Introduction
Hearing verbal auditory hallucinations (AHs), commonly described as ‘real voices’, in the absence of any real external auditory stimulation, can be a disturbing and disabling experience. Despite over 20 years of focused research, the aetiological basis of AHs still remains unresolved. Cognitive approaches have suggested that AHs are the consequence of faulty monitoring of inner speech (Frith & Done, Reference Frith and Done1988), problems with reality discrimination (Bentall, Reference Bentall1990) or impaired auditory feedback (David, Reference David2004), with only limited support for each of these models (Seal et al. Reference Seal, Aleman and McGuire2004).
Members of our group in McKay et al. (Reference McKay, Headlam and Copolov2000) employed a battery of auditory tests routinely used in a clinical environment to investigate whether AH patients show specific auditory processing deficits. Compared with healthy controls and patients with psychosis but without AH (non-AH), the AH patients performed significantly more poorly on a frequency tone pattern test and a staggered spondaic words test. These findings were argued to be indicative of either right auditory cortex dysfunction and/or interhemispheric pathway deficits at the level of corpus callosum. Other neuropsychological studies, using dichotic listening tasks, have also reported attenuated performance in AH compared with non-AH patients and again interpreted their findings as reflecting abnormal trans-callosal transfer (Green et al. Reference Green, Hugdahl and Mitchell1994; Bruder et al. Reference Bruder, Rabinowicz, Towey, Brown, Kaufmann, Amador, Malaspina and Gorman1995). Finally, functional magnetic resonance imaging (fMRI) studies have established a reversal of auditory cortex activity in AH patients, that is, right greater than left brain activation whilst listening to speech (Woodruff et al. Reference Woodruff, Wright, Bullmore, Brammer, Howard, Williams, Shapleske, Rossell, David, McGuire and Murray1997); the expected pattern in healthy controls is left larger than right activation. Further, this reversal was only significant within the secondary auditory cortex.
Thus, the literature indicates that an auditory interhemispheric connectivity problem, that is, dysfunctional communication between the left and right auditory cortices, may lead to difficulties in central auditory processing (McKay et al. Reference McKay, Headlam and Copolov2000) and also predispose individuals to unusual auditory perceptions and/or AH (Rossell et al. Reference Rossell, Shapleske, Fukuda, Woodruff, Simmons and David2001; Woodruff, Reference Woodruff2004; Rossell & Boundy, Reference Rossell and Boundy2005; Shea et al. Reference Shea, Sergejew, Burnham, Jones, Rossell, Copolov and Egan2007). The present study was designed to clarify directly whether abnormalities in AH patients are at the level of interhemispheric connectivity. We used functional connectivity in fMRI as a measure of interhemispheric connections at the level of auditory cortices. The interhemispheric connectivity was separately estimated for the primary (A1) and secondary (A2) auditory cortices since both these brain areas have been previously indicated to abnormally activate in patients with schizophrenia who experience AH (Woodruff et al. Reference Woodruff, Wright, Bullmore, Brammer, Howard, Williams, Shapleske, Rossell, David, McGuire and Murray1997; Dierks et al. Reference Dierks, Linden, Jandl, Formisano, Goebel, Lanfermann and Singer1999; Lennox et al. Reference Lennox, Park, Medley, Morris and Jones2000; van de Ven et al. Reference van de Ven, Formisano, Roder, Prvulovic, Bittner, Dietz, Hubl, Dierks, Federspiel, Esposito, Di Salle, Jansma, Goebel and Linden2005). Functional connectivity as investigated in neuroimaging is defined as the ‘temporal correlation between spatially separated neurophysiological measurements’ (Friston et al. Reference Friston, Frith, Liddle and Frackowiak1993). The source of these correlations in fMRI measures is presumed to be the correlated firing rates of interconnected neurons (Xiong et al. Reference Xiong, Parsons, Gao and Fox1999). The correlation between firing rates can be induced by task performance or it can be spontaneously observed, as in resting state fMRI data (Xiong et al. Reference Xiong, Parsons, Gao and Fox1999). Previous work has shown the existence of interhemispheric auditory cortex connectivity within resting state fMRI data (Cordes et al. Reference Cordes, Haughton, Arfanakis, Wendt, Turski, Moritz, Quigley and Meyerand2000). Further, Quigley et al. (Reference Quigley, Cordes, Turski, Moritz, Haughton, Seth and Meyerand2003) found diminished interhemispheric connectivity in the auditory cortices of patients with agenesis of the corpus callosum, suggesting that the temporal correlation between homologous auditory cortices time-courses depends upon the integrity of the white-matter interhemispheric tracts. Thus, the estimation of functional connectivity from resting state fMRI data is well suited to investigate interhemispheric connectivity of the auditory cortex.
