- Academic Editor
Background: Ovarian cancer (OC) is one of the most lethal gynecological
malignant neoplasms. The aim of this study was to use high-throughput sequencing
data to investigate the molecular and clinical characteristics of OC subtypes
related to lipid metabolism and glycolysis, thus providing a theoretical basis
for clinical decision-making. Methods: Molecular data and
clinicopathological characteristics of OC patients were extracted from the Cancer
Genome Atlas (TCGA), Genotype-Tissue Expression Project (GTEx), and the Gene
Expression Omnibus (GEO). Following analysis of genes involved in lipid
metabolism and glycolysis, OC was classified into subtypes by unsupervised
clustering. The molecular features and clinical outcomes of these subtypes were
then evaluated. Results: OC patients were divided into five subtypes
based on the analysis of nine genes of interest. Amongst these, patients in
subtype D had longer overall survival and more benign clinical features. Subtypes
B and E had shorter overall- and progression-free survival, respectively. Both
the B and E subtypes were closely related to lipid metabolism and to the
glycolytic process. Subtype D was positively correlated with the infiltration of
CD8
Ovarian cancer (OC) is one of the most common malignant neoplasms in women globally. Because of the insidious onset of OC and its rapid progress, most patients are diagnosed at an advanced stage [1]. The altered metabolism of OC and other cells within the tumor microenvironment is a critical factor that drives OC progression [2]. Recent genomic analysis has revealed that remodeling of metabolic pathways may play an important role in several tumor types [3, 4]. The classification of OC into different subtypes with distinct metabolic characteristics may therefore help with tumor diagnosis and with the prediction of patient outcomes.
Other recent work has revealed that cancer cells have unusual lipid metabolism and activation of related pathways [5]. Lipids generally regarded as being associated with cancer development and resistance to chemotherapy include fatty acids, glycerolipids, glycerophospholipids, sphingolipids, and sterol lipids [6]. Differences in lipid metabolism between benign and cancer tissues have long been considered to represent possible targets for cancer therapy [7, 8]. An association has also been reported between a high-fat diet, which can alter lipid metabolism, and the development of prostate cancer [9]. Moreover, exosomes originating from colorectal cancer cells can promote pre-metastatic niche formation and liver metastasis via aberrant lipid metabolism in cancer-associated fibroblasts [10].
Molecular subtypes that are based on lipid-metabolism-related signatures and have significant clinical value have been reported in several cancer types, including bladder, gastric, lung, and colon [11, 12, 13, 14]. Although associations between lipid metabolism and many different tumor types have been reported, the overall influence of lipid metabolism on OC development remains poorly understood.
Similar to lipogenesis, glycolysis is often aberrantly activated in cancer [15]. This supplies cancer cells with abundant energy while also suppressing oxidative stress by avoiding the electron transport chain responsible for the generation of reactive oxygen species [16, 17]. Moreover, it has been shown that lactate, which is the final product of glycolysis, mediates the reprogramming of immune cells. This helps to establish disease-specific conditions via post-translational histone lactylation [18]. Elevated glycolysis is a prominent feature of OC, and the modulation of glucose metabolism has been reported to increase drug resistance [19].
Glucose is the direct source of lipid synthesis in most tumor cells. Glucose-derived acetyl-CoA is converted to citrate via the tricarboxylic acid cycle, which is then exported by mitochondria to the cytoplasm, where it is involved in lipid synthesis [20]. In prostate cancer, androgen can promote the utilization of glucose for de novo lipid synthesis by upregulating hexokinase 2 (HK2) and 6-phosphofructo-2-kinase/fructose-2,6-biphosphatase 2 (PFKFB2), thus demonstrating the relationship between glycolysis and lipid synthesis [21]. However, little is known about the crosstalk between lipid metabolism and glycolysis that can drive the aggressive features of OC, such as motility, invasiveness, and tumor-initiating capacity. It seems important, therefore, to identify potential biomarkers and OC subtypes related to lipid metabolism and glycolysis.
The aim of the present study was to comprehensively investigate the metabolic signatures in OC that are associated with altered metabolic transcriptional profiles. To achieve this, we analyzed genomic data from The University Of Cingifornia Santa Cruz (UCSC) Xena and Gene Expression Omnibus (GEO). Three metabolic subtypes of OC were identified based on nine signatures associated with lipid metabolism or glycolysis. In addition, we found some unique clinicopathological features associated with two other subtypes, although patient survival was not significantly different. The clinical features of patients with distinct metabolic features revealed the occurrence of tumor-specific molecular events within these subtypes. Our results led to the construction of a clinically useful OC classification scheme that could help to further clarify the relationship between lipid metabolism and glycolysis, as well as guide the design of targeted therapy for OC.
