Establishment of insulin resistance-related ten-gene signature in endometrial cancer and identification of ACTL8 as a prognostic and immunological biomarker

Abstract

Background

Endometrial cancer (EC) is a common gynecological tumor. Insulin resistance (IR) increases the risk of EC. However, the common molecular basis between the two remains unclear. This study aims to screen the common differential expression genes (DEGs) between the two diseases and construct a prognostic risk model.

Methods

We obtained gene expression profiles and clinical information of patients with IR and EC from GEO and TCGA datasets. We performed differential analysis to discover the shared DEGs between IR and EC. Subsequently, the interactions among overlapping DEGs, along with their biological functions and genetic mutations in EC, were comprehensively analyzed via protein–protein interaction (PPI) network, function enrichment analyses, and genetic mutation analyses. Then, machine-learning algorithms were employed to figure out genes significantly associated with survival. For clinical application, we constructed a prognostic risk model and also compared tumor-infiltrating immune cells (TIICs) and genetic mutation between high- and low-risk groups. Finally, we screened one of the most important markers in the prognostic signature to investigate its expression-prognosis pattern, biological function, and underlying mechanism.

Results

Our analysis identified 20 co-upregulated genes and 32 co-downregulated genes of IR and EC. In addition, the two subnetworks and the top 20 top genes were obtained through PPI analysis, while the construction of extracellular matrix and immune response were the most enriched functions of DEGs. Filtered by random forest, gradient boosting machine, and extreme gradient boosting, six upregulated markers (ACTL8, WNT7A, CTSV, MMP9, CNIH2, and PLAUR) and four downregulated markers (COL6A6, MYOC, PHLDB1, and FIBIN), were defined as the characteristic genes for the prognosis of EC patients. The risk prediction model constructed by these ten genes had good predictive value in prognosis of EC patients and was related to immune regulation and genetic mutation. ACTL8 was further studied as the most significant marker among 10-gene signature. The correlation between the upregulation of ACTL8 and the poor prognosis of EC patients suggested its carcinogenic effect, which was correlated to its regulation of cilium movement.

Conclusion

Our findings suggest that there are common molecular profiles between IR and EC. IR-related prognostic model represents an excellent prognosis predictor and immune-related biomarker, which can be applied to risk stratification and precise treatment of EC patients with IR.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12094-026-04230-x.

Introduction

Endometrial cancer (EC) is the most common gynecological tumor, ranking fourth in incidence and fifth in mortality among the female population in the United States [1]. In 2022, 77,700 new EC cases was reported in China, which made up more than 20% of the new female cancer cases in the country [2]. On a global scale, the epidemiological data of EC in 2020 were striking, with 417,367 new diagnoses and 97,370 new deaths registered across the world [3]. Its lifetime incidence rate is approximately 3%, and the median age of diagnosis is 64 years [4]. Traditionally, EC is considered to have a good prognosis, with an overall 5-year survival rate of 81% [5]. However, this might be due to the fact that EC often occurs in postmenopausal women. Abnormal postmenopausal bleeding gives patients the opportunity for early diagnosis, and the survival rate of early patients is as high as 80–90% [5]. Unfortunately, the prognosis of patients in stage III and stage IV is still not optimistic, with 5-year survival rates are 50–65% and 15–17%, respectively [5]. Furthermore, the incidence of EC and its related mortality rates are expected to rise generally worldwide. According to data from the International Agency for Research on Cancer (IARC), it is projected that by 2045, the number of new cases of endometrial cancer worldwide will increase by 65% compared to the record in 2022 [6]. The escalating patient count and the dismal prognosis for those in the middle and advanced stages of EC catapulted it to the forefront of critical concerns in the global medical arena, forcing the medical community to pursue biomarkers for early diagnosis and prognosis prediction to allow patients to receive early onset and precisely calibrated treatment without delay.