The aim of this study was to compare the interhemispheric functional connectivity in the auditory cortex across three groups of participants: (1) AH patients; (2) non-AH patients; (3) healthy controls. Regions of interest (ROI) were defined using individual functional activation maps obtained from a task requiring participants to listen to the presentation of single words. Resting state fMRI data were recorded for the participants and the functional connectivity was estimated for the ROI as defined in the stimulus-related data. To investigate the reliability of the connectivity estimates, we also used a method that removed the stimulus-induced effects (Arfanakis et al. Reference Arfanakis, Cordes, Haughton, Moritz, Quigley and Meyerand2000; Whalley et al. Reference Whalley, Simonotto, Marshall, Owens, Goddard, Johnstone and Lawrie2005; Gavrilescu et al. Reference Gavrilescu, Stuart, Rossell, Henshall, McKay, Sergejew, Copolov and Egan2008) and estimated the functional connectivity for the same ROI from the stimulus-related data.
Methods
Participants
Patients and healthy volunteers registered on a participant registry held at the Mental Health Research Institute (Melbourne) were recruited into three groups of participants: (1) 14 schizophrenia patients with a current history of AHs; (2) 13 schizophrenia patients with no lifetime history of AHs; (3) 16 healthy controls. All participants on the registry were from the same inner metropolitan region of Melbourne (North-West). The study was approved by Melbourne Health Human Research Ethics Committee (North Western Mental Health) and all participants gave written consent prior to scanning.
Patients were diagnosed with schizophrenia or schizoaffective disorder according to the Structured Clinical Interview for DSM-IV (First et al. Reference First, Spitzer, Gibbon and Williams1996). Diagnostic and symptom ratings were performed by a trained clinical research assistant (E.N.) and confirmed with a senior clinical researcher (S.R.). Exclusion criteria in all groups included a neurological disorder or trauma that may affect cognition, poor command of English, the diagnosis of a substance use disorder, an estimated pre-morbid IQ of <70 as measured by the National Adult Reading Test, 2nd edition (Nelson & O'Connell, Reference Nelson and O'Connell1978) and poor hearing as assessed by a trained audiologist (K.H.). One AH patient was excluded due to a mild to moderate hearing loss bilaterally. Additionally, the healthy controls had no psychiatric history or first-degree relatives with a psychiatric history.
Current psychotic symptoms were recorded on the Positive and Negative Syndrome Scale (PANSS, Kay et al. Reference Kay, Fiszbein and Opler1987). Five symptom ratings are reported: AH severity (as measured by item P3 of the PANSS scale; this P3 score was adjusted to reflect auditory–verbal hallucinations only rather than general hallucinations and called P3-aud), global positive, negative and general psychopathology scores (Table 1). As global positive PANSS includes the P3 values, we also reported a modified score (PANSS Pmod) calculated as the sum of the first seven PANSS components excluding P3, to assess whether the two patient groups were matched for other positive symptoms. All AH patients had a current score of ⩾3 on P3 and a history of P3⩾3 for longer than 1 month (this criterion was also used by Shapleske et al. Reference Shapleske, Rossell, Simmons, David and Woodruff2001). All non-AH patients had never experienced AHs (n=11) or only had a minor experience of AHs at P3⩽3 for <1 week in the past (>18 months ago n=2). Despite the AH patients having a current history of AHs none of the patients reported experiencing hallucinations during scanning. Current medication dosage was calculated based on chlorpromazine equivalents (Davis, Reference Davis1976). The handedness of participants was assessed using Coren's Laterality Index (Coren, Reference Coren1993) and was matched across groups.
