Microstructural and Microvascular Alterations in Psychotic Spectrum Disorders: A Three-Compartment Intravoxel Incoherent Imaging and Free Water Model

Abstract

Background and Hypothesis

Microvascular and inflammatory mechanisms have been hypothesized to be involved in the pathophysiology of psychotic spectrum disorders (PSDs). However, data evaluating these hypotheses remain limited.

Study Design

We applied a three-compartment intravoxel incoherent motion free water imaging (IVIM-FWI) technique that estimates the perfusion fraction (PF), free water fraction (FW), and anisotropic diffusion of tissue (FAt) to examine microvascular and microstructural changes in gray and white matter in 55 young adults with a PSD compared to 37 healthy controls (HCs).

Study Results

We found significantly increased PF, FW, and FAt in gray matter regions, and significantly increased PF, FW, and decreased FAt in white matter regions in the PSD group versus HC. Furthermore, in patients, but not in the HC group, increased PF, FW, and FAt in gray matter and increased PF in white matter were significantly associated with poor performance on several cognitive tests assessing memory and processing speed. We additionally report significant associations between IVIM-FWI metrics and myo-inositol, choline, and N-acetylaspartic acid magnetic resonance spectroscopy imaging metabolites in the posterior cingulate cortex, which further supports the validity of PF, FW, and FAt as microvascular and microstructural biomarkers of PSD. Finally, we found significant relationships between IVIM-FWI metrics and the duration of psychosis in gray and white matter regions.

Conclusions

The three-compartment IVIM-FWI model provides metrics that are associated with cognitive deficits and may reflect disease progression.

Introduction

Psychotic spectrum disorders (PSDs) are a group of often devastating psychiatric syndromes that share the common symptoms of hallucinations and/or delusions and affect nearly 4% of the population.¹ Despite over a century of investigation, the PSD etiology remains poorly understood.² A number of theories have been proposed to explain the diverse deficits noted; among them inflammation and microvascular dysfunction have emerged as potentially relevant pathological mechanisms.^3–9 This is supported by accumulating immunochemical and genetic evidence which includes: (1) increased serum levels of pro-inflammatory cytokines^10,11 and chemokines¹² in patients with PSD compared to healthy individuals, and (2) increased representation of genes involved in vascular function, vasoregulation, and post-ischemic repair among candidate genes for schizophrenia, a primary psychotic disorder.¹³ Additionally, microvascular changes in PSD have been supported by studies that have found morphological abnormalities of the blood venules in the retina¹⁴ and nailfold capillary bed.¹⁵ Theoretical developments have aimed to unify these diverse findings by proposing that psychosis may be a result of alterations of the brain microvasculature initiated or caused by inflammation.^16–18

The involvement of inflammation in PSD has received support over the last decade from a growing number of studies employing a diffusion imaging-based putative marker of inflammation, the free water fraction (FW).¹⁹ The free water imaging (FWI) model estimates the fraction of an isotropic water pool with the diffusivity of free water (i.e., 3 × 10⁻³ mm²/s). The model also includes a tissue compartment, which is anisotropic and is usually described by the fractional anisotropy (or anisotropic diffusion of tissue (FAt)) of the corresponding diffusion tensor. Free water is assumed to arise from the extra-cellular compartment and reflect edema-like tissue inflammation and/or glial activation.^19,20 The FW has been consistently shown to be increased in the early stages of PSD (i.e., first episode) and decreased thereafter.^21–23 Recent work has also shown that FW associates with peripheral markers of inflammation^24,25 as well as with magnetic resonance spectroscopy imaging (MRSI)-based brain metabolites reflective of inflammation, such as glutathione.²⁶

Whereas single-shell diffusion acquisitions have been originally used to calculate FW, recent work has shown that its derivation may be improved by multishell acquisitions that include lower b-values and integrate a pseudodiffusion microvascular high-diffusivity compartment.²⁷ Typically estimated using the intravoxel incoherent imaging (IVIM) diffusion-based approach, the microvascular compartment accounts for the diffusion-like signal arising from blood protons flowing through randomly oriented tissue capillaries.²⁸ The fraction of the microvascular compartment derived using IVIM, the perfusion fraction (PF), has been shown to reflect capillary density and vasodilatation.²⁹ The combined FWI-IVIM model was shown to substantially improve the FW estimation in addition to estimating PF.^27,30

