Heterogenous Patterns of Brain Atrophy in Schizophrenia Localize to A Common Brain Network

doi:10.21203/rs.3.rs-3470220/v1

Download PDF

Research Article

Heterogenous Patterns of Brain Atrophy in Schizophrenia Localize to A Common Brain Network

https://doi.org/10.21203/rs.3.rs-3470220/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Understanding the neuroanatomy of schizophrenia remains elusive due to heterogenous findings across neuroimaging studies. Here, we investigated whether patterns of brain atrophy associated with schizophrenia would localize to a common brain network. Using the human connectome as a wiring diagram, we identified a connectivity pattern, a schizophrenia network, uniting heterogenous results from 90 published studies of atrophy in schizophrenia (total n>8,000). This network was specific to schizophrenia, differentiating it from atrophy in high-risk individuals (n=3038), normal aging (n=4,195), neurodegenerative disorders (n=3,707), and other psychiatric conditions (n=3,432). The network was also stable with disease progression and across different clusters of schizophrenia symptoms. Patterns of brain atrophy in schizophrenia were negatively correlated with lesions linked to psychosis-related thought processes in an independent cohort (n=181). Our results propose a unique, stable, and unified schizophrenia network, addressing a significant portion of the heterogeneity observed in prior atrophy studies.

Computational Neuroscience

Schizophrenia

Neuroimaging

Brain Networks.

Schizophrenia is among the major causes of disability worldwide (Vigo et al., 2016). While neuroimaging has contributed to our understanding of the neural correlates of schizophrenia (Keshavan et al., 2019; Kraguljac et al., 2021), there remains significant heterogeneity across different studies and modalities (Dabiri et al., 2022). For example, some studies have reported decreases in regional gray matter volume in the prefrontal cortex of patients with schizophrenia (Pomarol-Clotet et al., 2010; Ortiz-Gil et al 2011), while others have also found reductions in regions of the temporal lobe, occipital cortex, and subcortical structures (Huang et al,, 2015; Chang et al., 2017; Delvecchio et al., 2017). Similarly, while some studies have found increased functional connectivity within specific brain networks (default mode network, salience network, and frontoparietal network), others have reported decreased connectivity (Hu et al., 2017; Menon et al., 2023). The heterogeneity in neuroimaging findings has made it challenging to develop reliable biomarkers or neuroanatomically targeted interventions for schizophrenia. For instance, while therapeutic brain stimulation is effective for various neuropsychiatric disorders, treatment targets for schizophrenia remain unclear (Siddiqi et al., 2023).

Amidst this heterogeneity, there may be common threads. Such threads may be identified using coordinate network mapping (CNM), which identifies common networks linked to heterogeneous coordinates of neuroimaging abnormalities (Darby et al., 2019). This method has been successfully applied to a range of brain disorders, including depression (Zhukovsky et al., 2021; Cash et al., 2023), addiction (Stubbs et al., 2023), Alzheimer's disease (Darby et al., 2019), Parkinson's disease dementia (Weil et al., 2019), migraine (Burke et al., 2020), and even transdiagnostic mental illness (Taylor et al., 2023). While traditional neuroimaging meta-analytic approaches such as activation likelihood estimation (ALE) identify spatially consistent brain regions across studies (Eickhoff et al., 2012), CNM enables the mapping of brain disorders to functionally-defined interconnected brain networks.

To address the heterogeneous literature on brain atrophy in schizophrenia, we applied CNM to findings from brain atrophy studies in schizophrenia. We hypothesized that structural atrophy coordinates reported in published studies of individuals with schizophrenia would converge to a common brain network that is sensitive and specific to the disorder. We also hypothesized that structural changes corresponding to different symptom clusters in schizophrenia could be localized to distinct brain networks.

Data were sourced from a published activation likelihood estimation (ALE) meta-analysis of patients with schizophrenia and individuals deemed to be at high risk of developing the disorder based on genetics or prodromal symptoms (Liloia et al., 2021). This study identified 113 peer-reviewed articles comprising 11,270 participants in four clinical groups and 6,007 healthy controls. 1,636 patients across 41 studies had recently-diagnosed schizophrenia (illness duration < 2 years), 2,120 patients across 49 studies had chronic schizophrenia (illness duration > 2 years), 927 subjects across 18 studies were at high risk based on genetics, and 580 subjects across 16 studies were at high risk based on prodromal symptoms.

Given the heterogeneity of atrophy locations (Fig. 1), we estimated the normative functional connectivity of atrophy locations identified in cohorts of patients with schizophrenia using a normative connectome database (n = 1000). This analysis yielded a connectivity map of each study’s reported pattern of atrophy. We thresholded these study-level maps at |t|>5 and combined them into one composite sensitivity map representing the voxel-wise overlap of connectivity maps across all studies (Fig. 2A), as in prior CNM studies (Taylor et al., 2023). To test for peak regions that were significantly consistent across all studies, we also combined the unthresholded maps into a composite t-map using a voxelwise one-sample t-test with threshold-free cluster enhancement (TFCE) to correct for multiple comparisons (Fig. 2B). Finally, we combined the results of the sensitivity and consistency analyses in one map (sensitivity-consistency map) showing only areas that were statistically significant in both analyses (sensitive and consistent) after correcting for multiple comparisons (p_FWE < 0.05) (Fig. 3A). The peak regions in this map included the left insula and bilateral anterior cingulate cortex. Other statistically significant regions are reported in Supplementary Table 1.

