Unsupervised Clustering of ACC and Hippocampal Metabolites Identifies Subgroups in Antipsychotic-Naïve First-Episode...

Glutamatergic abnormalities are among the most consistent neurometabolic findings in psychosis spectrum disorders, but individual variability suggests biologically distinct subgroups. Proton magnetic resonance spectroscopy (1H-MRS) enables in vivo quantification of glutamate (Glu) and related metabolites but has rarely been applied to subgroup discovery in early psychosis. We hypothesized that metabolite-based clustering of hippocampal and anterior cingulate cortex (ACC) measures would identify biologically distinct subtypes differing in symptom severity and cognition.

Single-voxel 1H-MRS data were acquired from the left hippocampus and ACC in 90 antipsychotic-naïve first-episode psychosis (FEP) patients. Metabolites included Glu, composite creatine, composite choline, and N-acetylaspartate, quantified in both regions (eight total measures). K-means clustering was applied to these variables. Classification trees identified the most discriminative features, guiding simplified two-dimens-ional (2D) clustering models. Subgroup differences in demographics, symptoms, and cognition were evaluated using ANOVAs.

Clustering of all eight metabolites revealed two stable subgroups differing in baseline positive symptoms but not cognition or treatment response. Across models, classification trees consistently identified hippocampal Glu as the primary split, with secondary variables varying by method. These pairings guided three 2D clustering models, of which the hippocampal Glu + ACC Cr solution yielded four metabolically distinct subgroups. One subgroup, characterized by intermediate hippocampal Glu and low ACC Cr, showed significantly better cognitive performance after adjusting for age and sex.

Data-driven clustering of hippocampal and ACC metabolites identified subgroups with distinct cognitive profiles in antipsychotic-naïve FEP, supporting 1H-MRS-based biotyping in early psychosis.