Neuroinflammation has been assessed in vivo using MRSI, which quantifies a number of inflammation-sensitive metabolites, including myo-inositol (mIns) and choline (Cho).^31–34 While mIns is the primary metabolite used to investigate neuroinflammation as it is found primarily in astrocytes, Cho is also more abundant in glial cells than neurons, and therefore can quantify the increase in microglia and astroglia which occurs during an inflammatory reaction.^33,34 Additionally, N-acetylaspartic acid (NAA) is a representative marker of neuroinflammation as its reduction represents neuronal injury or loss.³⁵ mIns, Cho, and NAA changes have been reported in PSD, primarily using localized single-voxel assessments.^36–38 However, only one study to date has examined relationships between mIns and free-water metrics in PSD.²⁶

The primary aim of this research was to apply the three-compartment IVIM-FWI model to describe PSD microvascular and microstructural changes in both gray and white matter. Additionally, we tested associations between several MRSI-derived metabolite metrics of inflammation and neuronal and cell membrane health and IVIM-FWI metrics to provide additional support for these measures as microvascular and microstructural biomarkers. Finally, we examined the relationships between the IVIM-FWI-derived PF, FW, and FAt and cognition and the duration of psychosis.

Methods

Participants

Individuals with PSD and healthy control (HC) participants aged 18–34 years old were recruited through advertisement and clinical referral, and interviewed under the supervision of a licensed clinical psychologist using the Diagnostic Interview for Genetic Studies (DIGS),³⁹ a semistructured interview that records information regarding a subject’s functioning and psychopathology with primary emphasis on information relevant to PSD diagnoses. The DIGS was employed to confirm diagnostic in patients or lack of symptoms in HC as well as obtain information on medication, substance use, and medical history. The MATRICS Consensus Cognitive Battery (MCCB),^40,41 which includes tests that examine working memory and processing speed, and the American version of the National Adult Reading Test, which provides a measure of premorbid intelligence, were also administered to all participants.

Inclusion criteria for patients included a diagnosis of schizophrenia (SZ), schizoaffective disorder (SZA), or bipolar disorder with psychotic features (BPF). Both patients and HC participants were excluded for substance abuse within the last 6 months, history of head trauma with loss of consciousness longer than 30 min, organic brain disorder and substance-induced psychotic disorder or psychotic disorder due to a general medical condition as determined by the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) criteria. In addition, HC participants were excluded for psychotropic medication usage, presence of prodromal symptoms, or a first-degree family member with a psychotic condition. The study was approved by the Institutional Review Board of the NYU Grossman School of Medicine. All participants provided informed consent and were given a full explanation of the research protocol before being enrolled in the study.

Five participants whose imaging data displayed significant motion and eddy-current artifacts and for whom repeated data acquisition was not successful or not possible were not included in the analysis. After inclusion and exclusion criteria and image quality checks were applied, 55 patients with a diagnosis of PSD and 37 HC individuals were retained in the study. The PSD group was additionally separated into two subgroups: a schizophrenia spectrum disorder (SSD) (SZA and SZ groups, N = 37) subgroup and a BPF (N = 18) subgroup. This further partition of the PSD group was motivated by recent research demonstrating that the BPF subtype can have significantly less severe neurobiological and cognitive deficits than SZA and SZ subtypes,^42–44 and different changes in FW at first episode.⁴⁵ Demographic characteristics for both groups and medication status for the patient group are described in Table 1.

Table 1.

Summary of Demographic, Cognitive Test Scores and Medication of the Participants in Patient and Healthy Control Groups