Next, we performed a specificity analysis. We compared the connectivity maps of atrophy patterns in schizophrenia to 10 control groups comprising connectivity of atrophy patterns in normal aging and nine brain disorders namely major depressive disorder, bipolar disorder, anxiety disorders, obsessive-compulsive disorder, substance use disorders, mild cognitive impairment, Alzheimer’s disease, Parkinson’s disease, and Huntington’s disease. We used a two-sample t-test to identify voxel clusters that are specific to schizophrenia. After multiple comparisons correction using TFCE (p_FWE < 0.05), this analysis yielded peak clusters in the left superior temporal gyrus and right somatosensory cortex (Fig. 3B). Other significant regions are listed in Supplementary Table 2. To confirm that the results were not driven by a specific disorder in the control group, we repeated the analysis sequentially excluding each of the 10 control groups and the result remained statistically significant (p_FWE < 0.05 using TFCE) for all comparisons. We also repeated the analysis using only psychiatric disorders or only neurodegenerative disorders and normal aging as controls. The peak region for schizophrenia versus psychiatric disorders was the left anterior cingulate cortex, bordering the pregenual and dorsal regions (p_FWE<0.05, Fig. S1 panel A). The peak region for schizophrenia versus neurodegenerative disorders and aging included the left amygdala (p_FWE<0.05, Fig. S1panel B). Each individual psychiatric and neurological disorder showed significant peaks that were distinct from schizophrenia, with the exception of bipolar disorder and major depression (Table S3). Finally, we combined the sensitivity/consistency map and the specificity map into a conjunction map, which we will heretofore refer to as the ‘schizophrenia network’. By definition, this network includes the connectivity of atrophy profiles that are sensitive and specific to schizophrenia. Peak regions in this map included the dorsal anterior cingulate cortex and mid-insula bilaterally (Fig. 3C).

To investigate whether the connectivity of atrophy patterns changes with disease progression or treatment, we divided the schizophrenia dataset into two groups representing recently diagnosed and chronic schizophrenia, respectively. We assessed for specific anatomical differences between the connectivity of atrophy patterns in recently-diagnosed versus chronic schizophrenia. We found no statistically significant voxel clusters (p_FWE > 0.22) that were different between the two groups. Further, the consistency maps were nearly identical between recently-diagnosed and chronic schizophrenia (spatial r = 0.89). This similarity was stronger than expected by chance, as demonstrated by a permutation test in which this spatial correlation was recomputed after randomly shuffling group assignments in the one-sample t-test (p = 0.00005 with 10⁵ permutations). (Fig. 4).

To test for symptom specificity, we divided our dataset into patients with predominant positive versus negative symptoms. Of note, none of the studies investigated patients who exclusively had positive or negative symptoms. Contrary to our hypothesis of distinct symptom cluster localization, no voxel clusters emerged as significantly different (p_FWE > 0.13). Sensitivity-consistency maps for positive versus negative symptoms were highly similar to each other (spatial r = 0.63, p = 0.02 with 10⁴ permutations) (Fig. 4).

Next, we assessed for specificity to schizophrenia relative to related clinical syndromes or genetic risk factors. We compared the schizophrenia network to atrophy coordinates from two high risk groups: individuals at clinical high risk (showing non-specific prodromal symptoms) and individuals at genetic high risk (monozygotic twins, siblings, and first/second-degree relatives of patients with schizophrenia). Relative to individuals at clinical high risk, atrophy patterns associated with schizophrenia were preferentially connected to a white matter tract in the medial temporal lobe (p_FWE< 0.05) (Fig. 5A). Using the Neurosynth platform (Yarkoni et al., 2011) for automated meta-analysis of neuroimaging data, this coordinate was most associated with the amygdala (z = 19.66) and weakly associated with the hippocampus (z = 4.97). Relative to individuals at genetic high risk, patients with schizophrenia showed atrophy patterns with significantly greater connectivity to the anterior cingulate cortex bilaterally and the right pre- and postcentral gyri (Fig. 5). Spatial correlation was r = 0.38 (p = 0.049 with 10⁴ permutations) between schizophrenia and the clinical high risk group and r = 0.59 (p = 0.049 with 10⁴ permutations) between schizophrenia and the genetic high risk group.

