A predictive metabolomic model for FLT3 and NPM1 mutations in Acute Myeloid Leukemia patients

Abstract

Cytogenetic abnormalities and gene mutations are essential for planning AML treatment. However, in Turkey, test results typically take 14–30 days. This delay emphasizes a critical need for rapid methods to deliver clinical data in urgent cases requiring immediate treatment decisions. To address this need, our objective was to develop a quick prediction method for NPM1 (Nucleophosmin-1) and FLT3 (FMS-like tyrosine kinase 3) mutations using LC-MS/MS (Liquid Chromatography-Tandem Mass Spectrometry) targeted metabolomics to detect these common and clinically important mutations in de novo AML patients (n = 42) through patient groups and a healthy group. We analyzed metabolic patterns using LC-MS/MS measurements of amino acids and acyl carnitines, key components critical to AML prognosis. The data were then subjected to multivariate analysis techniques. Principal Component Analysis (PCA) revealed that the model explained 79 % of the total variance among the sample groups. To further enhance class discrimination, we conducted Partial Least Squares-Discriminant Analysis (PLS-DA), resulting in R2Y and Q2 values of 0.845 and 0.619, respectively. Using the PLS-DA model, VIP (Variable Importance Projection) identified key metabolites with scores > 1.5, including C0 carnitine, glutamic acid, aspartic acid, tryptophan, histidine, isoleucine, and alpha-aminobutyric acid, respectively, highlighting their potential significance in distinguishing mutation groups. To ensure the validity of the PLS-DA model and evaluate potential overestimation, we validated the model using cross-validation and permutation test, demonstrating its robustness and reliability. Our preliminary model, developed through a targeted metabolomics approach, shows strong fit and predictive capability in determining the mutation status of NPM1 and FLT3 in AML patients.