Integrative transcriptomic analysis identifies lactylation subtypes and a lactylation-related prognostic signature in acute myeloid leukemia

Abstract

Background: Histone and non-histone lysine lactylation is an emerging metabolic–epigenetic modification linking glycolysis to gene regulation. Its prognostic and immunological significance in the highly glycolytic disease acute myeloid leukemia (AML) is incompletely characterised. Methods: We curated 57 lactylation-related genes (LRGs) and analysed The Cancer Genome Atlas acute myeloid leukemia cohort (TCGA-LAML) (training, n = 132), GSE106291 (n = 250) and GSE71014 (n = 104; external validation). Consensus clustering defined lactylation subtypes, with the cluster number selected primarily by the proportion of ambiguous clustering (PAC); a ten-gene prognostic signature was then developed using univariate Cox screening and least absolute shrinkage and selection operator (LASSO)-Cox regression on lactylation-related differentially expressed genes. Subtype transferability was tested by nearest-centroid classification, and the signature was validated in both external cohorts and via a training/validation-cohort-swap sensitivity analysis. Results: Two lactylation subtypes showed distinct Kaplan–Meier overall survival in TCGA-LAML (median 2.25 vs 1.00 years; log-rank p = 0.018), recapitulated in GSE106291 by nearest-centroid transfer (3.30 vs 1.24 years; p = 0.023), with hazard ratios (HRs) of 1.70, 1.45 and 1.48 across the three cohorts, although it did not reach significance in GSE71014. The ten-gene signature (CBR1, MX1, CCND3, LGALS1, CD52, PLA2G4A, LSP1, UCP2, HGF, SLC24A3-AS1) stratified training-set patients into high-/low-risk groups (log-rank p < 0.0001; 1/2/3-year area under the curve (AUC) 0.83/0.84/0.85) and remained an independent predictor of overall survival after adjusting for age and recurrent mutations (HR 2.67 per standard deviation (SD) of risk score, 95% confidence interval (CI) 1.89–3.77, p = 2.2×10⁻⁸). In external validation (nine of ten genes measured; SLC24A3-AS1 unavailable), the risk score predicted survival in GSE106291 (HR 1.25 per SD, 95% CI 1.06–1.48, p = 0.008) and separated risk groups in GSE71014 (log-rank p = 0.015; HR 1.36 per SD, 95% CI 0.99–1.87); a training/validation-swap sensitivity analysis yielded consistent results. A nomogram combining the risk score and age achieved a concordance index of 0.756. High-risk patients showed an inflamed microenvironment with high checkpoint expression and lower BCL2 expression. Conclusions: Lactylation-related gene expression is associated with molecular subtypes, immune contexture and outcome in AML. The proposed signature has independent prognostic value and reproduces, with modest effect sizes, in external cohorts; it is best regarded as a hypothesis-generating candidate biomarker for metabolism- and immune-directed research rather than a validated clinical tool. As a purely computational study, the findings warrant experimental and prospective validation.