Multi-omics reveal critical roles of phosphatidylcholine and sphingomyelin in antipsychotic efficacy for schizophrenia

Abstract

Nearly 30% of patients with schizophrenia respond inadequately to current antipsychotics, with unclear markers and mechanisms of antipsychotic efficacy. A total of 208 patients with schizophrenia treated for 6 weeks with oral paliperidone were analyzed through genotyping, mass spectrometry proteomic, and metabolomic profiling to explore underlying markers and mechanisms of antipsychotic efficacy. Machine learning analysis identified 20 proteins and 20 metabolites at baseline predictive of treatment response. Proteomic and metabolomic models achieved a cross-site mean AUC of 0.923 and 0.816, respectively. A multi-omics ensemble model achieved 0.941. GWAS and differential analyses identified 32 loci (P < 5 × 10^-5), 83 proteins, and 31 metabolites associated with efficacy (P < 0.05). Trans-omics analysis of these efficacy-related molecules across three omic layers highlighted glycerophospholipid metabolism (P = 3.25 × 10^-5) and sphingolipid metabolism (P = 0.039). Key molecules within these pathways exhibited a consistent direction of effect in regulating phosphatidylcholine (PC) and sphingomyelin (SM) metabolism, and higher PC and SM levels were found to correlate with better efficacy. These associations were further genetically validated using polygenic risk scores in two independent cohorts (2281 and 449 patients, respectively). In conclusion, multi-omics modeling is able to accurately identify antipsychotic efficacy, and higher PC and SM levels correlate with better antipsychotic efficacy, suggesting that variations in phospholipid metabolism may underlie the response to antipsychotics.

Fig. 1

Flow diagram and flowchart. a Flow diagram of the multi-omics cohort. b Flowchart of the study. First, in the multi-omics cohort, subjects treated with 6-week paliperidone were divided into treatment responder and non-responder groups. Baseline genomic, proteomic, and metabolomic markers associated with antipsychotic efficacy were then explored using GWAS, differential analysis, and machine learning. Multi-omics enrichment was employed to investigate biological pathways affecting the response to antipsychotics. Subsequently, detailed analysis of these pathways identified key molecules related to antipsychotic efficacy. Finally, the association between these key molecules and treatment efficacy was validated in external cohorts by plasma and brain metabolite PRS. GWAS genome-wide association study, PC phosphatidylcholine, PRS polygenic risk score, SM sphingomyelin

Fig. 2

Genomic, proteomic, and metabolomic analyses identified molecules related to antipsychotic efficacy, with multi-omics enrichment highlighting glycerophospholipid and sphingolipid metabolism pathways, including PC and SM as key molecules. a LASSO regression identified proteins and metabolites associated with treatment efficacy. b ROC curve for cross-site validation of proteomic, metabolomic, and stacking models. c Decision curve analysis plot of the stacking model. d Differentially expressed proteins and metabolites between TR and NTR. e Manhattan and Q-Q plots of the GWAS, and RAB7A brain expression based on eQTL summary data. GWAS genome-wide association study, LASSO least absolute shrinkage and selection operator, NEG negative effect on treatment response, NTR non-responders, PC phosphatidylcholine, PC-O ether-linked phosphatidylcholine, PE phosphatidylethanolamine, POS positive effect on treatment response, SM sphingomyelin, TR treatment responders

Fig. 3

Trans-omics analysis highlighted glycerophospholipid and sphingolipid metabolism pathways involved in antipsychotic efficacy. a KEGG analysis based on the GPMI network. Blue bars and borders indicate glycerophospholipid-related pathways, and red bars and borders indicate sphingolipid pathways. b GPMI network plot. Blue circles represent genes, green represents proteins, and yellow represents metabolites. Solid small circles indicate the genes, proteins, and metabolites associated with treatment response identified in this study. Solid large circles indicate the genes, proteins, and metabolites associated with treatment response identified in this study and also in the phospholipid pathway. Translucent small circles indicate genes, proteins, and metabolites predicted based on Protein–protein interactions and KEGG databases. Solid connections indicate those generated by the OmicsNet database. Dashed connections indicate those generated by previous literature. Red borders and connections indicate Sphingolipid metabolism and signaling pathways. Blue borders and connections indicate Glycerophospholipid metabolism pathways. GPMI Gene-Protein-Metabolite Interaction, KEGG Kyoto Encyclopedia of Genes and Genomes

Fig. 4

Genetic validation of PC and SM using plasma and brain metabolite PRS in independent datasets. a Plasma phospholipids PRS between TR and NTR. Data are shown as mean ± SD (n = 2281). b Brain phospholipids PRS between TR and NTR. Data are shown as mean ± SD (n = 449). Brain phospholipids PRS for DLPFC-Parietal-Temporal and DLPFC-Temporal cortices are shown separately. DLPFC dorsolateral prefrontal cortex, PC phosphatidylcholine, PRS polygenic risk score, SM sphingomyelin. *P < 0.05

Fig. 5

Hypothesized mechanisms for differential phospholipid metabolism and response to antipsychotics. a Potential difference in the hydrolysis of phosphatidylcholine and sphingomyelin recycling between treatment responders and non-responders. Red arrows indicate increases, blue arrows indicate decreases. Large flow in the Sankey diagram represents increased recycling, and small flow indicates decreased recycling. b Hypothetical molecular mechanism of phosphatidylcholine and sphingomyelin metabolism differences with varied antipsychotic responses. For non-responders, the upregulation of phosphatidylcholine hydrolysis and the decrease in sphingomyelin recycling may lead to potential neuroinflammation and impaired neurotransmission, potentially related to poor antipsychotic response. And treatment responders change in reverse. Red text indicates genes, proteins, and metabolites associated with treatment response identified in this study. Red background indicates the critical steps in phosphatidylcholine hydrolysis, and blue background indicates critical steps in sphingomyelin recycling. ASMase acid sphingomyelinase, C1P ceramide 1-phosphate, Cer ceramide, CERT ceramide transfer protein, ER endoplasmic reticulum, Golgi golgi apparatus, GSL glycosphingolipid, LPC lysophosphatidylcholine, PC phosphatidylcholine, PM plasma membrane, SM sphingomyelin