Integrated proteomic and metabolomic profiling reveals immuno-metabolic signatures in antipsychotic-naïve first-episode psychosis

Background

First-episode psychosis is a key period for identifying biological alterations before prolonged illness and medication exposure obscure disease-associated signals. Peripheral biomarkers remain poorly validated, and it is unclear whether immune and metabolic abnormalities converge into reproducible molecular signatures in antipsychotic-naïve patients.

Aims

We aimed to identify serum and plasma molecular signatures distinguishing antipsychotic-naïve adults with first-episode psychosis from matched healthy control subjects, and to test whether integrated proteomic and metabolomic profiling reveals coordinated immuno-metabolic alterations.

Methods

The study included 66 adults with first-episode psychosis and 51 age-, sex-, and body mass index-matched healthy control subjects. Patients were recruited during hospitalization at first clinical contact for psychotic symptoms. Fasting blood samples were collected before antipsychotic treatment; therefore, no neuropsychopharmaceutical treatment had been started at sampling. Diagnoses followed International Classification of Diseases, Tenth Revision criteria and were confirmed by longitudinal follow-up. Targeted liquid chromatography–mass spectrometry quantified 206 serum molecules; after quality control, 54 patients and 32 control subjects remained. Targeted plasma proteomics provided 153 proteins after filtering, with 60 patients and 49 control subjects retained. Data were integrated with age, sex, body weight, body mass index, and waist circumference as covariates. Group differences were assessed using Mann–Whitney U tests, Benjamini–Hochberg false discovery rate correction, and effect sizes. Multivariate classification used random forest models and elastic net–penalized logistic regression across 50 repeated stratified 70/30 train-test splits. Elastic net stable predictors were features selected in at least 60% of iterations.

Results

Out of all measured molecular features, 90 differed significantly between patients and control subjects after false discovery rate correction, with effect sizes up to |r| = 0.72. Molecular models outperformed covariate-only models. The covariate-only random forest model showed limited discrimination: mean area under the receiver operating characteristic curve = 0.551; accuracy = 0.527, whereas random forest using the top 30 molecular features plus covariates achieved a mean area under the curve of 0.919 and accuracy of 0.849. Elastic net–penalized logistic regression showed the highest and most stable performance: mean area under the curve = 0.945 (+/- 0.026); accuracy = 0.852 (+/- 0.045). Stability selection yielded a 19-feature multi-omic panel. Epidermal growth factor, taurine, EN-RAGE (S100 calcium-binding protein A12), oncostatin M, and interleukin-4 contributed positively to patient classification, whereas FMS-like tyrosine kinase 3 ligand, tumour necrosis factor beta, cathepsin C, kynurenic acid, and multiple phosphatidylcholines contributed toward control classification. A random forest model restricted to this panel achieved a mean area under the curve of 0.957 and accuracy of 0.88.

Conclusions

Integrated proteomic and metabolomic profiling identified a reproducible immuno-metabolic signature in antipsychotic-naïve first-episode psychosis, characterized by immune activation, growth-factor-related signaling, amino-acid pathway changes, and reduced phosphatidylcholines. These findings support stability-guided multi-omic modelling for characterizing early psychosis biology. However, the cross-sectional single-cohort design, modest sample size, and absence of independent validation mean that clinical utility cannot yet be inferred. Future studies should validate this multi-omic signature in independent and longitudinal cohorts to determine its temporal stability and generalizability.

References

No references.