Insulin resistance (IR) is positioned as the weakened response of peripheral target tissues to insulin and the reduced ability to inhibit hepatic gluconeogenesis. The result is with the consequence of a compensatory increase in circulating insulin levels to maintain normal blood glucose and the insidious onset of type 2 diabetes mellitus (T2DM) [7]. Insulin resistance (IR) is not only intricately associated with a wide spectrum of metabolic disorders, including obesity [8], cardiovascular disease [9], metabolic syndrome (MetS), and non-alcoholic fatty liver disease (NAFLD) [10], but also potentially serves as a key mechanism involved in the occurrence and development of multiple tumors. Hyperinsulinemia accompanied by IR leads to an increased risk of colorectal cancer [11], breast cancer [12], liver cancer [13], and gynecological tumors [14, 15]. The upregulation of the expression of Insulin-like Growth Factor 1 (IGF-1) and its receptor, the IGF-1 Receptor (IGF-1R), was firmly established across a wide spectrum of malignant tumors, and notably associated with an unfavorable prognosis for patients with bladder urothelial carcinoma (BLCA), cholangiocarcinoma (CHOL), lung adenocarcinoma (LUAD), and hepatocellular carcinoma (LIHC) [16]. IR was a prevalent condition among patients with EC, and its significant association with tumor progression and staging has already been well documented [17]. However, despite this recognition, the pivotal molecular players and the underlying mechanisms through which IR drives the advancement of EC and contributes to the unfavorable prognosis of patients remain largely elusive and demand more in-depth exploration.

In this study, comprehensive multi-scale bioinformatics analysis was used to find common key factors and prognostic markers of EC and IR. Differentially expressed genes in TCGA-UCEC and GSE20950 datasets for EC and IR were obtained, and their functions were analyzed via protein–protein interaction (PPI) network and gene enrichment analysis. Leveraging the power of weighted gene co-expression network analysis (WGCNA), we were able to discern the genes most closely correlated with the onset of IR. Multiple machine-learning algorithms were used to get relevant EC prognosis markers and construct a prognostic risk model. Additionally, we explored the disparities in gene mutations and the infiltration patterns of immune cells among patients belonging to different risk groups within this model, shedding light on the complex interplay between genetics, immunity, and disease progression. Finally, we selected a gene that has both prognostic predictive value and functional regulatory role in EC for further analysis. Our work identified co-regulatory genes with EC and IR, which paved the way for revealing the potential mechanism by which IR promoted the progression of EC.

Materials and methods

Data acquisition and processing

Single-nucleotide variant (SNV) data of 519 tumor samples and copy-number variations (CNVs) data of 558 tumor samples were acquired from The cancer genome atlas (TCGA, https://tcga-data.nci.nih.gov/tcga/). RNA sequencing and related clinical data of 35 normal and 549 tumor samples for TCGA uterine corpus endometrial carcinoma (UCEC) were downloaded from UCSC Xena (https://xenabrowser.net/datapages/). For IR, the gene expression profiling by array dataset GSE20950, comprising 39 subcutaneous and omental adipose tissue samples from insulin-sensitive and insulin-resistant patients undergoing gastric bypass surgery, was obtained from the Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo/). We performed normalization analysis on the TCGA-UCEC and GSE20950 datasets using the “limma” R package, aiming to eliminate technical variations between different platforms. We set the adjusted P < 0.05 and |log2FC|≥ 0.585 as the criteria for screening differentially expressed genes (DEGs). The DEGs were classified as upregulated or downregulated according to the positive or negative values of their log2FC. To better visualize these DEGs, “heatmap” and “ggplot2” R packages were utilized to create heatmaps and volcano plots, respectively. The overlapping DEGs of the two datasets were visualized using Venn diagrams.

Protein–protein interaction (PPI) network

The Search Tool for the Retrieval of Interacting Genes (STRING version 12.0, https://cn.string-db.org/) is an online tool for constructing analyses of direct and indirect PPI networks. STRING was used to evaluate the potential PPI relationships among the common DEGs with a minimum required interaction score of 0.4. The relationship network was imported into Cytoscape software version 3.8.0 for visualization. The Molecular Complex Detection (MCODE) and CytoHubba plugins were used for further module analysis and selection. MCODE is used to identify important modules in the PPI network and select potential functional modules, while CytoHubba is widely used to identify the most important genes in different biological networks. The top 20 Hub genes were chosen by CytoHubba app with degree method.

Machine learning

Three machine-learning models, Random Forest (RF), Gradient Boosting Machine (GBM), and extreme gradient boosting (XGBoost), were applied to rank feature genes by importance to accurately predict the prognosis of UCEC patients. The RF model, a supervised machine-learning algorithm based on classification trees, improves the prediction accuracy of the model by aggregating a large number of classification trees. We performed the RF algorithm using the “randomForest” package (ntree = 1000). GBM is a boosting-based learning algorithm that combines a sequence of base learners. Each base learner focuses on the error (residuals) of the previous learner, and this process is repeated until the error is less than a predefined threshold. A final prediction is made based on the combination of the response of all learners. XGboost algorithm, performed via “xgboost” packages, combines hundreds of tree models with low classification accuracy into a predictive model with high accuracy. We intersected the prognostic signature genes identified by the three machine-learning models and used the “car” R package to calculate the variance inflation factor (VIF) for genes to exclude collinearity. Finally, we constructed a risk score prediction model visualized based on the characteristic gens. The formula for the risk score prediction model is: Risk score = ∑(EXPi × COEFi) (EXP: expression level of the gene; COEF: regression coefficient obtained from COX analysis; n: number of intersecting genes).