RNA-Seq data and clinical information from OC patients were derived from the Cancer Genome Atlas (TCGA)-TARGET- Genotype-Tissue Expression Project (GTEx) cohort in the UCSC Xena database. GEO data was downloaded from GEO Series (GSE)18520, GSE18521, GSE27651, GSE26193, GSE14764, GSE26712, GSE32062, GSE63885, and GSE26942. Lipid metabolism and glycolysis-related gene sets were obtained from the Molecular Signatures Database (MSigDB v7.0) and the Kyoto Encyclopedia of Genes and Genomes (KEGG).
RNA-Seq data from the UCSC Xena and GEO cohorts were processed as follows: (1) samples without full clinical information were excluded; (2) the Ensemble or probe IDs were converted to Gene Symbol; (3) the mean value was recorded if there were multiple Gene Symbol expressions; and (4) Open source softwares including Linear Models for Microarray Data (limma v3.56.2, Victoria, Australia in Bioconductor v3.17, Heidelberg, Germany) and Surrogate Variable Analysis (sva v3.48.0, Baltimore, MD, USA in Bioconductor v3.17) operating in R 4.3.0 (Vienna, Austria) were used to remove batch effects and to normalize the data.
The R package “ConsensusClusterPlus” v1.64.0 (Chapel Hill, NC, USA) was used to identify different subtypes based on lipid metabolism and glycolysis-related genes. Metabolism subtypes were obtained using the following parameters: reps = 50, pItem = 0.8, pFeature = 1, and distance = pearson. After performing unsupervised hierarchical clustering with the same parameters according to the expression of critical genes obtained from lipid metabolism and glycolysis pathways, 5 molecular-based subtypes were obtained.
The functions of critical genes associated with lipid metabolism and glycolysis
were investigated using the online tool DAVID (Database for Annotation,
Visualization and Integrated Discovery). This was used to annotate signatures
with a potential role in the development of OC based on Gene Ontology (GO) terms.
Terms with p values
In addition, gene sets associated with lipid metabolism were downloaded from the MSigDB v7.0, while genes involved in glycolysis were downloaded from KEGG. Each gene set was comprehensively analyzed using the gene set variation analysis (GSVA v1.48.1, Catalonia, Spain) algorithm, with an evaluation of the specific variation in biological processes between subtypes. In order to visualize the differences in pathways between different subtypes, heat maps were constructed using the “pheatmap” R package.
The ESTIMATE (Estimation of Stromal and Immune cells in Malignant Tumor tissues using Expression data) method was used to calculate the stromal score, immune score, and tumor purity. The CIBERSORT (Cell-type Identification By Estimating Relative Subsets Of RNA Transcripts) algorithm was then used to analyze the RNA-Seq data of OC patients in order to determine the relative proportions of 22 types of infiltrating immune cells.
The critical genes identified from both the TCGA and GEO databases were selected for least absolute shrinkage and selection operator (LASSO) regression analysis via the “glmnet” package. The selected genes were used to calculate the risk score by adding the gene expression multiplied by the corresponding coefficient.
Following the analysis of data from TCGA-TARGET-GTEx and GEO, 71 lipid
metabolism-related genes were found to be significantly upregulated in OC (Fig. 1A, Supplementary Tables 1–5). GO results revealed that in addition to
lipid metabolism, these genes were also involved in ion transport, cell response
to peroxides, and adenosine triphosphate (ATP) binding. This piqued our interest
since they are also involved in glycolysis (Fig. 1B, Supplementary Table
6). Therefore, we obtained 25 key components of the glycolysis pathway derived
from KEGG and performed subsequent analysis (Supplementary Table 7).