Finally, we explored the causal relevance of the schizophrenia network by comparing it to brain lesion data, following recommended standards for drawing causal conclusions from atrophy studies (Siddiqi et al, 2022). Lesion analyses were conducted on data from the Vietnam Head Injury Study (VHIS), in which participants underwent a detailed behavioral assessment after focal brain lesions from combat-related penetrating head trauma during the Vietnam War. We hypothesized that our network would match the connectivity of lesions associated with increased psychosis-like thought processes. Specifically, the VHIS included two items from the clinician-rated neurobehavioral rating scale (NBRS) that pertain to psychotic processes, “unusual thought content” and “suspiciousness.” We mapped the connectivity of brain lesions for 181 participants for whom the NBRS was completed during their one-week inpatient evaluation and compared it to the schizophrenia network using spatial correlations across all participants. Negative connectivity of lesions to the schizophrenia network was significantly associated with higher scores on suspiciousness and unusual thought content. As a control, we repeated this analysis for the remaining 24 NBRS items; all of these yielded negative results except for the “somatic concern” item (Fig. 6A). The “hallucinatory behavior” item was excluded from this analysis because only 2 participants reported positive scores (a score of 1 on a scale of 0 to 6). For the remaining items, at least 15 participants reported positive scores. We also repeated the analysis after controlling for lesion size, Beck Depression Inventory (BDI) scores, posttraumatic stress disorder (PTSD) diagnosis as measured by the structured clinical interview for the DSM-IV, MMSE (mini mental status exam) scores, and total number of psychiatric diagnoses. The significant association with suspiciousness scores remained statistically significant after the aforementioned controls, while the association with unusual thought content remained statistically significant (p < 0.05) in all conditions except when controlling for PTSD (p = 0.07) and BDI (p = 0.06). Of note, lesion similarity to the schizophrenia network was not associated with total number of psychiatric diagnoses.

Using lesion network mapping (overlap of normative functional connectivity maps of lesion locations using a 1000-subject connectome database), we mapped the functional connectivity of brain lesions associated with suspiciousness and unusual thought content and compared these maps to the schizophrenia network. The results showed marked overlap between the schizophrenia network and brain regions with negative functional connectivity to lesions associated with both traits (Fig. 6B).

This study yielded several important findings that provide insights into the patterns of brain connectivity in schizophrenia. First, widely heterogeneous published atrophy coordinates in schizophrenia were united by a specific pattern of connectivity to one common brain network which we refer to as the schizophrenia network. Second, the schizophrenia network is relatively stable with disease progression, as there were no significant differences between recently diagnosed and chronic patients. Third, the schizophrenia network was specific to the disorder compared to normal aging, and patients with other brain disorders. Fourth, different symptom clusters in schizophrenia localized to similar brain networks. Fifth, a similar network was associated with high-risk groups, and atrophy patterns in patients who progressed to schizophrenia showed more connectivity to the medial temporal lobe and anterior cingulate. Finally, brain atrophy patterns associated with schizophrenia were negatively correlated with lesions associated with suspiciousness and unusual thought content.

While there is marked heterogeneity in structural neuroimaging findings in schizophrenia, this does not necessarily imply that previous research is inaccurate. Instead, it may reflect methodological diversity, as researchers are incentivized to employ novel methods in every study (Cash et al., 2023). Our findings help reconcile this diversity by showing that atrophy patterns in schizophrenia are connected to a common network that involves the insula, dorsal cingulate cortex, and superior temporal gyrus (Heschl's gyrus and planum temporale). This is consistent with the hypotheses that the neuroanatomy of schizophrenia is not limited to a single brain region (van den Heuvel and Fornito, 2014; Shafieie et al., 2020). A prominent example is the disconnection hypothesis which posits that the disorder is characterized by aberrant integration and coordination of functionally specialized brain regions (Friston & Frith, 1995; Silbersweig et al., 1995). Another possible explanation for the heterogeneity in neuroimaging studies of schizophrenia is the clinical and likely underlying pathophysiological heterogeneity of schizophrenia (Silbersweig and Loscalzo, 2017).

The stability of the schizophrenia network across first-episode and chronic psychosis is consistent with several studies that have shown stable neuroanatomical abnormalities irrespective of disease progression or therapeutic interventions (van den Heuvel & Fornito, 2014; Sheffield & Barch, 2016). This stability may point towards a trait-like characteristic of schizophrenia. However, while the overall pattern of connectivity might be stable, our study does not rule out the possibility that the severity and distribution of structural and functional alterations may vary with disease progression, individual symptoms, or treatment (Sheffield & Barch, 2016; Dong et al., 2018).

The results of our study revealed differences in the functional connectivity patterns associated with atrophy in patients with schizophrenia as compared to other psychiatric disorders (connectivity to the left anterior cingulate cortex, for example). However, this distinction does not negate the existence of similarities. For instance, the studies included in our systematic review were partly overlapping with those reported by Taylor et al. (2023), which found similarities across multiple psychiatric disorders. In the present study, we updated the systematic review with newer studies and searched specifically for differences rather than similarities. Together, these findings suggest a nuanced landscape of both distinct and shared neuroanatomical patterns across different psychiatric conditions.

We also expand on previous studies that have solely focused on hallucinations (Kim et al., 2021) or delusions (Darby et al., 2017) by demonstrating that brain networks linked with different symptom clusters in schizophrenia show a high degree of similarity. This could support the longstanding Kraepelinian model of schizophrenia as a single, unified disorder (Kraepelin, 1919). It would also support the notion of core processing disturbances or final common pathways of symptom expression (Silbersweig and Loscalzo, 2017). The singular nature of the schizophrenia network highlight the potential of using network-based approaches as a therapeutic target and potentially a diagnostic tool for schizophrenia as a whole.

