Skip to main content
NCBI home page
Cerebral Cortex (New York, NY) logoLink to Cerebral Cortex (New York, NY)
. 2019 Dec 9;30(5):2939–2947. doi: 10.1093/cercor/bhz285

Associating Psychotic Symptoms with Altered Brain Anatomy in Psychotic Disorders Using Multidimensional Item Response Theory Models

Ana D Stan 1, Carol A Tamminga 1,✉, Kihwan Han 2, Jong Bae Kim 7, Jaya Padmanabhan 3, Neeraj Tandon 3, Matthew E Hudgens-Haney 1, Matcheri S Keshavan 3, Brett A Clementz 4, Godfrey D Pearlson 5, John A Sweeney 6, Robert D Gibbons 7
PMCID: PMC7197072  PMID: 31813988

Abstract

Reduced cortical thickness has been demonstrated in psychotic disorders, but its relationship to clinical symptoms has not been established. We aimed to identify the regions throughout neocortex where clinical psychosis manifestations correlate with cortical thickness. Rather than perform a traditional correlation analysis using total scores on psychiatric rating scales, we applied multidimensional item response theory to identify a profile of psychotic symptoms that was related to a region where cortical thickness was reduced. This analysis was performed using a large population of probands with psychotic disorders (N = 865), their family members (N = 678) and healthy volunteers (N = 347), from the 5-site Bipolar-Schizophrenia Network for Intermediate Phenotypes. Regional cortical thickness from structural magnetic resonance scans was measured using FreeSurfer; individual symptoms were rated using the Positive and Negative Syndrome Scale, Montgomery-Asberg Depression Rating Scale, and Young Mania Rating Scale. A cluster of cortical regions whose thickness was inversely related to severity of psychosis symptoms was identified. The regions turned out to be located contiguously in a large region of heteromodal association cortex including temporal, parietal and frontal lobe regions, suggesting a cluster of contiguous neocortical regions important to psychosis expression. When we tested the relationship between reduced cortical surface area and high psychotic symptoms we found no linked regions describing a related cortical set.

Keywords: bifactor model, cortical thickness, high-dimensional datasets, numerical taxonomies

Introduction

Cortical thickness alterations are well established in psychosis syndromes (Gutierrez-Galve et al. 2010; Winkler et al. 2010; Buchy et al. 2011; Buchy et al. 2011; van Haren et al. 2011; Gong et al. 2016; Hanford et al. 2016; Knochel et al. 2016). The high heritability and regional nature of cortical thickness measures (Winkler et al. 2010) make them potentially informative brain imaging phenotypes. Cross-sectional examination of cortical thickness in schizophrenia (SZ) characteristically shows cortical thinning, particularly in the temporal and frontal lobes (van Haren et al. 2011). This pattern also has been reported in subjects at ultra-high risk for psychosis (Jung et al. 2011), in first-episode psychosis (Gutierrez-Galve et al. 2010; Buchy et al. 2011; Gong et al. 2016) and in chronic stages of illness (Knochel et al. 2016). Similarly, bipolar disorder has been associated with cortical thinning in the left anterior cingulate and para-cingulate cortices and other regions of the temporal and frontal lobes (Hanford et al. 2016; Knochel et al. 2016). We sought to examine whether there were specific regions of cortical thinning that were associated with a profile of high psychosis symptom ratings, using cortical thinning as a marker of brain regions important to psychosis symptom manifestations.

The relationship between morphometric alterations and clinical symptom severity typically has been found to be modest across psychosis syndromes. However, previous efforts to examine these relationships have typically examined associations of global thickness measures with total scores on psychiatric rating scales and have done so individually for each test and in individual psychosis syndromes (Hoptman et al. 2014; Tamminga et al. 2017).

We selected an alternative statistical approach to investigating brain–behavior relations by implementing a model-based measurement approach to examine the relations between mixtures of psychosis items (phenomenology) and regional cortical thickness measurements (neurobiology), independent of associations within each subset of measures. To achieve this aim, we used a multidimensional item response theory (MIRT) model (Bock and Aitkin 1981) known as the bifactor IRT model (Gibbons and Hedeker 1992). The bifactor model integrates data across different phenomenology and biology domains in a single model using all data simultaneously. The model can be used to determine the nature and magnitude of relevant brain–behavior relationships. Importantly, this approach identifies secondary dimensions that preserve domain-specific associations between symptoms and cortical thicknesses within specific brain regions. To our knowledge this is the first example of an IRT model of any kind that has synthesized information across phenomenological and biological domains in psychosis.

On the one hand, traditional unidimensional IRT ignores the nesting of symptoms within conventional diagnostic syndromes (i.e., SZ, bipolar disorder, and depression) and the nesting of cortical thickness measures within brain areas. On the other hand, more advanced correlational techniques such as canonical correlation and traditional factor analysis ignore the discrete nature of individual or item-level ratings of psychiatric symptoms, and segment variables into unique subsets typically defined by the type of biological versus psychological domain from which the item was drawn (brain region or diagnostic group). By contrast, the bifactor model can span characteristics of neuroimaging biomarkers and psychiatric symptomatology, while preserving the domain-specific correlation among the measures.

The goal of this analysis was to associate regional brain anatomic changes with symptoms of psychosis, each analyzed at an item level, to recognize regions of brain associated with psychosis manifestations. By examining both domains simultaneously while maintaining item-level symptom ratings and regional cortical thickness measures as inputs to the model simultaneously in the MIRT, our aim was to enhance precision in characterizing this relationship over what has been possible with traditional correlational approaches.

