The therapeutic levodopa formulation Stalevo®, used in human Parkinson’s disease treatment, has raised concerns of illicit use in equine racing to enhance dopaminergic performance. Current regulatory detection relies on a 3-methoxytyramine threshold (4 μg/mL), which identifies levodopa administration but is limited in detecting micro-dosing, complex Parkinson’s medications, and suffers from a narrow detection window. This study investigated whether a metabolomic workflow, incorporating machine learning, could identify more robust urinary biomarkers to extend detection windows for levodopa misuse.
Samples analysed were obtained from a 12-horse administration (six mares, six geldings; two controls per sex), conducted by Charles Sturt University with ethics approval (A20277). Each treated horse received eight Stalevo® 100 tablets (800 mg levodopa, 200 mg carbidopa, 1600 mg entacapone) via nasogastric tube. Urine was collected daily 7 days prior administration, multiple hourly time points post-administration, and up to 7 days (20 samples total). Samples underwent protein precipitation, LC-QTOF-MS analysis (positive/negative ESI, DIA acquisition), and subsequent processing with MS-DIAL, MetaboAnalyst, and Python scripts. Four machine learning classification models, (random forest, PLS-DA, SVM, and XGBoost) were compared for best treated/non-treated group differentiation results.
Metabolomic screening identified a group of phase I and II conjugated metabolites as potential biomarkers with feature selection based on area under the curve (AUC), adjusted p-value, and fold-change. Among four tested classification models, the XGBoost algorithm provided the best predictive accuracy and F1 scores, successfully differentiating treated from untreated horses while extending the detection window.
This study demonstrates that a combined metabolomic and machine learning approach can identify novel biomarkers with the potential to extend the detection window of dopaminergic manipulation in racehorses. The open-source workflow is also readily adaptable to other doping scenarios, offering broad applicability. While further validation is required, this workflow represents a promising advancement for equine anti-doping surveillance.