Individuals at high risk for developing schizophrenia—due to genetic predispositions or prodromal symptoms—exhibited similar patterns of atrophy. However, it is particularly noteworthy that those who progressed to develop schizophrenia were more likely to demonstrate a pattern of atrophy specifically connected the medial temporal lobe or anterior cingulate cortex. Another possible explanation for this finding is that with progression to schizophrenia, more damage occurs in brain regions connected to the medial temporal lobe and anterior cingulate. These two brain regions are integral neural regions implicated in the clinical phenomenology of schizophrenia. The medial temporal lobe, which includes the hippocampus and amygdala, is central to memory processing and emotional regulation (Ragland et al., 2015). Disruptions in the medial temporal lobe have been associated with the hallmark memory deficits and emotional dysregulation observed in schizophrenia (Tamminga et al., 2012). Meanwhile, the dorsal anterior cingulate cortex is involved in diverse functions such as conflict monitoring, error detection, and emotional processing (Shenhav et al., 2016).. Our observations are consistent with studies that reported that high risk individuals who ultimately progressed to schizophrenia showed significant neuroanatomical alterations most notably in the medial temporal lobe (Wood et al., 2008) and the anterior cingulate cortex (Fornito et al., 2008; Wood et al., 2008; Takayanagi et al., 2017; Collins et al., 2023). This may be an initial step towards the longstanding goal of predicting which high-risk patients are likely to develop schizophrenia. Future research may be warranted to longitudinally study these regions in high-risk individuals to establish a temporal sequence between focal atrophy and development of schizophrenia. Such an approach may provide a basis for early identification and intervention.

The negative correlation between atrophy-derived and lesion-derived findings presents a counterintuitive and intriguing perspective on the nature of brain atrophy in schizophrenia. Intuitively, one might expect lesions and brain atrophy patterns to align and contribute to similar symptomatology in brain disorders. Given that both processes can lead to neuronal loss or reduced activity, it is reasonable to expect that they would contribute to brain disorders in similar ways (Ferguson et al., 2019). However, some studies (Taylor et al., 2023; Stubbs et al., 2023) have shown that locations of brain atrophy show negative functional connectivity with the locations of lesions that cause the same symptom – that is, when spontaneous activity is increasing in the atrophy-derived circuit, it is simultaneously decreasing in the lesion-derived circuit. One possible explanation for this phenomenon is that while atrophy could be causal in some disorders, it might be a compensatory response in others (Siddiqi et al., 2022). An alternative explanation could be that atrophy is an epiphenomenon of the disorder or its treatments. It is also possible that schizophrenia and penetrating head trauma are affecting the network in different ways; for instance, in the case of epilepsy, atrophy and abnormal brain activity do not necessarily co-localize (Galovic et al., 2019).

Strengths of this study include meta-analytic convergence, causal validation using lesion analysis, multiple independent cohorts, and the assessment of disease stages and symptom clusters in schizophrenia. There are also several limitations, primarily related to the large group-level analysis, which limits the ability to extend results to individual patients. For instance, we used a normative connectome database (the average connectivity patterns in healthy individuals) instead of a patient-specific connectivity measurements, as the analysis was done using group-level atrophy coordinates. Normative connectome databases provide better spatial resolution and signal-to-noise ratio due to large sample size, but future studies may build upon our findings by replicating them using individualized connectivity. Our study is also retrospective, thus limiting our ability to test longitudinal effects of disease progression or to understand the mechanisms of the counterintuitive finding that atrophy and lesions associated with psychosis were negatively connected to each other. Future prospective studies could improve upon this with longitudinal follow-up to understand the dynamic nature of schizophrenia. Next, while we observed intriguing similarities and differences between patients with predominant positive symptoms and others with predominant negative symptoms of schizophrenia, our primary analysis was limited by the use of study-level data. Granular at the individual-level analyses of symptoms, prodromal states, genetic risk, and illness severity could yield a more nuanced understanding. Of note, most of these limitations add noise to the analysis and increase the probability of a false negative result, which is reassuring given that our primary analyses yielded positive results; however, future studies may identify additional components in the schizophrenia network. Finally, although our lesion-based validation may be promising for clinical translation (Siddiqi et al., 2022), this has not been shown specifically in schizophrenia. Future studies may test this hypothesis by directly targeting our network with focal brain stimulation or testing its utility as a biomarker.

Overall, this study identified a common brain network that links heterogeneous atrophy patterns associated with schizophrenia. The schizophrenia network showed relative stability across disease progression, underscoring its potential as a trait-like characteristic. It was also consistent across positive and negative symptoms, suggesting that it may be core to schizophrenia as a whole. This network may be useful in developing therapeutic targets and diagnostic tools for schizophrenia

Data Extraction

Data were extracted from each of the original studies cited in Liloia et al. (2021) using a predefined template. The template included sample size, atrophy coordinates, and clinical data regarding duration of illness and predominant symptom clusters (positive versus negative symptoms).