Materials and Methods

Study Participants and Clinical Characterization

We included data from a total of 2450 individuals collected by the B-SNIP consortium. Inclusion criteria, as well as procedures for subject recruitment and diagnosis have been previously published (Tamminga et al. 2013). All participants underwent a thorough psychiatric characterization, including a SCID diagnostic interview and clinical symptom assessments [including symptom ratings on the Positive and Negative Syndrome Scale (PANSS) (Kay et al. 1987), the Young Mania Rating Scale (YMRS) (Young et al. 1978), and the Montgomery-Asberg Depression Rating Scale (MADRS) (Montgomery and Asberg 1979)].

Out of the 2450 subjects in the full B-SNIP sample, we excluded the 560 persons who lacked both complete symptom score ratings and cortical thickness measurements. The remaining sample of 1890 participants included 1) 865 probands with a psychotic disorder, including SZ, schizoaffective disorder (SZA), and psychotic bipolar I disorder (BP), 2) 678 of their first-degree relatives, and 3) 347 healthy controls. First-degree relatives with and without a psychiatric diagnosis were included. All probands were clinically stable and receiving consistent psychiatric medication treatment in the community during the previous month (Tamminga et al. 2013) (Table 1.).

Table 1.

Clinical characteristics of the study sample

Probands Relatives Normal controls Statistics
SZP SZAP BPP SZR SZAR BPR
Number of participants 357 217 291 277 177 224 347 N/A
Age Mean, (StdDev) 35.8 (12.8) 36.6 (12.0) 36.2 (12.7) 42.2 (14.9) 39.3 (13.8) 39.1 (15.6) 36.9 (12.4) Three proband group ANOVA: F(2,862) = 0.31, P = 0.73 Three relative group ANOVA: F(2, 676) = 3.2, P = 0.04 All seven group ANOVA: F(6, 1884) = 8.12, P < 0.0001 post hoc Dunnett’s multiple comparison tests of the seven group ANOVA revealed significant difference between NC and SZR.
Gender 242 M, 115F 90 M, 127F 103 M, 188F 86 M, 191F 58 M, 119F 79 M, 145F 160 M, 187F Three proband group Chi-square: P < 0.0001 Three relative group Chi-square: P = 0.6 All seven group Chi-square: P < 0.001
Ethnicity nonHispanic/Hispanic 332/25 196/21 269/22 248/29 160/17 206/18 319/28 Three proband group Chi-square: P = 0.54 Three relative group Chi-square: P = 0.64 All seven group Chi-square: P = 0.77
YMRS Mean, (StdDev) 5.7 (5.8) 7.7 (6.7) 6.1 (7.0) 3.5 (4.9) 4.4 (5.2) 4.2 (5.8) 1.4 (2.1) Three proband group ANOVA: F(2, 829) = 6.78, P = 0.0012 Three relative group ANOVA: F(2,195) = 0.62, P = 0.53 All seven group ANOVA: F(6, 1071) = 10.58, P < 0.0001 post hoc Dunnett’s multiple comparison tests of the seven group ANOVA revealed significant difference between NC and SZP (P < 0.0001), NC and SZAP (P < 0.0001), and NC and BPP (P < 0.0001)
MADRS Mean, (StdDev) 8.3 (8.2) 14.7 (10.0) 10.4 (9.6) 6.5 (8.4) 9.3 (9.6) 7.6 (9.0) 2.0 (2.9) Three proband group ANOVA: F(2, 834) = 31, P < 0.0001 Three relative group ANOVA: F(2, 205) = 1.637, P = 0.2 All seven group ANOVA: F(6, 1083) = 20.45, P < 0.0001 post hoc Dunnett’s multiple comparison tests of the seven group ANOVA revealed significant difference between NC and each of the proband and relative groups: SZP (P < 0.0001), SZAP (P < 0.0001), BPP (P < 0.0001), SZR (P = 0.025), SZAR, (P = 0.0002), BPR (P = 0.007)
PANSS Positive Mean, (StdDev) 16.8 (5.8) 18.1 (5.2) 13.0 (4.6) 15.2 (5.9) 14.4 (5.2) 13.4 (5.3) 8.6 (1.6) Three proband group ANOVA: F(2, 836) = 68.77, P < 0.0001 Three relative group ANOVA: F(2, 128) = 1.26, P = 0.28 All seven group ANOVA: F(6,969) = 26.17, P < 0.0001 post hoc Dunnett’s multiple comparison tests of the seven group ANOVA revealed significant difference between NC and SZAP (P < 0.0001), NC and SZP (P = 0.0011), and NC and SZR (P = 0.02)
PANSS Negative Mean, (StdDev) 16.8 (5.9) 16.0 (5.1) 11.9 (3.9) 14.2 (5.3) 13.1 (5.2) 12.5 (4.8) 8.0 (1.2) Three proband group ANOVA: F(2, 835) = 77.58, P < 0.0001 Three relative group ANOVA: F(2,128) = 1.23, P = 0. 29 All seven group ANOVA: F(6, 948) = 30.22, P < 0.0001 post hoc Dunnett’s multiple comparison tests of the seven group ANOVA revealed significant difference between NC and SZAP (P = 0.0003), NC and SZP (P = 0.0001), and NC and SZR (P = 0.0137)
PANSS General Mean, (StdDev) 32.4 (9.1) 35.2 (18.6) 28.9 (8.2) 29.4 (9.1) 30.6 (9.4) 29.6 (8.1) 18.6 (3.0) Three proband group ANOVA F(2,835) = 32.62, P < 0.0001 Three relative group ANOVA F(2, 128) = 0.12, P = 0.78 All seven group ANOVA F(6, 968) =14.14, P < 0.001 post hoc Dunnett’s multiple comparison tests of the seven group ANOVA revealed significant difference between NC and each of the proband and relative groups: SZP (P = 0.008), SZAP (P < 0.0001), BPP (P = 0.0238), SZR (P = 0.0224), SZAR, (P = 0.0089), BPR (P = 0.0224)
PANSS Total Mean, (StdDev) 64.6 (16.0) 68.1 (14.6) 53.8 (14.3) 58.9 (17.4) 57.1 (16.2) 55.5 (15.2) 35.3 (5.1) Three proband group ANOVA F(2, 816) = 63.78, P < 0.0001 Three relative group ANOVA F(2, 127) = 0.49, P = 0.61 All seven group ANOVA F (6, 968) = 14.14, P < 0.0001 Posthoc Dunnett’s multiple comparison tests of the seven group ANOVA revealed significant difference between NC and each of the proband and relative groups: SZP (P < 0.0001), SZAP (P < 0.0001), BPP (P = 0.018), SZR (P = 0.0021), SZAR, (P = 0.0059), BPR (P = 0.0138)
Open in a new tab

