Associations Between Structural Covariance Network and Antipsychotic Treatment Response in Schizophrenia

Abstract

Background and Hypothesis

Schizophrenia is associated with widespread cortical thinning and abnormality in the structural covariance network, which may reflect connectome alterations due to treatment effect or disease progression. Notably, patients with treatment-resistant schizophrenia (TRS) have stronger and more widespread cortical thinning, but it remains unclear whether structural covariance is associated with treatment response in schizophrenia.

Study Design

We organized a multicenter magnetic resonance imaging study to assess structural covariance in a large population of TRS and non-TRS, who had been resistant and responsive to non-clozapine antipsychotics, respectively. Whole-brain structural covariance for cortical thickness was assessed in 102 patients with TRS, 77 patients with non-TRS, and 79 healthy controls (HC). Network-based statistics were used to examine the difference in structural covariance networks among the 3 groups. Moreover, the relationship between altered individual differentiated structural covariance and clinico-demographics was also explored.

Study Results

Patients with non-TRS exhibited greater structural covariance compared with HC, mainly in the fronto-temporal and fronto-occipital regions, while there were no significant differences in structural covariance between TRS and non-TRS or HC. Higher individual differentiated structural covariance was associated with lower general scores of the Positive and Negative Syndrome Scale in the non-TRS group, but not in the TRS group.

Conclusions

These findings suggest that reconfiguration of brain networks via coordinated cortical thinning is related to treatment response in schizophrenia. Further longitudinal studies are warranted to confirm if greater structural covariance could serve as a marker for treatment response in this disease.

Introduction

Schizophrenia is a debilitating mental illness believed to be present in approximately 0.5% of the global population¹; the 2016 Global Burden of Disease Study ranked schizophrenia as the 12th most disabling disorder.^2,3 The main treatment for schizophrenia is antipsychotic medication which mitigates positive symptomatology in patients with schizophrenia through the common property of striatal dopamine D2 receptors blockade.^4,5 Nonetheless, approximately 20%–40% of patients with schizophrenia are deemed to be resistant to first-line antipsychotics.^6,7 While patients with treatment-resistant schizophrenia (TRS) appear to be distinct from patients who respond to first-line antipsychotic medication,^8–12 a group of patients referred to as treatment nonresistant schizophrenia (non-TRS), the neurobiological mechanisms of both groups of patients remains unknown.

Extensive evidence supports morphological changes in schizophrenia.^13,14 The topographic distribution of widespread cortical thinning, particularly within fronto-temporal brain regions, suggest that schizophrenia is a disorder of brain network organization.¹⁵ Thus, understanding these structural alterations in terms of brain network may provide new insights into the mechanism of treatment resistance. Structural covariance analysis examines the group-level intercorrelations of brain structure indicators such as cortical thickness between brain regions and shows promise for investigating pathological bases of alteration in brain structure indicators common to those regions.¹⁶ This approach is based on the observation that functionally related regions co-vary in morphometric characteristics. A postmortem study showed that anatomically related components of the visual systems co-vary in terms of volume across individuals.¹⁷ Moreover, individuals with greater cortical thickness of Broca’s area also generally had greater cortical thickness of Wernicke’s area.¹⁸ It is hypothesized that neural connections can propagate pathological processes through several systems such as synaptic pruning, mutually trophic factor, excitotoxicity, and neurodegenerative progression, which leads to elevated structural covariance at the macroscale level. There is evidence that patients with schizophrenia have increased structural covariance compared with healthy controls (HC),^19,20 in relation to impaired theory of mind²¹ or anhedonia.¹⁹

Notably, patients with TRS exhibit widespread and stronger cortical alterations compared with patients with non-TRS.^22–25 Fan et al. also reported that cognitive impairment in TRS was largely mediated by cortical thinning of frontal, temporal, and parietal areas.²⁶ However, few studies have thus far examined the association between structural covariance network and antipsychotic treatment effect. Two studies of first-episode psychosis observed that responders to short-term antipsychotic treatment had greater structural covariance compared with non-responders. These results suggest facilitating effects of antipsychotics on the reconfiguration of the brain in responders.^27,28 However, no study has examined the difference in structural covariance among TRS, non-TRS, and HC, while only 1 study compared structural covariance between patients with TRS and HC and did not find any significant group difference.²³ It is still unclear whether structural covariance is related to treatment response confirmed by long-term antipsychotic treatment in schizophrenia. Furthermore, while structural covariance network analysis thus far uses group-wide statistics, advanced statistical methods such as individual differential structural covariance (IDSC) analysis have recently been developed to estimate the individual values of structural covariance.^19,29,30 This method enables us to examine relationships between structural covariance and clinico-demographic factors. It warrants further investigation whether there are any clinical indicators that are influenced by or affect abnormal coordinated change of brain structure related to treatment resistance in schizophrenia.