Coordinate Network Mapping

All schizophrenia studies were first combined into a single analysis. For each study, standard 4-mm spherical seeds were created at each identified coordinate of peak atrophy. Studies that reported Talairach coordinates were converted into MNI coordinates using Yale BioImage Suite. (Lacadie et al., 2008) Then, the spherical seeds were combined to generate a study-level seed. Resting-state functional connectivity data from a 1000-participant large normative connectome database (Yeo et al., 2011) were utilized to compute seed-based functional connectivity between study-level atrophy coordinates and the rest of the brain. Specifically, fMRI time courses were extracted for each brain region and for all coordinates of the study-level seed map. Then, Pearson’s correlation coefficient between time-courses was calculated. Using Fisher’s r to z transform, r values were converted to a normal distribution. Afterwards, Fisher z values were averaged across all 1000 subjects in the normative functional connectivity dataset. This process generated 1,000 functional connectivity maps for each study. These maps from the connectome were then merged to form a unified connectivity map per study, referred to as a ‘coordinate network’ for each study-level seed.

Sensitivity and Consistency Analyses

To conduct a sensitivity analysis, study maps were thresholded at |t|>5 and combined into one composite coordinate network overlap map, as in prior published work (Taylor et al., 2023). The analysis was also repeated using higher thresholds (|t|>7 and |t|>9) to confirm that the identified map was not threshold-dependent. To perform a consistency analysis, all unthresholded coordinate networks were incorporated in a voxel-wise one-sample t-test using FSL's randomise. This yields a map of connectivity of study-level seeds that describes how consistently they differ from zero. This map was corrected for multiple comparisons using threshold-free cluster enhancement (TFCE) (Smith and Nichols, 2009). The voxel threshold was established at 500 cubic millimeters, equivalent to 71 voxels, and the initial t value threshold was determined as t greater than 3.2. However, to improve the visibility of voxel clusters, the t value threshold was subsequently increased to be greater than 7 (Supplementary Table 1). Finally, the results of the sensitivity and consistency analyses were combined in one map (sensitivity-consistency map) showing the sensitivity of regions that survived multiple comparisons correction in the consistency analysis.

Specificity Analysis

The specificity analysis was performed via voxel-wise two-sample t-tests using FSL Permutation Analysis of Linear Models (PALM) and corrected for multiple comparisons using TFCE. The unthresholded maps that made up the schizophrenia network were compared to coordinate networks derived from atrophy locations associated with nine brain disorders and normal aging published in two meta-analyses (Goodkind et al., 2015; Minkova et al., 2017). The 10 control groups (coded separately in PALM) included 43 studies of individuals with normal aging, 39 studies of patients with MCI, 59 with Alzheimer’s disease, 40 with Parkinson’s disease, 16 with Huntington’s disease, 18 studies with bipolar disorder, 26 with major depressive disorder, 14 with anxiety disorders (including generalized anxiety disorder and panic disorder), 16 with obsessive-compulsive disorder, and 19 with a substance use disorder. Control analyses were also conducted to demonstrate that no specific disorder was responsible for primarily driving the results of our specificity analysis. This was accomplished by repeating the specificity analysis while excluding a different group each time (for instance, excluding AD, then excluding bipolar disorder, etc.). To test the robustness of the results, two-sample t-tests were performed comparing schizophrenia with only psychiatric disorders and with neurodegenerative disorders, exclusively. Finally, a specificity map was created showing statistically significant voxels (p_FWE <0.05) that were specific to schizophrenia using TFCE to correct for multiple comparisons. The t value threshold was set at t greater than 3.2 (Supplementary Table 2).

Conjunction Analysis

A conjunction analysis was performed on the results of the sensitivity-consistency analysis and the two-sample t-test of specificity to identify voxels that exhibited both consistency and specificity at a significance level of p_FWE < 0.05. This was achieved by applying a threshold of p_FWE < 0.05 to each map, binarizing them, and generating a conjunction map, which we termed the ‘schizophrenia network.’

Within Category and Across Category Analyses

The schizophrenia dataset was divided into two sub-datasets representing patients with recently diagnosed (illness duration < 2 years) or chronic schizophrenia (illness duration > 2 years), respectively. Sensitivity and consistency analyses were conducted on both groups following the same methods as above. Afterwards, sensitivity-consistency maps were compared using a two-sample t-test in FSL PALM and spatial correlation with permutation testing.

After excluding studies (30 out of 90 studies) that did not report specific Positive and Negative Syndrome Scale (PANSS) scores, the schizophrenia dataset was then divided according to predominant symptom cluster. The first dataset included 22 studies of patients with predominant positive symptoms, while the second group included 38 studies of patients with predominant negative symptoms. The sensitivity-consistency maps for each group were then contrasted using a two-sample t-test in FSL PALM and spatial correlation with permutation testing.