Variable	HC (n = 37)	PSD (n = 55)	P-Value	SSD (n = 37)	P-Value	BPF (n = 18)	P-Value
Male/Female	21/16 [0]	32/23 [0]	0.892	24/13 [0]	0.475	8/10 [0]	0.391
Handedness (R/L/both)	34/2/1 [0]	45/5/3 [2]	0.605	29/4/3 [1]	0.359	16/1/0 [1]	0.790
Age: Mean ± SD	25.6 ± 3.5 [0]	25.3 ± 3.2 [0]	0.733	25.2 ± 3.2 [0]	0.658	25.5 ± 3.2 [0]	0.979
Range	18–31	19–34		21–34		19–31
Education: Mean ± SD	15.2 ± 2.1 [0]	13.6 ± 1.9 [2]	<0.001*	13.2 ± 1.8 [1]	<0.001*	14.4 ± 2.0 [1]	0.203
Range	9–18	8–18		8–18		12–18
Race	15/11/6/0/5 [0]	11/25/3/1/10 [5]	0.093	6/17/3/1/7 [3]	0.118	5/8/0/3 [2]	0.228
Hispanic (Y/N)	13/24 [0]	15/38 [2]	0.491	12/24 [1]	0.871	3/14 [1]	0.191
AMNART IQ: Mean ± SD	116.8 ± 6.8 [1]	111.5 ± 7.9 [3]	0.001*	108.7 ± 6.8 [2]	<0.001*	117.3 ± 6.7 [1]	0.802
Range	97–126	95–126		95–125		98–126
Smoking (Y/N/Past)	0/29/3 [5]	7/27/6 [15]	0.026*	5/15/4 [13]	0.013*	2/12/2 [2]	0.111
Anxiety (Y/N/Past)	1/28/4 [4]	20/27/3 [5]	0.002*	13/17/3 [4]	0.004*	7/10/0 [1]	.001*
Psych duration: Mean ± SD	–	5.4 ± 4.2 [10]	–	5.2 ± 4.2 [1]	–	6.3 ± 4.5 [9]	–
Range		1–14		1–14		1–13
SANS: Mean ± SD	–	17.2 ± 17.1 [6]	–	19.9 ± 18.3 [6]	–	12.7 ± 14.3 [0]	–
Range		0–89		4–89		0–48
SAPS: Mean ± SD	–	22.9 ± 18.6 [6]	–	27.5 ± 17.7 [6]	–	14.9 ± 17.9 [0]	–
Range		0–76		0–76		0–52
PSI: Mean ± SD	46.6 ± 8.7 [0]	37.1 ± 13.8 [3]	<0.001*	32.5 ± 11.1 [2]	<0.001*	46.5 ± 14.1 [1]	0.977
Range	28–64	13–72		13–62		24–72
WMI: Mean ± SD	49.6 ± 11.2 [0]	41.7 ± 12.7 [3]	0.003*	37.7 ± 11.3 [2]	<0.001*	49.7 ± 11.7 [1]	0.973
Range	24–75	9–71		9–60		32–71
HVLTtr: Mean ± SD	45.9 ± 9.5 [0]	39.8 ± 9.2 [4]	0.003*	37.8 ± 8.8 [3]	<0.001*	43.7 ± 9.1 [1]	0.424
Range	21–63	20–58		20–58		23–58
HVLTdr: Mean ± SD	46.4 ± 11.2 [1]	35.6 ± 13.4 [4]	<0.001*	32.5 ± 11.5 [3]	<0.001*	41.9 ± 14.9 [1]	0.234
Range	21–60	20–63		20–59		20–63
HVLTrp: Mean ± SD	48.1 ± 8.7 [1]	40.4 ± 14.1 [4]	0.004*	37.6 ± 13.2 [3]	<0.001*	45.9 ± 14.6 [1]	0.503
Range	28–71	18–75		18–56		20–75
HVLTd: Mean ± SD	51.6 ± 8.4 [3]	46.3 ± 10.9 [4]	0.020*	45.4 ± 11.5 [3]	0.015*	48.1 ± 9.5 [1]	0.194
Range	20–58	21–58	–	21–58	–	29–58	–
Typical antipsychotic Current (Y/N)	–	2/52	–	2/34	–	0/18	–
Typical antipsychotic Past (Y/N)	–	4/50	–	3/33	–	1/17	–
Atypical antipsychotic Current (Y/N)	–	32/22	–	25/11	–	7/11	–
Atypical antipsychotic Past (Y/N)	–	20/34	–	15/21	–	5/13	–
Antidepressant Current (Y/N)	–	20/34	–	11/25	–	9/9	–
Antidepressant Past (Y/N)	–	15/39	–	11/25	–	4/14	–
Anxiolytic Current (Y/N)	–	12/42	–	10/26	–	2/16	–
Anxiolytic Past (Y/N)	–	7/47	–	5/31	–	2/16	–
Mood stabilizer Current (Y/N)	–	13/41	–	6/30	–	7/11	–
Mood stabilizer Past (Y/N)	–	7/47	–	4/32	–	3/15	–
Anticholinergic Current (Y/N)	–	4/50	–	3/33	–	1/17	–

