Altered Cortical Gyrification Is Associated with Cortical and White Matter Microstructure in Individuals with Early Psychosis: A Multimodal Neuroimaging Study

Abstract

Background and Hypothesis

Deviated brain folding patterns in psychosis may reflect dysfunctional brain development, more specifically cortical growth and white matter myelination. However, mechanisms that underlie the cortical folding disturbances and their relationship to clinical phenotypes in psychosis remain unknown.

Study Design

This cross-sectional multimodal neuroimaging study combines structural and diffusion magnetic resonance imaging (MRI) in 203 individuals with early psychosis and 73 healthy individuals. The primary goal is to use local gyrification index (LGI), which quantifies cortical folding, to assess possible cortical abnormalities in subjects with early psychosis. The secondary goal is to combine gyrification measures with advanced diffusion MRI indices, namely, heterogeneity of fractional anisotropy (FA), a proxy for cortical cytoarchitecture, and free-water corrected FA, the latter associated with the tissue-related white matter microstructure. Finally, the relationship between these imaging measures with psychiatric symptoms and cognitive abilities are examined.

Study Results

Compared to healthy individuals, the early psychosis group exhibited a significant LGI reduction, potentially reflecting sulcal widening and shallowing in the left temporo-parietal region. Local gyrification index was significantly associated with heterogeneity of FA in the corresponding region. Lower LGI was also associated with reduced FA in the white matter tracts connecting this region to the frontal lobe. Finally, lower LGI at the temporo-parietal junction was associated with more severe negative symptoms.

Conclusions

We present the first in vivo evidence that anomalies in cortical folding may be linked to alterations in both the underlying cortical cytoarchitecture and associated microstructure of ipsilateral long-range white matter connections in psychosis.

Introduction

Psychotic disorders, encompassing schizophrenia spectrum disorders and affective disorders with psychosis, represent psychiatric conditions characterized by hallucinations and delusions.¹ Some individuals with psychotic disorders, especially schizophrenia spectrum disorders, also exhibit negative symptoms and cognitive impairments, potentially leading to generalized functional deterioration.^2,3 Early-life brain disruptions may heighten vulnerability to psychosis, with subsequent aberrant brain maturation potentially precipitating its onset.^4-6

During early brain development, the cortical surface folds into gyri and sulci thereafter continuously transforming its sulcal–gyral pattern.^7,8 This process is guided by both local cortical growth and mechanical tension of the growing myelinated axons.⁹ The resulting gyrification pattern can be quantified using magnetic resonance imaging (MRI) and the gyrification index, a metric that quantifies the amount of cortex buried within the sulcal folds, relative to the amount of cortex on the exterior surface of the brain.¹⁰

To date, several MRI studies have reported anomalies in the gyrification pattern in psychotic disorders. One earlier study also showed sulco-gyral pattern anomalies in temporal lobe regions in individuals with chronic schizophrenia.¹¹ While these findings are inconsistent, most studies report an increased complexity of brain folding (ie, increased gyrification index) particularly in frontal and temporal regions during early course of psychosis, including first-episode schizophrenia^12,13 and at-risk mental state.¹⁴ Notably, there is a decreased complexity of brain folding (ie, decreased gyrification index) in later stages of the illness affecting several cortical regions (eg, medial parietal cortex).^15-17 In this context, decreased gyrification index can be viewed as possibly reflecting concurrent sulcal widening and diminished sulcal depth.⁷

Cortical folding is guided during brain development by the growth and expansion of the cortical layers, including cell differentiation, migration, synapse formation, and the mechanical tension of axons that involves the pushing and pulling of the cortical ribbon by myelinating axonal bundles.¹⁸ Indeed, changes in neuronal size, neuronal density, and neuropil reductions are reported in vitro in schizophrenia,¹⁹ and are linked to cortical neurodevelopmental pathology.²⁰ In addition, abnormalities in myelin-related genes²¹ and changes in the density of interstitial white matter neurons²² in post-mortem brains of individuals with schizophrenia support evidence for possible neurodevelopmental white matter pathology. However, there is limited evidence whether gyrification changes are aligned with cortical or white matter neurodevelopmental disorders, namely, disorganized cortical layers and poorly myelinated or coherent white matter tracts. Few multimodal studies have examined the association between LGI and conventional diffusion MRI metrics in psychiatric disorders,^23-25 failing to demonstrate any association with fractional anisotropy (FA), a primary measure of white matter microstructure.

In the current multimodal neuroimaging study, we measure the local gyrification index (LGI), a measure of the degree of cortical folding which has a methodological advantage over other gyrification indices because it takes into account the inherent three-dimensional nature of the cortical surface²⁶ in individuals with psychotic disorders. The LGI is calculated as the ratio of the surface area of the cerebral cortex, comprising folds and grooves to the surface area of a virtual smooth layer enveloping the cerebral cortex in the corresponding brain regions.²⁶ We also examined the microstructure of both underlying gray matter and associated white matter connections in the same sample. For estimation of gray matter microstructure, we used heterogeneity of fractional anisotropy (HFA), which is sensitive to variation in the microstructural characteristics of gray matter tissue.²⁷ To assess white matter microstructure, free-water corrected tractography was applied.²⁸ We used structural and diffusion MRI scans from the Human Connectome Project for Early Psychosis (HCP-EP) dataset,²⁹ which was collected using state-of-the-art HCP imaging protocols. Our aim in the current study was to explore the relationship between gyrification and cortical and white matter pathology in this population. This study represents the first to examine the relationship between LGI and advanced diffusion MRI indices derived from free-water imaging²⁸ in psychiatric disorders. We hypothesized that altered LGI would be associated with disruptions in HFA²⁷ and free-water corrected diffusivity indices,³⁰ which serve, respectively, as in vivo indicators of cortical cytoarchitecture and white matter integrity in psychosis.