In the current study, we investigated the whole-brain structural covariance of cortical thickness in a large-scale multicenter sample of patients with TRS, patients with non-TRS, and HC. We also explored relationships between structural covariance and clinical measures in the patient groups using IDSC. On the basis of previous research,^23,27,28 we hypothesized that the patient with non-TRS would exhibit increased structural covariance in widespread cortical regions compared with HC or patients with TRS. Also, we expected that altered structural covariance would be associated with clinical characteristics in the patient groups.

Materials and Methods

Participants

We used international multi-site cross-sectional neuroimaging datasets of 274 participants. Ninety participants were enrolled at Komagino Hospital, Tokyo, Japan, while 100 participants were enrolled at the Centre for Addiction and Mental Health (CAMH), Toronto, Canada, and 84 participants were enrolled at National Hospital Organization Shimofusa Psychiatric Medical Center, Chiba, Japan. These original studies were approved by the ethics committees at each site.^9–11 All participants were enrolled following the completion of an informed consent procedure and provided written assent. The details of inclusion criteria and clinical assessments were described in our previous studies.^9–11 Patients had a diagnosis of schizophrenia or schizoaffective disorder based on the DSM-IV or DSM-5. We assessed symptom severity with the Positive and Negative Syndrome Scale (PANSS)³¹ and the Clinical Global Impression Severity Scale (CGI-S).³² Antipsychotic treatment resistance was defined by the modified Treatment Response and Resistance in Psychosis Working Group Consensus criteria.³³ The inclusion criteria for TRS were as follows: (a) treated for at least 6 weeks with at least 2 antipsychotic drugs with a chlorpromazine equivalent of at least 400 mg/day and CGI-S score of at least 4 (moderate) and (b) current severity of illness defined as a score of 4 or more (moderate) on 2 positive symptoms of PANSS. The criteria for non-TRS included the following (a) current use of a non-clozapine antipsychotic, (b) response to treatment with this antipsychotic, a score of 3 or less (mild) on all positive symptoms of the PANSS and a score of 3 or less on the CGI-S, and (c) no symptomatic relapse in the previous 3 months. Response to past antipsychotic treatment was determined based on medical records. HC were assessed to confirm if they had no history of psychiatric illness. Exclusion criteria for all the groups were as follows: (a) history of alcohol or drug abuse within the last 6 months, (b) a positive urine drug screen at inclusion or before MRI scan, and (c) current organic brain disease.

MRI Data Acquisition and Processing

Participants underwent 3D T1-weighted structural MRI scans with similar scanning parameters at the 3 sites: (a) at the Komagino Hospital, 3 T Signa HDxt scanner (GE Healthcare) with an 8-channel head coil [BRAVO, echo time (TE) = 2.8 ms, repetition time (TR) = 6.4 ms, inversion time (TI) = 650 ms, flip angle = 8°, field of view (FOV) = 230 mm, matrix size = 256 × 256, slice thickness = 0.9 mm], (b) at CAMH, a 3 T GE Discovery R750 scanner (GE Healthcare) with an 8-channel head coil (BRAVO, TE = 3 ms, TR = 6.74 ms, TI = 650 ms, flip angle = 8°, FOV = 230 mm, matrix size = 256 × 256, slice thickness = 0.9 mm), (c) at the Shimofusa Psychiatric Medical Center, a 1.5 T Signa Explorer (GE Healthcare) with a 12-channel head coil (FSPGR, TE = 5.1 ms, TR = 12.2 ms, TI = 913 ms, flip angle = 25°, FOV = 256 mm, matrix size = 256 × 256, slice thickness = 1.0 mm).

