-UAUGUUUCUUCAACUGCUCTT-3’) and the siRNAs for negative control (si-NC sense 5’-UUCUCCGAACGUGUCACGUTT-3', si-NC antisense 5’-ACGUGACACGUUCGGAGAATT-3’) were purchased from GenePharma (Shanghai, China).
TE-1 cells were inoculated into 6-well plates with a density of 1.5 × 105 cells per well. Transfection mixtures of si-MTHFD2 and siNC were added to the cells, respectively, according to the manufacturer's instructions (Thermo Fisher Scientific), at a final concentration of 30 nM/well. 48 h after transfection, the transfection effect was detected by qRT-PCR and western blot.
Gene expression analysis by real-time qPCR
RNA was extracted using TRIzol protocol. Next, cDNA synthesis was performed using the PrimeScript RT reagent Kit (Perfect Real Time) following the provided instructions. Reverse transcription of RNA into cDNA was carried out using reverse transcriptase enzyme. The expression levels of target genes, namely SLC5A1, MTHFD2, and NT5DC4, were determined using real-time PCR with the TB Green® Premix Ex Taq fluorescent II kit (Tli RNaseH Plus). The reaction mixture was prepared according to the manufacturer's instructions, and PCR amplification was carried out using a Roche LightCycler 480II amplification instrument. The reaction conditions included an initial denaturation step at 95°C for 30 seconds, followed by 40 cycles of denaturation at 95°C for 5 seconds and annealing/extension at 60°C for 30 seconds. The primer sequence is shown in Supplementary Table
2.
Western blot
The cells were lysed using a protein lysis buffer supplemented with protease inhibitors, and the protein concentration was determined using a protein assay kit. Equal amounts of protein were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). The proteins were then transferred onto PVDF membrane, followed by blocking using a blocking buffer to prevent non-specific binding and incubated with the primary antibody against MTHFD2 (ab151447, abcam, UK) overnight at 4°C. After washing, the membrane was incubated with a secondary antibody conjugated with horseradish peroxidase (HRP) for one hour at room temperature. Protein bands were visualized using an enhanced chemiluminescence (ECL) substrate, and the signals were captured using a chemiluminescence imaging system. The protein expression level of MTHFD2 was normalized to a loading control GAPDH.
Cell counting Kit-8 (CCK-8) assay
Briefly, the transfected TE-1 cells were inoculated in a 96-well plate with approximately 5,000 cells per well and incubated under appropriate conditions for 24 hours. Then, the 10 µL of CCK-8 reagent was added to each well according to the manufacturer's instructions, and the cells were further incubated at 37°C for 1 h, the absorbance value of each well at 450 nm was detected using a microplate analyzer (Multiskan EX, Lab systems, Helsinki, Finland). Each experiment was repeated thrice.
After transfection, the cells were plated in six-well plates at a density of 1000 cells per well and incubated in RPMI-1640 medium supplemented with 10% FBS at 37°C and allowed to grow undisturbed for 14 days to view the visible colonies formed by each cell group. After the incubation period, the colonies were fixed with 4% paraformaldehyde to preserve their structure, stained with 0.2% crystal violet (Sigma), and photographed using an appropriate imaging system. Lastly, the number of cell colonies was recorded for both the siNC group and the si-MTHFD2 group. Each experiment was repeated thrice.
Transwell assay
Matrigel (BD Biosciences) was dissolved overnight at 4°C and then diluted in complete medium at a ratio of 1:3. Around 50 µL of the diluted Matrigel mixture was added to the upper and lower compartments of each well in a 24-well plate (Millipore). The plate was then placed in a 37°C cell incubator for 30 minutes to allow the Matrigel to coagulate. Next, The cells were suspended in 200 µL serum-free medium and added to the upper compartment of the Transwell insert, and 600 µL whole medium was added to the lower cavity. After incubation for 24 hours, the matrix and upper compartment cells were carefully removed. Cells invading and settling in the submembrane cavity through matrix coating are stained with crystal violet. Images of the stained cells were captured using a Nikon frontal microscope. Ten random fields of view were selected, and the number of invaded cells was counted in each field. Each experiment was repeated thrice.