MRI Structural Brain Imaging

T1-weighted structural images were acquired on 3-Tesla magnets across five sites. T1-weighted Magnetization Prepared Rapid Gradient Echo (MPRAGE) or Inversion Recovery-Prepared Spoiled Gradient-Echo (IR-SPGR) sequences, as appropriate for scanner brands, were administered following the Alzheimer’s Disease Neuroimaging Initiative (ADNI1) protocol (http://adni.loni.usc.edu/methods/documents/mri-protocols/). Different magnets were used across sites: 1) GE Signa, 2) Philips Achieva, 3) Siemens Allegra, 4) Siemens Trio, and 5) GE Signa HDxt; Siemens Trio. Scan protocols were acquired using similar sequences across sites, optimized to provide equivalent measures following the ADNI protocol; protocols are detailed in the Supplementary Material. All images that passed the quality control step were run through auto-recon 1 in FreeSurfer v5.1 (Fischl 2012). Images were edited by trained raters who had >95% reliability to remove any remaining nonbrain tissue (dura or sinus). An independent rater determined whether images were adequately cleaned for segmentation, and images were then processed through auto-recon 2 and 3. FreeSurfer v5.1 software was used to extract 68 standard regional cortical thickness measurements (Padmanabhan et al. 2015).

Statistical Methods

Item response theory is a model-based approach for describing the latent structure of underlying variation among variables. The bifactor model (Gibbons and Hedeker 1992) is a multidimensional extension of the traditional unidimensional IRT model. First, variables are assigned to a priori or data-driven subdomains (“secondary dimensions”), and within-subdomain correlations are assessed. Each variable then loads on a primary dimension, representing the relationship among all items. Because bifactor models preserve associations within subdomains, the primary dimension describes a latent structure independent from correlations within these subdomains. In terms of multiple comparisons, the MIRT provides simultaneous estimation of parameters for all variables in the model and therefore does not require adjustment for multiple comparisons.

Statistical analysis consisted of four steps. First, a maximum likelihood exploratory factor analysis model was fitted to the 68 cortical thickness measures from FreeSurfer, in order to identify subdomains for the bifactor model. Nine biological subdomains were identified. Second, the 51 symptom items were assigned to 5 subdomains corresponding to their respective scales, defined by depression (MADRS; 10 items), positive psychosis (PANSS Positive 7 items); negative psychosis (PANSS Negative 7 Items); general psychosis (PANSS General 16 items), and mania (Young Mania Scale, YMRS; 11 items). Third, a bifactor model was fit with the 68 cortical thickness measures and 51 symptom ratings. The bifactor model included 15 total dimensions: the primary dimension, 9 cortical thickness subdomains (from step 1 above), and the 5 symptom subdomains. Finally, in order to provide a more parsimonious and interpretable model, we identified a subset of 16 biological variables (brain regions) that loaded highly (>0.7) on the primary dimension, indicating a strong relationship between brain (cortical thickness) and behavior (symptom ratings). We then refit the bifactor model including only those 16 biological variables and all 51 of the symptom ratings. In summary, we began with 119 variables (68 cortical thicknesses and 51 clinical variables). We used factor analysis to identify 9 underlying dimensions for the 68 cortical thicknesses, and used clinical practice to categorize the 51 symptoms into 5 domains. In the bifactor model we, therefore, began with 119 variables with a primary dimension, 9 biological subdomains, and 5 clinical subdomains for a 15 dimensional bifactor model. The analysis revealed that 16 of the 68 biological variables loaded strongly on the primary dimension and we used those 16 variables and the 51 clinical variables to estimate the final bifactor model and identify which of the clinical variables were associated with this reduced set of 16 brain region thicknesses.