Methods

Below, we first describe the recruitment criteria and demographic characteristics of the participants, along with a description of the clinical symptoms and cognitive measures used in this study. Next, we outline the MRI acquisition and preprocessing steps, including the diffusion MRI harmonization. This is then followed by a description of the LGI calculation and the identification of the brain region of interest (ROI). The LGI calculation is performed using a surface-based analysis that evaluates LGI at each vertex across the cortical surface model. The brain region of interest consists of all vertices showing significant LGI differences between early psychosis and controls and will henceforth be referred to as the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex).

This is then followed by an explanation of the HFA measurements in the corresponding brain region and the extraction of free-water corrected diffusivity indices in the fiber tracts originating or terminating at the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex). The former served as an index of cortical cytoarchitecture, while the latter reflects white matter integrity. Finally, we describe the statistical methods employed to examine group difference in LGI, and we investigated the associations between LGI and HFA, as well as LGI and other diffusivity indices. Finally, we explored the associations of LGI, HFA, and diffusivity indices with clinical and cognitive variables. As noted previously, detailed information regarding the participants, assessments, and imaging procedures is published in the HCP-EP protocol paper,²⁹ although some of the more detailed descriptions have been relocated to the Supplementary Materials.

Participants

The total sample for the Human Connectome Project for Early Psychosis (HCP-EP) included 223 individuals with early psychosis and 80 control subjects for a total of 303 subjects. The early psychosis participants were from 4 specialized clinical sites: Beth Israel Deaconess Medical Center (BIDMC), McLean Hospital, Massachusetts General Hospital (MGH), and Indiana University (IU). Enrollment occurred from August 2016 to April 2022.

Clinical and cognitive assessments were conducted by trained raters under the supervision of licensed psychiatrists or clinical psychologists. This transdiagnostic study utilized the Structured Clinical Interview for DSM-5 Research Version (SCID-5-RV)³¹ and medical records to establish Diagnostic and Statistical Manual of Mental Disorders (DSM)-5 diagnoses.¹ Based on these diagnoses, 136 participants with non-affective psychosis (eg, schizophrenia, schizoaffective disorder, schizophreniform disorder, brief psychotic disorder, delusional disorder, or unspecified psychosis) and 67 participants with affective psychosis (eg, major depressive disorder with psychosis or bipolar disorder with psychosis) were identified. Both subgroups were restricted to individuals with an illness duration of less than 5 years (mean duration = 1.9 years, standard deviation = 1.4 years). Eighty healthy controls were recruited from the same sites via locally approved advertisements, online platforms, and referrals, and matched for age, sex, and parental socioeconomic status to the HCP-EP sample at each site. The inclusion and exclusion criteria for study participants are detailed in Supplementary Method S1.

The study received ethical approval from the coordination center at Brigham and Women’s Hospital (BWH) via the Partners Institutional Review Board for the Boston area sites (BIDMC, McLean, and MGH). In addition, ethical approval was obtained from the Indiana University Institutional Review Board.

Demographic, Clinical, and Cognitive Assessments

Demographic data, including parental socioeconomic status,³² chlorpromazine-equivalent doses,^33-36 and full-scale intelligence quotient (IQ),³⁷ were collected. Clinical symptoms, encompassing positive, negative, and general domains, were evaluated using the Positive and Negative Syndrome Scale (PANSS).^38,39 Cognitive function was assessed through 7 NIH Toolbox tests,^40-42 with composite scores for fluid and crystallized cognition derived from age- and sex-adjusted T-scores.⁴³ Detailed information on the demographic, clinical, and cognitive assessments is provided in Supplementary Method S2.

Image Acquisition and Preprocessing

Imaging data for all Boston area sites were initially collected at BWH, with McLean added as an additional MRI site in April 2018. Indiana University has consistently conducted scans at its facility. All imaging data were acquired using Siemens MAGNETOM Prisma 3 T scanners (Siemens, Erlangen, Germany). At BWH and IU, scanners were equipped with 32-channel head coils, while the McLean scanner featured a 64-channel head and neck coil. The details of the imaging protocols are provided in Supplementary Method S3.

Our in-house pipeline, available on GitHub (https://github.com/pnlbwh/luigi-pnlpipe), integrates distinct processing modules for imaging data analysis.

For structural data, raw DICOM files for T1-weighted (T1w) and T2-weighted (T2w) scans were obtained from 289 participants (210 early psychosis participants and 79 controls), converted to NIFTI format, and then aligned to standard axes and centered. Brain masks for T2w images were generated using the high-definition brain extraction tool (HD-BET),⁴⁴ followed by visual quality inspection and manual editing when necessary. These masks were registered to T1w images using rigid registration of Advanced Normalization Tools (ANTs). The masked T1w and T2w images were then corrected for bias fields.

Diffusion weighted images were collected from 286 participants, converted from DICOM to NIFTI format, aligned, centered, and subjected to semi-automated quality control. The processed diffusion data underwent Gibbs unringing to remove ringing artifacts,⁴⁵ brain masking, the process of extracting the brain from the skull, was performed using a convolutional neural network-based approach.⁴⁶ In addition, corrections for motion artifacts and eddy currents were conducted using the Functional Magnetic Resonance Imaging of the Brain Software Library.⁴⁷ Echo planar imaging distortion was also corrected to mitigate intensity loss and voxel displacement caused by magnetic field inhomogeneity and interactions.⁴⁸

Harmonization of the diffusion signals at b = 1500 s/mm² across the 3 scanners was conducted to reduce scanner-related bias, using our retrospective method.^49,50 For this retrospective method, BWH was designated as the reference site, with McLean and IU as target sites. We first identified the scanner-related differences between the reference and target sites by generating rotation-invariant spherical harmonics feature templates from a subset of age- and sex-matched healthy controls (10-20 subjects per site). Subsequently, we applied the learned scanner-related differences to those templates to harmonize the diffusion data in all participants across the sites. Seven participants, including 3 individuals with early psychosis and 4 healthy controls, were excluded due to harmonization errors.