Note: SD, standard deviation; AMNART IQ, American version of the National Adult Reading Test intelligence quotient; PSI, processing speed index composite score; WMI, working memory index composite score; HVLTtr, Hopkins Verbal Learning Test Total Recall; HVLTdr, Hopkins Verbal Learning Test Delayed Recall; HVLTrp, Hopkins Verbal Learning Test Retention Percent; HVLTd, Hopkins Verbal Learning Test Discrimination; SANS, scale for the assessment of negative symptoms; SAPS, scale for the assessment of positive symptoms. Race Index: 1: White, 2: Black, 3: Asian, 4: Native American, 5: Other, *P < 0.05. The number of individuals with missing data is shown in brackets.

Magnetic Resonance Image Acquisition

MRI data were acquired on a 3T Prisma MRI (Siemens Medical Solutions, Erlangen, Germany) using a 64-element head-neck coil. Three-dimensional T1-weighted images (T1w) were acquired using the Human Connectome Project (HCP) Lifespan protocol using 0.8 mm isotropic voxels. Diffusion MRI data were acquired using the HCP multiband diffusion-weighted spin-echo EPI for AP and PA polarities with 1.5 mm voxels, TR of 3230 ms, TE 89.20 ms, and multiband factor of 4. Images with b-values between 0 and 1000 s/mm² were added to the HCP diffusion encoding scheme to detect the IVIM effect.⁴⁶ Only b-values up to 1000 s/mm² (ten b = 0 s/mm², three orthogonal encodings for each of b = 20, 50, 75, 100, 150, 200, 400, and 600 s/mm², and eleven b = 1000 s/mm²) were used here in order to avoid non-Gaussian diffusion effects known to affect higher b-values.⁴⁷ MRSI data were acquired using a whole-brain volumetric echo-planar imaging acquisition with spin-echo excitation and 0.56 × 0.56 × 0.75 cm³ voxels. A water reference measurement was obtained using identical parameters. Several MRSI data sets were lost due to motion or data transfer and processing issues, which left 86 data sets (34 HC and 52 PSD: 18 BPF, 34 SSD) available for MRSI analyses.

Diffusion Image Processing

Diffusion images were first denoised using the MRtrix3 dwidenoise⁴⁸ and corrected for Gibbs artifacts.⁴⁹ Subsequently, images were corrected for motion and distortions from B0 field inhomogeneities and eddy currents and integrated using FSL topup and eddy.⁵⁰

The three-compartment IVIM-FWI model was implemented using Diffusion Microstructure Imaging in Python (DMIPY) software to obtain FW, PF, and FAt metrics following previous work^27,51:

$${\displaystyle \begin{array}{l}{\displaystyle S=S_0[PF\cdot e^{-b_i{D}_{pf}}+FW\cdot e^{-b_i{D}_{fw}}+(1-PF-FW)\cdot e^{-b_{i}g{}_i^T{D}_t{g}_i}}\end{array}}$$ [1]

where the diffusivity of FW compartment was set to D_fw = 3 μm²/ms, and the diffusivity of the microvascular compartment to D_pf = 7 μm²/ms, per DMIPY implementation. An alternate value of D_pf = 10 μm²/ms,⁵² was also examined. The tissue compartment was modeled using a second-order diffusion tensor (D_t) with diffusivities restricted to values between 0.5 and 1.7 μm²/ms.⁵³ The tissue diffusion tensor, D_t, was used to derive the FAt as implemented in DIPY.⁵⁴ Our implementation of the IVIM-FWI model was tested using simulations similar to²⁷ (Supplementary information, Section 2).

The FreeSurfer software and the T1w images were used to parcellate cortical gray and subcortical white matter of each participant according to the Desikan–Killiany atlas and obtain regional cortical thickness values.⁵⁵ Each T1w-space parcellation was then registered to the diffusion space using the standard FreeSurfer diffusion registration pipeline with trilinear interpolation⁵⁶ and used to calculate mean FW, PF, and FAt for bilateral regions of interest (ROI).