Table 1. Demographic and clinical measures for the three groups
Non-AH, without auditory hallucination; AH, with auditory hallucination; PANSS, Positive and Negative Syndrome Scale; CPZe, chlorpromazine equivalent; n.s., non-significant.
The across groups comparison for age and IQ were based on one way analysis of variance, factor group; for gender on χ2 test; for positive, positive modified, negative, general PANSS, age of onset, duration of illness on two sample t test; for P3-aud on Mann–Whitney U test.
The three groups were matched for age, IQ and gender. The two patient groups were matched for positive, positive modified, negative, general PANSS, age of onset, duration of illness and medication dosage but differed significantly for P3-aud (Table 1). Thus, importantly, the two patients groups were matched for all demographic and clinical variables except for a significant history of and current experience of AH.
fMRI data description
Resting state data
Resting state data with identical acquisition parameters were acquired for a subset of each group of participants (12 AHs, 11 non-AHs and 14 controls) with the following imaging parameters: TR=2 s, TE=40 ms, FA=60, 3.5×3.5×5 mm3 voxel size. For each participant, 146 images were recorded in one scanning session. The participants wore MRI-compatible headphones enclosed within a thick foam cushion to attenuate the scanner noise and were instructed to relax with their eyes closed. The participants were debriefed after resting state data collection and none of them reported falling asleep.
Auditory stimulus data
Participants were also scanned while passively listening to auditory stimuli. The paradigm was presented as a block design with three active conditions and a resting state baseline. All four conditions were presented in each block. The baseline condition was always placed at the end of the block and the order of the active conditions was pseudo-randomized across blocks using a Latin square design. During the active conditions, emotionally neutral words (neutral-valence monosyllable words spoken in a male voice from ANEW word list; Bradley & Lang, Reference Bradley and Lang1999) were presented to the participants either monaurally (to the left or to the right ear) or binaurally (bilateral presentation). The stimuli were set at a comfortable listening level and presented using electrodynamic speakers compatible with the MRI environment (Baumgart et al. Reference Baumgart, Kaulisch, Tempelmann, Gaschler-Markefski, Tegeler, Schindler, Stiller and Scheich1998; http://www.mr-confon.de). The mean duration of the words was 516 ms and they were adjusted to produce equal loudness. Six words were presented per condition with an inter-stimulus interval of approximately 1 s (the total duration of each condition was 9 s, allowing the acquisition of three full brain images). The duration of the rest condition was also 9 s. Data were acquired for 14 blocks per participant. The participants were instructed to relax with their eyes closed and listen to the stimuli. During functional data acquisition, the AH participants were provided with a two button box to indicate the start and end of any AH. Gradient echo planar images were recorded using a 3 T GE Signa LX whole body scanner (General Electric, USA) with the following imaging parameters: TR=3 s, TE=40 ms, FA=60, 1.88×1.88×5 mm3 voxel size. A structural T1 image was acquired for each participant using a spoiled-gradient-echo acquisition (SPGR sequence) in sagittal orientation with an in-plane resolution of 0.98×0.98 mm2 and a slice thickness of 1.3 mm (TR=7 min 56 s, matrix size 256×256×123).
Data processing
Data pre-processing was performed using SPM2 (http://www.fil.ion.ucl.ac.uk/spm/). The images recorded during auditory stimulus presentation were motion corrected; the mean echo planar imaging (EPI) image after motion correction for each participant was then normalized to the EPI template. The normalization parameters were then applied to all motion corrected images, these images were spatially smoothed using an 8×8×8 mm3 Gaussian kernel and high pass temporally filtered (128-s cut-off). The statistical analysis of the stimulus activated data employed an AR(1) model to account for temporal autocorrelation in the data. Three regressors were defined corresponding to the three active conditions: bilateral, left ear and right ear presentation respectively. For each participant, the F-map of the effects of interest at p<0.001 uncorrected probability threshold was then used to define the ROI used in the functional connectivity analysis.