Functional and pathway enrichment analyses

The “clusterProfiler” R package was utilized to conduct Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses with P value < 0.05 as a statistically significant difference. These analyses were intended to delve into the biological functions and potential pathways linked to the overlapping differentially expressed genes (DEGs) obtained from the TCGA-UCEC project and the GSE20950 dataset, along with the hub genes and biomarker-related DEGs. The results of these analyses were respectively visualized through bubble plots, chord plots, and circle plots. Leveraging the “h.all.v2024.1.Hs.symbols.gmt” from the Molecular Signatures Database (MSigDB), Gene Set Variation Analysis (GSVA) was conducted with the “GSVA” R package. Moreover, tumor samples in TCGA-UCEC were categorized into high and low groups based on biomarker’s median expression level. GSEA was executed using default weighted enrichment statistics, ranking genes through the Pearson method. Gene sets significantly enriched in the high group were identified as positively associated gene sets, whereas those significantly enriched in the low group were recognized as negatively correlated gene sets. A gene set with NES > 1 and FDR < 0.05 was deemed statistically significant.

Mutation and CNV of DEGs in UCEC

Based on the SNV data downloaded from TCGA, the R package “maftools” was used to analyze gene mutation frequency and tumor mutation burden (TMB). The “oncoplot” function was used to generate a waterfall plot of gene mutation frequencies in TCGA-UCEC samples. Frequency analyses of chromosomal amplifications, chromosomal deletions, and normal gene diploids were performed on the CNV data. The R package “ggplot2” was used to generate bar charts. The “RCircos” package was used to draw a CNV frequency circle, where red dots represented high frequency of increased genes copy number and blue dots represented high frequency of decreased genes copy number.

TMB is defined as the total number of somatic gene coding errors, base substitutions, gene insertions, or deletions detected per million bases. Microsatellite instability (MSI) refers to the instability in the length of microsatellite sequences caused by insertion or deletion mutations during DNA replication. Mismatch repair (MMR) deficiency often leads to this situation. The TMB of each sample was obtained by reading the mutation data using the “maftools” package, and the MSI scores were obtained from the clinical information of TCGA-UCEC. The Wilcoxon test was used to compare the TMB and MSI status between the high- and low-risk score groups of samples.

Survival analysis

To confirm the prognostic value of risk model and prognostic biomarker in UCEC patients, survival curves were derived from the Kaplan–Meier survival analysis using the “survival” R package. Patient samples were divided into two groups by median cutoff of model risk score and biomarker expression. Outcomes, including overall survival (OS), recurrence-free survival (RFS), disease-specific survival (DSS), and progression-free survival (PFS), were investigated.

Tumor microenvironment (TME) and immune cell infiltration

The ESTIMATE algorithm was used to estimate stromal cells and immune cells in tumor tissues, and the “ESTIMATE” package was employed to analyze the proportion of immune-stromal components in UCEC TME. The stromal score, immune score, and ESTIMATE score of each sample in the TCGA-UCEC project were calculated to show the proportion of each component in the TME. The CIBERSORT algorithm was used for the immune infiltration scoring. The TME scores and the proportions of immune cell types were compared between the high- and low-risk groups of prognostic model, together with the high- and low-expression groups of characteristic genes. Spearman correlation analysis was used to detect the correlation between immune cells and the expression of characteristic genes, and the results were validated by XCELL, TIMER, QUANTISEQ, MCPCOUNT, EPIC, and CIBERSORT-ABS methods. Correlation analyses were performed between feature genes and immune checkpoint genes in TCGA-UCEC dataset.

Results

Genetic landscape and biological function of DEGs

According to the criteria of P < 0.05 and |log2FC|≥ 0.585, 7923 and 352 DEGs were identified in the GSE20950 (Fig. 1A and C) and the TCGA-UCEC datasets (Fig. 1B and D), respectively. Specifically, the GSE20950 dataset contained 189 upregulated genes and 163 downregulated genes, while the TCGA-UCEC dataset included 3748 upregulated genes and 4165 downregulated genes. By interacting with the DEGs of the two datasets, 20 co-upregulated genes and 32 co-downregulated genes were identified (Fig. 1E).