MRSI Image Processing

MRSI reconstruction was carried out using the Metabolite Imaging and Data Analysis System (MIDAS) package.⁵⁷ The posterior cingulate cortex was chosen for the MRSI analyses as it provides excellent MRSI signal and has been previously shown to have increased metabolic rate, increased FW, and altered microstructure in PSD (Supplementary figure 1).^21,58,59 The left and right posterior cingulate cortical ROIs from Desikan–Killiany atlas were imported and warped to the MRSI space in MIDAS. The concentrations of mIns, NAA, and Cho (choline, phosphocholine, and glycerophosphocholine) were calculated using the M-Int Map-Integrated technique.⁵⁷ The M-Int Map-Integrated technique derives regional metabolite concentrations by first integrating all spectra within an ROI from voxels with greater than 80% of the gray matter tissue fraction to obtain a regional averaged spectrum.⁵⁷ MIns and Cho have been previously linked to gliosis and neuroinflammation.³³ NAA decreases relate to neuronal distress and axonal damage, which may be observed as an effect of inflammation.³² Therefore, NAA is examined here as a secondary marker of neuroinflammation as well as a marker of microstructural integrity.

Statistical Analyses

Independent t-tests and analyses of covariance (ANCOVA) that accounted for age, sex, and medication type were used to evaluate differences in IVIM-FWI metrics between HC and PSD groups. Secondary analyses explored differences between HC and two PSD subgroups: the SSD and BPF groups. Medication status was covaried using both current and previous medications (Table 1). Each MRI measure (PF, FW, and FAt) was compared independently between subject groups. The effect size of between-group differences was calculated using Cohen’s d.

Further analyses used Pearson’s correlations, which accounted for age, sex, and medication type, to explore relationships between diffusion metrics and working memory, processing speed, and verbal learning as measured by: (1) MCCB working memory index (WMI) composite score; (2) MCCB processing speed index (PSI) composite score; (3) Hopkins Verbal Learning Test (HVLT) total score; (4) HVLT total recall; (5) HVLT delayed recall; (6) HVLT retention percent; and (7) HVLT discrimination in the PSD group and subgroups, and the HC group. These cognitive domains have been shown to be the most consistently impaired in PSD.⁶⁰

Pearson’s partial correlations, controlled for age, were additionally utilized to examine associations between IVIM-FWI metrics of the bilateral posterior cingulate cortex and the corresponding mIns, Cho, and NAA concentrations.

Further analyses examined whether IVIM-FWI metrics relate to the duration of psychosis using both linear and quadratic non-linear regressions, that accounted for age, sex, and medication status, to test whether IVIM-FWI metrics relate to the duration of psychosis in the SSD group. Significant nonlinear relationships were found between IVIM-FWI metrics and psychosis duration up to 14 years. Secondary analyses examined the IVIM-FWI metrics’ relation with psychosis duration in a subset of the participants with a psychosis duration below 10 years (36 PSD, 29 SSD).

The ANCOVA and Pearson’s correlation analyses that controlled for the effects of age, sex, and medication status described above showed only age and sex to have an effect on between group differences. Only age was found to affect correlations between IVIM-FWI metrics and cognitive tests, MRSI metrics and psychosis duration. Therefore, only analyses including age (and sex for between-group comparisons) were included in the results.

Analyses were conducted for four gray and white matter lobes (frontal, temporal, parietal, and occipital) and 68 cortical and subcortical gray and white matter regions delineated by the Desikan–Killiany atlas. To address multiple comparisons, P-values were considered significant when <0.01 for lobes and <0.005 for sublobar regions in order to reduce the false discovery rate while maintaining reasonable statistical power. For these analyses, tests with P < 0.05 were assumed to reach a trend-level significance. Associations between the diffusion and MRSI metrics in the posterior cingulate areas were considered significant when P < 0.05.

Results

Between-group Comparisons

Perfusion Fraction.