Symptom ratings were considered relevant to the model if their loading on the primary dimension had an absolute value greater than 0.25. To provide a full characterization of the entire latent dimension, we included data from probands, relatives and normal controls in our analysis. In addition, we conducted a canonical correlation analysis relating the 51 symptom ratings to the 68 cortical thickness measures to provide a comparison to more traditionally used statistical models for these calculations.

The thresholds of 0.7 for biological variables and 0.25 for clinical variables were not selected a priori, but rather were selected after looking at the distribution of the factor loadings on the primary dimension, which revealed these thresholds as natural cutoffs. As expected, the correlations among the biological variables were stronger than for the clinical variables, hence the different thresholds.

Results

The analytic outcomes, as represented in Figure 1, show experimentally derived brain regions where reduced cortical thickness correlates with high levels of psychosis manifestations. MIRT analysis answered the underlying study question, that is, psychosis manifestations (phenomenological) associate with which brain regions (neurobiological), here showing a contiguous cluster across the fronto-temporal-parietal brain bilaterally.

Figure 1.

Figure 1

Open in a new tab

Inflated representation (lateral, medial, and ventral views) of the cortical regions generated by the bifactor model analysis. Color was assigned automatically according to the method described by Desikan et al. (2006). Cortical regions are labeled as follows: 1L: left inferior parietal lobule; 2L, 2R: left and right pars opercularis; 3L: left precuneus; 4L, 4R: left and right supramarginal gyrus; 5L: left lateral orbitofrontal gyrus; 6R: right pars triangularis; 7L: left bank of the superior temporal sulcus; 8L: left fusiform gyrus; 9L, 9R: left and right middle temporal gyrus; 10L, 10 R: left and right insula; 11L, 11R: left and right superior temporal gyrus.

The numerical results of the bifactor model are presented in Figure 2, including the 51 symptom items and the 16 biological variables (regional cortical thickness) that had loadings of >0.7 on the primary dimension based on the model fitted to all 68 biological variables. The indicated column (highlighted in black) identifies the primary biobehavioral integration variable. The secondary dimensions in this organization reflect the effect of the subconstructs, namely the 5 clinical subdomains (MADRS, PANSS Positive, PANSS Negative, PANSS General, and YMRS) and the biological subdomains (Fig. 2). Loadings on the biological subconstructs were generally high with an average subdomain loading of 0.64.

Figure 2.

Figure 2

Open in a new tab

Schematic representation of the bifactor model, showing the effect of the overarching construct, the latent biobehavioral integration variable (black column) and the subdomains. The symptoms and the biological variables with high loading factors are shown in bold characters.

The subset of symptoms and cortical thickness areas with large loadings on the primary dimension describe the final biobehavioral latent variable. The following symptom items rendered loadings with absolute values greater than 0.25 on the primary dimension: 1) delusions, 2) hallucinatory behavior, 3) suspiciousness, persecution, 4) passive, apathetic social withdrawal, 5) depression, 6) unusual thought content, and 7) active social avoidance. Their loadings are shown in bold with a corresponding arrow to the primary dimension. This cluster of psychosis symptoms is unique in itself, including the reality distortion characteristics of psychosis, along with negative symptoms, depression and social avoidance.

The cortical regions identified in the analysis constitute a set of brain ROIs whose thicknesses inversely correlates with the severity of psychosis symptoms. These areas include the contiguous regions of middle temporal gyrus, superior temporal gyrus, insula, supramarginal gyrus, and operculum bilaterally, and on the left side, the fusiform gyrus, banks of the superior temporal sulcus, inferior parietal lobule, precuneus, and lateral orbitofrontal gyrus, and on the right, the pars triangularis (Fig. 1). The relatively high number of probands assessed in parallel and included in this MIRT analysis, provides a measure of confidence in the outcomes.

Interestingly, we repeated the same analysis substituting cortical surface area for the cortical thickness measures, motivated by the interaction of these two measures in describing cortical volume. The results of this analysis, using cortical surface area, were not informative in that the cortical regions where high psychosis features correlated with low cortical surface area were scattered around the neocortex and were not clustered into a single regional set, like the fronto-partietal-temporal region identified as related to cortical thickness. This result failed to implicate a cortical surface meta-region associated with psychosis and failed to provide clues to psychosis processing. The result suggests that these two features of volume, thickness and surface area distinctly associate with different aspects of psychosis neurobiology, as had been previously shown with genetic characteristics (Winkler et al. 2018).

Result Validation

The goodness of fit of the bifactor model relative to various alternative parameterizations was tested using Bayesian Information Criterion (BIC; Schwarz 1978) where smaller BIC values indicate better fit of the model to the data. The full final model with all clinical and neurobiological subdomains yielded the lowest BIC = 288 665 (best fitting model). BICs for the reduced bifactor models were BIC = 290 112 for a bifactor model with 1 psychological and all neurobiological subdomains, BIC = 291 972 for a model with all psychological and 1 neurobiological subdomains, and BIC = 293 251 for a model with 1 psychological and 1 neurobiological subdomains. The worst fitting model was the unidimensional model, BIC = 304 673. The full bifactor model with all psychological and biological subdomains outperformed both the unidimensional IRT model as well as all 3 reduced form bifactor models.