Fig. 1.

Identification of the overlapping DEGs between IR and EC. A, C Heatmap (A) and volcano map (C) showing the top 50 DEGs in adipose tissues of IR and insulin-sensitive patients in GSE20950 dataset, with adjusted p value < 0.05 and the absolute value of logFC > 0.585. B, D Heatmap (B) and volcano map (D) showing the top 50 DEGs between EC and normal tissues in TCGA-UCEC database, with adjusted p value < 0.05 and the absolute value of logFC > 0.585. E Venn map showing the 20 upregulated DEGs and 32 downregulated DEGs shared between IR and EC. DEG differential expression genes, IR insulin resistance, EC endometrial cancer

To further study the association between IR and UCEC, we explored the biological functions of these DEGs in UCEC. GO enrichment analysis revealed that extracellular matrix (ECM)-related biological process (Fig. 2A) and cell component (Fig. 2B) were significantly accumulated, such as ECM organization, collagen-containing ECM, and protein complex involved in cell adhesion. Endopeptidase activity was the most relevant molecular function of these DEGs (Fig. 2C). Moreover, KEGG analysis showed that these DEGs were apparently enriched in intracellular signal transduction pathways, including RIG-I-like receptor signaling pathway, IL-17 signaling pathway, Toll–like receptor signaling pathway, and TNF signaling pathway (Fig. 2D). In addition, we conducted general analysis of the somatic mutation frequency in these 52 DEGs, and the result showed a relatively high mutation frequency in tumor samples from TCGA-UCEC database (Fig. 2E). 203 (39.19%) of 518 UCEC samples had mutations in DEGs. FAT2 had the highest mutation frequency (17%), followed by COL6A6, C6, PCDHB12, RGS22, ADAMTS3, ANO5, PHLDB1, AMPH, BBS10, MMP9, ZNF781, GBP4, and NCAPH. Somatic copy-number alterations in these DEGs were calculated and CNVs were pervasive in 41 DEGs. Most genes showed relatively elevated CNV (Fig. 2F). Detailed locations of CNV alterations on chromosomes are depicted in Fig. 2G.

Fig. 2.

Biological function and mutational landscape of overlapping DEGs between IR and EC. A–C BP (A), CC (B), and MF (C) of 52 common DEGs in TCGA-UCEC program. D KEGG analysis of common DEGs in TCGA-UCEC program. E Oncoprint visualization of the mutation atlas of these shared DEGs across EC samples from TCGA cohort. F The frequency of CNV (gain/loss) status. G The position of variations on 23 chromosomes. BP biological process, CC cell component, MF molecular function, KEGG Kyoto Encyclopedia of Genes and Genomes

PPI network of DEGs

The interaction connections between DEGs were obtained from the STRING database. Two significant subnetworks were filtered from the MCODE plugin of Cytoscape software (Fig. 3A and B). The most significant cluster with the 6 nodes and 14 edges contains six genes including MAPK13, PLAUR, CXCL10, FKBP5, IRF7, and MMP9. ACTL8, AMPH, NAP1L3, CDCA8, PLP1, and CBLN1 consisted of the second most significant cluster with 6 nodes and 6 edges. Then, the top 20 genes with the highest degrees of connectivity were selected from cytoHubba plugin and defined as hub genes (Fig. 3C). Further, the 20 hub genes were analyzed by “enrichplot” package. The chord plot showed top 10 significant terms of the GO enrichment analysis. The major biological processes were “negative regulation of endopeptidase activity” and “regulation of release of cytochrome c from mitochondria”, and the major cellular component were “collagen-containing extracellular matrix” (Fig. 3D). The KEGG pathway analysis showed that these genes were mostly enriched in RIG-I-like receptor signaling pathway, IL-17 signaling pathway, and TNF signaling pathway, which were related to immune regulation.

Fig. 3.

The protein interaction network obtained by analyzing PPI with Cytoscape. A, B The two sub-modules obtained by MCODE plugin in Cytoscape software. C Top 20 hub genes obtained from cytohubba plugin in Cytoscape software. D, E GO (D) and KEGG (E) enrichment analyses of hub genes. GO gene ontology, KEGG Kyoto Encyclopedia of Genes and Genomes

