Introduction
Psychosis spectrum disorders (PSDs), including schizophrenia (SZ), schizoaffective disorder (SAD), and bipolar disorder with psychotic features (BP), are marked by hallucinations (perceptual experiences without external stimuli), delusions (fixed false beliefs resistant to evidence), disorganized thinking, and cognitive impairment (American Psychological Association, 2013). PSD affects millions of people worldwide and imposes substantial global health burdens (Jongsma, Turner, Kirkbride, & Jones, Reference Jongsma, Turner, Kirkbride and Jones2019; Moreno-Küstner, Martín, & Pastor, Reference Moreno-Küstner, Martín and Pastor2018). The symptom presentation and severity vary across diagnoses. While these differences may reflect distinct diagnostic entities, converging evidence from genetics, neuroimaging, and cognitive research points to shared biological mechanisms (Van Os, Reference Van Os2015; van Os & Tamminga, Reference van Os and Tamminga2007). Thus, examining PSD both as a combined group and by specific diagnoses can offer complementary insights into common and unique pathways underlying psychosis.
Cerebral perfusion, the process by which blood delivers oxygen and nutrients to brain tissue via the vasculature, represents a critical yet underexplored dimension for understanding PSD pathophysiology. Commonly quantified by cerebral blood flow (CBF), the volume of blood flowing to brain tissue each minute (ml/100 g/min) (Wintermark et al., Reference Wintermark, Sesay, Barbier, Borbély, Dillon, Eastwood, Glenn, Grandin, Pedraza, Soustiel, Nariai, Zaharchuk, Caillé, Dousset and Yonas2005) and perfusion abnormalities have been implicated in disrupted neuronal function and neurovascular dysregulation underlying PSD symptoms (Benedetti et al., Reference Benedetti, Aggio, Pratesi, Greco and Furlan2020; Chaves, Dursun, Tusconi, & Hallak, Reference Chaves, Dursun, Tusconi and Hallak2024; Dimick et al., Reference Dimick, Toma, MacIntosh, Grigorian, Fiksenbaum, Youngstrom, Robertson and Goldstein2022; Ermakov, Mednova, Boiko, & Ivanova, Reference Ermakov, Mednova, Boiko and Ivanova2025). Arterial spin labeling (ASL) is a noninvasive magnetic resonance imaging technique that measures resting-state or task-based global and regional CBF (Chappell, MacIntosh, & Okell, Reference Chappell, MacIntosh and Okell2018). ASL has been extensively validated and is a powerful tool to investigate structural and functional changes in PSD (Percie du Sert et al., Reference Percie du Sert, Unrau, Gauthier, Chakravarty, Malla, Lepage and Raucher-Chéné2023). Several ASL sequences exist (Chappell et al., Reference Chappell, MacIntosh and Okell2018), and pseudo-continuous ASL (pCASL) is particularly useful for clinical and research applications since it provides higher labeling efficiency and signal-to-noise ratio compared to other sequences (Alsop et al., Reference Alsop, Detre, Golay, Günther, Hendrikse, Hernandez-Garcia, Lu, MacIntosh, Parkes, Smits, van Osch, Wang, Wong and Zaharchuk2015).
Cerebral perfusion in PSD is an intricate and dynamic process that varies based on multiple factors, including brain region, specific symptoms, disease state, treatment, individual patient characteristics, smoking status, and statistical modeling methods (Bojesen et al., Reference Bojesen, Glenthøj, Sigvard, Tangmose, Raghava, Ebdrup and Rostrup2023; Corson, O’Leary, Miller, & Andreasen, Reference Corson, O’Leary, Miller and Andreasen2002; Dimick et al., Reference Dimick, Toma, MacIntosh, Grigorian, Fiksenbaum, Youngstrom, Robertson and Goldstein2022; Durazzo, Meyerhoff, & Murray, Reference Durazzo, Meyerhoff and Murray2015). For example, antipsychotic-naïve patients with psychosis showed no significant CBF alterations compared to neurotypical control (NC), but females exhibited lower regional CBF than males (Bojesen et al., Reference Bojesen, Glenthøj, Sigvard, Tangmose, Raghava, Ebdrup and Rostrup2023). Treatment with partial dopamine agonists and neuroleptic medications increased CBF of subcortical structures in patients with psychosis (Bojesen et al., Reference Bojesen, Glenthøj, Sigvard, Tangmose, Raghava, Ebdrup and Rostrup2023; Corson et al., Reference Corson, O’Leary, Miller and Andreasen2002). Furthermore, cortical fronto-limbic hypoperfusion was linked to negative symptoms while hyperperfusion in subcortical structures, especially the left putamen, was associated with positive symptoms in schizophrenia (Percie du Sert et al., Reference Percie du Sert, Unrau, Gauthier, Chakravarty, Malla, Lepage and Raucher-Chéné2023). Our group previously found that greater perfusion in the left lateral occipital cortex was linked to poorer cognition in patients with bipolar disorder, predominantly in the nonpsychotic subtype (Zeng et al., Reference Zeng, Lizano, Bolo, Lutz, Brady, Ivleva, Dai, Clementz, Tamminga, Pearlson and Keshavan2021). Additionally, greater perfusion fluctuations in the left fusiform and inferior temporal regions were identified in patients with BD; higher perfusion connectivity between the anterior cingulate cortex and the occipitoparietal cortex was negatively associated with positive symptoms (Dai et al., Reference Dai, Chen, Duan, Zhao, Bolo, Tamminga, Clementz, Pearlson, Alsop and Keshavan2020). Still, other ASL studies have reported mixed findings regarding the relationship between perfusion and cognitive or clinical measures in psychosis (Boisvert et al., Reference Boisvert, Lungu, Pilon, Dumais and Potvin2023; Douli et al., Reference Douli, Georgiou, Konstantinopoulou, Karampas, Plakoutsis, Sioka, Aretouli and Petrikis2025; Pinkham et al., Reference Pinkham, Loughead, Ruparel, Wu, Overton, Gur and Gur2011; Selvaggi et al., Reference Selvaggi, Jauhar, Kotoula, Pepper, Veronese, Santangelo, Zelaya, Turkheimer, Mehta and Howes2023; Sritharan et al., Reference Sritharan, Zeng, Petr, Mutsaerts, Hoang, Bolo, Ivleva, Dai, Gershon, Keedy, Parker, Trotti, McDowell, Clementz, Tamminga, Pearlson, Keshavan and Lizano2025; Zhu et al., Reference Zhu, Zhuo, Qin, Xu, Xu, Liu and Yu2015, Reference Zhu, Zhuo, Xu, Liu, Qin and Yu2017). Together, this highlights the need for large-scale, harmonized investigations that account for biological and technical heterogeneity.
