Background
Endometrial carcinoma (EC), one of the most common female reproductive malignancies, caused nearly 90,000 deaths worldwide each year [
1]. Women with metabolic disorders, including obesity and diabetes, have a markedly increased risk of developing endometrial cancer. While early-stage endometrial cancer has a favorable prognosis, nearly 30% of patients are still diagnosed at a late stage, and over 80% of these individuals die in 5 years [
2]. In addition, several EC patients present a high risk of cancer progression or recurrence with insensitivity to chemotherapy, which indicates poor outcomes [
3]. Therefore, it is imperative to emphasize the molecular changes that occur during endometrial cancer progression and develop novel predictive biomarkers to accurately estimate patient outcomes.
Since fundamental metabolic differences between cancer and adjacent normal cells were first uncovered, metabolic reprogramming has increasingly become a hot topic in cancer biology [
4]. The metabolic phenotype of cancer cells is heterogeneous in various cancer types; for example, while several malignant tumors mainly rely on glycolysis, others present a metabolic phenotype mediated by oxidative phosphorylation [
5,
6]. Overall, through reprogramming tumor microenvironments, catabolic and anabolic metabolism is essential for cancer cells to sustain energy supply and biomass synthesis [
7‐
9]. While the underlying processes and molecular alterations of metabolic programming in various cancers have been well elucidated, the expression patterns of metabolism-related genes in endometrial cancer are still unclear.
In this study, we focused on the metabolism-related gene expression alterations of The Cancer Genome Atlas (TCGA) EC patients and obtained prognostic dysregulated MRGs. In addition, we established and validated a multiple-MRG-combined expression signature for predicting EC patient outcomes. Moreover, we integrated the clinical features of patients and the MRG model to establish a novel nomogram model that could guide comprehensive EC therapeutic strategies.
Methods
Integration of gene expression profiles and clinical information
We downloaded the mRNA expression profiles of EC patients (in the FPKM format) from The Cancer Genome Atlas (TCGA) database (
https://tcga-data.nci.nih.gov/tcga/), which contains a total of 541 cases. The cBioPortal for Cancer Genomics (
http://cbioportal.org) enables researchers to explore, visualize, and analyze multi-dimensional cancer genomics data and clinical information. The corresponding clinical data of EC was retrieved from the cBioPortal [
10].
Genes enriched in metabolic pathways in the KEGG database were utilized in this study as metabolic genes (Supplementary Table
1) [
11]. The mRNA expression of metabolic genes in the TCGA database was extracted.
Identification of prognosis-associated differentially expressed metabolism-related genes (DE-MRGs)
With the cut-off criteria set as |logFC| > 1 and
P-value < 0.05, we screened the DE-MRGs via the “limma”R package [
12]. Then, univariate Cox regression analysis was performed to identify prognosis-associated DE-MRGs. Hazard ratio (HR) < 1 indicates better overall survival (OS) outcomes while HR > 1 indicates worse OS outcomes. Genes with
P < 0.05 were regarded as prognosis-associated metabolic genes. The expression levels of the prognosis-associated DE-MRGs in each patient and between cancerous and normal samples were displayed via the “pheatmap” and “ggplot” R packages, respectively.
Functional enrichment analysis of the prognosis-associated DE-MRGs
Gene ontology (GO) [
13] and Kyoto Encyclopedia of Genes and Genomes (KEGG) [
11] pathway enrichment analyses were performed to explore the biological functions of the prognosis-related DE-MRGs via the “clusterProfiler” R package. Adjusted
P-value < 0.05 was set as the significance threshold, and the enrichment analysis result maps were presented by the “ggplot2” and “GOplot” R packages.
Protein-protein interaction (PPI) network construction and hub DE-MRG alteration analysis
The Search Tool for the Retrieval of Interacting Genes/Proteins (STRING,
https://string-db.org/) database comprises the interaction information of given proteins [
14]. Based on the minimum required interaction score setting of 0.4, we utilized the STRING database to construct a PPI network, which reflected the interactions among the DE-MRG proteins. The network was visualized by Cytoscape software and the top 15 hub proteins were selected based on their connectivity degree in the PPI network [
15]. In addition, the alteration landscapes of the hub DE-MRGs in EC were visualized by cBioPortal.
