Endometrial cancer (EC) is the sixth most commonly diagnosed cancer in women worldwide, accounting for approximately 10,170 deaths annually in the United States alone. Like most cancers, EC is driven by disruptions across multiple molecular levels: abnormal DNA methylation, copy number changes in tumor suppressor genes and oncogenes, and dysregulated expression of messenger RNAs, microRNAs, and long noncoding RNAs.
EC is broadly divided into two types. Type I EC, comprising about 80% of new cases, is endometrioid in histology and typically expresses estrogen and progesterone receptors. Type II EC is non-endometrioid, often presenting as high-grade, poorly differentiated tumors with clear-cell or serous papillary morphology. Despite making up only 20% of cases, type II EC accounts for 44% of all EC-related deaths.
Despite the wealth of molecular information published about EC, no systematic resource had compiled all EC-implicated genes in one place. Researchers studying a gene of interest had to search dozens of separate databases and thousands of publications individually, with no unified view of how that gene fit into the broader EC molecular landscape.
This study describes ECGene, the first literature-based online database specifically designed to aggregate, annotate, and make searchable all genes implicated in endometrial cancer. By consolidating published evidence, expression data, protein interaction networks, and long noncoding RNA coexpression patterns into a single platform, ECGene aims to accelerate understanding of EC biology and the identification of new diagnostic and therapeutic targets.
ECGene was built through two parallel processes. The first drew from three established genetic databases: OMIM (Online Mendelian Inheritance in Man), which contributed 4 genes; GAD (the Genetic Association Database), which contributed 113 unique genes from 149 studies; and GWASCatalog, which contributed 14 candidate genes identified through genome-wide association studies. Together these sources provided a starting list of 127 non-redundant EC-implicated genes.
The second and larger contribution came from manual literature curation. Using keywords combining endometrial/uterine terms with cancer/tumor/carcinoma terms, researchers queried the GeneRIF (Gene Reference Into Function) database and retrieved 845 PubMed abstracts. Three curators manually reviewed these abstracts, grouped them by semantic similarity, extracted gene names from descriptions about EC, and mapped them to standardized Entrez gene identifiers to ensure consistency across synonymous gene names.
This literature curation process identified 414 Entrez human genes from 747 abstracts (98 abstracts did not yield usable gene names after review). Combining these with the 127 database-derived genes and removing overlaps produced the final ECGene list of 458 EC-implicated genes, comprising 423 protein-coding and 35 noncoding genes.
Each gene was comprehensively annotated with biological pathway information from BioCyc, KEGG, PANTHER, and Reactome; disease associations from KEGG Disease, OMIM, and NHGRI; post-translational modification data; transcription factor regulation; methylation site information; expression profiles from 184 tumor and 84 normal tissue samples; and protein-protein interaction data from PathwayCommons.
Because 458 genes is a large number for researchers to investigate, ECGene ranked the relative importance of all 423 protein-coding genes using the ToppGene bioinformatics tool. Twenty-five highly reliable genes (each with 10 or more supporting publications) served as a training set. ToppGene then used biological feature annotations including gene coexpression, gene ontology evidence, pathway memberships, and protein domain characteristics to rank the remaining 399 genes.
The ranking confirmed known EC biology: well-established genes such as EGFR and CDH1 ranked highly, validating the method. Importantly, the ranking also surfaced lesser-studied candidates including MUC8, TARP, and MAPK1, each supported by only a single EC publication but whose functional features closely resemble those of the training set. The authors flagged these as meriting further experimental investigation.
Functional enrichment analysis of the top 100 ranked genes revealed enrichment in cancer-related and reproductive biological processes. Key enriched pathways included the broad Pathways in Cancer category and the p53 pathway feedback loops. The single most significantly enriched pathway was Signaling by SCF-KIT, a system linking stem cell factor (SCF) to its receptor c-KIT (CD117) to promote cell growth. Prior work had shown that targeting this axis with imatinib sensitized EC cells to cisplatin, suggesting it as a potential therapeutic vulnerability.
To test whether EC-implicated genes carry relevance beyond endometrial cancer, the researchers examined the somatic mutation patterns of the top 99 ranked genes (TP53 was excluded to avoid its outsized influence) across all cancer types in the cBioPortal genomics database. The results were striking: these 99 EC-implicated genes were mutated in 98.8% of TCGA endometrial cancer patients and 98.9% of lung squamous cell carcinoma patients, with mutation rates above 90% in 31 other cancer datasets spanning 19 cancer types.