The resting state data images were motion corrected and co-registered to the mean EPI image (after realignment) of auditory stimulus activation data for each participant. The images were then smoothed using a 5×5×5 mm3 Gaussian kernel.
ROI definition
The T1 images for all subjects were normalized to the standard space. M.G. was trained by S.R. to delineate the A1 and A2 regions on the normalized T1 images following criteria described in Shapleske et al. Reference Shapleske, Rossell, Simmons, David and Woodruff2001. The functional activation F-map for each participant (effects of interest F-map at p uncorrected<0.001) was overlayed on to the normalized T1 image of the same participant. Spherical ROI (4 mm radius) were defined separately for A1 and A2 in the left and right hemisphere within the delineated anatomical borders. The spheres were centred on the peaks in the activation maps as indicated in Fig. 1 a. By confining the ROI delineation to voxels around the peak, we sought to include in the ROI only those voxels exhibiting a strong task-related response. Thus, the position of the ROI was variable across participants to account for: (a) anatomical variability in the location of A1 and A2 across participants (Penhune et al. Reference Penhune, Zatorre, MacDonald and Evans1996; Tervaniemi & Hugdahl, Reference Tervaniemi and Hugdahl2003); (b) imperfections in the spatial normalization (Nieto-Castanon et al. Reference Nieto-Castanon, Ghosh, Tourville and Guenther2003). The spherical ROI defined in the standard space (33 voxels) were then projected back into the native space of each subject, keeping the ROI volumes constant across subjects (nine voxels for resting state data and 18 voxels for auditory stimulus data). These voxels were used to extract time-courses from both resting state and auditory stimulus data in each participant (Fig. 1 b–d).
Fig. 1 (a) F-maps of the effects of interest at p<0.001 uncorrected probability threshold. The F-maps are overlaid onto the normalized structural T1 images for each participant. One participant from each group is represented; first column a control participant; second column a patient without auditory hallucinations (non-AH); third column an auditory hallucination (AH) patient. All images are in radiological orientation. The blue circles represent the regions of interest (ROI) defined as 4-mm radius spheres over the primary (A1) and secondary (A2) auditory cortices; (b) representative time-course for the ROI in A1L in each participant. These time-courses were extracted form the resting state data; (c) time-courses extracted from the stimulus-related data before the stimulus-induced effects were regressed out. The fitted canonical haemodynamic response function is represented on the top of the figures; (d) the time-courses extracted from A1L in each participant from the stimulus-related data after the stimulus-induced effects were regressed out. For simplicity, only one session is represented in (c) and (d). BOLD, Blood oxygen level dependent.
Time-course processing
The time-courses extracted from resting state and stimulus data were averaged over the voxels included in a given ROI to produce a mean time-course for each participant (Fig. 1 b, c). These time-courses were demeaned, variance normalized, high pass filtered (using SPM2 routines) with a high pass cut-off of 0.01 Hz (Lund et al. Reference Lund, Madsen, Sidaros, Luo and Nichols2006) and denoised primarily to remove head motion-related artefacts (the motion parameters were three translations and three rotations, which were regressed out of the mean time-courses). Contributions from physiological noise (cardiac rhythm and respiration) were eliminated following a method similar to Madsen & Lund (Reference Madsen and Lund2006) and Behzadi et al. (Reference Behzadi, Restom, Liau and Liu2007). These authors suggested that physiological noise contamination is best captured by voxels with high variance in time. Therefore, a standard deviation image was calculated for each participant separately for resting state data and stimulus-related data. Consistent with Lund et al. (Reference Lund, Madsen, Sidaros, Luo and Nichols2006), regions of large variance were identified around the brain stem in the Circle of Willis. Time-courses for six voxels were extracted from this region and were regressed from the mean time-courses for each ROI prior to connectivity estimation.
