Endometrial cancer is the most common gynecological malignancy worldwide, and its incidence has increased 21% since 2008. Despite good survival rates for early-stage disease, patients with advanced or recurrent endometrial cancer have limited treatment options and poor prognosis. A key challenge is identifying reliable molecular biomarkers that can predict which patients are at highest risk of death and might benefit from more intensive treatment.
The human genome produces many RNA molecules that do not encode proteins but still perform critical regulatory functions. Long non-coding RNAs (lncRNAs) are RNA transcripts longer than 200 nucleotides that do not produce protein but can regulate gene expression. MicroRNAs (miRNAs) are small RNA molecules that typically suppress genes by binding to messenger RNA (mRNA). The ceRNA (competing endogenous RNA) hypothesis proposes that lncRNAs can act as 'sponges' for miRNAs - soaking them up so they cannot suppress their target mRNAs. This creates an indirect regulatory network where lncRNA levels influence mRNA expression by competing for the same miRNA.
This study used the TCGA-UCEC (The Cancer Genome Atlas Uterine Corpus Endometrial Carcinoma) dataset to systematically identify differentially expressed lncRNAs, miRNAs, and mRNAs in endometrial cancer versus normal tissue. It then constructed a ceRNA network from the interactions among these molecules and combined this with an analysis of immune cells in the tumor microenvironment to build prognostic models capable of predicting patient survival.
The study analyzed RNA sequencing data from 461 endometrial cancer patients and 87 tumor-free control samples from TCGA-UCEC. Differentially expressed genes were identified using the edgeR method with thresholds of absolute fold change greater than 2 (log fold change above 1.0 or below -1.0) and false discovery rate below 5%. This yielded 252 unique lncRNAs, 249 miRNAs, and 3,636 mRNAs differentially expressed in endometrial cancer.
The ceRNA network was constructed by mapping known lncRNA-miRNA and miRNA-mRNA interaction pairs from validated databases (miRTarBase and PITA). Only pairs where both molecules were significantly differentially expressed were included. The final ceRNA network contained 19 lncRNA-miRNA pairs and 434 miRNA-mRNA pairs involving 247 genes. Kaplan-Meier survival analysis and Cox regression identified which network genes had prognostic value, and LASSO regression narrowed these to a 12-gene model used for nomogram construction.
Immune cell composition was estimated using the CIBERSORT algorithm, which deconvolutes bulk RNA sequencing data to estimate the proportions of 22 different immune cell types within each tumor sample. CIBERSORT works by comparing each sample's gene expression profile against a reference panel of known immune cell signatures. Only samples with a CIBERSORT p-value below 0.05 were included in downstream immune analysis, ensuring reliable immune cell estimates.
Among all genes in the ceRNA network, Kaplan-Meier analysis identified ten genes with the strongest prognostic significance, with LRP8 (p = 2.68e-05), COL4A4, DLC1, SCML2, and several others reaching high statistical significance. The multivariate Cox regression model based on 12 ceRNA network genes produced a nomogram with AUCs of 0.733 at 1 year, 0.795 at 3 years, and 0.820 at 5 years for predicting overall survival - indicating improving discriminative ability over longer time horizons.
Four ceRNA network genes emerged as particularly relevant: LRP8 (a lipoprotein receptor previously linked to poor survival in multiple cancers), HDGF (heparin-binding growth factor that promotes blood vessel formation and tumor invasion), PPARGC1B (a regulator of mitochondrial metabolism), and TEAD1 (a transcription factor in the Hippo signaling pathway that promotes cell invasion). Protein expression validation in the Human Protein Atlas database confirmed that RAPGEF4, LRP8, and HDGF were positively expressed in endometrial cancer tissues compared to normal tissue, while other genes showed the reverse pattern.