FreeSurfer Processing, LGI Calculation, and Designing Analysis Area

The masked T1w images were processed using FreeSurfer software (v7.1.0), employing a standardized automatic reconstruction pipeline.⁵¹ Subsequently, the preprocessed images were parcellated into 176 gray and white matter regions based on the Desikan–Killiany Atlas.⁵² To delineate the 4 cortical lobes (frontal, temporal, parietal, and occipital), corresponding gray matter regions defined by the Desikan–Killiany Atlas were merged.

As part of the standard FreeSurfer processing, three-dimensional cortical surface models comprising approximately 150 000 vertices were reconstructed. Using these surface models can resolve the issue of buried sulci, which causes inaccurate gyrification index measurements. The next step involved employing a matching algorithm to generate a smooth outer surface, which virtually envelops the exterior of the brain, aligning vertices on the outer surface with those on the cortical (inner) surface. Within this framework, spherical brain regions with a 25-mm radius were defined, and LGI maps were computed by calculating the ratio of the inner to outer surface area for each vertex. Inter-subject registration was then performed to enable vertex-level statistical group comparisons.²⁶

Once LGI was calculated, vertex-wise group differences were tested (see “Statistical Analysis”). The brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) was used for the subsequent correlational analyses with the advanced diffusivity indices (Figure 1). The analysis of group comparison in LGI was performed in the fsaverage (Montreal Neurological Institute 305; MNI305) space, which is commonly used to align individual cortical surfaces to a common template for group-level vertex-wise comparisons. The resulting brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) was then mapped to T1w space for each subject via the FreeSurfer function mri_label2label. This surface-based representation of brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) was then converted to a 3D volumetric representation using the FreeSurfer tools (mri_label2vol, mri_binarize, and mris_calc). An average LGI value was extracted by averaging the LGI values across all vertices within the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex). These LGI values from all participants (individuals with early psychosis and healthy controls) were harmonized jointly across the 3 scanning sites using the ComBat method,⁵³ with site treated as the batch variable and age and sex included as covariates. Prior to harmonization, we confirmed that there were no significant site-specific differences in demographic characteristics (ie, age, sex, and education). This approach employs Bayesian estimation to account for site-specific differences and to reduce measurement bias introduced by differences among multi-site scanners.

Figure 1.

Schematic diagram showing how this multimodal neuroimaging study was conducted. Firstly, a full-brain tractography with free-water model was prepared. Next, the brain region that consists of all vertices showing significant LGI differences between early psychosis and controls was set as a brain ROI exhibiting LGI group differences (ie, left perisylvian cortex). Its surface-based representation in MNI305 space were transformed into 3D volumetric representation in DWI space for each subject. For the HFA part, average HFA within the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) was calculated. For the tractography part, tracts that originate or terminate at the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) and connect to 4 cortical lobes were extracted from a full-brain tractography using white matter query language. This extraction enables the calculation of the tract measures. Abbreviations: DWI, diffusion-weighted imaging; ROI, region of interest; HFA, heterogeneity of fractional anisotropy; LGI, local gyrification index; MNI, Montreal Neurological Institute; T1w, T1-weighted; UKF, unscented Kalman filter.

Finally, the volumetric brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) was registered from T1w to diffusion weighted imaging space through rigid ANTs registration. This allowed for the calculation of the average HFA value for each individual within the same brain region and the extraction of white matter fiber tracts that either originated from or terminated in the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex).

Estimation of Cortical Cytoarchitectural Heterogeneity

To estimate gray matter cytoarchitecture, we used a diffusion-derived measure known as HFA.²⁷ This measure was developed to address limitations of traditional gray matter cytoarchitecture measures, which are susceptible to voxel-by-voxel signal fluctuations in gray matter⁵⁴ and partial volume effects from extracellular components.⁵⁵ Heterogeneity of fractional anisotropy has been shown to be sensitive to variations in the cytoarchitectural characteristics of gray matter tissue within selected regions, possibly reflecting cellular composition and their spatial distribution within the cortical layers.⁵⁶

Fractional anisotropy maps corrected for free-water were calculated for each voxel by fitting the free-water model using the code in MATLAB R2017a (Mathworks, Natick, MA, USA).²⁸ Free-water correction was applied in the FA calculation to avoid partial volume effects from the extracellular component,⁵⁵ especially around the border between gray matter and cerebrospinal fluid. The FA maps generated for all subjects underwent rigorous visual quality control, leading to the exclusion of 8 subjects-3 individuals with early psychosis and 5 healthy controls-due to low-quality FA maps. The average HFA²⁷ within the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) was then calculated (Figure 1) in MATLAB R2017a using the following formula:

where N is the total number of voxels within the region of interests, and m is the FA value in a voxel, indexed by i or j.

Estimation of White Matter Microstructural Measures

To investigate white matter microstructural properties associated with altered cortical folding in psychosis, FA values were analyzed along white matter tracts connecting regions exhibiting LGI changes.

