Energy metabolism reprogramming represents a hallmark of cancer that not only facilitates the growth, invasiveness, and metastasis of malignant cells[33] but also promotes cancer development and progression through complicated interactions with a tumour ecosystem[34]. The heterogeneity of tumours generates complicated metabolic patterns[35]. To further study the metabolic alterations in OV, a pancancer analysis was performed on 14 tumour types from a more comprehensive perspective based on the MPI networks and genes identified in a previous study[21], revealing the distinct metabolic pattern of OV compared with the other 13 cancer types. Moreover, OV-specific MIPros were screened to establish the PCA subtypes. Interestingly, despite the C1 type displaying an adverse survival compared with C2, more malignant biological behaviours were detected in the C2 group, such as younger age and higher HRD score, indicating that the survival difference between the PCA subtypes is caused by metabolic discrepancy to some degree.
To further establish a robust and accurate MPIscore model, the intersected DEGs and genes in the PCA subtypes and the key module genes of WGCNA associated with PCA phenotypes were analysed via lasso-logistic regression methods. After comprehensive consideration, 5 genes were filtered to construct the MPIscore model. Specifically, C1QTNF3 is a novel adipokine that regulates hepatic glucose output[36] and was found to be upregulated in bowel metastasis samples compared with primary samples in OV[37]; SFRP2 is correlated with recurrence and overall survival outcomes [38] and is upregulated in stem cells of OV[39]; FZD1 is an independent prognosticator in OV[40]; CILP2 plays an important role in regulating hepatic glucose production[41], and its overexpression correlates with tumour progression and poor prognosis in colorectal cancer patients[42], and high expression of MFAP4 predicts primary platinum-based chemoresistance and is associated with adverse prognosis in serous ovarian cancer patients[43]. Subsequently, the performance and effectiveness of the MPIscore in survival prediction were tested and confirmed by the TCGA-OV and GEO validation datasets, suggesting that a high MPIscore acted as an independent prognostic factor and is associated with poor prognosis. Additionally, we found that oxidative phosphorylation, TCA cycle, steroid biosynthesis, and pyrimidine metabolism pathways were enriched in the low-MPIscore group, whereas glycosaminoglycan biosynthesis, chondroitin sulfate, glycolysis, gluconeogenesis, fatty acid metabolism, and arachidonic acid metabolism processes were enriched in the high-MPIscore group. Although aerobic glycolysis, the Warburg effect, has been thought to be the dominant energy metabolism in cancer, oxidative phosphorylation (OXPHOS) is the more likely form of energy metabolism in some cancer cells[44]. Here, these two biological courses were significantly enriched in the two groups, separately exhibiting the metabolic differences depending on MPIscore grouping. Genomic mutation analysis uncovered the difference between the MPIscore groups; for example, ROBO3 mutations are more likely in the high-MPIscore group. In our study, gene mutations related to the prognosis of OV were also observed. The comutation analysis revealed that some genes, such as FLG2 and MDN1 and RYR2 and TTN, were comutated, whereas TP53 and DST mutations were mutually exclusive.
Solid tumours are characterized by a unique microenvironment formed by malignant and several nonmalignant cells that can modify tumour characteristics[45]. We subsequently analysed the heterogeneity of tumour-infiltrating immune cells in OV. Accordingly, we speculated that several immune checkpoints showing higher expression in the high-MPIscore subtype, such as TIGIT, CTLA4, CCR4, CD27 and HAVCR2, might serve as high-MPIscore subtype-specific therapeutic targets. CIBERSORT was used to estimate tumour-infiltrating immune cell subsets, revealing that CD8 + T cells, resting memory CD4 + T cells, M2 macrophages, dendritic cells, and neutrophils were mainly distributed in the high-MPIscore group. The TIMER2.0 dataset was assessed using Timer, Xcell and EPIC to further depict the immune landscape of MPIscore subtypes, indicating that the high-MPIscore group had more immune cell infiltration, especially T-cell infiltration. Although T-cell infiltration has shown positive prognostic impacts in OV[46], T-cell infiltration was linked to a high MPIscore with an adverse prognosis in this study, which may, in part, be due to the dysfunction and exclusion of immunity that can be inferred by the remarkable positive correlation between the TIDE score and MPI score. Meanwhile, neutrophils, Treg cell infiltration and an increased density of M2-like TAMs were negatively correlated with OS outcomes in OV patients and associated with high mortality and decreased survival[47, 48]. More importantly, emerging evidence indicates that cancer cells can suppress the antitumour immune response by competing for and depleting essential nutrients or otherwise reducing the metabolic fitness of tumour-infiltrating immune cells[49, 50], and the MPIscore groups demonstrated different metabolic pathways. Furthermore, by prediction, based on the previously reported cohort, we discovered that the higher the MPIscore was, the lower the response to immunotherapy was, which was consistent with the above results; that is, the high-MPIscore group had immune dysfunction and worse survival outcomes.
Although our study provides new insights into OV metabolism and classification, there are limitations to this study. First, our analysis was based on the TCGA training set and three GEO validation datasets. In addition, this is a retrospective study, and prospective, large-scale trials are warranted to verify the clinical application of our findings. Last, although bioinformatic analysis is a powerful tool to study and stratify OV for precision medicine, further validation is needed.