The R package “TCGAbiolinks” was used to download RNA-seq data (TPM and Counts values) and clinical data (comprising 554 cancer tissue samples and 35 healthy adjacent tissue samples) from The Cancer Genome Atlas (TCGA) dataset for UCEC patients (Colaprico et al., 2016). The inclusion criteria for UCEC patients, along with clinical information for the eligible patients, is listed in Supplementary Table S1. A total of 1,407 genes related to lipid metabolism (LMRGs) were downloaded from the Molecular Signatures Database (MSigDB) (http://www.gsea-msigdb.org/) (Supplementary Table S2), and 893 lysosome-relevant genes (LYRGs) were retrieved from the Gene Ontology (GO) (https://geneontology.org/) and MSigDB databases (Supplementary Table S3).

Using a |log2FC| > 1 and padj. < 0.01 as criteria, 5,507 TCGA-DEGs between UCEC and normal tissues were screened via the “DESeq2” R package (Love et al., 2014). Of these TCGA-DEGs, 418 were identified as differentially expressed LMRGs (DE-LMRGs). The effect of the 418 DE-LMRGs on UCEC prognosis was then assessed by Kaplan-Meier survival analysis using the “survival” and “survminer” R packages. Finally, 57 putative prognosis-related genes (PRGs) were selected for further investigation.

Based on the expression profiles of the 57 putative PRGs, UCEC patients were classified into distinct molecular subgroups via unsupervised clustering. To ensure the stability of the clustering results, the analysis was repeated 1,000 times using the “ConsensusClusterPlus” package (Wilkerson and Hayes, 2010). Additionally, principal component analysis (PCA) was performed using the “stats” R package to validate and visualize the clustering results.

Differential expression analysis was conducted using the “DESeq2″ package to identify genes exhibiting differential expression among the three lipid metabolism subtypes (LM-DEGs). Genes with padj. values of less than 0.01 and an absolute log2 fold change greater than 1 were considered significantly differentially expressed. The intersection of the 5,507 TCGA-DEGs, 893 LYRGs, and 1,085 LM-DEGs yielded 26 overlapping genes, which were selected for subsequent analyses. Univariate Cox regression analysis was performed on the 26 genes, leading to the identification of 11 genes with prognostic potential. Following least absolute shrinkage and selection operator (LASSO) regression analysis of the 11 candidate genes, the eight exhibiting the strongest association with UCEC prognosis were selected and were subsequently used to construct a prognostic model for UCEC.

To evaluate the prognostic capacity of the model in UCEC, the patients were initially ranked into low- and high-risk groups based on their calculated risk scores. Subsequently, potential disparities in survival outcomes between the two groups were investigated. Several analytical approaches were employed to assess the accuracy of the model, including K-M survival, ROC curve, and Cox regression analyses. Additionally, to assess the reliability of the model and thereby enhance its clinical applicability, a nomogram and a calibration curve were created. Gene Ontology (GO) term and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were conducted employing the KOBAS web server (http://kobas.cbi.pku.edu.cn) (Bu et al., 2021). Gene Set Variation Analysis (GSVA) was performed to identify sets of enriched genes among all samples within the UCEC cohort (Hänzelmann et al., 2013).

The R package “maftools” was used to compute the TMB for each patient, and the Spearman correlation coefficient between the risk score and the TMB was subsequently calculated (Mayakonda et al., 2018). The single-sample Gene Set Enrichment Analysis (ssGSEA) algorithm was employed to quantify immune cell infiltration using the “GSVA” package in R. The Tumor Immune Dysfunction and Exclusion (TIDE) score was used to estimate tumor immune escape (Jiang et al., 2018). The ability of the developed risk features to predict anticancer treatment efficacy was validated using Immunophenoscore (IPS) data and the “oncoPredict” software package (Charoentong et al., 2017; Maeser et al., 2021).