The highest mutational overlap was seen in bladder and colorectal cancers, both of which are anatomically adjacent to the uterus. The authors suggest that tumors arising in neighboring tissues may share common driver mutations related to their shared tissue of origin, an observation consistent with the TCGA pan-cancer mutational analysis. This raises the possibility that drugs already approved for bladder or colorectal cancer might be repurposed to treat endometrial cancer patients with matching mutations.
Among the 239 TCGA endometrial cancer samples, the most frequently mutated genes (at 20% or higher mutation frequency) were PTEN, PIK3CA, PIK3R1, CTNNB1, TP53, and KRAS. These six genes are already recognized as major drivers of EC and are the focus of numerous ongoing clinical trials, confirming that ECGene's curation captured the most clinically relevant genetic landscape.
To move beyond individual genes and understand how EC-implicated genes interact collectively, the researchers reconstructed an endometrial cancer pathway interactome. Starting from a human protein-protein interaction network of 3,629 genes and 36,034 connections from PathwayCommons, the 458 EC-implicated genes were mapped in and connected via their shortest paths to form a coherent EC-specific subnetwork.
The reconstructed EC interactome contained 290 genes and 769 gene-gene interactions. Of the 290 nodes, 246 were EC-implicated genes from ECGene, with the remaining 44 being linker genes required to bridge them into a fully connected map. Topological analysis revealed a tightly connected structure: the network's degree distribution follows a power law with an exponent of 1.3, indicating far denser connectivity than the general human interactome (exponent 2.9), and 75.9% of gene pairs are separated by only four steps.
Eleven hub genes were identified, defined as genes with 20 or more connections in the network: TP53 (47 connections), MYC (37), CTNNB1 (33), AKT1 (23), NFKB1 (23), AR (22), ESR1 (22), HDAC1 (22), FOS (21), PIK3CA (21), and PIK3R1 (20). All 11 hub genes were present in ECGene, and their enriched functions include cellular response to endogenous stimuli and enzyme-linked receptor protein signaling. Ten of the eleven are simultaneously involved in general cancer pathways, colorectal cancer, prostate cancer, and endometrial cancer KEGG pathways.
Long noncoding RNAs (lncRNAs) are RNA molecules longer than 200 nucleotides that do not encode proteins but play important regulatory roles in gene expression, chromatin organization, and cancer development. Despite growing recognition of their importance, lncRNA roles in endometrial cancer were poorly characterized at the time ECGene was built.
To fill this gap, the researchers calculated Spearman correlation scores between each of the 423 protein-coding EC-implicated genes and all 17,250 lncRNA transcripts available in the MiTranscriptome database, using matched TCGA endometrial cancer patient samples. Pairs were considered significantly coexpressed if they had a correlation score above 0.3 and a false discovery rate-adjusted p-value below 0.01.
This analysis identified 43 EC-implicated genes each coexpressed with 100 or more lncRNAs. These high-lncRNA-association genes were enriched in the nucleoplasm and associated with cell cycle regulation and chromosome organization processes. A striking example was TERF1 (Telomeric Repeat Binding Factor 1), which was positively coexpressed with 539 lncRNAs in TCGA EC samples. TERF1 functions at telomeres throughout the cell cycle by inhibiting telomerase elongation, a process directly relevant to the unlimited replicative potential of cancer cells.
These findings represent the first systematic mapping of lncRNA-gene coexpression relationships in endometrial cancer. The results are accessible directly through the ECGene web interface, where users can click on any lncRNA ID to view its expression pattern across all TCGA tissue types, providing a gateway for researchers to generate new hypotheses about lncRNA-mediated regulation in EC.
ECGene represents the first literature-based online knowledgebase dedicated to endometrial cancer genetics. By consolidating 458 EC-implicated genes from both systematic database mining and manual curation of hundreds of PubMed abstracts, it provides researchers with a single, annotated entry point into the molecular genetics of this disease. The database is freely accessible and built with a user-friendly web interface that supports gene symbol searches, BLAST sequence comparisons, chromosome browsing, and keyword-based literature retrieval.
The validation of the database content through mutation analysis, interactome reconstruction, and lncRNA coexpression demonstrates that ECGene captures genuine biological signal rather than simply cataloging genes. The finding that top-ranked EC genes are mutated in nearly all EC patients, and frequently in neighboring cancer types as well, opens a practical path toward drug repurposing strategies for patients whose tumors carry targetable mutations already addressed by existing approved therapies.
ECGene's lncRNA coexpression module is particularly forward-looking: at the time of publication in 2016, the functional roles of most lncRNAs in cancer were unknown. The precomputed correlations in ECGene provide a starting point for researchers to identify which lncRNAs may regulate specific EC-implicated genes, filling a critical knowledge gap. The authors committed to continuing curation and developing additional browsing tools as the EC literature grows.