Structural brain differences in patients with schizophrenia spectrum disorders with and without auditory verbal hallucinations

Schizophrenia spectrum disorders (SSD) are debilitating, with auditory verbal hallucinations (AVHs) being a core characteristic. While gray matter volume (GMV) reductions are commonly replicated in SSD populations, the neural basis of AVHs remains unclear. Using previously published data, this study comprises two main analyses, one of GMV dissimilarities between SSD and healthy controls (HC), and one of GMV differences specifically associated with AVHs. Structural brain images from 71 adults with ( n = 46) and without ( n = 25) SSD were employed. Group differences in GMVs of the cortex, anterior cingulate (ACC), superior temporal gyrus (STG), hippocampi, and thalami were assessed. Additionally, volumes of left Heschl ’ s gyrus (HG) in a subgroup experiencing AVHs (AVH + , n = 23) were compared with those of patients who did not (AVH-, n = 23). SSD patients displayed reduced GMVs of the cortex, ACC, STG, hippocampi, and thalami compared to HC. AVH + had significantly reduced left HG volume when compared to AVH-. Finally, a right-lateralized ventral prefrontal cluster was found to be uniquely associated with AVH severity. This study corroborates previous findings of GMV reductions in SSD cohorts. Chiefly, our secondary analysis suggests that AVHs are associated with language areas and their contralateral homologues.


Introduction
Schizophrenia spectrum disorders (SSDs) are chronic, debilitating illnesses that affect up to 1.8 % of the population and incur profound personal and societal consequences (Christensen et al., 20222;Ringeisen et al., 2023;Peritogiannis et al., 2022).According to The Diagnostic and Statistical Manual of Mental Disorders (5th ed., ; American Psychiatric Association [APA], 2013), SSDs have in common the occurrence of psychotic symptomatology, including hallucinations, delusions, and speech disturbances.
Of these disorders, schizophrenia (SZ) and schizoaffective disorder (SD) are among the most frequent and devastating, not in the least because they often develop in adolescence or early adulthood (Jung et al., 2021;Wy and Saadabadi, 2023), inflicting potentially catastrophic damage during a fragile period of development.Indeed, a systematic review found a significant association between early age of SZ onset and larger counts of subsequent hospitalizations (Immonen et al., 2017), highlighting the necessity for the development of robust objective biomarkers to aid diagnostic and interventive efforts (García-Gutiérrez et al., 2020;Hager and Keshavan, 2015;Kraguljac et al. 2021).While treated as separate disorders, the significant overlap in clinical phenotypes between SZ and SD, something known to complicate accurate diagnosis (Wy and Saadabadi, 2023), have made some researchers question their nosological distinctness, arguing that they should be placed on a continuum rather than treated as discrete categories (Allardyce, 2007;Kantrowitz and Citrome, 2011;Quattrone et al., 2019;Reininghaus et al., 2019).
Of the symptoms shared by SD and SZ, auditory verbal hallucinations (AVHs), defined as the experience of sensory verbal inputs in the absence of appropriate external stimuli (DSM-5), are some of the most frequent and treatment-resistant (Andreasen and Flaum, 1991;Shergill et al., 1998).While poorly understood, AVHs are widely believed to be partially facilitated by irregularities in brain areas related to language processing (Allen et al., 2012;Jardri et al., 2011).Indeed, a meta-analysis of functional magnetic resonance imaging (fMRI) studies by Kompus et al. (2011) showed paradoxical brain activation patterns in patients when experiencing AVHs when compared with external auditory stimuli.Specifically, increased activity in the left primary auditory cortex (A1) and right rostral prefrontal cortex (PFC) was significantly related to the experience of AVHs, with attenuation occuring in the presence of external stimuli (Kompus et al., 2011).Corroborating the notion that these symptoms are associated with areas involved in normal language and auditory processing, a study employing a dichotic listening paradigm by Hugdahl et al. (2012) found the severity of AVHs to be inversely correlated with the processing of speech input, particularly in the left, language-dominant hemisphere.Similarly, structural MRI studies of patients with AVHs have found negative associations between volumes of the superior temporal gyrus (STG), including Heschl's gyrus (HG, A1), and the severity of experienced hallucinations (Bohlken et al., 2017;Palaniyappan et al., 2012;Sumich et al., 2005).Overall, empirical evidence suggests that left-lateralized language areas should be prime targets of interest when investigating AVHs in SSDs.
In light of the above, this study aims to i) contribute to the growing knowledge base of neurostructural differences underlying SSDs by attempting to replicate the aforementioned overlapping findings previously demarcated in SD and SZ populations, and ii) investigate to which degree AVH symptomatology can be associated with structural anomalies in language processing areas, especially in the left hemisphere.Specifically, we hypothesize reduced GMV, both widespread (cortical) and isolated (ACC, hippocampus, STG, thalamus) in SSD patients when compared with demographically matched healthy controls (HC).Finally, we expect volumetric differences, i.e. size reductions or increases, to be apparent in left HG (A1) in patients with AVHs (AVH+) when compared to those without (AVH-), and that these scale with symptom severity.