Given the balance between sensitivity to fiber structure and signal-to-noise ratio,⁵⁷ diffusion data with a b-value of 1500 s/mm² were employed for whole-brain tractography. The harmonized diffusion data were processed using the Unscented Kalman Filter (UKF) tractography algorithm, which enhances fiber tracking accuracy by incorporating information from prior steps during the tracking process.^58,59 A combined model was used, incorporating the 2-tensor model for detecting crossing fibers^60,61 and the free-water model to preserve accurate fiber orientation by reducing contamination from extracellular free-water.⁶² The UKF tractography reconstructed whole-brain white matter fiber bundles. Consistent UKF parameters were applied as follows: Voxels within the brain mask with FA values above 0.1 were seeded to exclude regions with low diffusion directionality, ensuring more reliable fiber tracking. Tracking was terminated when FA dropped below 0.08 or the normalized mean signal fell below 0.06. The normalized mean signal, calculated as the sum of normalized signals across all gradient directions for each voxel, reflects the overall quality and integrity of the diffusion data. To avoid bias from excessively short fibers, only fibers longer than 40 mm were retained.^63-65

White Matter Query Language (WMQL),⁶⁶ an automated tool for segmenting fiber tracts based on anatomically defined gray matter regions in any anatomical atlas, was employed to identify tracts originating or terminating in the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex). In this study, after co-registering T1w images with diffusion data,⁶⁷ we used the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) along with frontal, temporal, parietal, and occipital lobes defined by FreeSurfer as origin and termination points of the streamlines, applying additional exclusion criteria to eliminate stray fibers (Figure 1). Finally, free-water corrected diffusivity metrics were extracted for each tract, including FA (a representative measure of white matter integrity), mean diffusivity (MD; an overall measure of diffusion), axial diffusivity (AD; reflecting axonal integrity), and radial diffusivity (RD; assessing myelin integrity).³⁰ Combining these metrics facilitates a comprehensive assessment of white matter microstructure, including axonal degeneration and myelin injury.⁶⁸

Statistical Analysis

Demographic differences between the early psychosis and control groups were examined using a one-way analysis of variance (ANOVA) for continuous variables or the chi-square test for categorical variables (P < .05, 2-tailed) using SPSS version 27.0.1.0 (IBM Corp., Armonk, NY).

Local gyrification index was compared between groups of 203 individuals with early psychosis and 73 healthy controls who had both preprocessed T1w and diffusion weighted images and for whom LGI could be successfully calculated. Three participants, including 1 individual with early psychosis and 2 healthy controls, were excluded due to LGI calculation errors. Each vertex-wise LGI value was mapped onto the fsaverage space and smoothed at a 5-mm Gaussian kernel. To estimate group differences (early psychosis vs control; non-affective vs control; affective vs control; and non-affective vs affective) in LGI values, a general linear model adjusted for age, sex, and scanning site was performed using the mri_glmfit command. The FreeSurfer program enables vertex-wise analysis of cortical metrics, including LGI, providing statistical maps to identify spatially localized group differences. A cluster-wise correction for multiple comparisons was applied using precomputed Monte Carlo simulations derived from 10 000 iterations, with a vertex-wise threshold of P < .05 (2-tailed), to identify vertices with significant group differences.

SPSS version 27.0.1.0 was also employed to conduct the following correlation analyses. Correlations between the average LGI values within the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) and the advanced diffusivity indices were performed separately for the early psychosis and control groups. The relationship between the average LGI and HFA values at the corresponding brain region was examined using Spearman’s rank correlation coefficients after regressing out the effects of age and sex, setting a statistical threshold at P < .05 (2-tailed), as both values were not normally distributed by the Kolmogorov–Smirnov test. Furthermore, to determine whether the association between LGI and HFA was region-specific, average LGI and HFA values in the whole brain cortex as a control region were also calculated and subjected to Spearman’s rank correlation analyses.

Similarly, the association between the average LGI values and the average FA values of the 4 connections (ie, frontal, temporal, parietal, and occipital tract) were examined using Spearman’s rank correlation coefficients, after adjusting for age and sex. The statistical threshold was set to P < .013, reflecting Bonferroni correction for the 4 tracts. As a post hoc analysis of the significant association between LGI and FA, we examined the relationship between average LGI values and other diffusivity indices (ie, average AD, MD, and RD) within the connection, applying a statistical threshold of P < .013 (Bonferroni correction for the 4 indices).

If a significant relationship between the average LGI values and these advanced diffusivity indices was found in the early psychosis group, we performed supplementary analyses. Their associations were further examined within the early psychosis subgroups using the same statistical thresholds.

Given the significant association of LGI with HFA and frontal FA, we further explored the relationship between average LGI values and average HFA values, both within the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex), as well as the average FA values in the frontal connection with clinical symptoms and cognitive performance. Variables included PANSS positive symptom score (n = 152), PANSS negative symptom score (n = 150), PANSS general symptom score (n = 151), PANSS disorganization symptom score (n = 150), NIH Toolbox Fluid Cognition Composite score (n = 142), and NIH Toolbox Crystallized Cognition Composite score (n = 143). Multiple linear regression analyses were performed with each clinical or cognitive variable as the dependent variable and average LGI, HFA, and FA values in the frontal connection as independent variables. Age and sex were included as covariates for PANSS scales, along with the antipsychotic medication dose. The statistical threshold was set at P < .0083 after Bonferroni correction for the 6 variables (ie, PANSS positive, negative, general, and disorganization symptom scores, as well as Fluid and Crystallized Cognition Composite scores). PANSS disorganization symptom scores were calculated by summing specific PANSS subscales (P4, N5, G10, and G11) based on previous literature.⁶⁹ A post-hoc analysis was performed if significant associations were identified between LGI, HFA, and FA in the frontal connection and clinical or cognitive variables. If statistically significant associations were identified, further analyses were conducted on the relevant subscales, with Bonferroni-corrected thresholds adjusted for the number of items. Such post-hoc analyses were also performed in both non-affective and affective subgroups.