Statistical analysis
R software (version 3.5.3,
http://www.R-project.org) was used for statistical analysis. The relationship between clinical features and different ESCA subtypes was determined using the chi-squared test or Fisher's exact test. Correlation coefficients were determined using Pearson's correlation and distance correlation analyses. The contingency table was checked using a two-sided Fisher exact evaluation. In addition, Kaplan-Meier evaluation was employed to generate cluster survival curves, whereas statistical differences were analyzed using the log-rank test. In multiple testing, FDR correction was performed to reduce the false-positive rate (FPR). All the basic research experiments were replicated at least three times. The data are expressed as mean ± standard deviation, and the means of two groups were determined using the Wilcoxon test for outliers and unpaired Student's t-test for normally distributed variables. One-way ANOVA and Kruskal-Wallis test were used separately for parametric and nonparametric approaches for comparison between different groups. A difference of
P < 0.05 (two-tailed) was considered statistically significant.
Discussion
Metabolic reprogramming is a well-established hallmark in cancer [
36], closely related to ESCA proliferation, migration, and invasion by interacting with immune cells, stromal cells, as well as immune cell interactions and stromal crosstalk [
37,
38]. It has been suggested that metabolic reprogramming reduces immune responses by starving T cells and secreting suppressive metabolites [
39,
40]. A pan-cancer study revealed that the expression profiles of genes associated with the metabolic pathway may indicate the actual metabolic activity of patients [
41]. Therefore, it is important to identify metabolic subtypes and associated immune landscapes to discover metabolic heterogeneities and analyze survival differences in ESCA cases. Additionally, constructing metabolic characteristic indicators can help evaluate and stratify patient outcomes and provide novel targets for treatment strategies.
In this study, we used bioinformatics to identify three metabolic subtypes (MC1-MC3) with different metabolic profiles. Among the subtypes, MC3 showed an increased proportion of immune cells and immune pathways, along with improved prognosis. However, further analyses suggest that MC3 increases the TIDE score, indicating a decrease in response to immunotherapy, and these findings were validated as signatures of immunosuppression. The MC3 subtype had increased stromal scores, which have been related to tumor development through the remodeling of anticancer immunity and immunotherapy response [
42,
43]. Overall, combining immunotherapy with anti-metabolites targeting metabolic pathways, such as glycosaminoglycan metabolism, may suppress immune dysfunction in MC3 subtype patients. On the other hand, the MC1 subtype samples had the poorest prognosis, with the second-highest immunosuppression score and immune cell infiltration level and the lowest TIDE score but the strongest response to immunotherapy. These results suggest that the tumor microenvironment in this subtype exhibits distinct immune activation status, emphasize the need for tailored treatment strategies and highlight the potential of combining metabolic and immunotherapeutic approaches for ESCA patients.
The present study employed WGCNA and the LDA algorithm to identify key metabolic genes across the three ESCA sample subtypes using a typical characteristic index. The results revealed that the genes associated with the typical index were clustered into 13 distinct modules. Notably, the yellow, green, and turquoise modules exhibited significant correlations with the typical metabolic indices. From these modules, 13 characteristic genes (SLC40A1, GSTP1, MTHFD2, NT5DC4, ABCF2, SLC5A1, CDS1, SLC37A1, IVD, DPEP2, TAP2, PDE1A, and FAR1) were selected as key metabolic genes for ESCA cases, their potential in predicting patient prognosis was assessed, and the results indicated that SLC5A1, NT5DC4 and MTHFD2 showed significant associations with ESCA prognosis. The robustness and reproducibility of the metabolic index across diverse datasets enhance its clinical potential for assisting in prognosis assessment and guiding personalized treatment decisions.