Functional connectivity estimation
The resting state interhemispheric functional connectivity was estimated separately for the primary (A1L–A1R) and secondary (A2L–A2R) auditory cortices as the correlation coefficients between the pre-processed time-courses of the respective ROI (Fig. 1 b). We also estimated rest-like connectivity in the stimulus-related dataset to investigate the reliability of the connectivity estimates. This method is described in detail in Gavrilescu et al. (Reference Gavrilescu, Stuart, Rossell, Henshall, McKay, Sergejew, Copolov and Egan2008). Briefly, a mean time-course for each ROI was averaged across the blocks of auditory stimulation (14 blocks over two sessions) with the haemodynamic delay modelled as a 6-s shift. The haemodynamic response was estimated separately for each ROI and each participant to account for inter-participant and within-participant spatial variation in the haemodynamic response (Handwerker et al. Reference Handwerker, Ollinger and D'Esposito2004) and then regressed from the mean time-course of each ROI (Fig. 1 d).
Statistical analysis of functional connectivity data
The group level analysis of functional connectivity data estimated by both methods was performed using a 3×2×2 analysis of variance (ANOVA) with factors group, method and regions. In general, an ANOVA analysis is followed by post-hoc t tests that discard scan level variance resulting in loss of statistical power (Jiang et al. Reference Jiang, He, Zang and Weng2004; Salvador et al. Reference Salvador, Suckling, Coleman, Pickard, Menon and Bullmore2005). Since we expected subtle differences between the groups, as in McKay et al. (Reference McKay, Headlam and Copolov2000), we employed statistical methods that avoided the loss of statistical power by taking into account the scan level variance and modelling the fMRI temporal autocorrelation (Bullmore et al. Reference Bullmore, Long, Suckling, Fadili, Calvert, Zelaya, Carpenter and Brammer2001). A vector autoregressive modelling strategy (VAR; Gujarati, Reference Gujarati1995; Gavrilescu et al. Reference Gavrilescu, Stuart, Rossell, Henshall, McKay, Sergejew, Copolov and Egan2008) was used to account for participant level autocorrelation in the time-courses, allowing the use of the meta-analytic Q test for homogeneity (Hedges & Olkin, Reference Hedges and Olkin1985) for post-hoc comparisons. The Q test combines participant level and group level variance to compare the connectivity estimates across groups. The value of Q to detect between group differences is χ2 distributed with one degree of freedom. VAR models were used to create non-correlated residuals both for the resting state and the stimulus-removed time-courses with the functional connectivity estimated as correlation between these residuals.
Results
Statistical analysis of the stimulus data
The auditory stimulus produced consistent widespread activation over the primary and secondary auditory cortices in each participant with significant effects of interests F-maps (Fig. 1 a). For all participants the ROI coordinates were in the range of coordinates previously reported for A1 and A2 cortices (Belin et al. Reference Belin, Zatorre, Hoge, Evans and Pike1999; Griffiths & Warren, Reference Griffiths and Warren2002).
The interhemispheric functional connectivity
The group average connectivity values are presented in Table 2. The ANOVA analysis indicated that only the group effect was significant (Table 3). Post-hoc tests showed that for resting state data the interhemispheric connectivity estimates for the primary auditory cortex (A1L to A1R) was significantly lower for the AH group compared with the healthy controls (Q=17.5, p<0.0001) and the non-AH patients (Q=10.1, p<0.001). Similar results were obtained for the secondary auditory cortex (A2L to A2R) connectivity (AH patients versus healthy controls Q=7.1, p<0.0001; AH patients versus non-AH patients Q=5.0, p=0.03). No significant difference was observed between the healthy controls and the non-AH patients (Q=0.4, p=0.5 for A1 connectivity; Q=0.08, p=0.8 for A2 connectivity).