Establishment of a prognostic model based on the DE-MRGs
We randomly classified all TCGA EC patients into training and testing cohorts. The least absolute shrinkage and selection operator (LASSO) and multivariate Cox regression analysis were then performed to select key prognosis-related DE-MRGs via the “glmnet” R package. The formula of the risk score for the prediction of EC patients’ prognosis was as follows: risk score = the sum of the multivariate Cox regression coefficient ratio of each mRNA multiplied by the expression level of each mRNA. Based on the median risk score, we divided the training cohort patients into high- and low-risk subgroups. In the two subgroups, each patient’s survival status, OS time, and gene expression profile were presented via the “pheatmap” and “survival” R packages. In addition, the Kaplan-Meier curve analysis was performed, and receiver operating characteristic (ROC) curves were drawn to estimate the sensitivity and specificity of the prognostic signature.
Validation of the efficacy of the prognostic DE-MRG signature
The prognostic signature was then introduced into the testing cohort and the entire cohort. Based on the median risk score from the training cohort, the patients in the testing cohort and entire cohort were separated into high- or low-risk groups. Kaplan–Meier curve analysis, time-dependent ROC analysis, and patient outcome distribution were performed.
A total of 30 RNA later-reserved EC specimens were collected from patients who underwent surgery at Jiangsu Province Women and Children Health Hospital (Nanjing, China) between September 2017 and September 2019. All samples were immediately snap frozen in liquid nitrogen and stored at − 80 °C until further analysis. Total RNA was isolated from fresh-frozen tissues using TRIzol reagent (Invitrogen) from fresh-frozen tissues and transcribed into cDNA using a TaqMan Reverse Transcription kit (Applied Biosystems) with random hexamer primers. qRT-PCR was performed using 2 × SYBR Green qPCR Master Mix (Selleck, Shanghai, China). The housekeeping gene GAPDH was used for normalization of the qRT-PCR data before calculation using the ΔΔCt method and Student’s t-test (two-tailed) was used for the comparison analyses. The primers used are listed in Supplementary Table
2.
Evaluation of clinical independence and construction of the nomogram
Next, we removed EC patients who lacked detailed clinicopathological information including survival status and time, age, weight, clinical stage, tumor grade, and lymph node status. The clinicopathological characteristics and the MRG expression data of the remaining patients were compared between the high- and low-risk subgroups and comprehensively displayed in the heatmap. Moreover, the clinical indexes and risk scores were included in univariate and multivariate Cox regression analyses to validate the independence of the risk model. ROC curves for the signature and other clinical features were used to assess the predictive efficacy of the model. In addition, the correlation between the MRGs from the risk model and the clinical index was also measured. Finally, we utilized the “rms” R package to consolidate the risk score and clinical characteristics for nomogram construction.
Discussion
Metabolic abnormalities have recently been widely studied and shown to play an important role in tumor development in various cancer types. Metabolic dysfunction in the tumor microenvironment could lead to various outcomes in patients, and metabolism-related genes can be used as prognostic markers of tumors. In this work, we thoroughly investigated the implications of metabolism-related genes in endometrial cancer progression. By analyzing the mRNA data of TCGA EC patients, we obtained 220 dysregulated MRGs, among which 47 were associated with EC patients’ OS. Functional enrichment analysis of these prognostic MRGs showed that they were closely associated with the cellular amino acid metabolic process, glycolysis, and glycerophospholipid metabolism. In accordance with our observation, Byrne et al. also found that glycolysis and lipogenesis are highly associated with endometrial cancer phenotypes and that the suppression of GLUT6 gene expression could inhibit glycolysis and the survival of EC cells, underlying the crucial role of energy metabolism in tumor progression [
16]. In addition, our results further revealed the exact dysregulated metabolic genes of these disordered metabolism-related pathways, which may provide a new perspective on the molecular mechanisms of metabolism alterations in tumor progression.