Structural imaging analysis was done in the minc format on high-performance computing clusters (SciNet). T1-weighted structural images were preprocessed using the bpipe pipeline (https://github.com/CobraLab/minc-bpipe-library), which included a signal intensity correction,³⁴ and procedures to exclude the neck and skull. Cortical thickness was estimated using the CIVET processing pipeline³⁵ (version 2.1.0; Montreal Neurological Institute). T1-weighted images were aligned linearly to the ICBM 152 average template through a 9-parameter transformation (3 translations, rotations, and scales)³⁶ and preprocessed to reduce intensity nonuniformity effects.³⁷ Next, images were segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF).³⁸ Hemispheres were then modeled as GM and WM surfaces using a deformable model strategy, which generates 4 separate surfaces, each defined by 40 962 vertices.³⁹ CT was determined in native space through nonlinear surface-based normalization that uses a midsurface between pial and WM surfaces. The mean cortical thickness was estimated in 31 bilateral regions of interest (ROI), as defined by the Desikan-Killiany-Tourville atlas.⁴⁰ All images were visually inspected by ST and EP. One non-TRS patient in the Komagino cohort, 1 non-TRS patient in the Toronto cohort, and 14 participants in the Shimofusa cohort were excluded from subsequent analyses due to poor imaging quality.

We used a CovBat harmonization method⁴¹ to control for the site differences in the cortical parameters. The CovBat harmonization method estimates and removes the site bias while retaining biological factors (eg, age, disease status, and sex). In this study, we performed harmonization to correct only for the site difference while considering disease status, age, and sex as biological variables in the CovBat.

Statistical Analysis

Demographic differences among the groups were tested using 1-way analysis of variance (ANOVA) tests for continuous variables or chi-square tests for categorical variables. The residuals of cortical thickness, controlled for age and sex, were compared among the groups for each of the 62 cortical regions using ANOVA and post hoc test by R (version 4.0.2). To correct for multiple comparisons, the Benjamini-Hochberg method was used to control the false discovery rate at 5% across the regions.

Using the NBS toolbox (https://www.nitrc.org/projects/nbs/) available in MATLAB R2020b (Mathworks, Natick, MA, USA), whole-brain structural covariance of cortical thickness was compared among the groups with the following procedures. Structural covariance was quantified with the partial correlation coefficient between paired thickness estimates across a population, correcting for age and sex. This approach yielded a separate connectivity matrix of dimension 62 × 62 for each group that quantified the connectivity strength between all pairs of regions. These structural covariances were subjected to Fisher transformation to obtain the z-transformed values. The delta value was calculated by dividing the difference in z-values by the standard deviation (SD) of the difference in z-values. This was performed independently for each pair of regions. The SD of the difference in z-values was estimated by bootstrapping the sample 1000 times. Edges in the network matrices with a delta value of 3 or more were identified as suprathreshold. This threshold of a delta value ≧ 3 was set according to the previous study by Wannan et al.²³ and is equal to P ≦ .0013 since the delta value follows normal distribution. The cluster of supra thresholds edges was regarded as significant when the number of edges was within the top 5% of the 50 000 permuted network matrices, in which individuals were permuted between the 2 groups and structural covariance was then re-estimated for the permuted groups. To evaluate characteristics of the extracted network, we estimated the degree of centrality and betweenness centrality of the nodes. Nodes with centralities greater than 1 SD from the mean are defined as hubs of the network.

Calculation of IDSC

We aimed to explore the relationship between clinical measurements and the connectome alterations where significant group differences were found by NBS as described above. We used IDSC values, which represents the contribution of each individual to their overall group structural covariance, as a measure of inter-regional association strength for that participant. We estimated IDSC values using a jack-knife bias estimation procedure based on the method developed by Ainakina et al.³⁰ For each of the 3 groups, a group-based structural covariance matrix is constructed as a reference matrix by the correlations between residuals of cortical thickness of 62 regions, with the exclusion of a single participant. This reference matrix has the normative covariance structure of that group’s cortical thinning pattern. Then, a perturbed matrix with all participants is constructed in the same way. The IDSC matrix for this participant is constructed using the difference between the perturbed matrix and the reference matrix. Higher IDSC values signify that the participant contributes to greater overall structural covariance in the group. We then averaged IDSC of edges where we found significant group differences of structural covariance between non-TRS and HC. Pearson correlation analysis was used to examine the association among chlorpromazine dose, duration of illnesses, PANSS total, positive, negative, and general scores, and the averaged IDSC values. The Benjamini-Hochberg method was used to control the false discovery rate at 5% for multiple comparisons of 6 tests.