Table 2. The average functional connectivity values for the three groups (mean±standard deviation)
Table 3. The 3×2×2 analysis of variance on the correlation values
The functional connectivity estimates from the stimulus-related data also demonstrated significant differences between AH patients and controls. The primary auditory cortex connectivity was significantly diminished for the AH patients (AH patients versus healthy controls Q=20.7, p<0.0001; AH patients versus non-AH patients Q=9.9, p=0.002) as was the secondary auditory cortex connectivity (AH patients versus healthy controls Q=22.0, p<0.0001; AH patients versus non-AH patients Q=16.5, p<0.0001). No differences were observed between non-AH patients and healthy controls (Q=1.6, p=0.2 for A1 connectivity; Q=0.2, p=0.7 for A2 connectivity).
Discussion
We investigated fMRI functional connectivity as a measure of the interhemispheric connectivity at the level of both primary and secondary auditory cortices using resting state and stimulus-related fMRI data. Our results indicate that patients with a history of AHs have significantly reduced interhemispheric connectivity in both their primary and secondary auditory cortex when compared with patients with no history of AH and healthy controls. The latter two groups did not show any difference in functional connectivity. Further, this pattern of findings was similar across two datasets, that is, the resting state data and the stimulus-related data, indicating the reliability of our estimates. These findings are particularly important in the face of: (a) two well-matched groups of patients with schizophrenia, the only clinical difference being a history of AH; (b) no difference in interhemispheric connectivity between non-AH patients and healthy controls. Thus, the deficits reported here are not symptomatic or state effects (as no AH patient hallucinated during scanning) but are trait effects specific to a lifetime history of AH.
Our previous study (McKay et al. Reference McKay, Headlam and Copolov2000) had provided evidence that AHs are associated with deficits that may be explained by either right auditory association area (A2) deficits or by interhemispheric deficits at the level of corpus callosum. Our current results do not eliminate the possibility of right auditory cortex deficits but indicate that there are interhemispheric deficits specific to AHs. Further, our results showed reduced interhemispheric connectivity in AH patients in both the primary and secondary auditory cortices, presumably reflecting structural disconnections within both auditory cortices. Since A1 and A2 are responsible for different functions, our results are indicative of extensive auditory function deficits associated with AHs. That is, alterations in the integration of the basic auditory information between hemispheres (via A1) and in higher-order auditory/language processing abilities (via A2).
No other study, to date, has distinguished between possible abnormalities in both A1 and A2 in the same set of AH patients. There are a number of studies that have shown that A1 is activated during AH (Dierks et al. Reference Dierks, Linden, Jandl, Formisano, Goebel, Lanfermann and Singer1999; van de Ven et al. Reference van de Ven, Formisano, Roder, Prvulovic, Bittner, Dietz, Hubl, Dierks, Federspiel, Esposito, Di Salle, Jansma, Goebel and Linden2005) and other studies that have implicated that reversal of language-related activity in AH patients occurs within A2 (Woodruff et al. Reference Woodruff, Wright, Bullmore, Brammer, Howard, Williams, Shapleske, Rossell, David, McGuire and Murray1997). Only studies like ours that investigate the interrelationship between auditory cortex structures at the level of A1 and A2 will allow for a detailed and adequate description of auditory cortex dysfunction in AH to be put forward. It is important to underline in this context that the observed reduction in interhemispheric connectivity at the level of A2 could in fact be driven by the impairment in connectivity observed at the level of A1 via the corticocortical projections from A1 to A2 (Pandya, Reference Pandya1995), or due to the spatial smoothing of the images. Since our current methodology was not able to investigate A1 to A2 projections using fMRI connectivity (these two regions are too close to avoid signal contamination via image registration and spatial smoothing), further investigations are necessary to distinguish between interhemispheric connectivity deficits at the level of both A1 and A2 and an A1 specific deficit that is projected onto A2.