Study design and participants
This study utilized an openly available dataset (https://openneuro.org/datasets/ds004302/), originally collected and shared by Soler-Vidal et al. (2022).The dataset contained both functional and anatomical scans, of which we used only the latter for our analysis.Data was acquired from a patient group consisting of 46 patients meeting DSM-5 criteria for either schizophrenia (SZ) or schizoaffective disorder (SD) and a control group consisting of 25 healthy individuals (HC).The patient group consisted of 23 AVH+ and 23 AVH-patients, all taking antipsychotic medication.All participants were between 18 or 65 years old and right handed, determined by a simple prompt (e.g."are you right-or left-handed?").For more information regarding recruitment, please see Soler-Vidal et al. (2022).

AVH
To measure the severity of auditory verbal hallucinations in SSD patients, the Psychotic Symptoms Rating Scale Auditory Hallucinations subscale (PSYRATS-AH) was used.The subscale consists of 11 items (see Haddock et al. (1999)) for elaboration on items) that probe the severity of AVH.Patients with positive scores were included in the AVH+ group, while patients in the AVH-group were required to have been hallucination-free for at least 6 months.For more information on clinical characteristics, please read Soler-Vidal et al. (2022).

Quality control protocol
All segmentations and parcellations were visually inspected postprocessing to ensure the integrity of the pipeline.In addition, two standardized ENIGMA consortium (http://enigma.usc.edu/)quality assessment protocols were implemented to assess data quality.Initially, a cortical quality control package was employed to extract summary statistics and outlier measures for each cortical parcellation across all subjects.Then, a workflow specifically developed for the assessment of hippocampal subdivision quality was utilized (Sämann et al., 2022), producing a similar report.Because no comparable pipeline exists for the thalamus, the quality of this structure's segmentation was evaluated through a standard stem-and-leaf outlier analysis and visual inspection.All outliers were closely scrutinized.Volumes were edited and re-run if 1) a clear, ameliorable reason seemed to cause the outlier status (e.g.failed skull stripping or pial boundary demarcation), 2) inter-rater consensus on the problem and its solution was achieved.

Volumetric regions of interest (ROIs)
Cortical parcellations were defined according to the Desikan-Killiany atlas (Desikan et al., 2006), included in the FreeSurfer analysis suite 7.4.1 (Fischl, 2012).Total volumes of the rostral and caudal sections of the anterior cingulate gyri were combined to make up the ACC measures of interest.To represent HG, the transverse temporal gyrus was employed.Unilateral volumes of the hippocampus and thalamus were derived from the subfield and -nuclei pipelines described above (Iglesias et al., 2015(Iglesias et al., , 2018)).Total cortical GMV was obtained from FreeSurfer's standard segmentation pipeline (Fischl, 2012).

