Pathways That Make Voices:  White Matter Changes in Auditory Hallucinations FREE

Auditory hallucinations (AH), one of the most common psychiatric symptoms,elude a compelling explanation. They have been discussed in nearly every conceivablecontext, ranging from a very private experience to abnormal brain functionin the frame of schizophrenia. In 1838, Esquirol¹ wasthe first to formulate the concept of a brain-based origin of hallucinations.Although AH occur with a lifetime prevalence of 10% to 15% in persons withoutneuropsychiatric diseases,² they are most commonin schizophrenia, with an average prevalence of 60%.³ Therefore,recent models of AH were generally based on results gained from investigationsof patients with schizophrenia.^4- 6 Neuropathologic,^7- 8 structural magnetic resonance imaging(MRI)^9- 12 andfunctional MRI^13- 19 studiessuggest that the superior temporal lobe is altered in patients with AH, creatingdysfunctions within brain regions that are important for language and auditoryprocessing.

The present hypotheses of the generation of AH, considering recent neuroimagingfindings, propose that AH derive from inner speech misidentified as externalby means of defective self-monitoring.¹⁵ Functionalimaging studies in schizophrenic AH revealed involvement of the frontal motorand premotor speech areas (the Broca area and supplementary motor area) andtemporoparietal speech areas (the Wernicke area) that are necessary for thedecoding and encoding of language.¹⁵ Additionalregions described as involved in AH were the primary¹³ andhigher-order auditory and association cortex located in the temporal lobe,mainly in the left hemisphere.^13- 18 Inright-handed individuals, the speech-relevant areas are predominantly locatedin the left hemisphere,²⁰ which may be relatedto the fact that the left hemisphere also appears to be more functionallyinvolved in the generation of AH than the right hemisphere.^4- 5 Fornormal speech functions, intact connections between speech-relevant regionsare necessary,²⁰ but recent functional MRI²¹ and electroencephalography²² studiesapplying language tasks suggest dysfunctional interactions between frontal,parietal, and temporal brain regions in patients with AH. This finding supportsthe more general theories that schizophrenia involves disturbed frontoparietotemporalconnectivity.^23- 25 Themicrostructural foundation of these connections was not yet investigated becauseof a lack of methodical feasibility. Magnetic resonance diffusion tensor imaging(DTI) assesses the directionality of water diffusion (anisotropy), which isrestricted by boundaries such as white matter (WM) fibers. The amount of anisotropycorrelates with the directionality and coherence of fiber tracts.²⁶ Thus, a loss of WM directionality or disruption ofthe microstructure is reflected in reduced anisotropy values.²⁷ Ofthe relatively few studies published using DTI in schizophrenia, the majorityreported reduced anisotropy in subjects with schizophrenia,^28- 34 whereasthe minority did not find differences between patients and control subjects.^35- 37 However, these studiesvary widely in terms of MRI methods and analysis strategies and are not directlycomparable among each other. The more recent studies^32,34 focusedon the WM fiber tracts connecting the frontal with the temporal and parietalcortex. Both studies demonstrated pathologic features in schizophrenia; alack of normal left-greater-than-right asymmetry in the uncinate fasciculusin patients with schizophrenia³² respectivelydecreased WM integrity in the left hemispheric arcuate and uncinate fasciclulus.³⁴ Those studies provide in vivo support of the aforementionedneuroanatomical and neurofunctional reports of disruption of the frontal andtemporal brain regions.

In an earlier functional MRI study, we demonstrated an increase of neuronalactivity in the primary auditory cortex and language-related areas duringhallucinations in patients with schizophrenia;¹³ however,the relation to structural cerebral alterations remained unclear. In thisstudy, we investigate whether altered neuronal activity during AH may be mediatedby altered WM connections in patients with schizophrenia with a history offrequent AH in comparison with patients with schizophrenia who reported neverhaving perceived AH and healthy control subjects. We expected the most prominenthallucination-related differences in the arcuate fasciculus. This frontoparietotemporalfiber tract connects important language-related areas³⁸ thatwere reported to be most affected in schizophrenia in studies of structuraland functional imaging.