Despite growing evidence for perfusion abnormalities, no study to date has systematically examined CBF patterns across both conventional diagnoses and biologically derived psychosis subtypes in a large-scale cohort. To address this gap, we present the largest study to date of resting-state global and regional CBF in PSDs, leveraging the robust, multi-site dataset of the Bipolar-Schizophrenia Network for Intermediate Phenotypes (BSNIP2) study. Beyond traditional diagnostic categories, the BSNIP2 study also enables examination of perfusion differences across biologically informed subtypes known as Biotypes. Derived through unsupervised clustering of neurophysiological data, these Biotypes capture three clinically and biologically distinct subgroups: Biotype 1 (BT1) is characterized by diminished neural activity, severe psychotic symptoms, and severe cognitive impairment; Biotype 2 (BT2) features heightened neural activity, significant psychotic symptoms, and significant cognitive deficits; and Biotype 3 (BT3) exhibits near-normal neural activity, milder symptomatology, and relatively preserved cognition (14–17).
We hypothesize that individuals with PSD will exhibit region-specific alterations in perfusion compared to NC, with distinct patterns linked to positive versus negative symptom domains. Given that our patient samples are not antipsychotic-naïve, we anticipate that some perfusion alterations may be associated with medication use. Additionally, we hypothesize that BT1 and BT2, representing the cognitively impaired biotypes, will demonstrate the most pronounced regional perfusion abnormalities, with specific patterns correlating with positive and negative symptom severity.
In sum, by integrating perfusion imaging with clinical outcomes in a large cohort of PSD, this work aims to uncover robust biological insights into the neurovascular underpinnings of these conditions.
Methods
Participants
This study includes 1,027 participants (241 SZ, 227 SAD, 158 BP, and 401 NC) from the Psychosis and Affective Research Domains and Intermediate Phenotypes (PARDIP) and the Bipolar Schizophrenia Network on Intermediate Phenotypes 2 (BSNIP2) studies conducted across five sites (Athens, GA; Boston, MA; Chicago, IL; Dallas, TX; and Hartford, CT). All studies were approved by the Institutional Review Board at the respective institutions, and all participants provided written informed consent. Diagnoses of SZ, SAD, and BP were made using the Structured Clinical Interview for DSM-IV Axis I Disorders – Patient Edition (SCID-I/P) (First, Reference First1997) and confirmed via clinical consensus by clinicians at respective sites. General inclusion criteria included adults aged 18–60 with SZ, SAD, or BP, and NC with no personal or family history of psychiatric illness. General exclusion criteria included significant neurological or medical illness, intellectual disability, substance abuse within the past 6 months or dependence within 2 years, and contraindications to MRI. A subset of participants with a complete set of 44 markers consisting of general cognition, motor inhibitory, saccadic system, and auditory electroencephalogram/ event-related potential tasks were biotyped based on their neurobiological signatures, resulting in 180 BT1, 195 BT2, 235 BT3, and 382 NC that are included in this study. Additional details regarding recruitment, assessments, and biotyping methods are provided in the Supplementary Material and prior publications (Clementz et al., Reference Clementz, Sweeney, Hamm, Ivleva, Ethridge, Pearlson, Keshavan and Tamminga2016, Reference Clementz, Trotti, Pearlson, Keshavan, Gershon, Keedy, Ivleva, McDowell and Tamminga2020, Reference Clementz, Parker, Trotti, McDowell, Keedy, Keshavan, Pearlson, Gershon, Ivleva, Huang, Hill, Sweeney, Thomas, Hudgens-Haney, Gibbons and Tamminga2022; Tamminga et al., Reference Tamminga, Ivleva, Keshavan, Pearlson, Clementz, Witte, Morris, Bishop, Thaker and Sweeney2013, Reference Tamminga, Pearlson, Keshavan, Sweeney, Clementz and Thaker2014; Zeng et al., Reference Zeng, Lizano, Bolo, Lutz, Brady, Ivleva, Dai, Clementz, Tamminga, Pearlson and Keshavan2021).
Clinical and cognition measures
All participants were assessed for handedness and socioeconomic status using the Hollingshead Four-Factor Index score (Hollingshead, Reference Hollingshead2011). In patients with PSD, psychotic symptoms were assessed using the Positive and Negative Syndrome Scale (PANSS) (Kay, Fiszbein, & Opler, Reference Kay, Fiszbein and Opler1987). Manic and depressive symptoms were measured using the Young Mania Rating Scale (YMRS) (Young, Biggs, Ziegler, & Meyer, Reference Young, Biggs, Ziegler and Meyer1978) and Montgomery–Åsberg Depression Rating Scale (MADRS) (Montgomery & Åsberg, Reference Montgomery and Åsberg1979), respectively. In all participants, smoking status was assessed using the Fagerström Test for Nicotine Dependence (FTND) (27), which includes FTND score and the question, “Have you ever been a smoker?” Responses were coded as 1 = Yes and 0 = No to indicate lifetime smoking status (Heatherton, Kozlowski, Frecker, & Fagerström, Reference Heatherton, Kozlowski, Frecker and Fagerström1991). Cognition was evaluated using the Brief Assessment of Cognition in Schizophrenia (BACS) (Keefe et al., Reference Keefe, Goldberg, Harvey, Gold, Poe and Coughenour2004). BACS scores were normalized for age and sex, z-transformed, and winsorized to ±4.0 standard deviations. Medication status was recorded; daily dose of chlorpromazine (CPZ) equivalents were calculated (CPZ equivalents × dose days × antipsychotic dose) (Andreasen et al., Reference Andreasen, Pressler, Nopoulos, Miller and Ho2010).
Of the 1,027 participants, some had missing data for clinical and cognitive measures: PANSS Total and subscores (n = 38–40), YMRS (n = 40), MADRS (n = 37), FTND (n = 277), BACS composite and subscores (n = 30–31), and CPZ dosage (n = 243). Sample sizes for each analysis were adjusted accordingly based on data availability.