The PSD and SSD groups demonstrated significantly higher PF than HC in the right inferior temporal lobe and in the left temporal lobe and left middle temporal gray matter areas, respectively (figure 1; Supplementary Table S1). No significant differences were noted in PF in the BPF subgroup compared to the HC group.

Fig. 1.

Significant analyses of covariance (ANCOVA) group differences in perfusion fraction (PF), free water fraction (FW), and fractional anisotropy of tissue (FAt) in gray matter (blue) and white matter (green), covaried for subjects’ age and sex. Color indicates significant differences in patient groups compared to healthy controls at regional- (light color) or lobar-level (dark color).

Free Water.

Compared to HCs, both the PSD group and the SSD subgroup showed significantly higher FW in the gray matter of the bilateral insular cortices. Additionally, the PSD group showed significantly increased FW in the gray matter of the left temporal pole. The BPF group revealed significantly increased FW in the left temporal lobe and left insula as well as in the left temporal pole and left entorhinal and lateral-orbitofrontal cortices (figure 1; Supplementary Table S2). In white matter, significant FW increases compared with HC were noted only for the BPF group in the left frontal pole and right medial-orbitofrontal cortex (figure 1).

Fractional Anisotropy of Tissue.

Compared to HC, the PSD and SSD groups demonstrated higher FAt in gray matter with differences reaching significance in the left parietal lobe and supra-marginal cortex for the SSD group (Supplementary Table S3). Conversely, in white matter, all groups showed lower FAt compared to the HC group. Significantly decreased FAt in white matter was noted in all groups in the left parietal lobe. Regions with decreased FAt also included the right transverse temporal and left inferior parietal areas in the PSD and BPF groups and the left occipital lobe in the PSD group (figure 1; Supplementary Table S3).

IVIM-FWI Metrics’ Associations with Cognitive Tests, MRSI Metabolites, and Psychosis Duration

In patients (both the overall PSD group and the SSD subgroup) but not in HC, increased PF in gray and white matter and increased FAt and FW in gray matter negatively correlated with the cognitive tests examined (figure 2; Supplementary Tables S4–S6).

Fig. 2.

Examples of gray and white matter regions that show increased perfusion fraction (PF), free water fraction (FW), and fractional anisotropy of tissue (FAt) in psychotic spectrum disorder (PSD) relates to poor working memory index (WMI) and the Hopkins Verbal Learning Test (HVLT) memory tests. Partial Pearson’s correlations, controlled for age, are included in each plot with confidence intervals. For full list of regions showing significant associations see Supplementary Tables S4–S6. Abbreviations: white matter (wm) gray matter/cortex (GM/ctx), left hemisphere (lh), right hemisphere (rh).

In PSD and SSD groups, but not in the HC group, the PF in the bilateral posterior cingulate cortices was significantly positively correlated with Cho and mIns (figure 3, Supplementary Table S7). Additionally, FW was significantly negatively correlated with NAA (figure 3; Supplementary Table S7).

Fig. 3.

Significant partial Pearson’s correlations in psychotic spectrum disorder (PSD), controlled for age, are observed between perfusion fraction (PF) and free-water fraction (FW) and MRSI metabolites: choline (Cho), myo-inositol (mIns), and N-acetylaspartic acid (NAA) in the left and right posterior cingulate gray matter regions. Confidence intervals are shown in gray.

In the SSD group, significant nonlinear associations with the duration of psychosis were found in both gray and white matter for PF and FAt (figure 4; Supplementary Table S8). Significant linear correlations between PF, FW, and FAt metrics with the duration of psychosis were only found analyzing patients with a psychosis duration under 10 years. For this secondary analysis, PF was positively related to psychosis duration in gray matter and white matter while FAt was positively associated in gray matter (figure 4; Supplementary Table S8). FW was negatively associated with psychosis duration in white matter.

Fig. 4.

Associations between the duration of psychosis, controlling for age, and perfusion fraction (PF), free-water fraction (FW, and fractional anisotropy of tissue (FAt) metrics in the schizophrenia spectrum disorder (SSD) group shown for several gray and white matter regions of interest. A quadratic polynomial (dotted) regression was fit for the entire SSD group, and a linear (solid) regression was fit for all SSD participants with a psychosis duration of up to 10 years. Confidence intervals are shown in gray. Abbreviations: white matter (wm), gray matter/cortex (ctx), left hemisphere (lh), right hemisphere (rh).