Predictive model constructed via machine learning

Our aim was to search for characteristic genes related to the prognosis of UCEC patients based on the 52 overlapping DEGs between IR and UCEC, and establish a clinically applicable classifier for conveniently predicting the clinical outcomes of UCEC patients. We applied three machine-learning algorithms based on TCGA-UCEC cohort. For the RF algorithm, we set the number of iterations of the random forest classifier to 1000, so that the out-of-bag (OOB) error was stable and less than 0.4 (Fig. 4A). Nineteen genes, led by ACTL8, were identified as liver metastasis-specific genes using the RF algorithm. As described in the bar chart, the top 15 important prognostic characteristic genes, including CTSV, ACTL8, WNT7A, MMP9, and COL6A6 obtained by the GBM algorithm, were presented (Fig. 4B). For the XGboost algorithm, the histogram of the importance of feature genes suggested that CTSV ranked first, followed by ACTL8, MYOC, WNT7A, and MMP9 (Fig. 4C). Based on the results of the above three algorithms, the ten intersection genes, including six upregulated markers (ACTL8, WNT7A, CTSV, MMP9, CNIH2, and PLAUR) and four downregulated markers (COL6A6, MYOC, PHLDB1, and FIBIN), were defined as the characteristic genes for the prognosis of UCEC patients (Fig. 4D and F). Pearson correlation analysis revealed that the absolute values of the correlation coefficients between any two of these ten genes were all less than 0.4. To exclude the concern of collinearity, we used the 'car' R package to calculate the variance inflation factor (VIF) for genes with statistical significance in univariate analysis. The results showed that the VIF of all genes was less (Supplementary Table S1). We used cox regression analysis to calculate the coefficients of each gene and obtained the risk scoring formulas based on the expression of these genes: risk score = ACTL8* (0.20) + WNT7A* (0.11) + FIBIN* (0.18) + COL6A6* (0.68) + MYOC* (0.02) + PHLDB1* (0.22) + CTSV* (−0.15) + CNIH2* (−0.01) + MMP9* (−0.13) + PLAUR* (−0.11) (Fig. 4E). Subsequent univariate and multivariate Cox regression analyses identified ACTL8 (HR = 1.177, 95%CI 1.043–1.329, Logrank p = 0.0084) and FIBIN (HR = 1.254, 95%CI 1.062–1.482, Logrank p = 0.0077) as independent prognostic risk factors for patients with EC. Finally, patients with UCEC were stratified into high- and low-risk groups based on the median risk score, and Kaplan–Meier plots indicated that patients in the low-risk group had a significantly better prognosis (Logrank p < 0.001, Fig. 4K). The risk distribution of patients, survival time of patients in high- and low-risk groups, and expression profile heat map of the ten characteristic markers that formed the risk model are illustrated in Fig. 4L–N.

Fig. 4.

The ten-gene signature screened by machine-learning algorithms predicted the prognosis of EC patients. A The error rate of random survival forest and out-of-bag variable importance ranking. B, C Top 15 genes selected by GBM (B) and XGBoost (C) algorithms. D Venn map showing the ten genes overlapping three methods. E Barplot showing the coefficients of 10 genes. F Expression of ten genes in TCGA-UCEC project. G Forest plot for multivariable analysis. H The Kaplan–Meier survival curve shows the OS rate between UCEC patients with high- and low- risk. I, J Survival status (I) and distribution (J) of UCEC patients with high and low risk. K Heatmap showing expression of ten genes in high- and low-risk group. RF random forest, GBM gradient boosting machine, XGBoost extreme gradient boosting, ROC receiver-operating characteristic, DCA decision curve analysis, OS overall survival. ***, p < 0.001

Immunological and mutation analysis across risk stratification

We compared the expression correlations between prognostic characteristic genes and immune checkpoint inhibitors (ICIs) using Spearman correlation analysis. Heatmap depicted that these risk genes were significantly associated with the upregulation of most ICIs (Fig. 5A). Then, the comparison of stromal score and immune score between the high- and low-risk groups showed that the proportion of tumor cells in the samples of the high-risk group was higher (Fig. 5B). The gene expression matrix of the TCGA database in UCEC was imported into CIBERSORT algorithm to estimate the fractions of 22 immune cells. The ten genes were also shown to be highly linked to the majority of immune cells (Fig. 5C). Subsequently, our analysis revealed a negative correlation between the risk score and several cell types, namely regulatory T cells, resting natural killer (NK) cells, activated mast cells, M0 macrophages, neutrophils, and resting dendritic cells, as illustrated in Fig. 5D. Conversely, the risk score exhibited a positive correlation with resting CD4 memory T cells, follicular helper T cells, M1 macrophages, monocytes, naive B cells, activated NK cells, and activated dendritic cells (Fig. 5E).

Fig. 5.