The core set of cortical regions identified in the analysis was further validated by calculating the sum of the thicknesses of all 16 of the regional components for each subject, then correlating the composite thickness with clinical, demographical and cognitive measures of psychosis severity (duration of illness, number of hospitalizations, number of unique psychotropic medications, and total dose of antipsychotic in chlorpromazine equivalents), social and functional impairment (Global Assessment of Functioning, Social Functioning Scale and its subscales) and cognition (Wide Range Achievement Test and the Brief Assessment of Cognition in Schizophrenia and its component tests). All of these measures significantly correlated with the composite cortical thickness measure (Table 2).

Table 2.

Correlations between the global cortical thickness of the identified structure and measures of disease severity

Pearson r P Number of observations
Global Assessment of Functioning −0.20280 <0.0001 1825
SFS Social Engagement/Withdrawal score −0.14488 <0.0001 1768
SFS Interpersonal Communication score −0.17272 <0.0001 1762
SFS Independence: Competence score −0.07851 0.0019 1564
SFS Independence: Performance score −0.05329 0.0277 1706
SFS Recreation: Performance score −0.09115 0.0001 1726
SFS Prosocial: Performance score −0.19055 <0.0001 1701
SFS Occupation: Employment score −0.23192 <0.0001 1710
SFS total score −0.20476 <0.0001 1422
Duration of illness 0.33749 <0.0001 948
Number of hospitalizations 0.12679 0.0004 786
Number of unique psychotropic medications 0.14197 <0.0001 1824
Daily dose of antipsychotic (in chlorpromazine equivalents) 0.09541 0.0213 582
Wide Range Achievement Test (WRAT) score −0.14489 <0.0001 1807
Brief Assessment of Cognition (BACS) verbal memory z −0.14433 <0.0001 1724
BACS digit sequence z −0.09207 0.0001 1722
BACS motor token z −0.04893 0.0435 1703
BACS verbal fluency z −0.09744 <0.0001 1720
BACS symbol coding z −0.11891 <0.0001 1722
BACS tower of London z −0.11975 <0.0001 1719
BACS composite z −0.13569 <0.0001 1724
Open in a new tab

Comparison with Canonical Correlation

The canonical correlation analysis (performed as a contrast to the IRT analysis) failed to reveal meaningful associations between the clinical symptom variables and the cortical thickness. The first three (of 51) canonical variates, corresponding to apparent sadness, reported sadness, and inner tension, were statistically significant; however, the eigenvalues were all well below 1.0 (0.62, 0.53, and 0.51) (see Supplementary Table 1). Only a single clinical variable had a loading (i.e., correlation) greater than 0.25 on any of the three significant biological canonical variates (poor impulse control 0.26 on the 1st canonical variate) (see Supplementary Table 2). The only biological variables related to the 1st clinical canonical variate were left caudal middle frontal (0.26) and left rostral middle frontal cortex (0.27) (see Supplementary Table 3). These results support the increased sensitivity of the MIRT analysis approach for investigating and defining the relationships between clinical symptoms and neocortical anatomic alterations associated in psychosis.

Discussion

Using B-SNIP data, these analyses aimed to identify the set of psychosis symptoms (phenomenology) most strongly linked together across psychosis syndromes that correlated with cortical thickness (neurobiology) and then to describe the association between their expressions with regional reductions in cortical thickness using a novel statistical approach based on MIRT. The results of the current analysis identified a contiguous neocortical region, where high symptoms in the psychosis group correlated with diminished cortical thickness; these fell within contiguous regions of frontal, temporal and parietal lobes, with bilateral findings but with left hemisphere prominence.

In a previous B-SNIP analysis, we used para-ICA to test the association of structural brain pathology with genetic information, and reported interesting associations between gene outcomes and cortical thinness brain-wide (Ansell et al. 2015); other analyses have found limited regions of association between cortical thickness and symptoms in SZ and SZ-spectrum populations, each in their own particular anatomic distribution (Barta et al. 1990; Watsky et al. 2016; Walton et al. 2017) and investigations of antipsychotic drug effects on cortical thinning (Ansell et al. 2015; Lesh et al. 2015; Gong et al. 2016). Given that cortical thinning reflects local neocortical pathology and genetic influences (Thomson et al. 2016; Tandon et al. 2017), the data reported here, consistent with the literature, implicate an association of neuroanatomic pathology with psychosis manifestations.

The brain regions identified by the bifactor model clustered around the fronto-temporal axis, and included contiguous temporal, parietal, and frontal cortical areas. These include: left and right pars opercularis, right pars triangularis (triangular part of the inferior frontal gyrus), left lateral orbitofrontal gyrus (part of lateral orbital gyrus) in the frontal lobes; left and right middle temporal gyrus, left banks of the superior temporal sulcus, left and right superior temporal gyrus and left fusiform gyrus in the temporal lobes; left and right supramarginal gyrus, left inferior parietal lobule, and left precuneus in the parietal lobes; left and right insula. Key findings shown by our analysis were that a large continuous region of heteromodal cortex was associated with psychosis symptoms, and that the alterations across these regions were sufficiently consistent to form a single dimension in our analysis. Notably, rostral prefrontal regions were not included in this outcome, suggesting that regions outside of these results might be associated with other characteristics of psychotic disorders, like cognitive dysfunction.

These regions subserve several functional neural systems, including: 1) broad aspects of language processing, including phonetic, phonological, lexical, syntactic, and semantic networks (Bonilha et al. 2017; Soderstrom et al. 2017); 2) self-awareness and internal monitoring—insula and precuneus (Cosentino et al. 2015; Sapara et al. 2015); 3) object categorization—fusiform gyrus (Chen et al. 2016; Lech et al. 2016); and 4) social judgment and emotional regulation—lateral part of the orbital gyrus (Kringelbach 2005; Rodrigues et al. 2015). Although these systems and their functions have been previously implicated in psychosis syndromes, they reflect the sum of these regions and of these functions in this analysis.