Spatial permutation tests were used to test whether the degree of similarity of networks between sub-groups and predominant symptoms clusters was stronger than expected by chance, as described in prior work (Siddiqi et al., 2021). Briefly, the maps were recomputed after randomly shuffling real study-level coordinates with hypothetical study-level coordinates, which were set to equal zero to simulate a one-sample t-test. The tests were computed in MATLAB with 10,000 permutations, excluding the comparison between recently diagnosed and chronic schizophrenia. This particular comparison required 100,000 permutations, as the initial p-value was nearly zero with only 10,000 permutations. P-values were calculated as the percentage of cases in which the real spatial correlation was stronger than the permuted correlation.

Next, the schizophrenia network was compared to functional connectivity maps of atrophy coordinates from two high risk groups: individuals at clinical high risk and individuals at genetic high risk using two-sample t-tests in FSL PALM, spatial correlation, and spatial permutation in MATLAB.

Neurosynth was used to identify coordinate locations via automated meta-analysis of neuroimaging data. We excluded any results that did not directly refer to or involve neuroanatomy. For instance, results focusing solely on psychological constructs or emotional states, such as 'fear', without a clear link to neuroanatomical changes or structures, were not included in our analysis.

Lesion-based Validation

We analyzed lesion locations in 181 participants from the Vietnam Head Injury Study (VHIS). The schizophrenia network was compared with the connectivity profile of each lesion, yielding an estimate of each lesion’s connectivity to the schizophrenia network. This estimate was compared to Neurobehavioral Rating Scale (NBRS) scores using Pearson correlations. The analysis was repeated after controlling for lesion size, Beck Depression Inventory (BDI) scores, diagnosis of post-traumatic stress disorder (given its high prevalence in the studied population) on the structured clinical interview for the DSM-IV, and Mini-Mental State Examination (MMSE) scores.

To perform lesion network mapping, we created normative functional connectivity of lesion locations associated with suspiciousness and unusual thought content using a 1000-subject connectome database. This process produced a connectivity map of each lesion location. These connectivity maps were compared with NBRS scores using voxel-wise Pearson correlation, yielding a map of the connectivity of lesions associated with greater scores on each NBRS item. Subsequently, the schizophrenia network was overlaid on the functional connectivity maps for NBRS items related to psychosis, specifically “suspiciousness” and “unusual thought content”.

Statistical Considerations

Statistical analyses were conducted in FSL (Jenkinson et al., 2012) and MATLAB (MathWorks, Natick, MA), as per the specific requirements of each step in the study. FSL was employed for its suite of neuroimaging analysis tools. We also made use of FSL's TFCE method for correcting multiple comparisons in our sensitivity and consistency analyses. MATLAB, on the other hand, was utilized for its robust capabilities in generating thousands of permutations, a necessity in spatial permutation tests.

TFCE was chosen because of its ability to preserve the spatial information of the statistical maps while also controlling for multiple comparisons (Smith and Nichols, 2009). This approach enhances the detectability of signal changes spread over space rather than localized to single voxels, making it particularly appropriate for our study of schizophrenia, where the disease pathology is likely to involve interconnected networks of the brain.

In contrast, spatial permutation tests were performed to evaluate the degree of similarity observed between sub-groups and predominant symptom clusters, as these tests allow for the comparison of spatial patterns under the null hypothesis of random spatial distribution. This method is useful to ensure the observed patterns were not due to chance, providing a robust statistical validation of our results (Siddiqi et al., 2021).

Potential conflicts of interest

S.H.S. is a scientific consultant for Magnus Medical, and a clinical consultant for Acacia Mental Health, Kaizen Brain Center and Boston Precision Neurotherapeutics. S.H.S. has received investigator-initiated research funding from Neuronetics and BrainsWay. S.H.S. has served as a speaker for BrainsWay (branded) and PsychU.org (unbranded, sponsored by Otsuka). S.H.S. owns stock in BrainsWay (publicly traded) and Magnus Medical (not publicly traded). S.H.S. owns intellectual property involving the use of functional connectivity to target TMS. M.D.F. is a consultant for Magnus Medical, Solaris and Boston Scientific, and has intellectual property using connectivity imaging to guide brain stimulation.

