Molecular and micro-architectural mapping of gray matter alterations in psychosis

Abstract

The psychosis spectrum encompasses a heterogeneous range of clinical conditions associated with abnormal brain development. Detecting patterns of atypical neuroanatomical maturation across psychiatric disorders requires an interpretable metric standardized by age-, sex- and site-effect. The molecular and micro-architectural attributes that account for these deviations in brain structure from typical neurodevelopment are still unknown. Here, we aggregate structural magnetic resonance imaging data from 38,696 healthy controls (HC) and 1256 psychosis-related conditions, including first-degree relatives of schizophrenia (SCZ) and schizoaffective disorder (SAD) patients (n = 160), individuals who had psychotic experiences (n = 157), patients who experienced a first episode of psychosis (FEP, n = 352), and individuals with chronic SCZ or SAD (n = 587). Using a normative modeling approach, we generated centile scores for cortical gray matter (GM) phenotypes, identifying deviations in regional volumes below the expected trajectory for all conditions, with a greater impact on the clinically diagnosed ones, FEP and chronic. Additionally, we mapped 46 neurobiological features from healthy individuals (including neurotransmitters, cell types, layer thickness, microstructure, cortical expansion, and metabolism) to these abnormal centiles using a multivariate approach. Results revealed that neurobiological features were highly co-localized with centile deviations, where metabolism (e.g., cerebral metabolic rate of oxygen (CMRGlu) and cerebral blood flow (CBF)) and neurotransmitter concentrations (e.g., serotonin (5-HT) and acetylcholine (α₄β₂) receptors) showed the most consistent spatial overlap with abnormal GM trajectories. Taken together these findings shed light on the vulnerability factors that may underlie atypical brain maturation during different stages of psychosis.

Fig. 1. Analysis pipeline.

Centiles and effect sizes were computed from regional brain volumes, which were then predicted based on combinations of neurobiological features. A Deviation of regional volume from the median volume of neurotypical population for a single patient (red dot). Resulting ranked deviations, known as centiles, were computed for individuals with the same diagnosis for each brain region defined in the Desikan–Killiany atlas. Centiles range from 0 to 1, with values below 0.5 indicating that an individual has a lower GM volume than the expected normative values for their age and sex. B MRI data analyzed in the present study included eight different psychosis-related diagnoses clustered into four groups according to their clinical profile. C Effect sizes were computed as the Cohen’s d between regional centiles for each pair of groups. D Associations between neurobiological maps and empirical centiles (or their effect sizes) were conducted using PCA-CCA modeling which resulted in a set of predicted centiles (or effect sizes) derived from linear combinations of neurobiological features.

Fig. 2. Regional brain volume centiles.

Regional MRI brain volumes were converted into centiles and subsequently averaged across individuals to generate a mean centile map for each diagnosis and group. The highlighted regions show those regional centiles that exhibit significant differences from HC after FDR correction (Wilcoxon rank-sum test, P < 0.05).

Fig. 3. Effect sizes of centiles between groups.

Cohen’s d was computed between regional centiles of each pair of groups to map the effect sizes of centiles between conditions. The highlighted regions show those regional effect sizes that exhibit significant differences between groups after FDR correction (P_perm < 0.05). Top-right panel represents the sum of regional centile squared differences (SSD) between groups. Asterisks (*) indicate significant differences in SSD between groups (FDR-corrected P_perm < 0.05).

Fig. 4. Empirical and predicted centiles, and associated loadings from PCA-CCA models.

A Maps of empirical MRI-derived centiles (top) and predicted PCA-CCA-derived centiles from neurobiological features (bottom). B Correlation between empirical and predicted regional centiles. (C) PCA-CCA significant loadings associated to each neurobiological map (P_spin < 0.05). Non-significant models are denoted as n.s (FDR-corrected P_spin > 0.05). Error bars represent the standard deviation.

Fig. 5. Shared neurobiological features among psychosis-related groups.

A Stacked neurobiological loadings of each group, regardless of their significance, were ranked from the most negative to the most positive average contribution. B Neurobiological similarity matrix obtained by correlating the regional patterns of neurobiological features in HC (left). Structural co-vulnerability to psychosis matrix constructed by correlating the regional patterns of the effect sizes of centiles across psychosis-related groups (middle). Association between neurobiological similarity and structural co-vulnerability to psychosis (right).