Image acquisition, processing, harmonization, and imputation
The detailed imaging and scanner specifications are described in Supplementary Table 1. Briefly, pCASL scans were acquired using 3 tesla scanners across sites: GE Signa HDxt (Athens and Boston), Philips Achieva (Chicago and Dallas), and Siemens Skyra (Hartford). All sites implemented pCASL sequences with comparable field of view, resolution, and slice thickness, with other parameters matched as close as possible given scanner-specific hardware and software constraints. At the Boston site, ASL data were acquired using three related protocols due to study-specific sequence development and a mid-study scanner upgrade. At the Hartford site, ASL data were acquired in two different post-labeling delay (PLD), and the change happened while the B-SNIP2 study was ongoing so that the PLD would match the PLD for other sites. While anatomical imaging parameters were harmonized across participants (1.2 mm isotropic T1-weighted scans using ADNI-1 or ADNI-2 protocols) (Jack et al., Reference Jack, Bernstein, Fox, Thompson, Alexander, Harvey, Borowski, Britson, Whitwell, Ward and Dale2008), ASL acquisition parameters, including PLD, varied across protocols. Each of the three Boston protocols and the two Hartford protocols were therefore modeled as a separate factor in harmonization to account for protocol-specific effects (Supplementary Materials). The T1 sequence used was MP-RAGE at the Chicago, Dallas, and Hartford sites, while the Athens and Boston sites used IR-SPGR (Supplementary Materials, Supplementary Table 1).
Raw pCASL scans (DICOM files [for GE and Siemens scanners] or PAR/REC files [for Philips scanners]) were processed using ExploreASL version 1.9.0, which enables standardized analyses across scanners and sites (29). The ExploreASL pipeline consists of three sequential modules: Structural, ASL, and Population, each performing distinct preprocessing steps, where (1) the Structural module performs tissue segmentation, gray matter and white matter partial volume correction (PVC) map creation, and spatial normalization by transforming structural images to Montreal Neurological Institute standard space; (2) the ASL module generates the quantitative CBF maps (otherwise referred to as ASL images) by conducting motion correction, outlier exclusion, and corrections for vascular artifacts; (3) the Population module creates an analysis mask based on regions with sufficient signal-to-noise ratio and/or statistical power, conducts the PVC using the gray matter and white matter PVC maps (Asllani, Borogovac, & Brown, Reference Asllani, Borogovac and Brown2008), and performs group-level template generation, region-of-interest (ROI) statistics, and quality control.
CBF was quantified using its standardized pCASL model, with protocol-specific acquisition parameters, including labeling duration, PLD, and readout type, incorporated on a per-site and per-protocol basis based on DICOM metadata and manual specification when needed. Thus, while acquisition parameters differed across sites and protocols, CBF quantification was performed within a consistent modeling framework with parameter values adapted as appropriate. Additionally, since PVC was performed at the ROI level on the ASL images in the Population module, all reported CBF measures therefore represent partial-volume-corrected gray-matter CBF, reducing potential bias related to between-group differences in tissue composition. The resulting ASL images underwent both qualitative and quantitative quality control methods to ensure high data quality. Specifically, each ASL image was visually inspected, and scans were excluded due to incidental findings on the structural or ASL images (e.g. lesions or enlarged ventricles, n = 21), severe head motion artifacts (n = 15), and ASL labeling failures (n = 4). Additional exclusions were based on outlier detection, with scans removed if the raw perfusion values or asymmetry index for total gray matter or lobe-wise perfusion exceeded four standard deviations from the site-specific mean (n = 10). To account for site differences, the resulting high-quality ASL images were harmonized using ComBat function from the sva R package (Johnson, Li, & Rabinovic, Reference Johnson, Li and Rabinovic2007; Leek et al., Reference Leek, Johnson, Parker, Jaffe and Storey2012), adjusting for age, sex, and diagnostic group (Leek et al., Reference Leek, Johnson, Parker, Jaffe and Storey2012). ComBat harmonization reduced inter-scanner variance while retaining diagnostic and biological variance. Further details regarding quality control, ComBat harmonization, and outlier removal procedures are described in the Supplementary Material.
This work investigated resting-state harmonized total gray matter CBF (GM CBF) and regional CBF. The former is the average CBF across all gray matter voxels in the brain, derived using a whole-brain gray matter mask. There were no missing data for total GM CBF values. In contrast, regional CBF values from 78 brain regions are used, in which 68 cortical regions are defined by the Desikan-Killiany-Tourville atlas (Klein & Tourville, Reference Klein and Tourville2012) and 10 subcortical regions are defined by the Aseg atlas (Fischl et al., Reference Fischl, Salat, Busa, Albert, Dieterich, Haselgrove, van der Kouwe, Killiany, Kennedy, Klaveness, Montillo, Makris, Rosen and Dale2002). These atlases were used because they provide standardized, well-validated parcellations with improved anatomical consistency that are utilized for FreeSurfer-based processing and facilitate comparison with existing literature. Eighteen regions with less than 85% available CBF data were excluded from analyses. Of the remaining regions, 21 regions had 100% completeness (i.e. no missing data), while 39 regions had 85%–99.9% completeness (i.e. minimal partial missing data). Missing values in these 39 regions were imputed using Amelia II (Honaker, King, & Blackwell, Reference Honaker, King and Blackwell2011), which has been shown to outperform other imputation approaches (Sarkar, Pramanik, Khatedi, & Maiti, Reference Sarkar, Pramanik, Khatedi and Maiti2020). Amelia II was selected because it uses bootstrapped expectation–maximization to preserve the multivariate covariance structure across brain regions, making it well-suited for regional CBF data where anatomical correlations are strong and missingness is sparse. In total, 60 regions were retained for analysis; a complete list of regions and their percent completeness is provided in the Supplementary Material. To assess the impact of missing data, group contrast analyses were conducted using both the original dataset with missing values and the dataset with imputed CBF values. Given that the results are comparable between using the imputed CBF data and non-imputed CBF data (Supplementary Tables 6 and 7), we used the imputed CBF data for all analyses to maximize statistical power.