The origin and culture methods of the four endometrial cancer cell lines used in this study (Ishikawa, ECC-1, HEC-1A, and HEC-1B) were as previously described (Yang et al., 2022). Human phospholipase A and acyltransferase 1 (PLAAT1)-specific siRNA and control siRNA were sourced from Gene Pharma (Shanghai, China). The sequences used for the silencing of PLAAT1 are presented in Supplementary Table S4.

Total RNA was extracted from logarithmically growing cells with Trizol reagent (Invitrogen) according to the manufacturer’s instructions. Total RNA (500 ng) was reverse transcribed into cDNA using PrimeScript RT Master Mix (ES Science, Shanghai, China). qPCR was performed with the primers listed in Supplementary Table S5 and SYBR Green Master Mix (ES Science). Proteins were separated by SDS-PAGE electrophoresis, initially at 80 V until the samples migrated through the stacking gel, followed by 120 V until the target protein markers were adequately resolved. The proteins were then transferred to a PVDF membrane at a constant current of 250 mA for 40 min. The membrane was blocked with 5% skim milk for 1 h, incubated with primary antibody (PLAAT1, 1:500 dilution) at 4°C overnight, and subsequently probed with a secondary antibody (1:5,000 dilution) at room temperature for 90 min, followed by TBST washes. Finally, the membrane was incubated with ECL substrate for 1–2 min and visualized using a chemiluminescence imaging system. The detailed information of the antibodies used is provided below: anti-ACTB (81115-1-RR, Proteintech, China), anti-PLAAT1 (YN3890, Immunoway, China).

To compare the proliferative capacity of endometrial cancer cells after different treatments, a CCK-8 assay (SB-CCK8S, Share-Bio, China) was performed following the manufacturer’s instructions. HEC-1A cells (3,000 per well) were seeded in a 96-well plate, and the absorbance at 450 nm was measured initially after cell adhesion, and then every 24 h for a total of 5 times. In the colony formation assay, cells were seeded in a 24-well plate at a density of 500 cells per well. After 12 days of incubation, the colonies were stained and imaged for subsequent analysis.

Cell suspensions (4 × 10⁴ HEC-1A cells) were added to the upper chambers (8-μm pore; coated or not with Matrigel) of 24-well Transwell plates for cell migration and invasion assays. The invasion assay was conducted for a total of 48 h, while the migration assay was conducted for a total of 36 h. After fixation and staining, the cells were imaged and counted.

Statistical analyses were conducted using R software (v.4.1.3) and GraphPad Prism 9.0. The presented results are based on a minimum of three independent experiments and are expressed as means ± SD. Statistical significance was determined by comparing p-values, with a threshold of <0.05 considered significant.*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, ns: not significant.

The overall workflow of the study is illustrated in Figure 1. We retrieved 1,407 LMRGs from the GSEA website and identified 418 DE-LMRGs in UCEC patients from TCGA cohort. A batch survival analysis of these 418 DE-LMRGs revealed that 57 of them were significantly associated with overall survival (OS). A consensus cluster consisting of these 57 genes was constructed with the “ConsensusClusterPlus” package. The results showed that TCGA-UCEC cohort was effectively stratified into three clusters—cluster A, with 133 samples; cluster B, with 149; and cluster C, with 228 samples (Figure 2A; Supplementary Figure S1A, B). A PCA plot of the three subgroups is illustrated in Figure 2B. Subsequent survival analysis uncovered notable disparities among the three clusters, with cluster A exhibiting a particularly notable survival advantage (Figure 2C). We also investigated the relationship between the subgroups and clinical-pathological parameters in UCEC patients (Figure 2D). Cluster A had the lowest proportion of G3 and stage III–IV patients, followed by cluster B, and then cluster C. The expression patterns of the 57 genes used for clustering are shown as a heatmap in Supplementary Figure S1C.