Discussion

In this study, we demonstrate for the first time, that the three-compartment IVIM-FWI model can be applied to PSD to describe microvascular, neuroinflammatory, and microstructural pathology and provide novel evidence of increased IVIM-derived PF in SSD in both gray and white matter. We also show that the PF increase in the PSD group significantly relates to memory impairment, MRSI-measured metabolic metrics related to inflammation, neuronal and membrane health, and psychosis duration.

Aside from increased perfusion in the hippocampus and temporal lobe, which has been more consistently reported, previous perfusion studies in PSD have generally provided inconsistent results with both hypo- and hyper-perfusion reported.^61–66 These studies were based on different techniques ranging from arterial spin labeling (ASL), dynamic contrast-enhanced (DCE), and dynamic susceptibility contrast (DSC) to Positron Emission Tomography. Therefore, the inconsistencies among literature reports could be due in part to the heterogeneity of the techniques used, in addition to the heterogeneity of the PSD groups studied.^66–69 Moreover, ASL, DSC, or DCE cerebral blood volume and flow estimates generally reflect the signal from arteriolar, capillary, and venular compartments of the vasculature rather than randomly flowing capillary blood as IVIM.²⁹ Therefore, IVIM provides a specific contrast that may be uniquely sensitive to microvascular pathology in PSD. Although no IVIM data have been published to date in PSD, the IVIM-derived PF was found to be increased in gray matter in individuals with type 2 diabetes,⁷⁰ mild vascular cognitive impairment,⁷¹ and Moyamoya disease,⁷² all disorders with known microvascular pathology.

Histological studies investigating PSD have suggested a number of cellular and microvascular neural tissue deficits, any of which may contribute to the increased PF observed here. These deficits include proliferation/malformation of microglia and astroglia covering the blood capillaries, leaky blood-brain-barrier (BBB), deformation of the basal lamina, and wide-spread small vessel disease with demyelination in white matter.^73–75 Animal studies have found that increased perfusion is associated with increased astrocyte immunoreactivity and a leaky BBB.^76,77 Therefore, we speculate that the increased gray and white matter perfusion in PSD likely reflects a combination of microvascular and glial pathologies. This interpretation is supported by the significant association found between PF and the MRSI markers of glial function, mIns, and Cho.

Our results are consistent with previous reports of significantly increased FW in the early disease stage in people with PSD.^21–23,78 Given that recent work has suggested that FW might be overestimated when not including a perfusion compartment,²⁷ our findings substantiate the credibility of previous FW findings, particularly in the gray matter, in people with PSD.

Changes in diffusion fractional anisotropy in white matter are one of the longest-standing findings in PSD.⁷⁹ Therefore, in line with previous research our findings of decreased white matter FAt in the PSD, SSD, and BPF groups may suggest either demyelination or increased fiber dispersion. Anisotropic diffusion in gray matter has been less investigated; the increased FAt we note in gray matter may reflect increased cortical myelin⁸⁰ as well as changes in the relative proportion or microstructural properties of the axial and radial cortical axonal and dendritic bundles. Additionally, inflammation may affect diffusion properties of the glial structures,⁸¹ and may be potentially detected as a change in anisotropy. Further research is needed to map how FAt relates to specific gray matter changes.

Increased PF in gray and white matter and increased FAt and FW in gray matter were linked to poor performance on tests of processing speed and several aspects of working memory. The negative relationship between PF and cognitive function suggests that microvascular functions, such as efficient blood flow to neurons,⁸² may be critical for supporting healthy cognitive function. Plausibly, PF may be a proxy for a dysregulated neuro-vascular unit and thus reflect a set of neural features involved in cognition. Of note, increased PF was found to relate to poorer verbal working memory in a cohort of individuals with type 2 diabetes,⁷⁰ akin to our findings. Although it is unclear which aspects of gray matter microstructure FAt may be measured, variations in myelination, size, or density of axons or neurons may both affect it and impact cognitive function in PSD. Increased extracellular inflammation, which is assumed to underlie increased FW, and the FW metric have been previously linked to poor cognition.^83–85 Activated glial cells may disrupt cognitive processes, by impeding growth or neuronal signaling, necessary for healthy memory and executive functioning.⁸³