Ancillary Analysis

We have examined reproducibility of the results across sites using the leave-one-site-out method. We repeated structural covariance network analysis and correlation analysis among averaged IDSC values and clinical values removing data of each site. Moreover, structural covariance network analysis and IDSC analysis were repeated with the addition of mean cortical thickness or years of education as a covariate.

Computer codes used in this study are available at github (https://github.com/stsugawa/TRS_SCN).

Results

Demographic Background

After the quality check of images, we included 102 patients with TRS, 77 patients with non-TRS, and 79 HC, for a total of 258 participants. Demographic and clinical characteristics of the final sample are summarized in table 1. Although the groups were comparable in age and sex, HC had higher years of education compared with patient groups. Patients with TRS had lower age of onset and higher PANSS total and subscale scores than patients with non-TRS (table 1).

Table 1.

Characteristics of study subjects.

	HC	Non-TRS	TRS	ANOVA		TRS vs non-TRS
				Test statistic	P value	Test statistic	P value
Number of subjects (Total), No. of subjects	79	77	102
Scanning site 1 (Komagino), No. of subjects	30	31	28
Scanning site 2 (Toronto), No. of subjects	26	20	53
Scanning site 3 (Shimofusa), No. of subjects	23	26	21
Age	41.0 (11.8)	43.0 (12.9)	43.5 (11.8)	F(2,255) = 0.98	.375
Sex	44/35	43/34	62/40	χ 2(2) = 0.636	.728
Education	15.4 (2.3)	13.1 (2.3)	12.6 (2.9)	F(2,255) = 27.85	<.001
Age of onset		25.8 (8.8)	23.3 (6.6)			F(1,172) = 4.56	.034
Duration of disease		17.4 (12.9)	19.93 (11.7)			F(1,172) = 1.9	.169
CPZ		455.8 (286.0)	717.41 (443.8)			F(1,177) = 20.37	<.001
Type of antipsychotics (first/second generation)		9/68	6/96	χ 2(2) = 1.926	.165
PANSS total		55.9 (14.1)	85.4 (25.2)			F(1,172) = 83.72	<.001
PANSS positive		11.1 (3.0)	20.9 (7.0)			F(1,172) = 129.8	<.001
PANSS negative		15.9 (5.5)	22.7 (7.8)			F(1,172) = 41.3	<.001
PANSS general		28.9 (7.3)	41.8 (12.6)			F(1,172) = 63.5	<.001

Group Differences in Cortical Thickness

Significant group differences in residuals of cortical thickness controlling for age and sex were found in 58 out of 62 cortical regions (figure 1). A total of 58 and 31 regions were found to show significant cortical thickness reductions in the TRS and non-TRS groups compared with HC, respectively. Also, cortical thickness in 26 regions was reduced in the TRS group than in the non-TRS group. Both in the TRS and non-TRS groups, regions with cortical thinning were located mainly in the frontal and temporal lobes and cingulate cortex while the TRS group had more widespread thickness reduction compared with non-TRS. Also, the mean effect sizes (Cohen’s d) of regions with significant reduction were larger in the TRS group (d = 0.71) than the non-TRS group (d = 0.51). No significant difference was found in the variances in residuals of cortical thickness among the 3 groups.

Fig. 1.

Whole-brain cortical thinning in patients with schizophrenia compared with HC. Cortical thickness was reduced in the TRS group than in HC (above), in the non-TRS group than HC (middle), and in the TRS group than in the non-TRS group (below). Regions for which the null hypothesis was rejected after controlling the false discovery rate at 5% are shown in color. The color bar represents t scores. Abbreviations: HC, healthy controls; TRS, treatment-resistant schizophrenia.

Group Differences in Whole-brain Structural Covariance

The null hypothesis of equality in structural covariance between the non-TRS and control groups was rejected. Patients with non-TRS had higher structural covariance than HC; the temporo-frontal (5/28 edges, 17.9%) and temporo-occipital (5/28 edges, 17.9%) connections were mainly altered, as shown in figure 2. There were no connections in which patients with non-TRS had lower structural covariance compared with HC. The null hypothesis could not be rejected for structural covariance between the TRS and the non-TRS groups or between the TRS and HC groups. For the topological features of the extracted network where non-TRS had elevated structural covariance than HC, the left inferior temporal gyrus, left insula, and right lateral orbitofrontal cortex had high degree centrality while left inferior temporal gyrus, left insula, left medial orbitofrontal cortex, and right pars orbitalis had high betweenness centrality (supplementary figures 1 and 2).

Fig. 2.