After building a protein-protein interaction (PPI) network via the STRING
database, we found interactions between four lipid metabolism-related genes
(Mini-chromosome Maintenance Complex Component 2 (MCM2), nucleolar and
spindle associated protein (NUSAP1), Isocitrate Dehydrogenase
(NADP
Distinct gene expression signatures associated with lipid
metabolism and glycolysis in ovarian cancer (OC). (A) The expression of 11,436
genes involved in lipid metabolism and obtained with Molecular Signatures Database (MSigDB v7.0) using three datasets
(the Cancer Genome Atlas (TCGA)-TARGET- Genotype-Tissue Expression Project (GTEx), GSE18520, and GSE27651) are displayed as a Venn diagram. Each
dataset is represented by distinct colors. A total of 71 genes were upregulated
in all three datasets (
We next investigated the clinical features of patients with OC subtypes
classified by expression signatures for lipid metabolism and glycolysis genes, as
well as the enrichment scores for metabolism-related pathways. The RNA sequencing
data of 1164 OC patients was extracted from TCGA (n = 378) and GEO datasets,
including GSE14764 (n = 80), GSE18521 (n = 53), GSE26193 (n = 107), GSE26712 (n =
185), GSE32062 (n = 260) and GSE63885 (n = 101). Unsupervised cluster analysis
classified the OC patients into five subtypes (maxK = 5) (Fig. 2A). After
deleting samples missing patient survival data, 1138 OC samples were available
for analysis. Subtype D patients had longer overall survival time compared to the
other four subtypes (66.56 months, Mantel-Cox p value
Clinical features of different OC subtypes identified by lipid
metabolism and glycolysis. (A) The ConsensusClusterPlus package was used to
perform unsupervised hierarchical clustering for 1164 OC samples. Five OC
subtypes were identified using this method. (B) Results from Kaplan–Meier
analysis revealed marked differences in overall- and progression-free patient
survival for the 5 metabolic subtypes of OC. The log-rank test was used to
statistically evaluate the survival differences. (C–J) Fisher’s exact test was
performed to evaluate differences in the following clinical features between
patients from subtypes B, D, and E: (C) age, (D) Gynecology and Obstetrics (FIGO) Stage, (E) grade, (F)
histology, (G) Breast Cancer Susceptibility Protein (BRCA) status, (H) Tumor Protein P53 (TP53) status, (I) platinum sensitivity, (J)
clinical status. The proportions of each clinical feature in the different OC
subtypes are shown (*p
We next analyzed the clinicopathological features of the five OC subtypes; the clinical information of patients was shown in Supplementary Table 9. Subtype D patients showed significant differences in age, The International Federation of Gynecology and Obstetrics (FIGO) stage, The World Health Organization (WHO) classification, Breast Cancer Susceptibility Protein 1 (BRCA1) mutation, platinum-based chemotherapy sensitivity, and clinical response compared to the other subtypes (Fig. 2C–E,G,I,J). The histological composition and Tumor Protein P53 (TP53) mutation status of subtype B patients were also significantly different (Fig. 2F,H). Overall, these results suggest that lipid metabolism and glycolysis processes may differ between OC patients and are significantly correlated with clinical features and outcomes. It is, therefore, important to be able to distinguish between the five subtypes.
The five subtypes described above showed different expression levels for lipid metabolism and glycolysis-related genes. The expression levels for the nine aforementioned genes were also significantly different between the five subgroups. We further analyzed the possible molecular mechanisms relating to these subtypes based on their different clinical features. The GSVA score of the metabolism-related gene set in subtype B was significantly higher than that of subtype D. Interestingly, in addition to a high metabolism-related gene set score, subtype E was significantly enriched for gene sets or pathways related to DNA replication and the cell cycle (Fig. 3). This suggests that subtype B and E patients may have a poor prognosis due to disruptions in their metabolic processes and that lipid metabolism and glycolytic metabolism may be necessary for OC cell division.
Significantly enriched gene sets in OC subtypes classified according to critical metabolic signatures. (A) and (B) Gene Set Variation Analysis (GSVA) results for OC samples in the TCGA and GEO cohorts. The heat map shows the normalized enrichment scores for subtypes B and D. (C,D) GSVA results for OC samples in the TCGA and GEO cohorts. The heat map shows the normalized enrichment scores for subtypes E and D.
Immunotherapy is safe and effective, especially for OC patients who do not
respond well to chemotherapy and poly ADP-ribose polymerase inhibitor (PARPi)
therapy. We, therefore, compared the immune microenvironment between different OC
metabolic subtypes in order to evaluate the correlation between lipid metabolism
and glycolysis processes. Not surprisingly, subtype D showed a higher
infiltration with CD8
The immune microenvironment in different OC subtypes. (A)
Cell-type Identification By Estimating Relative Subsets Of RNA Transcripts (CIBERSORT) scores for immune cell infiltration in OC subtypes are displayed as
heat maps. (B) Infiltration scores for three immune cell types in three OC
metabolic subtypes. **p
We next constructed risk models for three OC subtypes (B, D, and E) based on the
expression of nine critical genes and the survival outcome of patients. OC
samples were divided in a ratio of 7:3 into training and test sets, respectively.
The prognostic value of the nine genes in the different subtypes was analyzed in
the training set using LASSO-Cox analysis. Variables with Cox p
Subtype B: Risk score = MCM2
Subtype D: Risk score = MCM2
Subtype E: Risk score = MCM2
Construction of multigene risk models using LASSO-Cox regression analysis. (A) The trajectory of each independent variable, where the horizontal axis represents the log-value of the independent variable lambda and the vertical axis represents the coefficient of the independent variable. (B) The confidence interval under each lambda.
The TCGA and GEO databases were used in the present study to identify five metabolic subtypes of OC according to lipid metabolism and glycolysis. These were then linked to underlying gene expression patterns that play critical roles in tumor biology, clinical outcome, and the tumor immune microenvironment.