Examining functional connectivity within schizophrenia research has previously resulted in the disconnectivity hypothesis (Friston & Frith, Reference Friston and Frith1995). This hypothesis suggests that schizophrenia is the result of abnormal integration of a distributed network of brain regions. A large number of functional connectivity studies have indicated multiple connectivity deficits specific to schizophrenia patients, including abnormal frontotemporal-parietal connections (Friston & Frith, Reference Friston and Frith1995; Lawrie et al. Reference Lawrie, Buechel, Whalley, Frith, Friston and Johnstone2002; Kim et al. Reference Kim, Kwon, Park, Youn, Kang, Kim, Lee and Lee2003; Tan et al. Reference Tan, Sust, Buckholtz, Mattay, Meyer-Lindenberg, Egan, Weinberger and Callicott2006), cortico-subcortical connections (Honey et al. Reference Honey, Pomarol-Clotet, Corlett, Honey, McKenna, Bullmore and Fletcher2005), anterior cingulate connectivity (Spence et al. Reference Spence, Liddle, Stefan, Hellewell, Sharma, Friston, Hirsch, Frith, Murray, Deakin and Grasby2000; Boksman et al. Reference Boksman, Theberge, Williamson, Drost, Malla, Densmore, Takhar, Pavlosky, Menon and Neufeld2005; Mechelli et al. Reference Mechelli, Allen, Amaro, Fu, Williams, Brammer, Johns and McGuire2007), as well as brain level connectivity, revealed by aberrant default mode networks (Garrity et al. Reference Garrity, Pearlson, McKiernan, Lloyd, Kiehl and Calhoun2007) and disrupted small world networks (Liu et al. Reference Liu, Liang, Zhou, He, Hao, Song, Yu, Liu, Liu and Jiang2008). Based on these studies it can be concluded that schizophrenia can be associated with large-scale changes in both inter- and intra-hemispheric connectivity patterns. We want to emphasize here, however, that the investigation of brain level connectivity differences across groups using automated labelling tools to identify the ROI could be confounded by the variability in size and spatial location of functional regions across patients and controls, as well as by imperfections in spatial normalization. In the current study we confined our investigation to connectivity at the level of auditory cortices and overcame inter-subject variability by carefully identifying the ROI for each subject based on anatomical landmarks and functional activations. This approach allowed us to compare functionally similar structures across subjects.
Previous resting state connectivity studies have shown that brain level connectivity maps obtained for a seed voxel placed in the auditory cortex in one hemisphere only show significant connectivity with the contra-lateral auditory cortex (Cordes et al. Reference Cordes, Haughton, Arfanakis, Wendt, Turski, Moritz, Quigley and Meyerand2000; Quigley et al. Reference Quigley, Cordes, Turski, Moritz, Haughton, Seth and Meyerand2003). The data in the current study showed remarkable similarity to these previously published studies. However, for brevity of presentation and in order to directly address our hypothesis, the present study focused exclusively on the investigation of interhemispheric connectivity at the level of the auditory cortices with particular emphasis on comparing AH with non-AH patients.
Another important observation is that our study compared only the resting state connectivity across the three groups. Resting state connectivity is believed to originate from correlated fluctuations in the firing rate of the interconnected neurons (Xiong et al. Reference Xiong, Parsons, Gao and Fox1999). Leopold et al. (Reference Leopold, Murayama and Logothetis2003) suggested that this type of correlated fluctuations in fMRI data could be associated with low frequency fluctuations in the local field potentials. We found that this type of connectivity was impaired in the AH participants. In principle, the interhemispheric connectivity at the level of the auditory cortex could be further altered by an active hallucinating state (Sritharan et al. Reference Sritharan, Line, Sergejew, Silberstein, Egan and Copolov2005). Further research is therefore needed to investigate whether the difference in auditory cortex connectivity reported here between AH and non-AH patients can be additionally modulated by an auditory task (using a longer task duration that was used here) or during the hallucinating state.