Greater structural covariance in non-TRS compared with HC. Greater structural covariance was observed in the non-TRS group compared with the HC group, represented by circular connectograms (left), which were generated using NeuroMArVL (https://immersive.erc.monash.edu/neuromarvl/). Brain regions are grouped on the connectogram circumference according to lobes. The network with greater structural covariance consists of two communities divided using linkcomm package (https://alextkalinka.github.io/linkcomm/) (right). Nodes in the upper left network contain the regions of the salience network and nodes in the lower right network include the orbitofrontal cortex and temporal gyrus. Right pars orbitalis was an overlapping node. Abbreviations: HC, healthy controls; L, left; R, right; TRS, treatment-resistant schizophrenia.

Relationship Between Structural Covariance and Clinico-demographics

In this analysis, averaged IDSC values were used as an individual index of overall changes in the network where the non-TRS group had higher structural covariance compared with HC. Lower PANSS general score was associated with higher averaged IDSC values in the non-TRS group (r = −0.30, P = .008) while there was no relationship between these values in the TRS group (figure 3 and supplementary table 1). The other clinico-demographic variables were not associated with averaged IDSC values.

Fig. 3.

Relationship between averaged individual differentiated structural covariance and clinical variables in the patient groups. The scatter plots for the correlation of the averaged IDSC values and PANSS general scores in TRS (blue) and non-TRS (red). Abbreviations: IDSC, individual differentiated structural covariance; TRS, treatment-resistant schizophrenia.

Ancillary Analysis

Non-TRS group had a higher structural covariance network compared with HC only when we removed data of the Toronto cohort (supplementary figure 3) and the correlation between averaged IDSC values and PANSS general scores also remained significant (r = −0.33, P = .011; supplementary figure 4). On the other hand, the significance did not remain after removing Komagino or Shimofusa data. Also, the elevated structural covariance and negative correlation of PANSS general score and averaged IDSC values in the non-TRS group remained significant after controlling years of education (supplementary figures 5 and 6) while those significance were diminished when controlling mean cortical thickness values.

Discussion

We examined the structural covariance network and its related factors among patients with TRS, patients with non-TRS, and HC. Our main findings were 4-fold as follows: (a) patients with TRS showed greater cortical thinning mainly in the frontal and temporal lobes compared with HC and patients with non-TRS; (b) the non-TRS group had greater structural covariance compared with the HC group in a brain network centered on the inferior temporal gyrus, medial orbitofrontal cortex, lateral orbitofrontal cortex, pars orbitalis, and insula; (c) there was no structural covariance network with significant difference between the TRS and HC groups or between the TRS and non-TRS groups; and (d) IDSC was negatively associated with PANSS general score in the non-TRS group.

Previous studies have demonstrated greater structural covariance in various illness phases of schizophrenia, including subjects at risk of psychosis,⁴² patients with first-episode psychosis,^20,23 and chronic patients,¹⁹ in comparison with HC. These studies indicated that several factors such as abnormal neurodevelopment, treatment effect, and neurodegeneration progression, may contribute to coordinated brain structural alteration. On the other hand, few studies have thus far examined the association between structural covariance network and antipsychotic treatment response in patients with psychosis or schizophrenia. Saiz-Masvidal et al. focused on 3 cortical and 4 subcortical regions which represent the cortical and cortico-subcortical networks, such as the default mode network, salience network, and central executive network, as well as the cortico-limbic and cortico-thalamic networks. They found that responders with first-episode psychosis had higher structural covariance between the posterior cingulate cortex and precentral gyrus or middle occipital gyrus compared with non-responders.²⁸ Moreover, Jiang et al. investigated the short-term effects of atypical antipsychotics on brain morphometry in patients with drug-naive first-episode schizophrenia. They found that responders had stronger reduction of cortical thickness and higher covariance of cortical thickness change in the regions where cortical thinning were evident compared with non-responders after 12-week antipsychotic treatment. Thus, antipsychotic drugs may reconstruct cortical morphology by affecting cortical thickness in multiple regions through the common physiology when symptoms are controlled.²⁷ However, the criteria of responders in those studies are different from the criteria of non-TRS since responders were determined according to the percentage of PANSS score reduction through follow-up. Moreover, it is impossible to discuss the abnormality of structural covariance in those patients since these 2 studies did not include HC. Also, they only focused on first-episode patients and the long-term effect of antipsychotic treatment is still unknown. Thus, it is important to validate structural covariance networks in chronic patients with schizophrenia because a certain number of patients respond to antipsychotics during the phase of first-episode psychosis but become treatment-resistant in the long term.^43,44 Wannan et al. reported that structural covariance networks did not differ between the TRS and HC groups, while they did not include patients with non-TRS.^20,23