Metabolic prognostic risk signatures that combined the expression of multiple metabolism-related genes have been indicated to serve as powerful prognostic indicators in various malignant diseases, such as glioma, liver cancer, ovarian cancer, and papillary thyroid carcinoma. Zhou et al. identified a 29-energy metabolism-related gene signature, containing branched-chain amino acid transaminase 1 (BCAT1), interleukin-4 and carbohydrate sulfotransferases, to evaluate the prognosis of diffuse glioma [
17]. Wang et al. enrolled 6 risks and 2 protective metabolic genes into the prognostic metabolic model which effectively predicted ovarian cancer patients’ prognosis [
18]. Likewise, Ma et al. developed a metabolic gene signature as a biomarker for dedifferentiated thyroid cancer [
19], and Liu et al. built a four-metabolic gene signature for liver cancer patient outcome prediction [
20].
In the present study, we performed LASSO and multivariate Cox regression analyses and identified a nine-gene signature including CYP4F3, CEL, GPAT3, LYPLA2, HNMT, PHGDH, CKM, UCK2, and ACACB. Among them, HNMT was considered a protective factor while others were risk factors. The diagnostic and predictive effectiveness of these prognostic genes has already been reported in other studies. Cui et al. reported a significantly higher expression of carboxyl ester lipase (CEL) in breast cancer. The combination of CEL and other biomarkers could improve the diagnostic capability for breast cancer [
21]. Likewise, Richard et al. found that over 70% of estrogen receptor (ER)-negative breast cancers exhibited elevated phosphoglycerate dehydrogenase (PHGDH) protein expression, which is crucial for promoting serine pathway flux [
22]. Li reported downregulation of PHGDH caused by overexpressing LncRNA PlncRNA-1 mediated cell apoptosis rate in breast cancer [
23]. In addition, Zhang et al. discovered that PHGDH could define a metabolic subtype in lung adenocarcinomas with unique metabolic dependencies [
24]. In pancreatic ductal adenocarcinoma, CYP4F3, one isoform of the cytochrome P450 (CYP) superfamily, was shown to be upregulated in tumor tissues and could serve as a distinguishing marker [
25]. Uridine-cytidine kinase 2(UCK2) was positively correlated with early recurrence and poor prognosis in hepatocellular carcinoma. Overexpression of UCK2 increased MMP2/9 expression and further activated Stat3 signaling, mediating the metastasis of hepatocellular carcinoma cells [
26]. For ACACB, Lally et al. showed that humans with fatal HCC subtypes have increased acetyl-CoA carboxylase (ACC) expression and that the genetic activation of ACC promoted the formation of hepatic de novo lipids and induced subsequent liver carcinogenesis [
27].
Here, through bioinformatic analysis and outside validation, we innovatively reported that these metabolic genes are closely related to the prognosis of EC patients. In addition, the metabolic risk signature combining these genes could accurately categorize EC patients into high- or low-risk subgroups which represented patients’ long-term outcomes. Last, our study was the first to build a comprehensive nomogram that incorporated a metabolism-related signature with clinical features including age, stage, tumor grade and lymph node status to effectively predict the survival of EC patients. This prognostic scoring system could provide a precise method to help both physicians and patients perform individualized survival evaluations and select treatment options.
Conclusion
In conclusion, we identified 47 prognosis-related dysregulated metabolic genes in EC. The prognostic DE-MRGs were highly associated with amino acid, glycolysis, and glycerophospholipid metabolism. The top 15 hub genes in the PPI network were also identified and analyzed. We performed LASSO and multivariate Cox regression analyses to establish and validate a robust prognostic risk signature enrolling the nine dysregulated MRGs. In addition, a comprehensive nomogram that combined clinical characteristics and the risk model was constructed, and its efficacy in predicting EC patients’ prognosis was also demonstrated. The 9-MRG model and nomogram may guide the selection of rational therapeutic strategies for doctors in clinical practice.
Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.