Immunological role of ten-gene signature. A Correlation of ten characteristic genes and immune checkpoint inhibitor. B TME scores of high- and low- risk group. C Correlation between ten characteristic genes and TIICs. D TIICs negatively associated with prognostic risk scores. E TIICs positively associated with prognostic risk scores. TIIC tumor immune infiltration cell. *, p < 0.05. **, p < 0.01. ***, p < 0.001

We further analyzed the gene mutations in different risk subgroups. We identified the top 20 genes with the highest mutation rates in the high-risk subgroup (Fig. 6A) and low-risk subgroup (Fig. 6B). The results illustrated that missense mutation and Multi Hit were the most common mutation type in the high- and low- risk group, respectively. The mutation rates of PTEN, PIK3CA, ARID1A, and TTN were not only higher than 30% in both groups, but the most common in both groups. Interesingly, TP53 mutation was the most common in high-risk group, but not the top 20 mutation in low-risk group. Moreover, we analyzed the relationship between the risk score and TMB. The expression of TMB was significantly higher in the low-risk subgroup than in the high-risk subgroup (Fig. 6C). In addition, we revealed that a low-risk score was linked to MSI-H status, whereas a high-risk score was linked to microsatellite stable (MSS) status (Fig. 6D and E).

Fig. 6.

Genetic mutation between high- and low- risk group in TCGA-UCEC samples. A, B Oncoprint visualization of the mutation atlas of EC samples in high-(A) and low-(B) risk group from TCGA cohort. C TMB between high- and low-risk group. D Comparison of risk scores between MSS, MSI-L, and MSI-H samples. E Portion changes of MSS, MSI-L, and MSI-H samples between high- and low-risk group. TMB, tumor mutation burden. MSS, microsatellite stability. MSI-L, microsatellite Instability-Low. MSI-H, microsatellite Instability-high. *, p < 0.05. **, p < 0.01. ***, p < 0.001

Prognostic role and biological function of ACTL8

Using the relevant clinical data of TCGA-UCEC dataset, we grouped patients with UCEC according to the median level of the ten genes and compared the OS differences between different groups. According to the analysis result, only the ACTL8 gene level significantly affected DFS (Fig. 7A), PFS (Fig. 7B), DSS (Fig. 7C), and OS (Fig. 7D) of patients with UCEC, indicating that the IR-related gene ACTL8 may play an important role in the UCEC prognosis. To further research the functions and mechanisms of the ACTL8-related DEGs, all DEGs were assessed by GO term and KEGG analyses. The results of GO analysis demonstrated that the DEGs were significantly enriched with cilium movement, and cilium-dependent cell motility in the BP category (Fig. 7E). Furthermore, the neuroactive ligand-receptor interaction and calcium signaling pathway were significantly enriched in the signal transduction pathway category (Fig. 7F). To characterize the potential function of ACTL8 gene, GSEA was performed on the differential genes. As shown in Fig. 7G a total of four gene sets were found, including myc targets v1 (NES = 1.450, p = 0.002), E2F targets (NES = 1.395, p = 0.02), G2M checkpoint (NES = 1.383, p = 0.022), and oxidative phosphorylation (NES = 1.362, p = 0.046). Finally, through GSVA analysis, we found that ACTL8 was positively related to the wnt/beta-catenin signaling pathway, which indicates that ACTL8 expression increased with the enrichment of this pathway (Fig. 7H).

Fig. 7.

Prognostic role and biological function of ACTL8. A–D ACTL8 was related to poor clinical outcomes of EC patients in TCGA project in terms of DFS(A), PFS(B), DSS(C), and OS(D). E, F GO(D) and KEGG(E) enrichment analyses of ACTL8-related DEGs in TCGA-UCEC program. G GSVA analysis of high- and low-ACTL8 expression. GO gene ontology, KEGG Kyoto Encyclopedia of Genes and Genomes, DEG differential expression genes

Immunological role of ACTL8

TME plays an important role in the occurrence and progression of malignancies. In the TME scores obtained by the estimate method, including StromalScore, ImmuneScore, and ESTIMATEScore, each TME score of the ACTL8 high expression group was significantly lower than that of the counterpart group (Fig. 8A). To further decipher the relationship between ACTL8 expression and the immune cell infiltration, CIBERSORT analysis was performed to investigate the ratios of tumor-infiltrating immune subsets in UCEC patients in the TCGA database. A total of 22 kinds of immune cell profiles in UCEC samples were constructed. Seven of 22 cell types, including activated CD4 + memory T cells, regulatory T cells (Tregs), Macrophages M1, resting and activated dendritic cells (DCs), CD8 + T cells, and plasma cells exhibited significant differences between the high and low ACTL8 expression groups (Fig. 8B). Spearman correlation analyses showed when naive B cells and activated DCs correlated positively with ACTL8, plasma cells, Tregs, activated mast cells, CD4 + memory T cells, and resting DCs, neutrophils correlated negatively (Fig. 8C). Correlation between the ACTL8 expression and the TIICs was verified by another six methods, CIBERSORT-ABS, QUANTISEQ, MCPCOUNTER, EPIC, TIMER, and XCELL, demonstrating the potential influence of ACTL8 on T-cell infiltration (Fig. 8D).