Statistical procedures
Statistical analyses were conducted using SPSS v28 (IBM, Armonk, NY).First, normality of data distributions was assessed using a Shapiro-Wilk test (Shapiro and Wilk, 1965) and visually through Q-Q plots and histograms.Because inconsistencies exist as to whether SSD-related GMV findings are uni-or bilateral (Van Erp et al., 2016, 2018;Walton et al., 2017Walton et al., , 2018)), two analyses per ROI were conducted, one for each hemisphere.Univariate models were fit to assess volumetric differences between HC and patients.For all ten models (unilateral volumes of the cortex, ACC, STG, hippocampus, and thalamus, respectively), group (HC vs. patients) was the fixed effect while age, sex, IQ, and intracranial volume (ICV) were entered as covariates.Finally, to investigate whether patient subgroups deviated in volumes of HG, an identical model was employed but utilizing AVH+/AVH-as the grouping variable.Here, only volumes of the left hemisphere were tested on an a priori assumption that AVHs are related to language processing (Section 1.0), something we expect to be left-lateralized in the vast majority of our sample, which includes only right-handed individuals (Knecht et al., 2000).We applied the False Discovery Rate (FDR) method, set at q = 0.05 (5 %), to control for multiple comparisons (Benjamini and Hochberg, 1995), resulting in an FDR-adjusted p-value, p FDR .

Post-hoc analyses
Because a follow-up correlation analysis of left HG volumes and AVH severity was null (p > .05),we opted to investigate associations between symptom severity and deviations in cortical volume through an exploratory vertex-wise analysis.This approach was implemented within FreeSurfer on the 23 AVH+ subjects by fitting a general linear model on regional cortical volume estimates with individual PSYRATS-AH score as the explanatory variable and age, sex, IQ, and ICV as covariates.Cluster-correction was performed at vertex-wise and clusterdefining thresholds of 5 %.Smoothing was implemented using a 10 mm full-width-half-maximum gaussian kernel.Individual cluster scores were subsequently used, alongside PSYRATS-AH, in a bivariate Pearson correlation to illustrate the strength and direction of the association, solely for demonstrative purposes.Additionally, to assess whether reduced left-hemispheric HG volumes were likely to be language-related as hypothesized, and not a product of more general cortical deviations in AVH+, we explored whether volumes were different in the right hemisphere as well, utilizing an identical model as described above (Section 2.7).

Missing data protocol
Two observations of IQ were missing from the original dataset.To account for this, a monotone multiple imputation procedure was implemented (White and Carlin, 2010).The two resulting imputations were carried into the group-level analyses.To assess the robustness of results, all analyses were repeated utilizing only complete cases to ensure imputations had not skewed them (European Medicines Agency (EMA), 2010).

Sample characteristics
A summary of demographic measures and comparisons is reported in Table 1.

Quality assessment
Initial visual inspections revealed no gross processing errors.Quality control protocols delineated 145 cortical, 38 hippocampal subfield, and 66 thalamic nuclei outliers distributed across 59 unique subjects.Upon inspection, 40 cortical areas across 25 subjects warranted minor edits and subsequent re-processing, with the remaining 115 cortical outliers, along with every subcortical outlier, being deemed as occurrences of individual variation or errors of an incorrigibly small magnitude.Outliers did not systematically differ between the groups.The vast majority of cortical errors were related to the inclusion of non-brain tissue (e.g.skull, dura) within the pial surface border.

Post-hoc analyses
The vertex-wise analysis demarcated a right-lateralized cluster encompassing parts of the pars orbitalis and orbitofrontal cortex, the volume of which was significantly related to PSYRATS-AH score (Table 3, Fig. 2).The subsequent correlation analysis revealed the association to be strongly negative, r(21) = − 0.53, p = .010(Fig. 2), e.g. less volume in this area was related to increased symptom severity.The post-hoc assessment of whether reduced HG volume in AVH+ was present in the right hemisphere revealed no significant difference between patient subgroups, F(1, 40) = 0.18, p = .672.Note.Measures are given in the format: Mean (SD) where appropriate.T-tests and chi-square where appropriate (χ 2 ) were conducted between demographic measures to investigate group differences.Univariate analyses to investigate group differences in volumetric measures (mm 3 ) controlled for intracranial volume, age, IQ, and gender.HC = healthy controls, L = left, R = right, ACC = anterior cingulate cortex, STG = superior temporal gyrus.