In addition, the association of the imaging measures (ie, LGI, HFA, and frontal FA) with medication (antipsychotic medication dose [n = 179]) was assessed using Spearman’s rank correlation coefficients, after controlling for age and sex, with the statistical threshold set at P < .017 after Bonferroni correction for the 3 imaging measures.

Moreover, the relationship between the average LGI values within the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) and the number of past psychiatric hospitalizations, which partially reflect relapse history,⁷⁰ was assessed using Spearman’s rank correlation coefficients, after controlling for age and sex, with the statistical threshold set at P < .05.

Results

Demographic Background

The demographic characteristics of the study participants are summarized in Table 1 and Supplementary Table S1. While the healthy control group had a higher education level than the early psychosis group, the parental socioeconomic level was comparable between the 2, which is expected based on social drift observed in those with psychosis compared to their family origin.⁷¹ There was also a group difference in handedness. The early psychosis group also had a significantly lower full-scale IQ and lower fluid cognition ability compared with the healthy control group.

Table 1.

Characteristics of Study Participants

	Healthy controls	Early psychosis	Test statistic	P value
Number of subjects (total), no. of subjects [n = 276]	73	203
Scanning site 1 (Indiana University), no. of subjects [n = 126]	23	103
Scanning site 2 (Brigham and Women’s Hospital), no. of subjects [n = 76]	33	43
Scanning site 3 (McLean Hospital), no. of subjects [n = 74]	17	57
Sex, male/female, no. [n = 276]	41/32	123/80	χ2(1) = 0.436	.509
Age, years, mean (SD) [n = 276]	23.9 (3.6)	23.1 (3.8)	F(1,275) = 2.491	.116
Education level, mean (SD)a [n = 276]	8.7 (2.1)	6.5 (2.2)	F(1,275) = 55.388	<.001
Parental socioeconomic level, mean (SD)b [n = 264]	2.1 (1.0)	2.3 (1.2)	F(1,263) = 2.070	.151
Handedness, right/left/ambidextrous, no.c [n = 275]	60/13/0	177/17/8	χ2(2) = 7.411	.025
Duration of illness, years, mean (SD)d [n = 200]		1.9 (1.4)
Antipsychotic medication dose, chlorpromazine equivalent mg/day, mean (SD)e [n = 180]		308.3 (314.3)
The number of past psychiatric hospitalization, mean (SD)f [n = 200]		2.1 (1.8)
WASI-II Full Scale IQ score estimate, mean (SD)g [n = 269]	116.0 (11.3)	105.5 (15.6)	F(1,268) = 27.011	<.001
PANSS
PANSS positive symptom score, mean (SD)h [n = 197]		14.2 (5.2)
PANSS negative symptom score, mean (SD)i [n = 194]		12.0 (5.3)
PANSS general symptom score, mean (SD)j [n = 196]		10.8 (3.0)
NIH Toolbox
Fluid Cognition Composite score, mean (SD)k [n = 246]	50.8 (10.0) [n = 65]	38.7 (11.4) [n = 181]	F(1,245) = 57.361	<.001
Crystallized Cognition Composite score, mean (SD)l [n = 248]	57.8 (10.6) [n = 65]	56.0 (11.7) [n = 183]	F(1,247) = 1.304	.255
Intracranial Volume, cm3, mean (SD)m [n = 276]	1040.9 (114.6)	1021.4 (121.6)	F(1,275) = 1.430	.233

Abbreviations: PANSS, Positive and Negative Syndrome Scale; WASI-II, Wechsler Abbreviated Scale of Intelligence-II.

^aEducation level is the highest education level completed by the participant: 1 = elementary school, 2 = middle/junior high school, 3 = high school, no degree, 4 = GED, 5 = high school, degree, 6 = vocational school, 7 = some university courses, 8 = associate’s degree, 9 = bachelor’s degree, 10 = some graduate level courses, 11 = master’s degree, 12 = doctorate level courses, 13 = doctorate degree.

^bData missing for 12 individuals. Using the Hollingshead Four-Factor Parental Socioeconomic Status Scale, parent socioeconomic level was assessed. Education and occupation scores of parents and guardians were weighted to produce a single score (1 being the greatest SES classification, 5 being the lowest SES classification).

^cData missing for an individual.

^dData missing for 3 individuals.

^eDifferent antipsychotic dosages were converted into chlorpromazine equivalents. Data available for 180 individuals.

^fData missing for 3 individuals.

^gData missing for 7 individuals.

^hData missing for 6 individuals.

ⁱData missing for 9 individuals.

^jData missing for 7 individuals.

^kData missing for 30 individuals.

^lData missing for 28 individuals.

^mFor group comparisons, an analysis of covariance with age as a covariate was performed.

Group Comparison of LGI

Compared to the healthy control group, the early psychosis group had a significantly lower LGI in the superior temporal gyrus, middle temporal gyrus, banks of the superior temporal sulcus, and supramarginal gyrus in the left hemisphere (Figure 2; Supplementary Table S2). No brain regions were identified where the early psychosis group had a significantly higher LGI compared to the healthy control group. The vertices with significant LGI group difference are aggregated into the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex), which is subsequently used in the following correlational analyses. The average LGI of the vertices within this ROI was significantly reduced in the early psychosis group relative to healthy controls (Figure 2; diagnosis effect: F(1, 275) = 5.110, P = .025, partial η² = 0.018). This LGI reduction persisted when handedness or current comorbid substance use disorder was included as a covariate (Supplementary Figures S1 and S2). The non-affective psychosis subgroup, compared with healthy controls, exhibited a significantly lower LGI in a similar left temporo-parietal region (Supplementary Figure S3). No statistically significant LGI differences were observed between the affective psychosis subgroups and healthy controls, as well as the non-affective and affective psychosis subgroups.

