Machine learning-based integration identifies lactylation biomarkers for prognosis prediction and tumor microenvironment modulation in bladder cancer.

BACKGROUND

Lactylation, a metabolic-driven post-translational modification, plays a critical role in tumor microenvironment (TME) remodeling. Lactylation modifies histones and non-histone proteins to regulate gene expression and cell signaling, and recent studies have linked it to immune evasion and tumor progression in bladder cancer (BLCA). However, the prognostic value of lactylation-related genes (LRGs) and their underlying immunoregulatory mechanisms in BLCA have yet to be elucidated. This study aims to establish a lactylation-based molecular classification for BLCA and to construct a robust machine learning prognostic model, thereby clarifying the immunoregulatory role of lactylation and its clinical implications.

METHODS

We integrated multi-center transcriptomic datasets from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) cohorts. Unsupervised consensus clustering was employed to identify lactylation-associated molecular subtypes. Innovatively, we developed an ensemble machine learning framework using 107 combinations of 10 distinct algorithms to establish an optimal prognostic model, incorporating a SHapley Additive exPlanations (SHAP) analysis to interpret core gene contributions. We further evaluated the association between the model, the TME landscape, immune checkpoints, and drug sensitivity.

RESULTS

We identified two molecular subtypes with distinct survival outcomes based on 40 differentially expressed LRGs. The final 12-gene prognostic model-constructed using a combination of random survival forest (RSF) and Ridge regression-demonstrated superior predictive performance across multiple cohorts [with its concordance index (C-index) and area under the curve (AUC) values significantly exceeding those of existing models]. The SHAP analysis identified five core contributors, including PTGIS and SRP68. Immunological characterization revealed that the high-risk group exhibited a paradoxical combination of high immune infiltration (including activated CD4+ T cells, activated B cells, natural killer cells, macrophages, and dendritic cells) and high immune suppression, characterized by upregulated checkpoints [e.g., CD274/programmed cell death ligand 1 (PD-L1) and cytotoxic T-lymphocyte-associated protein 4 (CTLA4)] and elevated Tumor Immune Dysfunction and Exclusion (TIDE) scores, suggesting significant immune evasion potential. Conversely, this group showed enhanced sensitivity to conventional chemotherapeutics such as cisplatin and gemcitabine.

CONCLUSIONS

This study established a robust, lactylation-based prognostic tool for BLCA and provided mechanistic insights into how lactylation shapes the immunosuppressive microenvironment, offering a scientific basis for precision immunotherapy and chemotherapy strategies.