That these contiguous targets, identified by MIRT as associated with psychosis, are consistent with psychosis correlates that have been emerging recently from several laboratories (Bobilev et al. 2019; Tamminga et al. 2010). In the medial-temporal-cortex (MTL), contiguous with this newly identified neocortical set, molecular and functional changes have been identified in brain tissue from psychotic disorders (Li et al. 2015; Tamminga and Zukin 2015). The hippocampus has been found to be hyperactive in schizophrenic psychosis (Medoff et al. 2001), and the hyperactivity is particularly associated with the C1 subfield (Talati et al. 2016; Schobel et al. 2013). These defects are tied to abnormalities in cognition (Ivleva et al. 2012). A reverse translation animal psychosis model points up potential determinants of the hyperactivity in CA3 and CA1 and implicates specific molecular targets (Segev et al. 2018). Moreover, this animal model implicates a forward-directed excitatory circuit connecting the MTL with the parietal and prefrontal cortical areas, again reflected in the contiguous neocortical set reported in this paper. The current outcome points up the role of the neocortex adjacent to the hippocampal–frontal circuits in contributing to psychosis manifestations.

Another distinctive observation in these data is the unique symptom set that was related to the extended, temporo-parietal-frontal abnormalities. Many of these regions have been individually implicated in psychotic disorders. This advance from prior work is made possible by considering “item-level” data, model development considering brain and behavioral measures simultaneously, and considering common psychotic syndromes simultaneously. A unique feature of the bifactor model is that it accounts for domain-specific intercorrelations, leaving the primary dimension to represent the unique association between the neurobiological and psychological variables.

The comparison in this paper of MIRT with a traditional canonical correlation analysis revealed that the bifactor model identified a much richer structure both biologically and clinically, consisting of 7 clinical variables and 16 biologic variables that loaded strongly on the primary synthetic dimension. By contrast, the canonical correlation analysis revealed little evidence of a relationship between the clinical and biologic variables, with only a single clinical variable and two biologic variables showing association and all eigenvalues for the first three statistically significant canonical variates being well below 1.0.

In this cross-diagnostic psychosis study population, we used MIRT to associate all psychiatric symptoms from a broad set of rating scales, the PANNS, MADRAS and YMRS, without preselecting a psychosis set. The associated symptom set itself was interesting in that it included ‘reality disorientation’ psychotic symptoms (Carpenter 2004) along with depression and negative symptoms, especially social withdrawal and avoidance, but excluded cognitive ‘disorganization’. The data dissociate the localization of reality disorientation symptoms (hallucinations/delusions/paranoia) from cognitive disorganization, often considered coexpressed psychosis symptoms, suggesting a distinctive biology. Moreover, it localizes the expression of what are often considered secondary symptoms of psychosis, including depression and social withdrawal, within this temporo-fronto-parietal anatomic region with shared borders.

In part, these results implicate this empirically derived temporo-fronto-parietal neocortical network as a functional-anatomical unit for a specific set of psychosis symptoms. To the extent that previous structural imaging experiments, both in vivo and in vitro, have assumed that reduced cortical thickness marks cerebral pathology, we can further test where and how this anatomic region mediates these symptoms in psychosis. Experiments focused on this region with biobehavioral outcomes to understand their relations to cortical thinning would contribute to understanding the dynamics of this symptom-set, its pharmacology and, possibly focused treatments. Moving forward, we need to understand the ‘driver’-region and the cellular and molecular substrates for this anatomical change and whether it is intrinsic to the cortices represented. It can be used as a regional anatomical target for further studies of genetic associations, psychosis pathophysiology and possibly for novel psychosis drugs.

Financial Disclosures

Dr Carol Tamminga is Ad Hoc Consultant for Astellas, Autifony, Karuna, Kynexis and TAISHO Pharmaceutical Co LTD. Dr Matcheri Keshavan has received support from Sunovion and GlaxoSmithKline. Dr John Sweeney has consulted to Takeda. Dr Robert Gibbons is a founder of Adaptive Testing Technologies, a company that distributes computerized adaptive tests. The terms of this arrangement have been reviewed and approved by the University of Chicago in accordance with its conflict of interest policies. All other authors declare no conflict of interest.

Supplementary Material

Supplementary_Tables_copy_bhz285
Click here for additional data file. (27.1KB, docx)

Funding

The National Institute of Mental Health (through its support of the Bipolar and Schizophrenia Network for Intermediate Phenotypes: NIMH, MH077851 (to C.A.T.), MH078113 (to M.S.K.), MH077945 (to G.D.P.), MH077852 (to G.K.T.), MH077862 (to J.A.S.), and MH083888 (to J.R.B.)).