Absolute reductions
In order to assess whether the reductions in individual cortical gyri (ACC and STG) were greater than the general reduction in cortical GMV, we used estimated marginal means corrected for sex, age, total intracranial volume and IQ to calculate the GMV reductions in percent.Our final outcome showed cortical GMV reduction in SSD when compared to HC to be 8.8 %, for STG 8.5 %, and for ACC 11.5 %.

Sensitivity analyses
Analysis reiterations utilizing only complete cases resulted in marginal changes in significance.For SSD/HC analyses, the largest change was the significance of the hippocampal finding, which changed from p  = .014to p = .009.For AVH+/AVH-, the HG finding changed, from p = .049to p = .037.No repeat of the vertex-wise analysis took place since no subjects within the AVH+ group had missing data.

Discussion
The aims of the current study were to i) replicate neurostructural results previously discovered in SSD populations, and ii) investigate the degree to which AVH symptomatology is associated with structural anomalies in language processing areas.Indeed, the findings produced by this investigation were consistent with previous results, revealing significant GMV differences between SSD and HC, AVH+ and AVH-.SSD patients displayed cortical and subcortical GMV reductions spanning global (cortex) and regional (ACC, STG, hippocampi, thalami) measures, while AVH+ presented with unique irregularities of left HG, confirming our main hypotheses.While our hypothesized association between HG volume and symptom severity was null, an exploratory whole-brain cortical analysis delineated a right-lateralized ventral prefrontal cluster, the volume of which was significantly related to the severity of AVHs.

Main findings
The present findings of reduced widespread GMV in SSD patients are in line with two recent large-scale meta-analyses by the ENIGMA consortium (Gutman et al., 2022;Van Erp et al., 2018), wherein global and regional volume reductions were found in SZ cohorts.These anomalies tend to be apparent at diagnosis and worsen longitudinally, interpreted to reflect a neurodevelopmental or -progressive dimension to SZ (Gutman et al., 2022;Van Erp et al., 2018).Similar patterns have been found in SD populations (Amann et al., 2016;Van Lutterveld et al., 2014), supporting the assumption that overlapping pathophysiological processes subserve both disorders.However, the present study is unable to determine the degree to which global GMV reduction is a biomarker that is specific to SSDs, or whether it is a biomarker that is simply related to a general psychopathology factor.Indeed, a link between general psychopathology and reduced GMV have been suggested by empirical studies in children and adolescents (Kaczkurkin et al. 2019;Snyder et al. 2017).To our knowledge, no such studies have directly examined this issue in adult populations.
Additionally, the present analysis corroborated previous findings of volume reductions in both the ACC and STG.The ACC has been hypothesized to be a principal junction in a complex neurophysiological network implicated in numerous aspects of affective, social and executive cognition (Allman et al., 2001).In SSD cohorts, reduced activity of this area during emotional processing and its dysconnectivity with the insular cortex have been associated with poorer social functioning and increased aggression, respectively (Fortier et al., 2023;Nelson et al., 2015), making the ACC a conspicuous target for efforts hoping to understand and improve the long-term social outcomes of SSDs.Indeed, volumetric reductions of the ACC represent a common finding in SZ literature (Baiano et al., 2007), with some evidence to suggest an association between these and positive symptomatology (Choi et al., 2005).Similarly, morphometric differences of the STG have been reported in SZ cohorts (Bandeira et al., 2021;Walton et al., 2018), with some naming it as a future candidate for a prognostic biomarker given more research is undertaken to affirm its robustness (Bandeira et al., 2021).However, our post-hoc analysis on effect sizes raised questions on whether findings on STG reductions are explained by the overall GMV reductions or if the STG actually is reduced further.Indeed, the effect size on overall GMV reduction surpasses the regional reduction of the STG, which could imply that the reduction of the STG is part of the overall atrophy of GMV in SSD patients.Reductions of the ACC, however, were greater than those of the cortex as a whole, implying an accelerated atrophy in the  Note.Region defined according to the Desikan-Killiany atlas.Coordinates are given in MNI152-space.P-value is corrected for multiple comparisons at a vertex-wise and cluster-defining threshold of 5 %.
ACC in particular, which could indicate this as a potential region of interest for future examinations.
In addition to global GMV reductions and specific ACC differences, we also found volumetric reductions in the thalamus and hippocampus.The thalamus, frequently dubbed the 'central relay station' of the brain, has been argued to be related to SSD in numerous ways, such as by contributing to poor mental coordination (Andreasen et al., 1998), parkinsonisms (Fritze et al., 2021), and altered cognitive functioning (Pergola et al., 2015).Supporting this purported subcortical involvement, volumetric reductions of thalamic and hippocampal volumes are robustly found in studies of SZ and SD (Brosch et al., 2022;Smith et al., 2011;Van Erp et al., 2016).Like the thalamus, the hippocampus is commonly associated with broad aspects of cognition and sensory processing, and is thus a prime target in SSDs, disorders characterized in part by dysfunctions within these domains.Particularly, its role as a hub of glutamatergic neurotransmission is thought implicated in the pathophysiology of SZ through subfield-specific hypermetabolism, serving as a driver of psychotic symptoms and of the progressive hippocampal and cortical atrophy previously observed in patients (Lieberman et al., 2018).Together, our findings of volumetric reductions of the thalamus and hippocampus in patients bolster the hypothesis that irregularities within deep subcortical areas are associated with SSD pathophysiology.However, the cross-sectional design of this study prevents us from inferring on causality, see limitations.