Fig. 8.

Comparison of TME and TIICs between high- and low-ACTL8 expression. A The differences of TME between high- and low-ACTL8 expression. B The component differences of TIICs between high- and low-ACTL8 expression. C Scatter plots showing the TIICs related to ACTL8 expression via Cibersort methods. D The correlation analysis of TIICs and ACTL8 expression via multiple methods. TME, tumor microenvironment. TIIC, tumor infiltration immune cell. *, p < 0.05. **, p < 0.01. ***, p < 0.001

Discussion

EC is a malignant tumor that mainly occurs in postmenopausal women, and excessive exposure to estrogen is its most significant risk factor [18]. EC is classified into multiple histological subtypes. Among them, endometrioid EC, defined as type 1 tumor, is usually associated with long-lasting unopposed estrogen stimulation and accounts for more than 80% of newly diagnosed cases of endometrial cancer [19]. Recently, it has gradually been recognized that metabolic disorders, including obesity, MetS, and IR, have become the main factor contributing to the increased risk of EC. In a prospective study, Gunter, et al. found that fasting insulin levels in women who did not receive hormone treatment were positively correlated with the risk of EC, and hyperinsulinemia was reported as a risk factor for endometrial cancer, independent of estradiol [20]. Some studies demonstrated the relationship between IR and EC risks through peripheral blood markers, such as Homeostasis model assessment (HOMA) [21] and triglyceride-glucose(TyG) index [22]. Another prospective study confirmed the prevalence of IR in patients with EC [23]. In addition, insulin levels are significantly associated with the risk of lymph-node metastasis [24] and the recurrence rate in patients with EC [25]. Considering the clinical correlation between the two, there may be some common molecular mechanisms between IR and EC.

In the present study, we screened the common DEGs in the transcriptional profiles of IR and EC. PPI network and functional enrichment analysis suggested that these DEGs were closely related to the construction of extracellular matrix and the regulation of inflammatory immune-related signaling pathways. This result partly explained that IR increased the risk and progression of EC. To further confirm the impact of IR on the prognosis of EC patients, we extracted ten prognostic characteristic genes in the common DEGs using three machine-learning algorithms to construct an IR-related risk model. This model effectively distinguished the clinical outcomes of EC patients. Furthermore, the correlation of expression of prognostic characteristic genes with the TIICs and the expression of immune checkpoint genes were confirmed in the study. TME rich in TIICs plays a bidirectional effect of inhibiting or promoting tumor growth [26]. On the one hand, this bidirectional interaction was related to the types of TIICs, such as anti-cancer M1 macrophages and cancer-promoting M2 macrophages. On the other hand, some TIICs played a dual role. They acted as sentries capable of identifying and eliminating cancer cells, such as cytotoxic T cells and NK cells. They shifted from preventing tumor growth in the early stage of the tumor to promoting tumor progression regulated by cytokines secreted by the tumor in the late stage. Our study initially explored the correlation between IR-related prognostic risk models and the infiltration of different types of TIICs. However, the biogenic effects and underlying mechanisms caused by the changes in TIICs still need further investigation. In addition, different gene mutation landscapes were presented in the samples of the high- and low- risk groups. The most significant change among them lies in the obvious upregulation of the proportion of P53 mutations, TMB, and downregulation of MSI-H in the high-risk group. EC is divided into four molecular subtypes: microsatellite instability high (MSI-H), DNA polymerase ϵ (POLE) mutated, and copy number low, and copy number high [27]. The PFS rate of POLE-mutated tumors is the highest, while the risk of tumors with low copy number and MSI-H is moderate, and the prognosis of tumors with high copy number is the worst [27]. For the convenience of diagnosis, the immunohistochemical identification marker method was applied to convert these four molecular subtypes into: mismatch repair deficiency (MMRd), POLE exonuclease domain mutation (POLEmut), p53 wild-type/nonspecific molecular profile type (NSMP), and p53 abnormality (p53abn) [28]. The higher TP53 mutation rate and the lower proportion of MSI-H among patients in the high-risk group may be the molecular basis for their poor prognosis.

