Insula sub-regions across the psychosis spectrum: morphology and clinical correlates

The insula is a heterogeneous cortical region, comprised of three cytoarchitecturally distinct sub-regions (agranular, dysgranular, and granular), which traverse the anterior-posterior axis and are differentially involved in affective, cognitive, and somatosensory processing. Smaller insula volume is consistently reported in psychosis-spectrum disorders and is hypothesized to result, in part, from abnormal neurodevelopment. To better understand the regional and diagnostic specificity of insula abnormalities in psychosis, their developmental etiology, and clinical correlates, we characterized insula volume and morphology in a large group of adults with a psychotic disorder (schizophrenia spectrum, psychotic bipolar disorder) and a community-ascertained cohort of psychosis-spectrum youth (age 8–21). Insula volume and morphology (cortical thickness, gyrification, sulcal depth) were quantified from T1-weighted structural brain images using the Computational Anatomy Toolbox (CAT12). Healthy adults (n = 196), people with a psychotic disorder (n = 303), and 1368 individuals from the Philadelphia Neurodevelopmental Cohort (PNC) (381 typically developing (TD), 381 psychosis-spectrum (PS) youth, 606 youth with other psychopathology (OP)), were investigated. Insula volume was significantly reduced in adults with psychotic disorders and psychosis-spectrum youth, following an anterior-posterior gradient across granular sub-regions. Morphological abnormalities were limited to lower gyrification in psychotic disorders, which was specific to schizophrenia and associated with cognitive ability. Insula volume and thickness were associated with cognition, and positive and negative symptoms of psychosis. We conclude that smaller insula volume follows an anterior-posterior gradient in psychosis and confers a broad risk for psychosis-spectrum disorders. Reduced gyrification is specific to schizophrenia and may reflect altered prenatal development that contributes to cognitive impairment.

Youth with other psychopathologies were defined as those that had sub-threshold psychopathology symptoms-endorsed symptoms, frequency, and duration approximate with DSM-IV disorders or episode criteria, with significant distress (≥5) on the K-SADS-but did not meet psychosis criteria.
Typically developing youth were defined as those that did not meet criteria for either psychosis or other psychopathologies.
Of the 1601 participants with imaging data, 69 were excluded for serious medical conditions, 19 for a diagnosis of autism, 63 for insufficient clinical data to reach a diagnosis and 84 for poor scan and/or segmentation quality. A small number of individuals from each group were taking psychotropic medications (typically developing: n=16; psychosis spectrum: n=71; other psychopathologies: n=49).
In the PNC, cognitive ability was measured using the Penn Computerized Neurocognitive Battery (6), which consists of 14 tests covering five cognitive domains: executive function, episodic memory, social cognition, complex cognition, and sensorimotor ability. A general cognitive ability score was created by averaging accuracy z-scores across the four non-motor domains.

PNC Sensitivity Analysis
Differences in age, sex and race were notable between the three groups. In order to determine the impact of these variables on our significant results, we ran a sensitivity analysis by creating sub-samples of the psychosis spectrum and typically developing youth matched for age, sex and race. Matching was conducted using the MatchIt package (Version 3.0.2) in R (Version 3.6.1; R Core Team, 2019), using the 'exact' matching method, which matches each member of the typically developing youth with any psychosis spectrum youth that have the same values for age, sex and race. Race was considered as a binary variable, separated into Caucasian and non-Caucasian youth. This resulted in a final sample of 143 psychosis spectrum and 120 typically developing youth. See Table S10 for demographics.

MRI Acquisition
Psychosis cohort: Neuroimaging data were acquired on two identical 3T Philips Intera Achieva scanners (32 channel received head coil, single-band imaging) located at the Vanderbilt Institute for Imaging Sciences (VUIIS). In all three studies, a high resolution T1-weighted structural scans were collected with a 3D T1 fast field echo sequence with 1 mm 3

Voxel-Based Morphometry Quality Assurance
Segmented gray matter images were modulated by the normalization factors to preserve the original gray matter volume. CAT12 provides a quantitative measure of image quality that evaluates image parameters such as noise, inhomogeneities and image resolution. This measure is placed on a rating scale and assigned a letter grade, with any scan rated B-or above considered good quality and scans rated C-to C+ considered satisfactory. In our data, scans rated C+ or below were further visually inspected, blind to group membership, for gray matter segmentation quality. Scans with gray matter segmentation that did not capture the gray matter, or included non-gray matter voxels (e.g. skull), were excluded from further analysis.
Additionally, the modulated, normalized gray matter segmentations were checked using the CAT12 automated quality check protocol, which checks image inhomogeneity, defined as the mean correlation between gray matter volumes, with higher correlation indicating greater homogeneity. Flagged images were visually inspected, and segmented images with significant inhomogeneity were excluded from further analysis. The CAT12 pipeline was containerized using Singularity, built at SingularityHub (7) https://singularity-hub.org), and executed using the Vanderbilt University Institute of Imaging Science XNAT infrastructure (8).