Burke MJ, Joutsa J, Cohen AL, Soussand L, Cooke D, Burstein R, et al. Mapping migraine to a common brain network. Brain. 2020;143(2):541-53.
Cash, R.F.H., Müller, V.I., Fitzgerald, P.B. et al. Altered brain activity in unipolar depression unveiled using connectomics. Nat. Mental Health 1, 174–185 (2023). https://doi.org/10.1038/s44220-023-00038-8
Chang M, Womer FY, Bai C, Zhou Q, Wei S, Jiang X, et al. Correction: Voxel-Based Morphometry in Individuals at Genetic High Risk for Schizophrenia and Patients with Schizophrenia during Their First Episode of Psychosis. PLoS One. 2017;12(1):e0170146.
Collins MA, Ji JL, Chung Y, Lympus CA, Afriyie-Agyemang Y, Addington JM, et al. Accelerated cortical thinning precedes and predicts conversion to psychosis: The NAPLS3 longitudinal study of youth at clinical high-risk. Mol Psychiatry. 2023;28(3):1182-9.
Dabiri M, Dehghani Firouzabadi F, Yang K, Barker PB, Lee RR, Yousem DM. Neuroimaging in schizophrenia: A review article. Front Neurosci. 2022;16:1042814.
Darby RR, Joutsa J, Fox MD. Network localization of heterogeneous neuroimaging findings. Brain. 2019;142(1):70-9.
Darby RR, Laganiere S, Pascual-Leone A, Prasad S, Fox MD. Finding the imposter: brain connectivity of lesions causing delusional misidentifications. Brain. 2017;140(2):497-507.
Delvecchio G, Lorandi A, Perlini C, Barillari M, Ruggeri M, Altamura AC, et al. Brain anatomy of symptom stratification in schizophrenia: a voxel-based morphometry study. Nord J Psychiatry. 2017;71(5):348-54.
Dong D, Wang Y, Chang X, Jiang Y, Klugah-Brown B, Luo C, et al. Shared abnormality of white matter integrity in schizophrenia and bipolar disorder: A comparative voxel-based meta-analysis. Schizophr Res. 2017;185:41-50.
Eickhoff SB, Bzdok D, Laird AR, Kurth F, Fox PT. Activation likelihood estimation meta-analysis revisited. Neuroimage. 2012;59(3):2349-61.
Ferguson MA, Lim C, Cooke D, Darby RR, Wu O, Rost NS, et al. A human memory circuit derived from brain lesions causing amnesia. Nat Commun. 2019;10(1):3497.
Fornito A, Yung AR, Wood SJ, Phillips LJ, Nelson B, Cotton S, et al. Anatomic abnormalities of the anterior cingulate cortex before psychosis onset: an MRI study of ultra-high-risk individuals. Biol Psychiatry. 2008;64(9):758-65.
Galovic M, van Dooren VQH, Postma TS, Vos SB, Caciagli L, Borzì G, et al. Progressive Cortical Thinning in Patients With Focal Epilepsy. JAMA Neurol. 2019;76(10):1230-9.
Goodkind M, Eickhoff SB, Oathes DJ, Jiang Y, Chang A, Jones-Hagata LB, et al. Identification of a common neurobiological substrate for mental illness. JAMA Psychiatry. 2015;72(4):305-15.
Hu ML, Zong XF, Mann JJ, Zheng JJ, Liao YH, Li ZC, et al. A Review of the Functional and Anatomical Default Mode Network in Schizophrenia. Neurosci Bull. 2017;33(1):73-84.
Huang P, Xi Y, Lu ZL, Chen Y, Li X, Li W, et al. Decreased bilateral thalamic gray matter volume in first-episode schizophrenia with prominent hallucinatory symptoms: A volumetric MRI study. Sci Rep. 2015;5:14505.
Jenkinson M, Beckmann CF, Behrens TE, Woolrich MW, Smith SM. FSL. Neuroimage. 2012;62(2):782-90.
Keshavan MS, Collin G, Guimond S, Kelly S, Prasad KM, Lizano P. Neuroimaging in Schizophrenia. Neuroimaging Clin N Am. 2020;30(1):73-83.
Kim NY, Hsu J, Talmasov D, Joutsa J, Soussand L, Wu O, et al. Lesions causing hallucinations localize to one common brain network. Mol Psychiatry. 2021;26(4):1299-309.
Kraguljac NV, Lahti AC. Neuroimaging as a Window Into the Pathophysiological Mechanisms of Schizophrenia. Front Psychiatry. 2021;12:613764.
Lacadie CM, Fulbright RK, Rajeevan N, Constable RT, Papademetris X. More accurate Talairach coordinates for neuroimaging using non-linear registration. Neuroimage. 2008;42(2):717-25.
Liloia D, Brasso C, Cauda F, Mancuso L, Nani A, Manuello J, et al. Updating and characterizing neuroanatomical markers in high-risk subjects, recently diagnosed and chronic patients with schizophrenia: A revised coordinate-based meta-analysis. Neurosci Biobehav Rev. 2021;123:83-103.
Menon V, Palaniyappan L, Supekar K. Integrative Brain Network and Salience Models of Psychopathology and Cognitive Dysfunction in Schizophrenia. Biol Psychiatry. 2023;94(2):108-20.
Minkova L, Habich A, Peter J, Kaller CP, Eickhoff SB, Klöppel S. Gray matter asymmetries in aging and neurodegeneration: A review and meta-analysis. Hum Brain Mapp. 2017;38(12):5890-904.
Ortiz-Gil J, Pomarol-Clotet E, Salvador R, Canales-Rodríguez EJ, Sarró S, Gomar JJ, et al. Neural correlates of cognitive impairment in schizophrenia. Br J Psychiatry. 2011;199(3):202-10.