Statistical analyses
Statistical analyses were performed using R (version 4.1.2). Demographic and clinical variables were assessed using chi-squared tests (for categorical variables) or analysis of variance (ANOVA, for continuous variables). Group differences were computed using general linear models, and the covariates for these models were chosen based on univariate ANOVAs, which test moderator effects of age, sex, self-reported race, handedness, socioeconomic status, intracranial volume, ASL scanning site, duration of illness since first episode, daily dose of CPZ, FTND score, FTND lifetime smoking status on total GM CBF and raw and imputed regional CBF values. The results of these ANOVAs are shown in Supplementary Table 2. Following the moderator analyses, CBF values were examined using progressively complex models with different sets of covariates. For total GM CBF: Model 1 included covariates for age, sex, and intracranial volume; Model 2 included the Model 1 covariates and FTND lifetime smoking status. For regional CBF values: Models 1 and 2 used the same covariates as above, while Model 3 included the Model 2 covariates and self-reported race. The main results are shown using Model 2, unless otherwise specified. Results of models 1 and 3, as well as non-imputed model 2, are in Supplementary Tables 6 and 7. Multiple comparison correction was performed using the Benjamini–Hochberg method; the significance threshold (denoted as “q”) was set at <0.05 (Benjamini & Hochberg, Reference Benjamini and Hochberg1995). Covariate-adjusted Cohen’s d effect sizes were reported.
Partial Spearman correlations were computed between total GM and/or regional CBF values and cognition (BACS) and clinical measures (PANSS, MADRS, and YMRS) within patient groups. Correlations were analyzed separately between these regions and composite clinical scores, and between these regions and BACS scores. These correlations were examined across various groups: the PSD group, diagnostic groups (SZ, SAD, and BP), and biotypes (BT1, BT2, and BT3). For each correlation analysis, p-values from the analysis were corrected for multiple comparisons using the Benjamini–Hochberg FDR procedure across all ROI–cognitive or ROI–clinical measure pairs tested; only correlations surviving FDR correction (q < 0.05) were considered significant. Total GM CBF and regional CBF values were adjusted for age, sex, intracranial volume, and FTND lifetime smoking status; clinical measures were adjusted for age, sex, and socioeconomic status; BACS scores were adjusted for socioeconomic status. Tukey’s honest significant difference test was conducted as a post-hoc analysis to determine the directionality of medication effects on CBF values in patient groups that showed significant regional differences compared to NC (Tukey, Reference Tukey1949).
Results
Sample characteristics
Patients with PSD differed in age, sex, race, socioeconomic status, BACS, FTND lifetime smoking status, and FTND score but did not differ in handedness compared to NC (Supplementary Table 3). Diagnostic groups (SZ, SAD, and BP) differed in age, sex, race, handedness, socioeconomic status, BACS, FTND lifetime smoking status, and FTND score compared to NC; diagnostic groups differed from one another in duration of psychosis illness, PANSS, YMRS, MADRS, but not in daily CPZ dose (Supplementary Table 4). Biotypes differed in age, sex, race, socioeconomic status, BACS, FTND lifetime smoking status, and smoking score compared to NC; biotypes differed from one another in duration of psychosis illness, PANSS, and FTND lifetime smoking status (Supplementary Table 5).
Group contrasts
We first examined the difference in total GM CBF between NC and various patient groups. Compared to NC, PSD demonstrated significantly lower perfusion (d = −0.124, p = 0.037, q = 0.037); BP exhibited significantly lower perfusion (d = −0.259, p = 0.004, q = 0.004) (Figure 1a,c); BT1 exhibited significantly lower perfusion (d = −0.222, p = 0.006, q = 0.006) (Figure 1b,c). For regional CBF, there were no significant differences between PSD and NC. However, compared with NC, BP demonstrated significantly lower perfusion in 17 cortical and 8 subcortical regions, with hypoperfusion predominantly affecting the bilateral temporal, occipital, and inferior frontal lobes, and left parietal lobe, as well as subcortical structures, after adjusting for age, sex, intracranial volume, and lifetime smoking status as measured by the FTND (Figure 2, Supplementary Table 6). Specifically, the five brain regions demonstrating the largest effect sizes are: bilateral thalamus (left: d = −0.326, p < 0.001, q = 0.014; right: d = −0.318, p < 0.001, q = 0.014), left superior temporal (d = −0.298, p < 0.001, q = 0.017), left bank of the superior temporal sulcus (d = −0.295, p < 0.001, q = 0.017), and right putamen (d = −0.284, p = 0.002, q = 0.02). In biotypes, BT1 showed significantly lower perfusion in 19 cortical and 4 subcortical regions (Figure 3, Supplementary Table 7), with the left lingual (d = −0.263, p = 0.001, q = 0.033) and right pars triangularis (d = −0.252, p = 0.002, q = 0.033) showing the highest effect sizes. There were no significant findings for total and regional CBF within diagnostic groups or biotypes.
Total gray matter cerebral blood flow differences between neurotypical controls and patient groups: psychosis spectrum disorder (PSD), bipolar disorder with psychosis (BP), and Biotype 1 (BT1). (a, b) Violin and boxplot showing the distribution of total gray matter CBF; red diamond denotes mean total gray matter CBF per group; q-values of the main Model 2 are displayed. (c) Table displaying the Cohen’s d, p-value, and q-values for all group contrasts for both Model 1 and Model 2.
Figure 1. Long description
Panel A is a violin and boxplot graph. The y-axis is labeled Total Gray Matter C B F ranging from minus 2 to 3. The x-axis is labeled Groups with three categories: N C (light blue), P S D (light green), and B P (light pink). Each violin contains a boxplot and a red diamond representing the mean. Horizontal brackets above the plots indicate q-value equals 0.037 between N C and P S D, and q-value equals 0.004 between N C and B P.
Panel B is a similar violin plot. The y-axis is identical. The x-axis compares N C (light blue) and B T 1 (light orange). A bracket indicates a q-value equals 0.006 between these two groups.
Panel C is a table titled Model, Cohen’s d, p-value, and q-value.
* For N C (n equals 401) vs P S D (n equals 626): Model 1 has d equals 0.144, p equals 0.022, q equals 0.022. Model 2 has d equals 0.124, p equals 0.037, q equals 0.037.
* For N C (n equals 401) vs B P (158): Model 1 has d equals 0.271, p equals 0.004, q equals 0.004. Model 2 has d equals 0.259, p equals 0.004, q equals 0.004.
* For N C (n equals 382) vs Biotype 1 (n equals 179): Model 1 has d equals 0.244, p equals 0.005, q equals 0.005. Model 2 has d equals 0.222, p equals 0.006, q equals 0.006.