Figure 2.

Group comparison of lgi between the early psychosis and healthy control groups. (A) Cortical statistical maps displaying a reduced local gyrification index in the early psychosis group compared with the healthy control group. The maps are shown for the right and left hemispheres in lateral and medial views, respectively. The horizontal bar shows P values corrected for multiple comparisons. (B) Average LGI values of vertices within the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) for each group. Abbreviations: EP, early psychosis; HC, healthy controls; LGI, local gyrification index; ROI, region of interest.

Relationship between LGI and HFA

Average LGI values in the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) were positively correlated with average HFA values in the same brain region only in the early psychosis group, with statistical significance (r = 0.143, P = .044) (Figure 3; Supplementary Table S3). In the early psychosis subgroups, this positive correlation was observed only for the non-affective psychosis group (r = 0.274, P = .001) (Supplementary Table S3). However, average LGI values and average HFA in the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) did not show a significant correlation in the healthy control group (r = −0.103, P = .405) (Supplementary Table S3). In addition, there was no significant correlation between average LGI values and average HFA in the whole brain cortex used as a control region (early psychosis group: r = 0.089, P = .209; healthy control group: r = −0.009, P = .940). These findings suggest that the significant correlation between LGI and HFA is specific to early psychosis and to the regions exhibiting gyrification alterations.

Figure 3.

Relationships between lgi and advanced diffusivity indices in the early psychosis group. (A) Scatter plots for the relationship between LGI and HFA at the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) in the early psychosis group. (B) Scatter plots for the relationship between LGI in the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) and average FA of the frontal tracts in the early psychosis group. (C) Scatter plots for the relationship between LGI in the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) and average RD of the frontal tracts in the early psychosis group. Abbreviations: FA, fractional anisotropy; ROI, region of interest; HFA, heterogeneity of fractional anisotropy; LGI, local gyrification index; RD, radial diffusivity.

Relationship between LGI and Tract Measures

Lower LGI values in the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) were significantly associated with reduced FA values in the frontal tract, but only in the early psychosis group (r = 0.212, P = .005) (Figure 3; Supplementary Table S4). In a post-hoc analysis, lower LGI values in the same brain region were significantly associated with increased RD values in the frontal tract (r = −0.224, P = .003) (Figure 3; Supplementary Table S5).

In the early psychosis subgroups, a significant relationship between lower LGI values and reduced FA values in the frontal tract was seen only in the non-affective psychosis group (r = 0.261, P = .004) (Supplementary Table S4).

Relationships of LGI, HFA, and FA in the Frontal Tract with Clinical and Cognitive Variables

Multiple regression analyses showed a significant association between the average LGI values in the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) and PANSS negative symptom score (adjusted R² = 0.081, standardized β = −0.2340, T = −2.7820, P = .0061; semi-partial r = −0.219, sr² = 0.048) (Table 2). The association between average HFA values in the same brain regions and NIH Toolbox Fluid Cognition Composite score was detected (adjusted R² = 0.049, standardized β = −0.1810, T = −2.1570, P = .0330; semi-partial r = −0.177, sr² = 0.031), but it did not reach the Bonferroni-corrected significance level (Table 2).

Table 2.

Multiple Regression Analyses in the Early Psychosis Group: LGI, HFA, and FA in the Frontal Tract = > Clinical or Cognitive Variables

Clinical or cognitive variables		PANSS positive symptom score	PANSS negative symptom score	PANSS general symptom score	PANSS disorganization symptom score	NIH Toolbox Fluid Cognition Composite score	NIH Toolbox Crystallized Cognition Composite score
Adjusted R2		0.00700	0.08100	0.00700	0.00017	0.04900	0.01400
LGI	Standardized β	−0.0340	−0.2340	−0.0420	−0.0250	0.0260	0.1370
	T	−0.3900	−2.7820	−0.4840	−0.2870	0.3010	1.5660
	p	0.6970	0.0061**	0.6290	0.7750	0.7640	0.1200
HFA	Standardized β	0.0680	−0.0460	0.0350	0.0330	−0.1810	−0.0060
	T	0.8100	−0.5620	0.4120	0.3830	−2.1570	−0.0750
	p	0.4190	0.5750	0.6810	0.7020	0.0330*	0.9400
FA in the frontal tract	Standardized β	0.0150	0.0840	0.0590	0.0290	0.1480	−0.0290
	T	0.1740	1.0400	0.7000	0.3490	1.7380	−0.3300
	P	.8620	.3000	.4850	.7280	.0850	.7420

Abbreviations: FA, fractional anisotropy; HFA, heterogeneity of fractional anisotropy; LGI, local gyrification index; PANSS, Positive and Negative Syndrome Scale.

^* P < .05, uncorrected statistical significance level.

^** P < .0083, statistical significance level after Bonferroni correction for the 6 variables.

In the post-hoc analyses of early psychosis participants, the PANSS negative symptom subscale scores showed significant associations between average LGI values in the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) and blunted affect (adjusted R² = 0.0750, standardized β = −0.2310, T = −2.7670, P = .0060; semi-partial r = −0.216, sr² = 0.047) as well as lack of spontaneity and flow of conversation (adjusted R² = 0.0960, standardized β = −0.2480, T = −3.0040, P = .0030; semi-partial r = −0.232, sr² = 0.054) (Supplementary Table S6). Similar results were observed in both non-affective and affective subgroups (Supplementary Tables S7 and S8).

There were no significant relationships between average LGI values or average HFA values, either in the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex), nor in average FA values in the frontal tract, or antipsychotic medication dose (Supplementary Table S9).

