From metabolomics to a clinically accessible four-gene surrogate for risk stratification in B-ALL

Abstract Adult B-cell acute lymphoblastic leukemia (B-ALL) is clinically heterogeneous, and established prognostic markers such as minimal residual disease (MRD) do not fully explain survival differences. Untargeted metabolomics may capture biologic states linked to outcome, but it is not routinely available in clinical practice. Here, we developed a plasma metabolite-based risk model for overall survival (OS) and translated it into a transcriptomic surrogate for RNA-only cohorts. Using newly diagnosed patients with plasma metabolomics and OS follow-up ( n = 18), we built a five-metabolite metabolic risk score (MRS) through bootstrap stability selection and elastic net Cox regression. In paired metabolomic-transcriptomic samples with OS data ( n = 11), we then trained a four-gene RNA surrogate and evaluated agreement in an independent paired validation set ( n = 6), before applying the model to an RNA cohort with OS follow-up ( n = 52). The MRS significantly stratified OS, including in non-transplanted and MRD-positive subsets. Higher MRS was associated with cellular remodeling and KRAS signaling, whereas lower MRS was linked to interferon-related pathways. The RNA surrogate remained independently associated with OS and showed higher discrimination than individual clinical predictors. These findings support metabolite-informed risk stratification in B-ALL and warrant external validation.