Thus, our result is the first observation of a greater structural covariance network in chronic patients with non-TRS compared with HC. Moreover, the higher inter-regional association was related to lower severity of PANSS general score in the non-TRS group. This result is consistent with the study finding that the structural covariance between the left and right medial orbitofrontal thickness were positively correlated with anticipatory pleasure in schizophrenia.¹⁹ Therefore, observed structural covariance networks may contribute to treatment response in schizophrenia although structural covariance may associate with only a part of schizophrenia symptoms since only the correlation between averaged IDSC values and PANSS general symptoms were significant, and no correlations were found with other PANSS total and subscale scores. In line with previous reports, antipsychotic treatment may potentially have positive effects on the reconfiguration of the brain not only in the early stages of treatment, but also in the long term for the subgroup of patients showing symptomatic improvement. Importantly, the characterization of structural alteration in each individual using IDSC values can help to predict treatment response in the future.

In addition, the hub regions of the brain network where the non-TRS group had higher structural covariance than HC comprised the inferior temporal gyrus, medial orbitofrontal cortex, lateral orbitofrontal cortex, pars orbitalis, and insula. These regions have been previously reported as hubs of higher structural covariance networks in schizophrenia.^19,20Figure 2b shows that the extracted structural covariance in non-TRS is composed of 2 main networks. One of the 2 networks contains the regions of the salience network (ie, the insula and cingulate gyrus) and it has been reported that dysregulation of dopamine modulation centered on the salience network is a possible contributor to the core clinical phenotype of schizophrenia.⁴⁵ Another network includes the orbitofrontal cortex and temporal gyrus, which may be responsible for emotional and executive functioning, decision-making, reward-related behavior, and auditory hallucinations.^46–49 Notably, the pars orbitalis (a portion of the inferior frontal gyrus) links those 2 networks. Although studies on the pars orbitalis are limited, this region is involved in language function and interpretation^20,50,51 and also an important part of the fronto-temporal network, which has been hypothesized as being disrupted in schizophrenia.^52,53 Thus, reconfiguration of brain networks including these regions may contribute to antipsychotic effects on auditory hallucinations and formal thought disorder.

In the present study, the TRS group did not show a difference in structural covariance compared with HC, which is consistent with previous studies.²³ This result is conceivably due to the lack of antipsychotic effect on reconstruction of cortical morphology in the TRS group, contrary to the non-TRS group. However, while the mechanisms underlying alteration in structural covariance between brain regions have not yet been elucidated, cortical reduction and elevation of structural covariance can not only be caused by medication effect but also by other factors, such as excitotoxicity which is more pronounced in TRS.²⁵ Given that Wannan et al. demonstrated that regions with reduced cortical thickness in schizophrenia compared with HC have higher structural covariance than regions with nonsignificant group difference, it is unclear why higher structural covariance network was not found in the TRS group, which exhibited stronger and more widespread cortical thinning than the non-TRS group. The possible cause of our results may be greater heterogeneity in the pathophysiology of cortical thinning in patients with TRS compared with those with non-TRS. Since the structural covariance network analysis only detects alterations at the group level in coordinated brain structures, nonsignificant structural covariance may be observed if each individual presents a different pattern of cortical thinning. Previous studies indicated that the TRS group has greater heterogeneity compared with the non-TRS group in several aspects such as brain structure and glutamatergic neurometabolites.⁸ While the dopamine hypothesis can generally account for the pathophysiology of positive symptoms in non-TRS and the mechanisms of action of non-clozapine antipsychotics, several hypotheses have been proposed for the neurobiological mechanisms underlying TRS, including dopamine supersensitivity, hyperdopaminergic and normodopaminergic subtypes, glutamate dysregulation, inflammation and oxidative stress, and serotonin dysregulation.⁵⁴ These different underlying mechanisms might contribute to the structural alteration of different brain networks in each patient with TRS.