CIBERSORT analysis revealed significant differences in immune cell composition between tumor and normal tissue. Memory CD4+ T cells, mast cells, and M0 macrophages were significantly more abundant in tumors than in normal tissue. Activated natural killer (NK) cells (p = 0.010) and M2 macrophages (p = 0.030) showed significant associations with overall survival in Kaplan-Meier analysis, with activated NK cells being associated with better outcomes - consistent with their role as direct anti-tumor effectors.
A key finding was the direct correlation between specific ceRNA network genes and immune cell infiltration. Activated dendritic cells were positively associated with LRP8 (r = 0.25), HDGF (r = 0.25), and TEAD1 (r = 0.19) - meaning tumors with more active dendritic cells also tended to have higher expression of these pro-tumor genes. Conversely, activated NK cells were negatively correlated with PPARGC1B (r = -0.19), suggesting that tumors with high PPARGC1B expression had fewer NK cell infiltrates, potentially reflecting an immunosuppressive tumor environment.
These correlations suggest a mechanistic link between ceRNA-mediated gene regulation and the composition of the tumor immune microenvironment. The tumor immune environment in endometrial cancer is not random - it is shaped partly by the same regulatory RNA networks that control tumor cell behavior. Dendritic cells, which act as commanders of the immune system by presenting tumor antigens to T cells, B cells, and NK cells, may be both influencing and influenced by the expression of LRP8, HDGF, and TEAD1.
A second nomogram based solely on immune cell proportions - specifically activated NK cells and M2 macrophages - achieved AUCs of 0.656, 0.666, and 0.645 for 1-, 3-, and 5-year survival, which was lower than the ceRNA-based nomogram. This comparison demonstrates that the molecular ceRNA network captures prognostic information that immune cell composition alone does not fully reflect, and that combining both biological layers is the optimal approach.
The practical output of this study is two prognostic nomograms - visual scoring tools that allow clinicians to predict a patient's probability of surviving 1, 2, or 3 years based on measurable inputs. The ceRNA-based nomogram uses 12 gene expression values measured from the tumor; the immune-based nomogram uses proportions of NK cells and M2 macrophages estimated from tumor gene expression. Both tools are grounded in TCGA data from 461 endometrial cancer patients.
The ceRNA nomogram substantially outperformed the immune nomogram, with 5-year AUC of 0.820 versus 0.645. This performance advantage highlights that the molecular regulatory architecture of the tumor - encoded in its RNA interactions - provides more prognostic information than immune cell composition alone. However, the ideal model would combine both, since they reflect partially complementary aspects of tumor biology.
Limitations of the study include its reliance on a single retrospective dataset (TCGA-UCEC) and the absence of an independent external validation cohort. The study did not incorporate the TCGA molecular subtypes (POLE, MSI-H, CN-L, CN-H) into the analysis, which means the prognostic value of ceRNA genes within each molecular subtype is unknown. Future studies with prospective multi-center cohorts incorporating both molecular subtyping and ceRNA network analysis could substantially improve prediction accuracy and clinical applicability.
This study demonstrates that the competitive endogenous RNA network - the system by which lncRNAs indirectly regulate protein-coding gene expression by sponging up miRNAs - is a meaningful source of prognostic information in endometrial cancer. Genes like LRP8, HDGF, TEAD1, and PPARGC1B, embedded within this regulatory network, capture aspects of tumor biology that conventional clinical factors do not.
The integration of ceRNA network analysis with tumor immune microenvironment characterization using CIBERSORT represents a promising framework for understanding how tumors regulate their own immune environment through RNA-level mechanisms. The finding that ceRNA genes are directly correlated with immune cell fractions suggests that targeting these genes could simultaneously alter tumor behavior and the immune response - a dual benefit that immunotherapy-combined approaches might exploit.
The study identifies LRP8, PPARGC1B, HDGF, and TEAD1 as priority candidates for further investigation as therapeutic targets or non-invasive biomarkers in endometrial cancer. Their consistent appearance as both prognostic markers and immune regulators, validated at the protein level in independent databases, strengthens the case for their biological relevance. Prospective multicenter validation with standardized RNA sequencing protocols is the necessary next step before clinical translation.