Pomarol-Clotet E, Canales-Rodríguez EJ, Salvador R, Sarró S, Gomar JJ, Vila F, et al. Medial prefrontal cortex pathology in schizophrenia as revealed by convergent findings from multimodal imaging. Mol Psychiatry. 2010;15(8):823-30.
Ragland JD, Ranganath C, Harms MP, Barch DM, Gold JM, Layher E, et al. Functional and Neuroanatomic Specificity of Episodic Memory Dysfunction in Schizophrenia: A Functional Magnetic Resonance Imaging Study of the Relational and Item-Specific Encoding Task. JAMA Psychiatry. 2015;72(9):909-16.
Shafiei G, Markello RD, Makowski C, Talpalaru A, Kirschner M, Devenyi GA, et al. Spatial Patterning of Tissue Volume Loss in Schizophrenia Reflects Brain Network Architecture. Biol Psychiatry. 2020;87(8):727-35.
Sheffield JM, Barch DM. Cognition and resting-state functional connectivity in schizophrenia. Neurosci Biobehav Rev. 2016;61:108-20.
Shenhav A, Cohen JD, Botvinick MM. Dorsal anterior cingulate cortex and the value of control. Nat Neurosci. 2016;19(10):1286-91.
Siddiqi SH, Khosravani S, Rolston JD, Fox MD. The future of brain circuit-targeted therapeutics. Neuropsychopharmacology. 2023.
Siddiqi SH, Kording KP, Parvizi J, Fox MD. Causal mapping of human brain function. Nat Rev Neurosci. 2022;23(6):361-75.
Siddiqi SH, Schaper FLWV, Horn A, Hsu J, Padmanabhan JL, Brodtmann A, et al. Brain stimulation and brain lesions converge on common causal circuits in neuropsychiatric disease. Nat Hum Behav. 2021;5(12):1707-16.
Silbersweig DA, Stern E, Frith C, Cahill C, Holmes A, Grootoonk S, et al. A functional neuroanatomy of hallucinations in schizophrenia. Nature. 1995;378(6553):176-9.
Silbersweig D, Loscalzo J. Precision Psychiatry Meets Network Medicine: Network Psychiatry. JAMA Psychiatry. 2017;74(7):665-6.
Smith SM, Nichols TE. Threshold-free cluster enhancement: addressing problems of smoothing, threshold dependence and localisation in cluster inference. Neuroimage. 2009;44(1):83-98.
Stubbs, J.L., Taylor, J.J., Siddiqi, S.H. et al. Heterogeneous neuroimaging findings across substance use disorders localize to a common brain network. Nat. Mental Health (2023). https://doi.org/10.1038/s44220-023-00128-7
Takayanagi Y, Kulason S, Sasabayashi D, Takahashi T, Katagiri N, Sakuma A, et al. Reduced Thickness of the Anterior Cingulate Cortex in Individuals With an At-Risk Mental State Who Later Develop Psychosis. Schizophr Bull. 2017;43(4):907-13.
Tamminga CA, Stan AD, Wagner AD. The hippocampal formation in schizophrenia. Am J Psychiatry. 2010;167(10):1178-93.
Taylor JJ, Lin C, Talmasov D, Ferguson MA, Schaper FLWV, Jiang J, et al. A transdiagnostic network for psychiatric illness derived from atrophy and lesions. Nat Hum Behav. 2023;7(3):420-9.
van den Heuvel MP, Fornito A. Brain networks in schizophrenia. Neuropsychol Rev. 2014;24(1):32-48.
Vigo D, Thornicroft G, Atun R. Estimating the true global burden of mental illness. Lancet Psychiatry. 2016;3(2):171-8.
Wang R, Reddy PH. Role of Glutamate and NMDA Receptors in Alzheimer's Disease. J Alzheimers Dis. 2017;57(4):1041-8.
Weil RS, Hsu JK, Darby RR, Soussand L, Fox MD. Neuroimaging in Parkinson's disease dementia: connecting the dots. Brain Commun. 2019;1(1):fcz006.
Winkler AM, Ridgway GR, Webster MA, Smith SM, Nichols TE. Permutation inference for the general linear model. Neuroimage. 2014;92(100):381-97.
Wood SJ, Pantelis C, Velakoulis D, Yücel M, Fornito A, McGorry PD. Progressive changes in the development toward schizophrenia: studies in subjects at increased symptomatic risk. Schizophr Bull. 2008;34(2):322-9.
Yarkoni T, Poldrack RA, Nichols TE, Van Essen DC, Wager TD. Large-scale automated synthesis of human functional neuroimaging data. Nat Methods. 2011;8(8):665-70.
Yeo BT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, Hollinshead M, et al. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol. 2011;106(3):1125-65.
Zhukovsky P, Anderson JAE, Coughlan G, Mulsant BH, Cipriani A, Voineskos AN. Coordinate-Based Network Mapping of Brain Structure in Major Depressive Disorder in Younger and Older Adults: A Systematic Review and Meta-Analysis. Am J Psychiatry. 2021;178(12):1119-28.

CNMschizophreniaatrophypapersupplement10.06.23ATM.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Heterogenous Patterns of Brain Atrophy in Schizophrenia Localize to A Common Brain Network

Status:

Version 1

Abstract

Figures

Introduction

Results

Discussion

Methods

Data Extraction

Coordinate Network Mapping

Sensitivity and Consistency Analyses

Specificity Analysis

Conjunction Analysis

Within Category and Across Category Analyses

Lesion-based Validation

Statistical Considerations

Declarations

References

Supplementary Files

Status:

Version 1