References

  1. Ansell BR, Dwyer DB, Wood SJ, Bora E, Brewer WJ, Proffitt TM, et al. 2015. Divergent effects of first-generation and second-generation antipsychotics on cortical thickness in first-episode psychosis. Psychol Med. 45:515–527. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Barta PE, Pearlson GD, Powers RE, Richards SS, Tune LE. 1990. Auditory hallucinations and smaller superior temporal gyral volume in schizophrenia. Am J Psychiatry. 147:1457–1462. [DOI] [PubMed] [Google Scholar]
  3. Bobilev AM, Perez JM, Tamminga CA. 2019. Molecular alterations in the medial temporal lobe in schizophrenia. Schizophr Res. [DOI] [PubMed] [Google Scholar]
  4. Bock RD, Aitkin M. 1981. Marginal maximum likelihood estimation of item parameters: application of an EM algorithm. Psychometrika. 46:443–459. [Google Scholar]
  5. Bonilha L, Hillis AE, Hickok G, den Ouden DB, Rorden C, Fridriksson J. 2017. Temporal lobe networks supporting the comprehension of spoken words. Brain. 140:2370–2380. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Buchy L, Ad-Dab'bagh Y, Malla A, Lepage C, Bodnar M, Joober R, et al. 2011. Cortical thickness is associated with poor insight in first-episode psychosis. J Psychiatr Res. 45:781–787. [DOI] [PubMed] [Google Scholar]
  7. Carpenter WT., Jr 2004. Clinical constructs and therapeutic discovery. Schizophr Res. 72:69–73. [DOI] [PubMed] [Google Scholar]
  8. Chen Q, Garcea FE, Mahon BZ. 2016. The representation of object-directed action and function knowledge in the human brain. Cereb Cortex. 26:1609–1618. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Cosentino S, Brickman AM, Griffith E, Habeck C, Cines S, Farrell M, et al. 2015. The right insula contributes to memory awareness in cognitively diverse older adults. Neuropsychologia. 75:163–169. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Desikan RS, Segonne F, Fischl B, Quinn BT, Dickerson BC, Blacker D, et al. 2006. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage. 31:968–980. [DOI] [PubMed] [Google Scholar]
  11. Fischl B. 2012. FreeSurfer. Neuroimage. 62:774–781. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Gibbons RD, Hedeker DR. 1992. Full-information item bifactor analysis. Psychometrika. 57:423–436. [Google Scholar]
  13. Gong Q, Lui S, Sweeney JA. 2016. A selective review of cerebral abnormalities in patients with first-episode schizophrenia before and after treatment. Am J Psychiatry. 173:232–243. [DOI] [PubMed] [Google Scholar]
  14. Gutierrez-Galve L, Wheeler-Kingshott CA, Altmann DR, Price G, Chu EM, Leeson VC, et al. 2010. Changes in the frontotemporal cortex and cognitive correlates in first-episode psychosis. Biol Psychiatry. 68:51–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Hanford LC, Nazarov A, Hall GB, Sassi RB. 2016. Cortical thickness in bipolar disorder: a systematic review. Bipolar Disord. 18:4–18. [DOI] [PubMed] [Google Scholar]
  16. Hoptman MJ, Antonius D, Mauro CJ, Parker EM, Javitt DC. 2014. Cortical thinning, functional connectivity, and mood-related impulsivity in schizophrenia: relationship to aggressive attitudes and behavior. Am J Psychiatry. 171:939–948. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Ivleva EI, Shohamy D, Mihalakos P, Morris DW, Carmody T, Tamminga CA. 2012. Memory generalization is selectively altered in the psychosis dimension. Schizophr Res. 138:74–80. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Jung WH, Kim JS, Jang JH, Choi JS, Jung MH, Park JY, et al. 2011. Cortical thickness reduction in individuals at ultra-high-risk for psychosis. Schizophr Bull. 37:839–849. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Kay SR, Fiszbein A, Opler LA. 1987. The positive and negative syndrome scale (PANSS) for schizophrenia. Schizophr Bull. 13:261–276. [DOI] [PubMed] [Google Scholar]
  20. Knochel C, Reuter J, Reinke B, Stablein M, Marbach K, Feddern R, et al. 2016. Cortical thinning in bipolar disorder and schizophrenia. Schizophr Res. 172:78–85. [DOI] [PubMed] [Google Scholar]
  21. Kringelbach ML. 2005. The human orbitofrontal cortex: linking reward to hedonic experience. Nat Rev Neurosci. 6:691–702. [DOI] [PubMed] [Google Scholar]
  22. Lech RK, Gunturkun O, Suchan B. 2016. An interplay of fusiform gyrus and hippocampus enables prototype- and exemplar-based category learning. Behav Brain Res. 311:239–246. [DOI] [PubMed] [Google Scholar]
  23. Lesh TA, Tanase C, Geib BR, Niendam TA, Yoon JH, Minzenberg MJ, et al. 2015. A multimodal analysis of antipsychotic effects on brain structure and function in first-episode schizophrenia. JAMA Psychiat. 72:226–234. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Li W, Ghose S, Gleason K, Begovic A, Perez J, Bartko J, et al. 2015. Synaptic proteins in the hippocampus indicative of increased neuronal activity in CA3 in schizophrenia. Am J Psychiatry. 172:373–382. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Medoff DR, Holcomb HH, Lahti AC, Tamminga CA. 2001. Probing the human hippocampus using rCBF: contrasts in schizophrenia. Hippocampus. 11:543–550. [DOI] [PubMed] [Google Scholar]
  26. Montgomery SA, Asberg M. 1979. A new depression scale designed to be sensitive to change. Br J Psychiatry. 134:382–389. [DOI] [PubMed] [Google Scholar]
