Lysosomes are organelles inside cells that act as the cellular recycling center - they break down damaged proteins, old organelles, and cellular debris. Lysosome-dependent cell death (LDCD) refers to a form of programmed cell death triggered when lysosomes rupture and spill their enzyme contents into the cell interior. This mode of death is distinct from better-known processes like apoptosis and ferroptosis.
Cancer cells frequently develop abnormal lysosome function. They may upregulate lysosomal activity to survive stressful conditions, or they may accumulate lysosomal enzyme mutations that affect how readily LDCD can be triggered. The key question is whether the LDCD pathway plays a meaningful role in endometrial cancer biology - and whether targeting it could offer new therapeutic opportunities.
This study used data from TCGA-UCEC (The Cancer Genome Atlas, endometrial cancer cohort) combined with additional public datasets to build machine learning models that characterize LDCD-related gene expression patterns and their association with prognosis, immune microenvironment, and drug sensitivity in endometrial cancer.
Starting with a curated list of LDCD-related genes from published literature, the researchers used consensus clustering to identify molecular subtypes of endometrial cancer based on LDCD gene expression. Each subtype showed distinct survival outcomes and immune characteristics, validating that LDCD gene expression captures biologically meaningful variation.
Feature selection used multiple statistical filters: differential expression analysis between subtypes, univariate Cox survival analysis, and then a systematic comparison of 15 machine learning algorithms including LASSO, Ridge, ElasticNet, Random Forest, SVM, CoxBoost, and gradient boosting variants. This exhaustive approach identifies which combination of algorithm and features produces the most robust and reproducible prognostic signature.
Validation used both the internal TCGA dataset and independent GEO (Gene Expression Omnibus) datasets. Single-cell RNA sequencing data from dataset GSE139555 (containing over 175,000 individual cells) was used to confirm which cell types within the tumor actually express the key LDCD genes - an important step because bulk tumor analysis cannot distinguish which cell populations drive the signal.
The final prognostic signature centers on 6 LDCD hub genes: CTSV, LAMP3, STXBP1, STXBP2, FER, and GGA2. These genes are involved in lysosomal enzyme production (CTSV is cathepsin V, a lysosomal protease), membrane trafficking to lysosomes (LAMP3, GGA2), and vesicle fusion events (STXBP1, STXBP2). FER is a tyrosine kinase with roles in cellular stress signaling.
Patients with high expression of this gene set were classified as high risk, with significantly shorter overall survival. The model achieved AUC values of 0.7-0.8 across validation cohorts - in the range considered acceptable to good for a prognostic tool, though not exceptional. Importantly, the signature remained independently prognostic after adjusting for FIGO stage, age, and histological grade.
Single-cell validation revealed that these LDCD hub genes are expressed predominantly in malignant epithelial cells and certain immune cell populations within the tumor, rather than in the surrounding stromal tissue. This cell-type specificity is important for understanding potential therapeutic targeting - drugs would need to reach specific cell populations to affect LDCD pathway activity.
High-risk LDCD patients showed a distinctive immune microenvironment profile: lower overall immune and stromal scores, reduced infiltration of cytotoxic immune cells, and higher proportions of immunosuppressive cell types. Specifically, high-risk tumors had more regulatory T cells (Tregs) - cells that suppress immune responses and prevent the immune system from attacking cancer - and more M2-polarized macrophages, which promote tumor growth rather than fighting it.
This immune suppression pattern has important clinical implications. Tumors with Treg and M2 macrophage dominance tend to respond poorly to immune checkpoint inhibitors like pembrolizumab (anti-PD-1), which rely on having functional cytotoxic T cells present to be unleashed. The LDCD risk score could therefore help predict which patients are unlikely to benefit from standard immunotherapy regimens.
The LDCD high-risk group also showed differential sensitivity to specific chemotherapy drugs when analyzed using the TIDE (Tumor Immune Dysfunction and Exclusion) algorithm and drug sensitivity databases. This pharmacogenomic dimension suggests that identifying patients by their LDCD profile might help select alternative treatment strategies for the immunotherapy-resistant subgroup.
The LDCD pathway represents a relatively underexplored therapeutic avenue in endometrial cancer. While most targeted therapies focus on signaling pathways (PI3K, mTOR, HER2), the lysosomal cell death pathway is structurally distinct and may offer opportunities to sensitize resistant cancer cells to treatment - particularly those that have upregulated lysosomal activity as a survival mechanism.
The study's use of 15 different machine learning algorithms is a methodological strength - by testing many approaches and selecting the most robust, the risk of presenting an accidentally good result from a single algorithm is reduced. The single-cell validation is also a noteworthy addition that many purely computational prognostic studies lack.
Clinical translation would require developing this into a standardized gene expression assay, then prospectively validating it in a clinical trial where treatment decisions are guided by LDCD risk score. Given the immune profile data, a natural trial design would randomize high-risk LDCD patients to immunotherapy versus an alternative regimen to directly test whether the risk score predicts differential treatment benefit.