A note at the bottom defines Model 1 as adjusted for age, sex, and intracranial volume, and Model 2 as Model 1 plus Fagerström Test for Nicotine Dependence lifetime smoking status.
Regional cerebral blood flow differences between individuals with bipolar disorder with psychotic features and neurotypical controls. (a) Bar plot of regional effect sizes, ordered from lowest to highest Cohen’s d, with corresponding p-values and q-values. (b) Neuroanatomical visualization of regional differences, with regions colored by Cohen’s d (lighter blue indicates smaller effect sizes; darker blue indicates larger effect sizes).
Figure 2. Long description
Panel A is a horizontal bar plot titled Cohen d on the x-axis, ranging from negative 0.35 to 0.00. The y-axis lists 23 brain regions with their corresponding p and q values. The bars are colored in a gradient from light blue to dark blue as the effect size becomes more negative. The list begins at the top with Left Precuneus (p = 0.020, q = 0.048) and ends at the bottom with the largest effect sizes in the Right Thalamus (p < 0.001, q = 0.014) and Left Thalamus (p < 0.001, q = 0.014).
Panel B contains five brain visualizations. The top row shows Left Lateral and Right Lateral views. The middle row shows Left Medial and Right Medial views. The bottom row shows a coronal cross-section labeled Left and Right. Regions are shaded in blue based on the Cohen d scale shown in a legend on the right, where light blue represents negative 0.15 and dark blue represents negative 0.35. The darkest blue regions, indicating the most significant blood flow differences, are concentrated in the central thalamic and subcortical areas in the coronal view and the temporal and cingulate regions in the lateral and medial views.
Regional cerebral blood flow differences between Biotype 1 individuals and neurotypical controls. (a) Bar plot of regional effect sizes, ordered from lowest to highest Cohen’s d, with corresponding p-values and q-values. (b) Neuroanatomical visualization of regional differences, with regions colored by Cohen’s d (lighter blue indicates smaller effect sizes; darker blue indicates larger effect sizes).
Figure 3. Long description
Panel A is a horizontal bar plot. The x-axis represents Cohen’s d values ranging from negative 0.35 to 0.00. The y-axis lists 22 brain regions. From top to bottom, the regions show a trend of increasing effect size (more negative Cohen’s d). Key regions include the Right Cuneus (p equals 0.017, q equals 0.043) at the top with the smallest effect, and the Left Lingual (p equals 0.001, q equals 0.034) at the bottom with the largest effect. Other regions listed include the Hippocampus, Superior Temporal Sulcus, Pars Opercularis, Anterior Cingulate, Orbitofrontal cortex, Thalamus, Fusiform, Insula, and Isthmus Cingulate for both hemispheres.
Panel B contains five anatomical brain visualizations. Four cortical surface views (Left Lateral, Right Lateral, Left Medial, and Right Medial) show specific regions highlighted in shades of blue. Below these is a coronal cross-section showing subcortical regions. A color scale legend titled Cohen’s d indicates that lighter blue represents negative 0.15 and transitions to a dark blue representing negative 0.35. The highlighted regions in the brain maps correspond to the blue bars in Panel A, indicating lower cerebral blood flow in Biotype 1 individuals compared to controls.
Perfusion relationships with cognition, symptoms, and medications
Associations between total GM CBF and regional CBF with cognitive performance (BACS) and clinical outcomes (PANSS, YMRS, and MADRS) were assessed. No significant associations were found between CBF measures and cognitive performance in PSD, individual diagnostic group, or biotype subgroups. Within PSD, total GM CBF showed positive correlations with both PANSS Total scores and PANSS Negative subscale scores, indicating that higher global perfusion was associated with greater symptom severity (Figure 4). At the regional level, multiple brain areas demonstrated significant associations with clinical measures (Figure 4), with higher perfusion in the right bank of the superior temporal sulcus correlating with greater severity across multiple symptom domains: general psychotic symptoms (PANSS General Total, Spearman rho = 0.144, q = 0.015), positive symptoms (PANSS Positive Total, rho = 0.166, q = 0.006), negative symptoms (PANSS Negative Total, rho = 0.136, q = 0.017), total psychotic symptoms (PANSS Total, rho = 0.172, q = 0.005), mania (YMRS, rho = 0.108, q = 0.039), and depression (MADRS, rho = 0.11, q = 0.037). PANSS Negative scores showed the most widespread regional associations across all clinical measures. Biotype analyses revealed similar patterns to PSD, with total GM CBF positively correlating with PANSS Total and Negative scores, and PANSS Negative scores again demonstrating the most extensive regional associations, particularly in the left posterior cingulate (Figure 5). No significant associations were found between CBF measures and clinical measures in BP or BT1.
Associations of total gray matter and regional cerebral blood flow with clinical measures within individuals with psychosis spectrum disorders. Tile color reflects Spearman’s ρ (light purple = weaker; dark purple = stronger). Clinical measures: PANSS (Positive and Negative Syndrome Scale: Total, General, Positive, Negative), YMRS (Young Mania Rating Scale), and MADRS (Montgomery–Åsberg Depression Rating Scale). Text within tiles reports ρ and q-values.
Figure 4. Long description
A heatmap grid with clinical measures on the horizontal X-axis and brain regions on the vertical Y-axis.
Horizontal X-axis labels from left to right:
* P A N S S Total
* P A N S S General Total
* P A N S S Positive Total
* P A N S S Negative Total
* Y M R S Total
* M A D R S Total
Vertical Y-axis is split into two main sections: Left hemisphere regions (top) and Right hemisphere regions (bottom), with a final row for Total G M C B F.
Data Visualization:
* Purple tiles indicate significant correlations, with darker purple representing higher Spearman rho values up to 0.25.
* Gray tiles indicate non-significant associations.
* Each significant tile contains text for rho and q-values.
Key Observations:
* The Left Lateral Occipital and Left Inferior Temporal regions show consistent significant correlations across the four P A N S S measures.
* The Right Cuneus and Right Bank of the Superior Temporal Sulcus show the strongest correlations (darkest purple), with rho values ranging from 0.136 to 0.183 and q-values as low as 0.0035.