Strength

To the best of our knowledge, this is the first large-scale, multicenter study that examined whole-brain structural covariance of cortical thickness among the 3 groups including chronic patients with TRS, who had markedly severe positive symptoms, patients with non-TRS, and HC. Since previous studies examined structural covariance only in treatment responders and non-responders,^27,28 it was unclear whether group difference in structural covariance between the 2 groups was due to the elevated structural covariance in responders compared with HC or lower structural covariance in non-responders than HC. In this study, we revealed that the non-TRS group had an abnormally elevated structural covariance network by comparing patient groups to HC. In addition, previous studies were limited to patients with first-episode psychosis or schizophrenia, and it was unclear whether structural covariance is related to treatment resistance in chronic schizophrenia. Our findings suggest that higher structural covariance may also be associated with treatment response in chronic schizophrenia. Moreover, since the methodologies of structural covariance network analysis in previous studies varied by each study, we investigated the whole-brain structural covariance network using NBS which has higher statistical power for controlling family-wise error than conventional methods.⁵⁵ Finally, unlike the previous studies which examined only group-level structural covariance, we examined associations between IDSC values and clinical variables that possibly influences structural covariance networks in patients.

Limitation

Several limitations need to be addressed. First, the TRS group had more severe symptoms than non-TRS since the criteria of TRS and non-TRS were determined by severity of PANSS positive scores, which may affect the structural covariance. On the other hand, we did not find significant correlation between PANSS positive scores and averaged IDSC values in either group. Thus, elevated structural covariance may depend on grouping by treatment response rather than symptom severity since structural covariance was not associated with symptoms that determine treatment resistance. There were also possibilities that elevated structural covariance in the non-TRS group reflected the state of symptom severity except positive symptoms if lower symptom severity were associated with higher structural covariance in the whole patient group. However, the association between structural covariance and symptom severity does not seem to be simple because we only found the significant correlation between IDSC values and PANSS general in non-TRS group and TRS group did not have any significant correlation between IDSC values and PANSS total and subscale scores. The evidence about the association between structural covariance and symptom severity or antipsychotic doses were limited, which warrants further study. Second, there was a group difference in chlorpromazine doses between the TRS and non-TRS groups. However, greater structural covariance was found in patients with non-TRS who had lesser antipsychotics compared with patients with TRS. According to the results of Jiang et al., antipsychotic medication may increase structural covariance.²⁷ Also, mean IDSC values in the patient groups were not associated with chlorpromazine doses in the present study. Thus, our results do not seem to be influenced by the antipsychotic doses because structural covariance should increase in the TRS group if more antipsychotic doses increase structural covariance. However, further study is needed to examine the associations between structural covariance and antipsychotic administration. Third, multicenter MRI findings may be hindered by image differences acquired from different scanners/parameters. Therefore, we harmonized data from 3 scanners using the CovBat method, which has advantages over the frequently used Combat method, particularly in removing the site effect of covariance.⁴¹ Although we could not find any literature that studied harmonization between 1.5 T data and 3 T data, Reynold et al. reported that ComBat, the predecessor of CovBat, was able to effectively remove the effect of magnetic field strength using 3 T and 1.5 T test–retest scans.⁵⁶ Since CovBat is a modified version of ComBat and the values calculated by the 2 methods were almost identical, we believe that CovBat may also be effective on harmonization of scanner strength. Fourth, we examined the site effect in ancillary analysis and found significant differences in structural covariance between non-TRS and HC only when we removed data of the Toronto cohort. The elevated structural covariance did not remain after removing the Komagino or Shimofusa data. It is possible that the Komagino and Shimofusa data may have had a similar trend of higher structural covariance in non-TRS than HC, but no such relationship may have been observed in Toronto data. Among the 3 sites, there are several differences in factors that could affect the results such as race, antipsychotic treatment, cannabis use in HC, and diagnosis criteria. Thus, it is uncertain whether the elevation of structural covariance can be replicated in a diverse population, which warrants further investigation. Fifth, we repeated structural covariance network analysis adding mean cortical thickness as covariate to ensure that the rise in structural covariance is not simply due to innate individual differences in cortical thickness (ie, some people are born with thicker or thinner overall cortical thickness, which makes the thickness of 2 regions appear to be correlated). As a result, the significant elevation of structural covariance in non-TRS compared with HC did not remain after controlling for mean cortical thickness. Although we cannot deny the possibility that the results were due to innate individual differences, the mean cortical thickness might reflect thickness reductions related to the progression of the disease since patient groups had widespread cortical thinning. Thus, it is possible that even pathophysiology-related changes in structural covariance were removed by including mean cortical thickness as a covariate. Seventh, since we did not have the full information about treatment duration or whether the diagnosis is schizophrenia or schizoaffective disorders, we could not examine the influence of those factors on our findings. Finally, although structural covariance may be affected by several factors including neurodevelopmental abnormality, excitotoxicity, neurodegenerative progression, and antipsychotic treatment, we could not determine which of these has a significant impact on our results.