  27. Padmanabhan JL, Tandon N, Haller CS, Mathew IT, Eack SM, Clementz BA, et al. 2015. Correlations between brain structure and symptom dimensions of psychosis in schizophrenia, schizoaffective, and psychotic bipolar I disorders. Schizophr Bull. 41:154–162. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Rodrigues TP, Rodrigues MA, Paz DA, Costa MD, Centeno RS, Chaddad Neto FE, et al. 2015. Orbitofrontal sulcal and gyrus pattern in human: an anatomical study. Arq Neuropsiquiatr. 73:431–435. [DOI] [PubMed] [Google Scholar]
  29. Sapara A, Ffytche DH, Cooke MA, Williams SC, Kumari V. 2015. Is it me? Verbal self-monitoring neural network and clinical insight in schizophrenia. Psychiatry Res. 234:328–335. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Schobel SA, Chaudhury NH, Khan UA, Paniagua B, Styner MA, Asllani I, et al. 2013. Imaging patients with psychosis and a mouse model establishes a spreading pattern of hippocampal dysfunction and implicates glutamate as a driver. Neuron. 78:81–93. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Schwarz G. (1978). Estimating the dimension of a model. The Annals of Statistics, 6:461–464. [Google Scholar]
  32. Segev A, Yanagi M, Scott D, Southcott SA, Lister JM, Tan C, et al. 2018. Reduced GluN1 in mouse dentate gyrus is associated with CA3 hyperactivity and psychosis-like behaviors. Mol Psychiatry. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Soderstrom P, Horne M, Mannfolk P, van WD, Roll M. 2017. Tone-grammar association within words: concurrent ERP and fMRI show rapid neural pre-activation and involvement of left inferior frontal gyrus in pseudoword processing. Brain Lang. 174:119–126. [DOI] [PubMed] [Google Scholar]
  34. Talati P, Rane S, Donahue MJ, Heckers S. 2016. Hippocampal arterial cerebral blood volume in early psychosis. Psychiatry Res Neuroimag. 256:21–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Tamminga CA, Ivleva EI, Keshavan MS, Pearlson GD, Clementz BA, Witte B, et al. 2013. Clinical phenotypes of psychosis in the bipolar-schizophrenia network on intermediate phenotypes (B-SNIP). Am J Psychiatry. 170:1263–1274. [DOI] [PubMed] [Google Scholar]
  36. Tamminga CA, Pearlson GD, Stan AD, Gibbons RD, Padmanabhan J, Keshavan M, et al. 2017. Strategies for advancing disease definition using biomarkers and genetics: the bipolar and schizophrenia network for intermediate phenotypes. Biol Psychiatry Cogn Neurosci Neuroimag. 2:20–27. [DOI] [PubMed] [Google Scholar]
  37. Tamminga CA, Stan AD, Wagner AD. 2010. The hippocampal formation in schizophrenia. Am J Psychiatry. 167:1178–1193. [DOI] [PubMed] [Google Scholar]
  38. Tamminga CA, Zukin RS. 2015. Schizophrenia: evidence implicating hippocampal GluN2B protein and REST epigenetics in psychosis pathophysiology. Neuroscience. 309:233–242. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Tandon N, Nanda P, Padmanabhan JL, Mathew IT, Eack SM, Narayanan B, et al. 2017. Novel gene-brain structure relationships in psychotic disorder revealed using parallel independent component analyses. Schizophr Res. 182:74–83. [DOI] [PubMed] [Google Scholar]
  40. Thomson PA, Duff B, Blackwood DH, Romaniuk L, Watson A, Whalley HC, et al. 2016. Balanced translocation linked to psychiatric disorder, glutamate, and cortical structure/function. NPJ Schizophr. 2: 16024. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. van Haren NE, Schnack HG, Cahn W, van den Heuvel MP, Lepage C, Collins L, et al. 2011. Changes in cortical thickness during the course of illness in schizophrenia. Arch Gen Psychiatry. 68:871–880. [DOI] [PubMed] [Google Scholar]
  42. Walton E, Hibar DP, van Erp TG, Potkin SG, Roiz-Santianez R, Crespo-Facorro B, et al. 2017. Positive symptoms associate with cortical thinning in the superior temporal gyrus via the ENIGMA schizophrenia consortium. Acta Psychiatr Scand. 135:439–447. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Watsky RE, Pollard KL, Greenstein D, Shora L, Dillard-Broadnax D, Gochman P, et al. 2016. Severity of cortical thinning correlates with schizophrenia Spectrum symptoms. J Am Acad Child Adolesc Psychiatry. 55:130–136. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Winkler AM, Greve DN, Bjuland KJ, Nichols TE, Sabuncu MR, Haberg AK, et al. 2018. Joint analysis of cortical area and thickness as a replacement for the analysis of the volume of the cerebral cortex. Cereb Cortex. 28:738–749. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Winkler AM, Kochunov P, Blangero J, Almasy L, Zilles K, Fox PT, et al. 2010. Cortical thickness or grey matter volume? The importance of selecting the phenotype for imaging genetics studies. Neuroimage. 53:1135–1146. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Young RC, Biggs JT, Ziegler VE, Meyer DA. 1978. A rating scale for mania: reliability, validity and sensitivity. Br J Psychiatry. 133:429–435. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary_Tables_copy_bhz285
Click here for additional data file. (27.1KB, docx)

Articles from Cerebral Cortex (New York, NY) are provided here courtesy of Oxford University Press

RESOURCES