* The M A D R S Total column has the fewest significant associations, limited primarily to the Left Rostral Anterior Cingulate, Left Posterior Cingulate, Right Posterior Cingulate, and Right Bank of the Superior Temporal Sulcus.
* Total G M C B F shows significant correlations with P A N S S Total (rho equals 0.108, q equals 0.041) and P A N S S Negative Total (rho equals 0.11, q equals 0.038).
Associations of total gray matter and regional cerebral blood flow with clinical measures within Biotypes Tile color reflects Spearman’s ρ (light purple = weaker; dark purple = stronger). Clinical measures: PANSS (Positive and Negative Syndrome Scale: Total, General, Positive, Negative), YMRS (Young Mania Rating Scale), and MADRS (Montgomery–Åsberg Depression Rating Scale). Text within tiles reports ρ and q-values.
Figure 5. Long description
The heatmap is organized with clinical measures on the x-axis and brain regions on the y-axis. The x-axis includes P A N S S Total, P A N S S General Total, P A N S S Positive Total, P A N S S Negative Total, Y M R S Total, and M A D R S Total. The y-axis is divided into three sections: Left hemisphere regions, Right hemisphere regions, and Total G M C B F. A color scale on the right indicates Spearman rho values from 0.00 (white) to 0.25 (dark purple).
* Left Hemisphere Section: Contains 16 regions. Significant correlations (purple tiles) are most dense in the P A N S S columns. Notable regions with multiple correlations include Left Superior Parietal, Left Precuneus, Left Posterior Cingulate, Left Middle Temporal, Left Lateral Occipital, and Left Bank of the Superior Temporal Sulcus. Rho values typically range from 0.106 to 0.165.
* Right Hemisphere Section: Contains 12 regions. High density of correlations is seen in P A N S S Total and P A N S S Negative Total. The Right Cuneus and Right Bank of the Superior Temporal Sulcus show the strongest correlations, with rho values reaching 0.179 and 0.172 respectively.
* Total G M C B F Row: Shows significant correlations with P A N S S Total (rho = 0.108, q = 0.046) and P A N S S Negative Total (rho = 0.115, q = 0.033).
* General Trends: The Y M R S Total and M A D R S Total columns have the fewest significant correlations across all brain regions compared to the P A N S S measures.
In PSD, the use of anticholinergic medications was significantly associated with lower total GM CBF (Tukey’s HSD = −0.234, q = 0.019), while using psychotropic medications (Tukey’s HSD = 0.282, q = 0.031) was associated with greater total GM CBF (Supplementary Table 8). Regionally in BP, the use of first-generation antipsychotics, lithium, analgesics, antibiotics, psychotropic medications, and vitamins was associated with greater regional CBF across various cortical and subcortical areas, with lithium showing the most widespread effects; conversely, using mood-stabilizing anticonvulsant, anxiolytic, or gastrointestinal medications was associated with lower CBF in the right amygdala and other regions (Supplementary Table 8). In BT1, using second-generation antipsychotics, lithium, immunologic, antidepressants, mood stabilizers, endocrine medications, selective serotonin reuptake inhibitors, or immunosuppressants was associated with higher regional CBF across multiple regions. Conversely, using anticholinergic, antiallergic, and antibiotic medications was associated with lower CBF (Supplementary Table 8).
Discussion
Using the largest PSD cohort to-date, this study examined total GM CBF and regional CBF patterns across a combined psychosis group, traditional diagnostic categories, and biotypes and revealed three key findings. First, individuals with PSD, BT1, and BP exhibited significantly lower total GM CBF compared to NC. Second, BT1 and BP had hypoperfusion in various brain regions, primarily the parietal, temporal, insular, and subcortical areas, with the most pronounced effects occurring in the bilateral thalamus, left superior temporal, left banks of the superior temporal sulcus, and right putamen. Third, higher total GM CBF and regional CBF were associated with greater psychotic symptom severity within patient groups, particularly for negative symptoms.
Widespread hypoperfusion across PSD aligns with previous neuroimaging studies (Boisvert et al., Reference Boisvert, Lungu, Pilon, Dumais and Potvin2023; Percie du Sert et al., Reference Percie du Sert, Unrau, Gauthier, Chakravarty, Malla, Lepage and Raucher-Chéné2023; Toma, MacIntosh, Swardfager, & Goldstein, Reference Toma, MacIntosh, Swardfager and Goldstein2018), though no significant differences in SZ or SAD groups were observed in this study. This discrepancy could be partially explained by the chronic nature and psychopharmacological landscape of our SZ and SAD populations, which are known to affect regional CBF (Gitelman & Prohovnik, Reference Gitelman and Prohovnik1992; Goozée, Handley, Kempton, & Dazzan, Reference Goozée, Handley, Kempton and Dazzan2014; Miller et al., Reference Miller, Andreasen, O’Leary, Rezai, Watkins, Boles Ponto and Hichwa1997; Mori et al., Reference Mori, Yamashita, Nagao, Horiguchi and Yamawaki2002), combined with our rigorous control for age, sex, and lifetime smoking status. The most pronounced regional alterations emerged in BT1, with hypoperfusion in the right caudal anterior cingulate, left lingual gyrus, left pars triangularis, right posterior cingulate, and right lingual cortex. Because BT1 reflects a biologically derived subtype, these regional deficits carry translational relevance: they may pinpoint conserved neural circuits underlying symptom dimensions that cut across traditional diagnoses and could guide future biomarker or treatment-target development. BP also showed distinct hypoperfusion, concentrated in regions involved in emotional processing, sensory integration, and salience detection, including the left superior temporal sulcus/gyrus, bilateral thalamus, and left insula (Lahnakoski et al., Reference Lahnakoski, Glerean, Salmi, Jääskeläinen, Sams, Hari and Nummenmaa2012; Sherman, Reference Sherman2007; Uddin et al., Reference Uddin, Nomi, Hébert-Seropian, Ghaziri and Boucher2017). This hypoperfusion pattern in BP is consistent with previous studies (Luciw, Toma, Goldstein, & MacIntosh, Reference Luciw, Toma, Goldstein and MacIntosh2021; Mio et al., Reference Mio, Dimick, Toma, MacIntosh, Goldstein and Strakowski2022; Toma et al., Reference Toma, MacIntosh, Swardfager and Goldstein2018) but differs from reports of elevated or mixed CBF findings in other BP samples (Dai et al., Reference Dai, Chen, Duan, Zhao, Bolo, Tamminga, Clementz, Pearlson, Alsop and Keshavan2020; Delvecchio, Gritti, Squarcina, & Brambilla, Reference Delvecchio, Gritti, Squarcina and Brambilla2022; Sritharan et al., Reference Sritharan, Zeng, Petr, Mutsaerts, Hoang, Bolo, Ivleva, Dai, Gershon, Keedy, Parker, Trotti, McDowell, Clementz, Tamminga, Pearlson, Keshavan and Lizano2025; Zeng et al., Reference Zeng, Lizano, Bolo, Lutz, Brady, Ivleva, Dai, Clementz, Tamminga, Pearlson and Keshavan2021). Much of this inconsistency likely reflects differences in analytical modeling (e.g. inclusion of total GM CBF as a covariate) and sample composition: our BP cohort was substantially larger and consisted exclusively of individuals with a psychosis history, unlike earlier work that combined BP patients with and without psychosis (Zeng et al., Reference Zeng, Lizano, Bolo, Lutz, Brady, Ivleva, Dai, Clementz, Tamminga, Pearlson and Keshavan2021). This distinction may explain the contrasting directional effects and supports emerging evidence that psychotic versus non-psychotic BP represent distinct neurobiological subtypes, warranting further investigation (Lutz et al., Reference Lutz, Lizano, Mothi, Zeng, Hegde, Hoang, Henson, Brady, Tamminga, Pearlson, Clementz, Sweeney and Keshavan2020). Together, these findings highlight shared and diagnosis-specific cerebrovascular signatures across the psychosis spectrum, underscoring the need for future work to disentangle how psychosis history, biological subtypes, and analytic choices shape observed perfusion patterns.