Two of the three genes mentioned above have been implicated in cancer-related processes. SLC5A1, also known as sodium/glucose transporter 1 (SGLT1), is highly expressed in various tumors and contributes to the uptake of glucose by tumor cells, supporting their glycolytic metabolism. It has been associated with cancer cell growth, metastasis, and poor survival outcomes [
44]. Methylenetetrahydrofolate dehydrogenase 2 (MTHFD2) is a key enzyme involved in the m1C cycle, which plays a role in metabolic reprogramming, immune evasion, and disease progression in multiple cancers, including prostate cancer, lung adenocarcinoma, and ESCA [
45‐
49]. However, the association of NT5DC4 with patient survival has not been previously reported, and its specific role in regulating the biological functions of tumor cells needs to be further investigated in in vivo and in vitro studies.
The gene expression analysis conducted in ESCA samples and cell lines revealed distinct expression patterns of the three promising genes (MTHFD2, SLC5A1, and NT5DC4) compared to normal tissues and control cell lines. Specifically, MTHFD2 expression was significantly increased in ESCA samples, indicating its upregulation in the disease context, suggesting its potential involvement in promoting tumor growth and progression. The subsequent protein expression analysis validated the gene expression findings of MTHFD2, reinforcing its relevance in ESCA and demonstrating its potential as a biomarker for the disease. To evaluate the functional impact of MTHFD2 downregulation, functional assays were performed. Overall, the functional characterization assays provide important insights into the functional role of MTHFD2 in ESCA, demonstrating its involvement in cell viability, proliferation, colony formation and invasion. These findings are in light with previous literature [
50], support the significance of MTHFD2 as a potential therapeutic target in ESCA and suggest that targeting MTHFD2 could be a promising approach to inhibit tumor growth and progression, as well as a link between the identified metabolic subtypes and their associated molecular mechanisms, adding to the overall understanding of ESCA heterogeneity and its clinical implications.
Overall, the findings of this study have potential clinical impacts. The identification of metabolic subtypes and their associated metabolic profiles can aid in refining prognosis assessment and personalizing treatment strategies for ESCA patients [
51]. The differential immune cell infiltration observed among the subtypes suggests the potential for tailored immunotherapeutic approaches. The construction of a metabolic index with high accuracy in classifying ESCA cases can assist in clinical decision-making and contribute to precision medicine [
52]. Moreover, the identification of prognostic metabolic genes, such as MTHFD2, provides potential targets for therapeutic interventions and opens avenues for developing novel treatment strategies. To further advance the field, future studies could explore the functional mechanisms through which the identified metabolic genes, including MTHFD2, contribute to ESCA progression and treatment response. Additionally, investigations into the crosstalk between metabolic reprogramming and immune modulation within the tumor microenvironment could provide deeper insights into the interplay between metabolism and the immune system in ESCA. Longitudinal studies involving larger patient cohorts and prospective clinical trials are needed to validate the clinical utility of the identified metabolic subtypes, metabolic index, and prognostic genes, ultimately translating these findings into improved patient outcomes and more effective therapeutic interventions for ESCA.
Despite the interesting findings reported, there were several limitations that should be acknowledged. Firstly, although the inclusion of 150 cases from RNA-seq and microarray platforms enhances the reliability and robustness of the findings, prospective studies are necessary to validate whether the identified key metabolic genes can serve as prognostic indicators and predictors of response to immunotherapy in patients with ESCA. Secondly, this study relied on bioinformatic analysis of publicly accessible cancer databases, and further studies involving clinical samples are warranted to validate the identified hallmarks. Thirdly, additional investigations are needed to elucidate the underlying mechanisms linking metabolic modulation and ESCA prognosis for the identified metabolic genes. Lastly, exploring whether intratumoral metabolic features undergo changes following treatment in ESCA patients and whether such changes influence patient response to immunotherapy is important.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.