Identification of fluid endocannabinoid and sphingolipid biomarkers in schizophrenia. A machine learning approach toward clinical value.

Abstract

Schizophrenia (SCZ) diagnosis relies on subjective clinical evaluation, with symptoms often overlapping between psychiatric conditions, hindering early intervention, and creating an urgent need for objective, accessible biomarkers. Lipids, especially endocannabinoid system (ECS) and sphingolipid metabolism members, are promising candidates given their implication in SCZ pathogenesis and measurability in accessible biofluids, with minimally invasive procedures. Our study conducted the first comprehensive crossover of endocannabinoids (eCBs) and sphingolipids in matched cerebrospinal fluid (CSF) and serum samples from antipsychotic-naïve SCZ patients (SCZ_naïve, n = 49), antipsychotic-treated patients (SCZ_treat, n = 56), and healthy volunteers (HVs, n = 51). The machine learning Integrated Disease Explanation and Risk Scoring (IDEARS) platform was applied to eCBs, N-acylethanolamines, ceramides, and sphingomyelins quantified using liquid chromatography-tandem mass spectrometry, identifying key discriminatory features across biofluids and conditions (CSF AUC: SCZ_naïve = 0.87, SCZ_treat = 0.94; serum AUC: SCZ_naïve = 0.87, SCZ_treat = 0.90). Anandamide emerged as the top CSF feature in SCZ_naïve patients, showing significantly elevated levels (q = 0.010) compared to HVs. Notably, though not statistically significant under conventional statistics, IDEARS identified anandamide as a top ten serum feature. Additionally, elevated sphingadiene backbone (d18:2) sphingolipids were conserved between the CSF and serum of SCZ_naïve (q < 0.05), highlighting their potential as peripheral surrogate markers. Treatment-specific ceramide patterns were also evident, with elevated C16-C18 and reduced long-chain (>C24) species potentially reflecting antipsychotic metabolic effects. Our results support the involvement of the ECS and sphingolipid imbalances in SCZ, highlighting the potential of machine learning for novel biomarker discovery, including stratification across biofluids, and therapeutic monitoring toward more objective psychiatric care.