Although the specific mechanisms remain unclear, the abnormal energy metabolism of tumor cells is related to their aberrant proliferation, invasion, and metastasis [22]. Furthermore, the reprogramming of lipid metabolism and glycolysis affects the normal response to tumors, as well as the body’s sensitivity to chemotherapeutic drugs [23, 24, 25]. A pan-cancer study found that gene expression profiles associated with metabolic pathways can indicate whether important metabolites in the body are altered [26]. The study of closely related molecular subtypes and associated clinical characteristics of OC patients can shed light on the metabolic differences in OC and lead to a better understanding of patient outcomes. Moreover, the development of a risk prediction model based on metabolic features should provide a novel approach to clinical diagnosis and treatment.
We used bioinformatic methods in the current study to identify five OC subtypes with distinct metabolic features. Subtype D had an inactive profile for lipid metabolism and glycolysis-relevant pathways and better patient prognosis than subtypes B and E. Other analyses revealed that subtype D also displayed a high level of immune cell infiltration, which is known to be associated with immune activation [27, 28, 29]. Subtype D is also correlated with TP53 mutation, which is indicative of stromal invasion and mesenchymal activation [30]. Overall, subtype D OC, therefore, exhibits an ideal metabolic phenotype, and targeting metabolic pathways might potentially reverse the poor clinical status of patients in subgroups B and E.
Although we focused on three subtypes in the present study, further investigation of the remaining two subtypes could prove worthwhile. Examination of the clinical characteristics for the five OC subtypes revealed a higher proportion of non-serous carcinoma in subtype A patients. The evaluation of treatment response in this study was based on platinum-based therapy. Therefore, our results suggest that subtype A could be more sensitive to treatment with other drugs, although more in-depth research is required. Subtype C showed a lower percentage of TP53 mutations, which could result in fewer mutations and, therefore, fewer changes in cancer-associated antigens, thereby explaining the small number of immune cell infiltrates. Subtypes A and C can be further classified in more detail in future studies.
The present study also developed a risk model for predicting patient outcomes. The nine metabolism-related genes used to identify the B, D, and E subtypes showed good predictive value. Thus, lipid metabolism and glycolysis are key factors in the body’s resistance to cancer.
Previous studies have also implicated the nine metabolic genes in tumorigenesis.
Mini-chromosome Maintenance Complex Component 2 (MCM2) was first implicated in
chromosome initiation in eukaryotic cells, and the inhibition of MCM2
promotes the sensitivity of OC to carboplatin [31]. Hiramatsu K et al. [32]
reported that knockdown of MCM2 via siRNA interference significantly decreased
the proliferation rate of ovarian cancer cell line. Nucleolar and spindle
associated protein (NUSAP1) is an important chromosome-chromosome interaction
protein that also has an important role in the OC cell cycle [33]. Moreover, the
tumor-promoting effects of NUSAP1 in gastric cancer are mediated mainly through
Yes-associated protein 1 (YAP1), with the aberrant expression of NUSAP1 and YAP1
being highly correlated in gastric cancer cells and tissues [34]. Isocitrate
Dehydrogenase (NADP
There are several shortcomings to this study. First, public gene chip and RNA-seq data were used to screen 1164 cases, with the results possibly influenced by the platform used. Moreover, this analysis was carried out on retrospective data, and more prospective trials and tests are therefore needed. In addition, the results of this project are based on publicly available data obtained from OC tissue samples and are discussed in relation to the relevant molecular mechanisms and possible impact on cancer tissues. Furthermore, the roles of several metabolic factors closely related to the molecular mechanisms of OC development were studied.
Overall, this study describes several different OC subtypes from the perspective of lipid metabolism and glycolysis, thereby providing novel information on the pathogenesis and clinical classification of OC. These results provide a theoretical foundation for the prevention and treatment of OC and for further drug research and development in this area.
The datasets used or analysed during this study are available from the corresponding author on reasonable request.
All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by XW, WX, ML, WL, LC, RW, HY and DZ. The first draft of the manuscript was written by XW and all authors commented on previous versions of the manuscript. All authors contributed to editorial changes in the manuscript. All authors read and approved the final manuscript. All authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.
Not applicable.
The authors are grateful to Li Sun for her help with the preparation of this paper.
This work was supported by the Shandong Province Medicine and Health Science and Technology Development Program [grant number 202105010436]; The Development Foundation of the Second Hospital of Shandong University [grant number 2022YP03]; The Qingmiao Foundation of Shandong Cancer Hospital and Institute [grant number CH-SFMU-QM20210004] and Natural Science Foundation of Shandong Province [grant number ZR2022QH179].
The authors declare no conflict of interest.
Publisher’s Note: IMR Press stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.