Despite the overall pattern of decreased perfusion relative to NC, we examined the critical relationship between CBF and clinical symptom severity across psychotic symptoms (PANSS), mania (YMRS), and depression (MADRS) within our patient groups. Higher total GM CBF was associated with greater psychotic symptom severity within both PSD and biotype groups, particularly for negative symptoms. One interpretation is that relatively higher perfusion within a hypoperfused group may reflect an attempt by the brain to compensate for dysfunction, yet this compensatory response is maladaptive as it is associated with more severe symptoms. Alternatively, hypoperfusion itself could represent a compensatory process, such that individuals who fail to exhibit this hypoperfusion exhibit worse clinical presentations. To test which hypothesis is true, longitudinal CBF and clinical data are highly useful, ideally in samples where medication effects can be minimized or rigorously controlled. Regional analyses revealed that higher perfusion in the right banks of the superior temporal sulcus was significantly associated with worsening across all clinical domains, suggesting this region may serve as a critical hub where cerebrovascular dysfunction intersects with diverse symptom presentations. Negative symptoms demonstrated the most extensive associations with perfusion alterations across multiple brain regions, with the right cuneus and left lingual being the most impacted, indicating that regions linked to negative symptomatology may be particularly vulnerable to cerebrovascular dysfunction, consistent with prior findings (Gangl et al., Reference Gangl, Conring, Federspiel, Wiest, Walther and Stegmayer2023). Importantly, these associations were present at the PSD and biotype level but were not detected within BP or BT1 individually, suggesting that the relationships of CBF and symptoms may reflect broader psychosis pathophysiology rather than diagnosis- or subtype-specific mechanisms. Together, these findings highlight the complex relationship between cerebral perfusion and clinical phenotypes, where regional higher perfusion within PSD may signal maladaptive processes rather than healthy brain function.
Four interrelated mechanisms may contribute to cerebrovascular dysfunction in PSD: neuroinflammation, neurovascular coupling dysfunction, neurotransmitter dysregulation, and structural brain abnormalities. Neuroinflammation, well-documented in psychosis (De Picker, Morrens, Chance, & Boche, Reference De Picker, Morrens, Chance and Boche2017; Hoang et al., Reference Hoang, Xu, Lutz, Bannai, Zeng, Bishop, Keshavan and Lizano2022; Lizano et al., Reference Lizano, Keshavan, Tandon, Mathew, Mothi, Montrose and Yao2016, Reference Lizano, Lutz, Ling, Lee, Eum, Bishop, Kelly, Pasternak, Clementz, Pearlson, Sweeney, Gershon, Tamminga and Keshavan2019, Reference Lizano, Lutz, Xu, Rubin, Paskowitz, Lee, Eum, Keedy, Hill, Reilly, Wu, Tamminga, Clementz, Pearlson, Gershon, Keshavan, Sweeney and Bishop2021, Reference Lizano, Kiely, Mijalkov, Meda, Keedy, Hoang, Zeng, Lutz, Pereira, Ivleva, Volpe, Xu, Lee, Rubin, Kristian Hill, Clementz, Tamminga, Pearlson, Sweeney and Bishop2023), involves pro-inflammatory cytokines disrupting vascular function by altering vasoconstrictor-vasodilator balance and compromising blood–brain barrier integrity, leading to increased or decreased perfusion, depending on whether inflammation is acute or chronic (Cyrino, Delwing-de Lima, Ullmann, & Maia, Reference Cyrino, Delwing-de Lima, Ullmann and Maia2021; Van Kesteren et al., Reference Van Kesteren, Gremmels, De Witte, Hol, Van Gool, Falkai, Kahn and Sommer2017; Yang et al., Reference Yang, Ran, Li, Lin, Ma, Yang, Fu and Yang2022). This inflammatory cascade directly affects neurovascular coupling, the critical process matching neuronal activity with appropriate blood supply to meet metabolic demands (Cyrino et al., Reference Cyrino, Delwing-de Lima, Ullmann and Maia2021; Van Kesteren et al., Reference Van Kesteren, Gremmels, De Witte, Hol, Van Gool, Falkai, Kahn and Sommer2017), resulting in insufficient or excessive perfusion that compromises brain function in psychotic disorders (Chen et al., Reference Chen, Xue, Yang, Wang, Xu, Wen, Cheng, Han and Wei2022; Najjar et al., Reference Najjar, Pahlajani, De Sanctis, Stern, Najjar and Chong2017). Neurotransmitter systems, particularly glutamate and dopamine, are intimately connected to perfusion regulation through astrocyte-mediated mechanisms (Koehler, Roman, & Harder, Reference Koehler, Roman and Harder2009; McEntee & Crook, Reference McEntee and Crook1993; O’Connor, Finger, Flor, & Cryan, Reference O’Connor, Finger, Flor and Cryan2010; Speranza et al., Reference Speranza, di Porzio, Viggiano, de Donato and Volpicelli2021). Glutamate hypofunction, specifically NMDA receptor dysfunction, triggers excessive glutamate release that disrupts neurovascular coupling and creates inadequate blood flow, correlating with negative symptoms in psychosis (Kruse & Bustillo, Reference Kruse and Bustillo2022; Mei, Wu, & Zhou, Reference Mei, Wu and Zhou2018). Finally, gray and white matter structural abnormalities and connectivity disruptions in psychotic disorders directly impact CBF distribution and regulation (Hoang et al., Reference Hoang, Lizano, Lutz, Zeng, Raymond, Miewald, Montrose and Keshavan2021; Keshavan et al., Reference Keshavan, Collin, Guimond, Kelly, Prasad and Lizano2020; Van Erp et al., Reference Van Erp, Walton, Hibar, Schmaal, Jiang, Glahn, Pearlson, Yao, Fukunaga and Hashimoto2018; White et al., Reference White, Magnotta, Bockholt, Williams, Wallace, Ehrlich, Mueller, Ho, Jung, Clark, Lauriello, Bustillo, Schulz, Gollub, Andreasen, Calhoun and Lim2011). These mechanisms are likely interdependent rather than mutually exclusive, necessitating future research to elucidate their complex interactions in psychotic spectrum pathophysiology.