We identified a total of 1,085 DEGs between the three lipid metabolism subgroups (Figure 3A). As illustrated in the Venn diagram in Figure 3B, the intersection of TCGA-DEGs, LYRGs, and LM-DEGs yielded a total of 26 overlapping genes, defined as LMRG-LYGs. To construct the UCEC prognostic model, we further screened the LMRG-LYGs using univariate Cox regression analysis, which revealed that 11 genes were significantly associated with UCEC prognosis (Supplementary Figure S2A). Subsequently, employing LASSO Cox regression analysis based on these 11 candidate genes, we developed a prognostic model consisting of eight genes (Figures 3C,D). The risk score was calculated based on these eight genes using the following formula: risk score = [expression (LAMP3) × (0.16454)] + [expression (RNF183) × (−0.00337)] + [expression (EEF1A2) × (0.04929)] + [expression (PLAAT1) × (0.06792)] + [expression (ELAPOR1) × (−0.00696)] + [expression (B4GALT1) × (−0.12904)] + [expression (ATP10B) × (0.19877)] + [expression (PLA2G10) × (−0.20899)]. Additionally, gene expression analyses showed that all eight risk genes were upregulated in tumor samples compared to normal samples (Supplementary Figure S2B). Meanwhile, we used immunohistochemistry data from The Human Protein Atlas database to confirm the protein levels of these eight risk genes (Supplementary Figure S3), and we presented the eight risk genes’ biological functions and annotations (Supplementary Table S6). To understand the prognostic value of these genes, survival curves were plotted (Supplementary Figure S2C). The training set was divided into low-risk and high-risk UCEC groups based on the median risk score, with the latter exhibiting markedly higher mortality than the former. The differential expression of the eight genes between the two groups is depicted in a heatmap in Figure 3E. Subsequent Kaplan-Meier survival curve analysis revealed that a high-risk score was correlated with an unfavorable prognosis (Figure 3F). ROC plots were generated to evaluate the diagnostic performance of the model, yielding area under the curve (AUC) values of 0.680 (1-year OS), 0.717 (3-year OS), and 0.727 (5-year OS) (Figure 3G). The prognostic model was tested in both the UCEC testing set and the entire UCEC cohort, consistently confirming the robust predictive value of the LMRG/LYG-based signature (Supplementary Figure S2D–G).

To comprehensively evaluate the clinical applicability of our LMRG/LYG-based signature, box plots were used to illustrate the correlation between risk scores and various clinicopathological features. We observed that higher risk scores were associated with older age, advanced histologic grade, and advanced clinical stage (Figures 4A–C). Further analysis using a forest plot identified significant independent associations between clinical characteristics (age, grade, and stage) and the OS of UCEC patients. Moreover, the risk score itself, in addition to being associated with these clinical features, also served as an independent prognostic factor, underscoring the high accuracy of our features (Figure 4D). To determine the probability of OS (1-, 3-, and 5-year) for each patient individually, a nomogram was established by integrating clinicopathological features and risk scores (Figure 4E). Meanwhile, the calibration curves for 1-, 3-, and 5-year survival demonstrated that there was strong agreement between the survival rates predicted by the nomogram and the observed survival rates (Figure 4F). This suggested that our nomogram could accurately predict the prognosis of UCEC patients.

The TMB can greatly influence anticancer immunity, implying that it holds predictive value for both clinical responses to immune checkpoint inhibitor therapies and the prognosis of UCEC patients. To further evaluate the predictive ability of our constructed LMRG/LYG-based signature and elucidate the tumor mutation landscape in the different risk groups, we first identified the 10 genes exhibiting the highest mutation frequency (Figure 5A). The low-risk group demonstrated a significantly higher prevalence of mutations in most of these genes compared to the high-risk group, and subsequent survival analysis revealed a strong association between a high TMB and improved OS (Figure 5B). To explore the synergistic effect of the TMB and risk score on outcome prediction in more detail, we divided patients into subgroups based on both metrics and observed notable statistical differences in survival rates across the different groups. The cohort with a high TMB in combination with a low risk score had the highest OS rate, while patients with a low TMB and a high risk score demonstrated the lowest OS rate (Figure 5C). Overall, our analysis showed that the risk score was negatively correlated with the TMB, suggesting that the former can serve as a reliable indicator of the latter (Figure 5D). These findings underscore the potential of the LMRG/LYG-based risk score as a predictive tool for responses to immune checkpoint inhibitor (ICI) therapy.