Conclusion

Our finding of greater structural covariance only in the non-TRS group suggests that TRS and non-TRS may have a different topographic distribution centered on inferior temporal gyrus and insula. The coordinated cortical thinning in the brain network may underlie pathophysiology of treatment response in schizophrenia. Further longitudinal studies are warranted to confirm if greater structural covariance could serve as a marker for treatment response in this disease.

Funding

This study was supported by Japan Society for the Promotion of Science (18H02755 and 22H03002), Takeda Science Foundation, Watanabe Foundation, Uehara Memorial Foundation, Inokashira Hospital Research Foundation (SN), Ontario Mental Health Foundation (OMHF) Type A grant (AGG) and by Canadian Institutes of Health Research (CIHR) Grant Nos. MOP-142493 (AGG) and MOP-141968 (AGG).

Conflict of interest

S.N. has received grants from Japan Society for the Promotion of Science (18H02755), Japan Agency for Medical Research and development (AMED), Japan Research Foundation for Clinical Pharmacology, Naito Foundation, Takeda Science Foundation, Uehara Memorial Foundation, and Daiichi Sankyo Scholarship Donation Program within the past 3 years. S.N. has also received research support, manuscript fees or speaker’s honoraria from Dainippon Sumitomo Pharma, Meiji Seika Pharma, Otsuka Pharmaceutical, Shionogi, and Yoshitomi Yakuhin within the past 3 years. GR has received research support from the Canadian Institutes of Health Research (CIHR), University of Toronto, and HLS Therapeutics Inc. Y.N. has received a Grant-in-Aid for Scientific Research (B) (21H02813) from the Japan Society for the Promotion of Science (JSPS), research grants from Japan Agency for Medical Research and Development (AMED), investigator-initiated clinical study grants from Teijin Pharma Ltd, and Inter Reha Co., Ltd. He also receives research grants from Japan Health Foundation, Meiji Yasuda Mental Health Foundation, Mitsui Life Social Welfare Foundation, Takeda Science Foundation, SENSHIN Medical Research Foundation, Health Science Center Foundation, Mochida Memorial Foundation for Medical and Pharmaceutical Research, Taiju Life Social Welfare Foundation, and Daiichi Sankyo Scholarship Donation Program. He has received speaker’s honoraria from Dainippon Sumitomo Pharma, Mochida Pharmaceutical Co., Ltd, Yoshitomiyakuhin Co., Ltd, Qol Co., Ltd, Teijin Pharma Ltd, and Takeda Pharmaceutical Co., Ltd within the past 5 years. He also receives equipment-in-kind support for an investigator-initiated study from Magventure Inc., Inter Reha Co., Ltd, Brainbox Ltd, and Miyuki Giken Co., Ltd H.U. has received grants from Daiichi Sankyo, Eisai, Mochida, Otsuka, and Sumitomo Dainippon Pharma; speaker’s fees from Eisai, Janssen, Lundbeck, Meiji Seika Pharma, Otsuka, and Sumitomo Dainippon Pharma; and advisory board fees from Lundbeck, Sumitomo Pharma and Boehringer Ingelheim Japan. F.U. has received fellowship grants from Discovery Fund, Nakatani Foundation, and the Canadian Institutes of Health Research (CIHR); manuscript fees from Dainippon Sumitomo Pharma; and consultant fees from VeraSci, and Uchiyama Underwriting within the past 3 years. M.M. has received speaker’s honoraria from Biogen Japan, Byer Pharmaceutical, Daiichi Sankyo, Dainippon Sumitomo Pharma, Demant Japan, Eisai, Eli Lilly, Fuji Film RI Pharma, Hisamitsu Pharmaceutical, H.U. Frontier, Janssen Pharmaceutical, Mochida Pharmaceutical, MSD, Mylan EPD, Nippon Chemipher, Novartis Pharma, Ono Yakuhin, Otsuka Pharmaceutical, Pfizer, Shionogi, Takeda Yakuhin, Teijin Pharma, and Viatris within the past 3 years.