Secondary findings (AVH related findings)
Previously, GMV alterations have been posited to contribute to psychotic symptomatology by disrupting corollary discharge (Whitford et al., 2012;Parlikar et al., 2019), an integral system believed responsible for the accurate distinction between self-and other-generated sensory inputs (Parlikar et al., 2019).Bolstering this interpretation is a recent fMRI study that found evidence of altered corollary discharge specific to the auditory cortex in patients with AVH (Stephane et al., 2022).While we are not currently able to directly associate our finding of reduced left HG volume in AVH+ patients with corollary discharge, it could be interpreted to be connected to this phenomenon by reflecting atrophy related to disruptions in the ability to separate internally and externally generated verbal auditory stimuli.This could potentially align with Soler-Vidal et al.'s (2022) interpretation that reduced left HG activity in the present SSD sample may reflect early auditory information processing deficits.Evidently, our findings contribute to the growing evidence that left-hemispheric auditory pathways are implicated in the presentation of AVHs (Hugdahl et al., 2012;Kompus et al., 2011).
Conversely, the cluster demarcated by our exploratory cortical analysis was right-lateralized, centered on the pars orbitalis (PO) of the inferior frontal gyrus (IFG), an area the contralateral homologue of which is associated with language processing (Ardila et al., 2017).Interestingly, irregular activity of the contralateral counterparts of language areas has previously been found in AVH+ patient cohorts (Bohlken et al., 2017), with one meta-analysis reporting reduced language lateralization in AVH+, but not AVH-, SSD patients (Ocklenburg et al., 2013).The right IFG specifically has been found to be uniquely engaged during AVHs when compared to regular inner speech (Sommer et al., 2008), with a study reporting reduced right PO thickness in clinical, but not non-clinical, participants experiencing AVH (Van Lutterveld et al., 2014).Connecting our left HG and right PO findings, a meta-analysis of fMRI studies in AVH cohorts revealed a specific pattern of increased and decreased activity in left auditory areas and right rostral prefrontal areas in the absence and presence of external auditory stimuli, respectively (Kompus et al., 2011).Thus, while our findings bode well for future investigations of left-hemispheric auditory areas and right-hemisphere language region homologues in AVH+ populations, it is important to stress that language lateralization and its relation to AVHs is still poorly understood.Indeed, lateralization irregularities can encompass both reduced and inverted language lateralization, as well as dysfunctional inter-hemispheric communication (Ocklenburg et al., 2015), complicating their interpretation.On a general level, we believe our findings support the notion that brain areas associated with different aspects of auditory and language processing, as well as their contralateral counterparts, are functionally implicated in the mediation of AVHs, reflected in structural irregularities when comparing AVH+ to AVH-and non-patient groups.Overall, while our analyses comparing AVH+ and AVH-corroborated evidence from the literature, more research is needed to elucidate the relationship between language lateralization and AVHs in patient and non-patient groups.