Lower LGI values in the brain ROI exhibiting LGI group differences (ie, left perisylvian cortex) were significantly associated with a higher number of past psychiatric hospitalizations (r = −0.132, P = .030).

Discussion

Our results demonstrate alterations in the cortical folding patterns assessed by LGI. These alterations were localized to the left temporo-parietal region in early psychosis participants. Moreover, the temporo-parietal LGI changes were significantly associated with temporo-parietal altered HFA (ie, non-uniform tissue organization within the cortical layer). Furthermore, lower temporo-parietal LGI was associated with reduced FA and increased RD (ie, disturbed myelin integrity) in the white matter tracts that connect this region to the frontal lobe. Finally, decreased temporo-parietal LGI was related to more severe negative symptoms. These findings provide in vivo evidence of the interrelationships between alterations in cortical folding, cortical cytoarchitecture, and ipsilateral long-range white matter microstructure, which may be involved in the symptomatology of psychosis.

Specifically, the current study demonstrates lower LGI in the left temporo-parietal region in the early psychosis group compared with healthy controls. The lower LGI may correspond to both sulcal widening and diminished sulcal depth.⁷ Previous findings in schizophrenia reported both higher and lower LGI. In the first-episode schizophrenia group, the majority of studies report higher LGI.^12,13 Notably, our HCP-EP dataset had longer mean duration of illness (1.9 years) and a higher illness duration as defined by inclusion criteria (within 5 years of illness onset) than those in the previous study cited (mean = 0.9 years; maximum was within 1 year of illness onset except for first psychiatric hospitalization).¹³ Nonetheless, a lower LGI in schizophrenia has been reported in subjects with a similar period of illness duration to those of the HCP-EP dataset.^72,73 Interestingly, our study also revealed an association between the reduced LGI and increased number of past psychiatric hospitalization. Taking these observations into account, a progressive LGI reduction after illness onset^73,74 may partially explain the variability in gyrification findings in psychotic disorders, potentially reflecting cortical plasticity changes around the onset of psychosis. The LGI shifts from hypergyria to hypogyria along the schizophrenia time course may reflect remodeling or compensatory processes,⁷⁵ and may be analogous to cortical thickness changes previously reported.⁷⁶ Such LGI trajectories may be observed across broad neurodevelopmental disorders,¹⁶ including very preterm babies with cognitive deficits in their adulthood.⁷⁷ Another consideration is the transdiagnostic sample composition in the present dataset, given the patterns of gyrification deviations, as well as those of cortical and white matter deterioration, in schizophrenia and bipolar disorder might differ.^78,79 On the other hand, it is also noteworthy that other psychiatric conditions (eg, obsessive-compulsive disorder) do not report low LGI pattern in the same left temporo-parietal region.⁸⁰

The brain region exhibiting lower LGI encompassed the primary auditory cortex and temporo-parietal junction (TPJ). Similarly, an earlier study on schizophrenia reported sulco-gyral pattern anomalies in the temporal lobe,¹¹ supramarginal gyrus,⁸¹ and parietal operculum⁸² of the left hemisphere. The affected brain region was localized exclusively in the left hemisphere, possibly reflecting inherent vulnerability of language-related networks, predominantly centered on the left perisylvian cortex, to psychosis.⁸³ Functional brain changes in these brain regions have also been implicated in auditory hallucinations⁸⁴ and delusions of reference⁸⁵ in chronic schizophrenia cases. However, the present study found no significant association between altered cortical folding and positive symptoms. Given that psychotic symptoms fluctuate throughout the illness course⁸⁶ and specific auditory verbal hallucination scores were unavailable,⁸⁷ the absence of an association between LGI and positive symptoms must be interpreted with caution. Likewise, genuine biological heterogeneity within the psychosis spectrum⁸⁸ may have contributed to the lack of correlation, indicating that future validation through subtyping studies is warranted.

An abnormal cortical folding pattern in psychotic disorders, as reported here, may be underpinned by gray matter cytoarchitectural changes. The postmortem findings in individuals with schizophrenia report smaller pyramidal cells and fewer dendritic spines in layer III of frontal and temporal cortices.^89-91 In addition, transcriptomic differences in layer III pyramidal cells of the prefrontal cortex relative to healthy controls were minimal in bipolar disorder, yet pronounced in schizophrenia.⁹² Since our study demonstrates a significant association between LGI and HFA in the temporo-parietal region, this may indeed suggest that gyrification changes in psychosis, notably in non-affective cases, arise from pathological changes involved in the formation of cortical layers.

In addition to local cortical abnormalities, our study also demonstrates that the abnormal cortical folding observed especially in non-affective psychosis, might be related to white matter connections between the TPJ and the frontal lobe. These connections are composed primarily of the arcuate fasciculus⁹³ and superior longitudinal fasciculus II, which are involved in language processing and verbal working memory.^94,95 Given that the cortical layer III includes many long-range projection neurons,⁹⁶ the aforementioned alterations in pyramidal cell morphology^89,91 may implicate an involvement of fronto-temporal circuitry.^97,98 Specifically, the present finding of a significant association between lower LGI and reduced white matter microstructure could imply that fewer, or less myelinated/less coherent axons might lead to reduced cortical folding. Such a finding is also consistent with a tension-based theory,¹⁸ proposing that during brain development, the group of axons, by pulling together, generates mechanical force to induce folding of the cortical sheet.