Regarding medication effects on CBF, anticholinergic agents were associated with decreased perfusion, while psychotropic medications and lithium were linked to higher total GM perfusion in PSD patients. The former findings align with established evidence that these drugs reduce CBF and impair cognition in PSD (Gitelman & Prohovnik, Reference Gitelman and Prohovnik1992; Mori et al., Reference Mori, Yamashita, Nagao, Horiguchi and Yamawaki2002). Although we did not observe direct correlations between perfusion and cognitive performance, these findings suggest that anticholinergic-induced vascular alterations may represent one pathway through which these medications exert their known cognitive effects (Zhang et al., Reference Zhang, Ivleva, Parker, Hill, Lizano, Keefe, Keedy, McDowell, Pearlson, Clementz, Keshavan, Gershon, Tamminga, Sweeney and Bishop2025). Conversely, the association of psychotropic medications with higher perfusion may potentially reflect therapeutic effects on neurovascular coupling or compensatory mechanisms, though the heterogeneity within this broad category necessitates careful interpretation (Goozée et al., Reference Goozée, Handley, Kempton and Dazzan2014; Miller et al., Reference Miller, Andreasen, O’Leary, Rezai, Watkins, Boles Ponto and Hichwa1997). The association of lithium with higher perfusion is particularly noteworthy given its established neuroprotective properties and mood-stabilizing effects in BP (Malhi et al., Reference Malhi, Tanious, Das, Coulston and Berk2013; Young, Reference Young2009), potentially suggesting improved cerebrovascular health and neuroplasticity. Given the broad spectrum of medications and their variable effects on perfusion, future studies should prioritize detailed pharmacological profiling to elucidate specific drug-perfusion relationships and their clinical implications for optimizing therapeutic outcomes in PSD.
The primary strengths of this study include collecting and analyzing CBF in the largest PSD cohort to-date and minimizing scanner-dependent differences through rigorous quality control and ComBat harmonization procedures, which effectively reduce technical variance while preserving biological differences. Several limitations are important while interpreting the results of this study: first, the cross-sectional design limits causal inference. Second, medication effects cannot be ruled out as explanatory factors, and our categorical medication groupings mask dose-dependent relationships. Third, the absence of task-based CBF measurements limits our ability to assess dynamic neurovascular responses. Fourth, the predominance of findings in BP patients requires independent replication. Fifth, PLDs vary across sites, with some lower than ideal. Although within published ranges and fixed within site, PLD differences may affect CBF sensitivity, especially in regions with prolonged arterial transit times. ComBat reduced site- and protocol-related differences, but residual acquisition heterogeneity may remain. Future work should prioritize longitudinal designs, compare perfusion differences between BP with and without psychosis history, include task-based ASL paradigms, and apply medication-specific analyses and greater standardization of ASL acquisition parameters in multi-site studies. Additionally, mechanistic studies examining interactions between neuroinflammation, neurotransmitter dysfunction, and cerebrovascular alterations will be crucial for developing targeted interventions that address both vascular and symptomatic aspects of PSD. In conclusion, our findings demonstrate that altered cerebral perfusion is an important feature of psychosis and underscore the potential of CBF metrics to guide biologically grounded stratification and treatment development.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726105844.
Data availability statement
The datasets presented in this article are not publicly available as the B-SNIP consortium principal investigators have determined that ASL data sharing is not yet available. Requests to access the datasets should be directed to the corresponding author at lizanopl@gmail.com.
Acknowledgments
The authors thank the participants and families who took part in this study. The authors also thank Neha Iska for her thoughtful review of the manuscript.
Author contribution
Conceptualization: D.H., P.L.; Data curation: V.Z., N.B.; Formal analysis: D.H.; Funding acquisition: E.G., G.P., M.K., C.T., B.C.; Methodology: V.Z., J.S., H.-J.M., J.P., R.T., N.B., P.L.; Project administration: M.K., P.L.; Software: V.Z.; Supervision: M.K., P.L.; Visualization: D.H.; Writing – original draft: D.H.; Writing – review & editing: V.Z., J.S., H.-J.M., J.P., R.T., E.I., W.D., E.G., S.K., B.C., C.T., G.P., N.B., M.K., P.L.
Funding statement
This study was supported by: E.I.: K23-MH102656; S.K.: R01-MH103368; B.C.: R01- MH103366; C.T.: R01-MH077851, R01-MH096913; G.P.: R01-MH077945, R01-MH096957; M.K.: R01-MH078113, R01-MH096942.
Competing interests
The authors declare no conflicts of interest related to this study or the preparation of the manuscript.
Ethical standard
This study was conducted in accordance with the Declaration of Helsinki and approved by the institutional review boards at UT Southwestern Medical Center, Beth Israel Deaconess Medical Center, Yale University, The University of Chicago, and The University of Georgia. All participants provided written informed consent prior to study participation.
Open access