Among the ten prognostic characteristic genes, ACTL8 emerged as a gene of particular interest. It functioned as a pivotal gene within the PPI network of the overlapping genes of IR and EC. Additionally, it possessed significant prognostic value. Intriguingly, OS, DFS, PFS, and DSS of EC patients with high ACTL8 expression were all significantly shorter, suggesting a strong association with an unfavorable prognosis. ACTL8, belonging to the cancer/testicular antigen (CTA) family, is limited in expression in normal tissues such as the testicles and placenta, but is elevated in various tumors, such as gastric cancer [29], colorectal cancer [30], head and neck squamous cell carcinoma [31], triple-negative breast cancer [32], lung adenocarcinoma [33], etc. The upregulation of ACTL8 expression in these tumors was associated with poor outcomes of patients, and its oncogenic role in the pathogenesis of cancers were confirmed. For instance, Han, et al. found that silencing ACTL8 in HT29 and SW620 colon cancer cell lines led to reduced proliferation, migration, and invasion [30]. This pro-cancer effect was related to its activation of the PI3K/Akt pathway, which was confirmed in multiple studies [29, 31, 32]. Our research found that the high expression of ACTL8 was strongly associated with cilium movement and wnt/beta-catenin pathway, suggesting that other mechanisms by which ACTL8 exerted its pro-cancer effect. Cilia, microtubule-based organelles extending from the cell surface into the extracellular space, functions as spatially restricted hubs and displays receptors through which cues from soluble ligands, or from ligands tethered to the extracellular matrix (ECM) can be received [34]. Primary cilia are the main signaling centers for growth factors, including insulin-like growth factor (IGF) [35]. A large group of secreted wnt ligands influence the balance between cellular differentiation, polarity controls, and proliferation to regulate tissue homeostasis [36]. Multi effectors of wnt/beta-catenin pathway, including Disheveled (DVL), Frizzled 2 (FZD2), and GSK3, are found in the ciliary axoneme, suggesting the presence of intraciliary Wnt signalling [37]. In addition, we analyzed the relationship of ACTL8 with TME and TIICs. In EC, the upregulation of ACTL8 expression led to a decrease in stromal cells and immune cells in the TME. Multi-algorithm analyses of TIICs revealed that ACTL8 was negatively correlated with the infiltration of various immune cells.

There are still some flaws in this study. First, this study was mainly based on public databases and retrospectively analyzed the clinical outcomes of EC patients and constructed a prognostic model, which requires prospective studies for verification. Second, we further studied the prognostic significance and biological functions of ACTL8 in EC. However, the potential mechanisms involved in this marker in IR and EC still require experiments to explore. Finally, we indirectly evaluated the regulatory effects of the prognostic risk model and ACTL8 on TME and TIICs in EC, but these analyses merely utilized bioinformatics techniques. In conclusion, IR-driven transcriptomic features play a crucial role in the progression of EC and are also highly correlated with the immune microenvironment and immune cell infiltration.

Conclusion

In summary, our research initially identified DEGs shared between EC and IR. Leveraging machine-learning algorithms, we meticulously screened out prognostic characteristic genes from these common DEGs, subsequently utilizing them to construct a robust risk model. Patients with EC assigned to different risk groups exhibit distinct and significant disparities in clinical outcomes, gene mutations, and immune regulation. Moreover, we devoted substantial attention to a comprehensive analysis of the prognostic value, biological functions, and potential underlying mechanisms associated with ACTL8. In conclusion, the IR-related risk model established in this study played a pivotal role in facilitating the risk stratification of EC patients. It also served as a valuable tool for enabling precise therapeutic interventions, thereby offering crucial insights and guidance for subsequent research endeavors focused on exploring the functional aspects and molecular mechanisms involved.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

JQ and JC designed the study. SL, JX, and XS executed data analysis and visualization. SL wrote the manuscript. JQ, YJ, and JC made a revision on this manuscript. JQ and SL were responsible for confirming whether the data were reliable and authentic. The manuscript has been read by all named authors and obtained their approval.

Funding

This work got support from Shanghai Science and Technology Commission Social Development Technology Research Project (No. 22DZ1202304). The funding sponsor reviewed and approved the study design and the final manuscript.

Data availability

Data are available upon reasonable request from the corresponding author.

Declarations

Conflict of interest

No financial conflicts of interest to disclose.

Ethical approval and informed consent

The research followed the Declaration of Helsinki.