We found that PF was significantly positively related to Cho and mIns in the PSD and SSD groups. Both mIns and Cho increases have been associated with inflammation.^33,86 Therefore, PF’s positive correlations with mIns and Cho further support the hypothesis that glial dysfunction is a core part of the underlying microvascular pathology in PSD suggested by previous research.^33,77,87 Although we hypothesized that FW will positively associate with mIns, we did not find such relationship in the examined ROI, which is consistent with previous findings.²⁶ This may suggest that neural substrates of early/acute inflammatory states (described by increased FW) may differ from those characterizing chronic states (described by increased PF, which associates with increased mIns). Decreases in NAA are related to neuronal cell damage or loss; therefore, the negative correlation between FW and NAA may be a representation of the destructive effect inflammation has on the neuronal structural integrity in PSD.^77,88 Overall, the close associations between the relatively novel diffusion markers and the more established MRSI metrics substantiate the validity of PF, FW, and FAt as meaningful descriptors of tissue microstructural and microvascular properties.

Our data also suggest that microvascular and microstructural alterations in SSD evolve with disease duration. Our data points to an increase in FW in early-stage SSD which then tapers off, alongside a steady rise in PF up to 10 years since the emergence of psychotic symptoms (figure 4). Additionally, white matter FAt appears at HC levels in early SSD participants but then steadily decreases, while FAt in gray matter mimics a PF-like pattern increasing up to 10 years since the first episode (figure 4). Increased FW at the first episode is often noted in PSD and is assumed to reflect edema-like inflammation. This pathology appears to subside with increased disease duration with vascular and glial alterations⁸⁹ appearing to become increasingly prevalent. This evolution is consistent with recent large-scale reports suggesting progressive microstructural changes in schizophrenia⁹⁰ and proposed neurovascular-inflammation psychosis models, which suggests that an initial inflammatory response may act as a catalyst for the microvascular disruption which in turn affects the integrity of the neurovascular unit and creates a cascade of neuropathological changes in PSD.^82,91 The significant nonlinear pattern of change we note in PF, FW, and FAt metrics over the duration of illness may be due to several factors including: (1) the fluctuation in the severity of illness over time, (2) distinct phases of clinical illness, and/or (3) effects of treatment.^92–94 The patterns of PF, FW, and FAt changes with disease duration may have important implications for the diagnosis and prediction of disease progression in SSD and PSD. Further longitudinal studies will be necessary to confirm our findings.

Limitations

Several limitations of this study need to be considered. First, although IVIM is a long-standing and well-tested method for detecting perfusion, it may be sensitive to CSF/glymphatic flow.^29,95 Second, the study had a purely cross-sectional design and a relatively small sample size for the subgroups examined, particularly the BPF group. As a psychosis duration could not be established for many BPF patients, who had milder and less defined psychotic symptoms, an assessment of the effects of psychosis duration was not possible in this group. Therefore, it will be important to replicate and extend these results using larger longitudinal future studies. Although we did examine the effect of medication type and found no effect, dose effects were not assessed as doses were not reported by all patients. Previous research has found anti-psychotics can have both an anti-inflammatory and protective effect on brain microstructure.^93,94 Therefore, future studies that parse out medications’ potential effects and examine medication-naïve patients are necessary. Future studies may also use more advanced biophysical diffusion models to describe the tissue compartment beyond a simple tensor representation.

Conclusions

We demonstrate, for the first time, to the best of our knowledge, that the three-compartment IVIM-FWI model can be applied to PSD to quantify abnormalities in perfusion, free water, and anisotropy of tissue. These abnormalities appear to associate with poorer memory and processing speed and progress with the duration of psychosis over the first decade from the onset. In patients, the novel diffusion metrics strongly correlate with several established MRSI markers of inflammation and neuronal and axonal health, which suggests complex disease mechanisms that likely involve both glial and vascular components, adding support to the recently proposed inflammatory-vascular models.⁷⁷ IVIM-FWI metrics may provide useful biomarkers of abnormal vasculature and microstructure and help elucidate neuropathology in psychiatric and neurological disorders.

Funding

This work was supported by the National Institute of Mental Health R01 MH108962 award.