Applying the criteria of padj. < 0.01 and |log2FC| > 0.8, we identified 1,514 DEGs between the high-risk and low-risk groups. These DEGs were visualized using a volcano plot and a heatmap (Figures 6A,B). To better understand the significance of the LMRG/LYG-based signature, KEGG and GO enrichment analyses were performed based on the DEGs between the two risk groups. KEGG analysis identified significant enrichment in several metabolic pathways, including Arachidonic acid metabolism, Cortisol synthesis and secretion, and Glycerophospholipid metabolism (Figure 6C; Supplementary Table S7). Moreover, GO analysis results indicated that these DEGs were significantly associated with terms such as microtubule bundle formation, extracellular matrix, and gated channel activity (Figure 6D; Supplementary Table S8). Furthermore, through GSVA, we identified gene enrichment in pathways such as DNA replication, cell cycle, and mismatch repair in the high-risk group. In contrast, the low-risk group exhibited enrichment of genes linked to pathways such as fatty acid metabolism, glycerolipid metabolism, and linoleic acid metabolism (Figure 6E; Supplementary Table S9).

Next, we used the ssGSEA algorithm to compare immune cell infiltration between the two risk groups. The high-risk group exhibited substantial infiltration of multiple helper T-cells, whereas the low-risk group demonstrated high infiltration of activated CD8⁺ T cells, CD56^bright natural killer cells, and plasmacytoid dendritic cells (Figure 7A). These findings reflect a more active immune state in the low-risk group, highlighting clear differences in immune profiles between the two groups. To further confirm the association between the LMRG/LYG-based signature and immune infiltration, immune cell fractions were estimated using TIMER, EPIC, CIBERSORT, and quanTIseq algorithms (Supplementary Figure S4). Then, a series of evaluation metrics was employed to determine whether our risk model was correlated with the response to immunotherapy. The high-risk group had a higher TIDE score along with a lower microsatellite instability (MSI) score than the low-risk cohort, indicative of stronger immune dysfunction and resistance (Figure 7B). In contrast, patients in the low-risk group exhibited higher IPS values, signifying that they had a superior response to immunotherapy (Figures 7C–F). These findings indicated that patients classified as low-risk are more suitable candidates for immunotherapy. To further explore the potential utility of the LMRG/LYG-based signature in guiding individualized UCEC treatment, we investigated the relationship between risk scores and sensitivity to chemotherapeutic drugs and small-molecule inhibitors (Figure 7G). We noted that the low-risk group exhibited greater sensitivity to commonly used chemotherapeutic drugs for UCEC, including docetaxel, paclitaxel, and topotecan, as well as three commonly used small-molecule inhibitors, namely, afatinib, sorafenib, and talazoparib. Other drugs with sensitivity differences between the risk groups are listed in Supplementary Table S10.

Considering the important role of CD8⁺ T cells in antitumor immune responses, we investigated the association between CD8⁺ T cells and the eight risk genes identified in this study using the TIMER web server (Supplementary Figure S5). Because PLAAT1 showed the strongest negative correlation with the CD8⁺ T-cell proportion among the eight risk genes, we focused on exploring its biological role in UCEC. First, we found that PLAAT1 was ubiquitously expressed in the four endometrial carcinoma cell lines at both the mRNA and protein levels (Figures 8A,B). Given that HEC-1A cells had the highest PLAAT1 protein level, they were selected for subsequent cytological experiments. To investigate the function of PLAAT1 in endometrial cancer cells, we downregulated its expression in HEC-1A cells through siRNA (p < 0.05, Figure 8C). The knockdown of PLAAT1 significantly suppressed cell proliferation, as determined by colony formation and CCK-8 assays (all p < 0.05; Figures 8D,E), while also inhibiting cell migration and invasion, as observed in the Transwell assay (all p < 0.05; Figure 8F). In conclusion, our data suggested that PLAAT1 may play a carcinogenic role in endometrial cancer. This makes PLAAT1 a promising therapeutic target that merits further study, potentially opening new directions for UCEC treatment.