Finally, more severe negative symptoms may be partly explained by the degree of diminished complexity of brain folding. Conversely, LGI reductions were partly accounted for by negative symptoms (Supplementary Table S10). Our findings are consistent with a prior study that also reports temporal LGI associations occur with negative symptoms in antipsychotic-naïve first-episode schizophrenia individuals.⁹⁹ Among negative symptoms, blunted affect has been particularly associated with frontal dysconnectivity.¹⁰⁰ Assuming that changes in gyrification and white matter microstructure reside on the same etiological pathway since they were interrelated, the fronto-temporal dysconnectivity may provide anatomical support for the disconnection hypothesis of psychosis,¹⁰¹ possibly being responsible for the type II syndrome,¹⁰² which is mainly composed of negative symptoms.^103,104 Accumulating neuroimaging evidence linking connectivity metrics to clinical manifestations, as exemplified by the present study, may facilitate the development of novel neuromodulation targets.¹⁰⁵ Indeed, recent transcranial direct current stimulation protocols targeting the TPJ-frontal circuit have been shown to alleviate overall psychopathology, including negative symptoms, in individuals with schizophrenia.¹⁰⁶

Limitations

A few limitations should, nonetheless, be noted. First, the majority of early psychosis individuals were taking antipsychotics at the time of scanning. The possibility that antipsychotic medications may have confounded the brain structural findings in psychotic disorders cannot be ruled out, particularly given some prior findings.¹⁰⁷ However, we did not find any significant associations between these measures and antipsychotic medication dosage, which is consistent with previous studies showing no effect of antipsychotic medication on LGI,¹⁰⁸ HFA,¹⁰⁹ or FA.¹¹⁰ Second, while we report subtle imaging metric alterations in early psychosis using large sample size in a multi-site design, the effect sizes for associations among imaging metrics were small to moderate, which aligns with previous studies.^78,111-113 While these statistically significant correlations reflect subtle developmental signatures, future investigations should examine additional brain features underlying anomalous gyrification. Third, the present vertex-wise analysis focused on identifying localized LGI abnormalities. However, given that schizophrenia is underpinned by whole-brain connectivity disturbances,¹¹⁴ this investigation should be complemented by methods optimized to detect network-level LGI anomalies.^115,116 Accordingly, network-wise analyses examining the relationship between LGI and white matter indices in psychosis may be also useful.¹¹⁷ Fourth, although HFA is a suggested surrogate marker of cytoarchitecture and has been validated in an animal study,⁵⁶ diffusion imaging has been primarily applied to white matter microstructure, and in gray matter it is confounded by partial volume effects much more than in white matter. We minimized these effects by using high-resolution data and free-water correction, but the model limitations (eg, unmodeled exchange between compartments) still exist.¹⁰⁹ Moreover, while HFA does capture cellular composition and spatial distribution, it does not encompass other biological factors that partly contribute to cytoarchitecture (eg, synapse, dendrite, myelin, and axon).⁵⁶ Therefore, future multimodal studies should validate our findings using more sophisticated models for cytoarchitecture of gray matter. Lastly, this study employed a cross-sectional design, making it difficult to determine whether the observed brain structural alterations reflect static deficits from early neurodevelopment, progressive brain pathology in later stages, or a combination of both.¹¹⁸ Notably, variations in the direction of LGI abnormalities, inconsistencies in their locations across studies, and our own findings on the association between LGI and white matter tracts suggest that LGI may be dynamically influenced by white matter maturation. Nevertheless, it is imperative to establish whether white matter maturation drives the LGI decline. Future multimodal neuroimaging studies with a longitudinal framework are warranted to examine the causal relationships between progressive changes in cortical folding and diffusivity measures along the disease course, while restricting subjects to more stratified cohorts (ie, affective or non-affective subgroups, DSM-based diagnostic categories).

Conclusions

By integrating advanced MRI techniques into our multimodal neuroimaging study, we provide the first in vivo evidence suggestive of specific mechanisms underlying anomalous cortical folding in psychosis. Particularly in non-affective psychosis, diminished complexity of cortical folding is underlain by cortical layer microstructural disorganization and long-range white matter tracts with less myelination and coherence. These relationships partially support the tension-based cortical morphogenesis.¹⁸ Further longitudinal multimodal neuroimaging studies are required to better understand the nature and dynamics of fronto-temporal circuitry abnormalities across psychosis subgroups, as well as their functional consequences along their time-course.

Funding

The present study was supported by JSPS KAKENHI Grant Numbers JP18K15509 and JP22K15745 to D.S., JP24K10704 to T.T., JP19H03579 and JP24K02378 to S.K., JP20KK0193, JP21H02851, and JP23K21414 to Y.H., and JP20H03598 to M.S., AMED Grant Number JP24wm0625302 to T.T., Y.H., and S.K., and the Health and Labour Sciences Research Grants for Comprehensive Research on Persons with Disabilities Grant Numbers JP16dk0307029 and JP20dk0307094 to M.S. The current study was also supported by SENSHIN Medical Research Foundation to T.T. and by NIMH Grant Numbers U01MH109977 to M.E.S., Y.R., K.E.L., R.I.M., J.W., O.P., M.S.K., D.H., D.Ö., A.B., M.J.C., S.B., R01-MH136503 to M.K., O.P., R01MH119222 to Y.R., R01MH117012 to K.E.L., and K99MH131850 to J.S.-H. We also acknowledge funding provided by the Brain and Behavior Research Foundation (BBRF): Young Investigator Award to S.C.-K. and J.S.-H., and a Distinguished Investigator Award to M.K. This research was additionally supported by BBRF Young Investigator grant (funded by Mary and John Osterhaus and the Brain & Behavior Research Foundation) to J.S.-H. and Women’s Brain Initiative Pilot Award from Brigham and Women’s Hospital to J.S.-H.

Conflicts of Interest

The all authors declare no competing interests.

Data Availability

The datasets generated and analyzed during the current study are not publicly available due to reasons of data privacy and regulatory/ethical approvals, but aggregate-level data are available from the corresponding author upon reasonable request and pending local and national data protection and ethics regulations.