It is difficult based solely on functional connectivity measurements to decide on the mechanisms responsible for the AH specific deficits reported in our study. One possible explanation would be that there is a disruption of structural connectivity at the level of trans-callosal white-matter tracts connecting the auditory cortices in the two hemispheres (as suggested by McKay et al. Reference McKay, Headlam and Copolov2000). Recent studies have reported structural connectivity deficits in schizophrenia as well as deficits specific to AH patients. For example, Hubl et al. (Reference Hubl, Koenig, Strik, Federspiel, Kreis, Boesch, Maier, Schroth, Lovblad and Dierks2004) reported significantly reduced fractional anisotropy in schizophrenia patients when compared with healthy controls localized in both hemispheres at the level of arcuate fasciculus, uncinate fasciculus, inferior longitudinal fasciculus and some parts of the corpus callosum. The comparison of AH with non-AH patients revealed increased fractional anisotropy values specific to AH patients in the left hemispheric fibre tracts, including the arcuate fasciculus and the cingulate bundle. Shergill et al. (Reference Shergill, Kanaan, Chitnis, O'Daly, Jones, Frangou, Williams, Howard, Barker, Murray and McGuire2007) found schizophrenia-related reduced fractional anisotropy in the genu of the corpus callosum and in the superior longitudinal fasciculi for both hemispheres. Although these authors did not directly compare the AH and non-AH groups, they found that the fractional anisotropy values on superior longitudinal fasciculi and the anterior cingulum were correlated with the propensity to experience AHs. However, no study to date has measured the white-matter integrity for the trans-callosal fibres connecting A1 and A2 in AH and non-AH patients as compared with healthy controls. Such a study would be invaluable in helping to interpret the anatomical basis for the functional results of our current study.
In conclusion, the connectivity data in the current study suggest that there may be a disruption of the integration of multiple auditory functions (basic and higher order) in AH patients. We postulate that such a disruption may give rise to aberrant auditory events through poor communication between the left and right hemisphere; in turn, these aberrant auditory events are interpreted by the patient as external, and thus, a voice.
Appendix 1: Q statistics definition
Hedges & Olkin (Reference Hedges and Olkin1985) defined the homogeneity statistic test Q as:
![Q \equals \sum\limits_{i \equals \setnum{1}}^{N} {w_{i} \cdot ES_{i}^{\setnum{2}} } \minus {{\left[ {\sum\nolimits_{i \equals \setnum{1}}^{N} {w_{i} \cdot ES_{i} } } \right]^{\setnum{2}} } \over {\sum\nolimits_{i \equals \setnum{1}}^{N} {w_{i} } }}\comma \hfill\vskip14](https://static.cambridge.org/binary/version/id/urn:cambridge.org:id:binary:20160927005040275-0596:S0033291709991632:S0033291709991632_eqn1.gif?pub-status=live){kind=small}
where N is the number of subjects in the sample, ES is the effect size, n is the number of independent observations per subject, and w=n−3. The Q statistic for the group is distributed as χ2 with N – 1 degree of freedom.
Q statistics can be used to compare effects sizes across groups. To calculate the between groups Q statistic, three Q statistics need to be calculated: Q 1 for group 1; Q 2 for group 2; Q total across groups 1 and 2 considered together (Q T). Then the Q statistic within groups is calculated as:
Q within is χ2 distributed with N – 2 degrees of freedom. The between groups Q statistic is defined as:
Q between is χ2 distributed with 1 degree of freedom. The probability value associated with Q between gives the significance of the differences between groups.
For our study the effect size for each subject was estimated as the Z score calculated with Fisher's Z transformation for correlations:
The number of independent observations per subject (n) was estimated using a vector autoregressive model as described in Gavrilescu et al. (Reference Gavrilescu, Stuart, Rossell, Henshall, McKay, Sergejew, Copolov and Egan2008).
Acknowledgements
This work was supported by grant NHMRC 236025 and funding from The Garnett Passe and Rodney Williams Memorial Foundation. G.E. acknowledges support from NHMRC fellowship 400317. S.R. and M.G. acknowledge support from MHRI core funding; and H.I-B. from Cognitive Neurobiology of Psychosis Platform at Neurosciences Victoria. We thank Ms Erica Neill for conducting the participant interviews and clinical testing.
Declaration of interest
None.