Strengths and limitations
This study has several strengths that are worthy of mention.First, our sample was relatively large considering current neuroimaging standards (Szucs and Ioannidis, 2020).Second, our results all survived multiple comparisons, as well as corrections for age, sex, IQ and intracranial volume (Sullivan et al., 2004;Wang et al., 2019).Finally, manual quality control of all parcellations and segmentations in Freesurfer by multiple people likely decreased the likelihood of our variables to be affected by gross imaging and computational errors (Monereo-Sánchez et al., 2021).
Regarding limitations, we consider a principal one to be the combined SZ/SD patient cohort.When categorizing patients with SZ or SD under the same group, findings risk being wrongly attributed to both phenotypes, when in reality their occurrence may be due to systematic differences between the groups, either in structural measures and/or in a general sample-wide overrepresentation of one diagnosis over the other.Conversely, other findings may be missed because they are too subtle to be discovered in a fused SSD-analysis.Indeed, some researchers have found morphological differences in the basal ganglia and thalamus between patients with SZ and SD (Cobia et al., 2021;Smith et al., 2011).However, as mentioned in the introduction, a growing body of empirical evidence suggests a neurobiological overlap between the disorders, and the fact that strong results have been found in data where SZ and SD participants grouped together is promising for the theory that SZ, SD, and related disorders exist on a spectrum.
Another limitation of this study is the lack of information we have on the severity and duration of the participants' illness courses.If these had been available, tests could have reaffirmed previous papers' results showing significant correlations between the duration and severity of patients' illness history and volumes in ROIs (Kim et al., 2017).Lastly, an inverse relationship between medication use and temporal pole (and potentially other regions) thickness in SZ has been shown previously (Van Erp et al., 2018), presenting another limitation in our ability to infer generally based on our results, since all participants were medicated.Additionally, due to our limited knowledge of the type of medication which our participants were on, we were unable to correct for potential differences across this dimension.This may be problematic since certain types of medication have been shown to be associated with specific subcortical differences (Vita et al., 2015).

Implications
Present findings mark another step towards solidifying evidence of volumetric biomarkers in individuals with SSDs.We have replicated research findings implicating global and regional GMV reductions in SSDs, and left-lateralized volumes of auditory areas in AVHs.Moreover, the novel discovery of a right-hemispheric cluster related to AVHs lends support to recent advances indicating a role of contralateral language area homologues in mediating this symptom dimension.Future studies should probe the robustness of these findings by attempting replication.Finally, with this study being the second analysis of the data acquired by Soler-Vidal et al. (2022), and by using the 'leftover' structural data from a functional analysis to analyze volumetric differences between groups, this paper is evidence that previously 'used' data may benefit from 'recycling', as hidden discoveries may yet be discerned from openly available neuroimaging repositories (Markiewicz et al., 2021).

Conclusion
This study constitutes another contribution to the growing body of evidence on overlapping neurostructural biomarkers in SSDs, particularly findings on general and regional GMV reductions.Additionally, our results implicate primary auditory areas and contralateral homologues of language regions in the mediation of AVHs.

Fig. 2 .
Fig. 2. Results of exploratory vertex-wise whole-brain volume analysis.A) Cluster (red, right hemisphere) significantly associated with auditory verbal hallucination symptom severity.B) Scatterplot of relationship between individual cluster volumes and symptom severity.PSYRATS-AH = The Psychotic Symptom Rating Scale Auditory Hallucination subscale.

Table 1
Demographics, volumetric descriptive statistics, and analyses between patients and healthy controls.

Table 2
Demographics, volumetric descriptive statistics, and analyses between patient subgroups.Measures are given in the format: Mean (SD) where appropriate.T-tests and chi-square where appropriate (χ 2 ) were conducted between demographic measures to investigate group differences.Univariate analyses to investigate group differences in volumetric measures (mm 3 ) controlled for intracranial volume, age, IQ, and gender.AVH+ = patients with present auditory verbal hallucinations, AVH-= patients without present auditory verbal hallucinations.PSYRATS-AH = Psychotic Symptoms Rating Scale Auditory Hallucinations subscale, L = left, R = right.

Table 3
Cluster table for vertex-wise analysis of PSYRATS score and right-hemispheric cortical volume.