Surface-Based Morphometry
Project-based thickness (PBT) uses a tissue segmentation to estimate white matter distance and then projects the local maxima onto gray matter voxels using a neighboring relationship (10). Topological defects of the surface reconstruction are corrected using a method based on spherical harmonics, in which the original MRI intensity values are used to identify and "fill" or "cut" each topological defect. These defects are patched using a low-pass filtered alternative reconstruction (12). The number of topological defects identified within-subject is used as a covariate in all surface-based analyses, following the removal of subjects with z>3.0 defects. To enable inter-subject analysis, a spherical map of a cortical surface is used to reparameterize the surface mesh into a common coordinate system (13). Finally, spherical registration is conducted by applying a multi-grid approach that uses reparameterized values of sulcal depth and shape index defined on the sphere (14).
Regarding the SBM measures calculated in CAT12, cortical thickness estimates the thickness of the gray matter surface. Gyrification estimates the cortical folding within a region of interest based on absolute mean curvature, with greater cortical folding contributing to greater local gyrification. Square-root transformed values of sulcal depth were also estimated in CAT12 based on the Euclidean distance between the central surface and its convex hull.

Insula sub-region definition
Insula volume was calculated for each sub-region based on masks provided by Farb and colleagues (9). These masks include hand-drawn insula parcellations based on wellcharacterized cytoarchitectonic divisions, which comprised the agranular, dysgranular and granular sub-regions. The mask was defined on a high-resolution T1-weighted template, based on well-characterized cytoarchitectonic divisions (15). Insula surface metrics were calculated based on the HCP-MMP1 atlas based on the guidance by Glasser (11). The HCP-MMP1 subregion designations were as follows: Agranular = AVI & AAIC; Dysgranular = Pol1, Pol2, & MI; Granular = Ig.

Insula Sub-Region Differences
Prior to analyzing group differences, we characterized differences in morphology (thickness, gyrification, sulcal depth) across insula sub-regions. Volume was excluded from this analysis due to differences in the sizes of the a-priori insula ROIs. A significant main effect of region was observed for all SBM metrics, in both the psychosis and PNC cohorts (all p<.001). The pattern of regional differences for each metric was the same across cohorts. Thickness: dysgranular>agranular>granular; Gyrification: granular>agranular>dysgranular; Sulcal Depth: granular>dysgranular>agranular. Interestingly, all of the patterns differed from one another, highlighting their relative independence in indexing different aspects of insula structure (full data in Table S1).

Illness Stage Effects
To determine the impact of illness stage on insula volume, patients were divided into early psychosis (illness duration ≤ 2 years; N=146) and chronic psychosis (illness duration >2 years; N=147).

Diagnostic Specificity (Schizophrenia-Spectrum vs. Psychotic Bipolar Disorder)
Volume: There was a main effect of group (F(2,492)=11.90, p<.001) and a main effect of region (F(2,984)=13.522, p<.001), but no significant group by region interaction (F(4,984)=1.83, p=.121). Overall insula volume was significantly smaller in schizophrenia participants compared to healthy (5.75% smaller, p<.001) and bipolar participants (3.97% smaller, p=.01). Bipolar participants did not significantly differ from healthy participants (1.86% smaller, p=.241). This pattern held across all three sub-areas, with schizophrenia participants demonstrating significant reductions in insula volume compared to both groups. Healthy and bipolar participants were not significantly different from one another.

Sensitivity Analysis for PNC volume results
To further mitigate the potential impact of demographic variables (age, gender, race) on the volume findings, we conducted a sensitivity analysis on a smaller group of PS (N=143) and TD (N=126) youth matched on demographic features using the MatchIt package in R (16) ( Table   S10). While the group differences were no longer statistically significant in the much smaller samples (F(1,226)

Age Effect Analyses
Age effects were examined in separate linear regression models with insula sub-area structural metrics as the outcome variables, age and age by group interactions as the predictors. Sex, TIV, topography defects (surface-metrics) and project (psychosis cohort) were included as covariates. Age effects are visualized in Figures S2 and S3.
Psychosis Cohort: Volume and sulcal depth were strongly negatively associated with age for all insula sub-areas. Results were more variable for cortical thickness and gyrification (full results presented in Table S7). No significant group by age interactions were observed for any metrics except dysgranular insula gyrification (p=.002). Investigation into this interaction revealed a significant positive association between age and dysgranular gyrification in healthy controls (r=.19, p=.009) that was not observed in psychosis patients (r=-.06, p=.303).
PNC Cohort: Volume and sulcal depth were negatively associated with age for all insula subregions, similar to the psychosis cohort, with more variable associations for thickness and gyrification (Table S8). The only significant group by age interaction across all metrics was between OP and TD youth for granular gyrification. In the TD youth, age and granular gyrification were negatively associated (r=-.10, p=.042) while they were positively associated in OP youth (r=.11, p=.005).

Sex Effects Analysis
Main effects and group by sex interactions were largely non-significant within both cohorts (Tables S7-S8). In the psychosis cohort, women had slightly greater granular gyrification than men (p=.003). In the PNC, boys had slightly greater dysgranular volume (p=.007), dysgranular thickness (p<.001), and agranular gyrification (p=.001) than girls.