We investigated 13 patients with acute schizophrenia (International Classification of Diseases, 10th Revision diagnosis criteria)³⁹ with frequent AH (group H), 13 patients with acuteschizophrenia who reported that they had never perceived AH (group N), and13 healthy control subjects (group C). All patients were patients of the UniversityHospital for Clinical Psychiatry in Bern, Switzerland. The groups were matchedfor age and sex, and all subjects were right-handed (Table 1). None of the patients reported substance abuse before hospitalizationexcept sporadic cannabis consumption by 4 patients (3 in group N; 1 in groupH). Only patients and subjects without relevant medical disorders (exceptschizophrenia) were included based on medical history and medical and neurologicalexamination. All patients but 2 received typical or atypical antipsychotictreatment in conventional dosages, in both patient groups. The investigationwas conducted in accordance with the Declaration of Helsinki and approvedby the local ethics committee. Before the investigation began, all patientsand healthy control subjects gave their written informed consent to participatein the study.

To assign each individual patient to 1 subtype of symptom (trait), anextensive semistructured interview concerning medical history, with a specialfocus on the perception of AH, was assessed before scanning. Furthermore,the files of every patient were studied to add missing data. The group withouthallucinations included only patients who had never experienced AH, neitherat the time of this investigation nor at any time in their previous history.The patients prone to hallucinations experienced verbal AH at the time ofthe present hospitalization as well as during all the prior exacerbationsof their disease.

The Oulis et al⁴⁰ AH rating scale wasused to document the character of AH. The hallucinations fulfilled the criteriaof the Schneiderian first-rank symptoms of comment or dialog voices in allpatients with AH. Eight of 13 patients reported that the voices were comingfrom outside their heads, whereas 3 of 13 experienced voices coming from insideand outside the head. Eight of 13 patients perceived the voices to be as loudas real voices, and 3 of 13 perceived them to be quieter or like a whisperingvoice. All patients were convinced that the voices were real. Two patientsrefused the interview for the hallucination assessment.

The Positive and Negative Syndrome Scale⁴¹ andClinical Global Impressions Scale⁴² were usedto assess psychopathologic symptoms and the severity and acuity of the disease(state). The Positive and Negative Syndrome Scale did not differ significantlybetween the patient groups for the total score and the negative subscore.The positive subscore was significantly higher in the group with AH (P≤.05). Higher values were because of higher scoreson the hallucination and delusion subscore (Table 2).

A 1.5 T Signa MR system (version 5.8, equipped with echospeed gradientsof 22 mT/m; General Electric Medical Systems, Milwaukee, Wis) was used forthe investigations. In 1 session, 3-dimensional structural images and diffusiontensors were assessed. Individual whole brain 3-dimensional anatomy was measuredwith a 3-dimensional gradient echo sequence, providing 124 axial slices with1.2-mm thickness, 240 mm × 240- mm field of view, and 256 × 128-pixelresolution. Further scanning parameters were as follows: repetition time,22 milliseconds; echo time, 8 milliseconds; and flip angle, 45°. The originalvoxel size of 0.94 × 1.88 × 1.20 mm³ was interpolatedto a voxel size of 1 × 1 × 1 mm³ with BrainVoyager2000software (BrainInnovation, Maastricht, the Netherlands). Total 3-dimensionalscan time was 9.04 minutes.

Diffusion tensor imaging was realized with a line-scan technique⁴³ that provides a line-by-line spin-echo sampling ofeach slice. In contrast to the more widespread echo planar (ie, gradient echo)method, this technique is less sensitive to susceptibility-related distortionsand, as with single-shot echo planar imaging, it is fairly motion insensitive,which is favorable when imaging is done on patients with schizophrenia. Eddycurrent–related distortions are also very small, in particular afteran eddy-current correction of the DTIs, which is based on cross-correlationwith T2-weighted images. In the line-scan technique, patient motion does notlead to ringing artifacts in phase direction but rather to discontinuitiesthat can be detected and eliminated in postprocessing. This leads to robustnesswith respect to patient motion. The reliability of the apparent diffusioncoefficient determined by line scanning was found to be very high in healthysubjects and patients who had had strokes.⁴⁴ Similarly,the fractional anisotropy (FA)⁴⁵ values aremore reproducible with the line-scan technique than with echo planar imagingsequences (S.E.M., unpublished data, November 2003). Twelve axial slices (parallelto the anatomical data set) with 5-mm slice thickness and 1-mm interslicedistance were measured. This volume covered a region around the Sylvian fissure,including the inferior frontal gyrus up to the angular gyrus (Figure 1). For each slice, 6 images with high diffusion weighting(β = 1000 s/mm²) in 6 different directions and 2 images withlow diffusion weighting (β = 5 s/mm²) were collected. Thefield of view was 220 mm × 165 mm; scan matrix, 128 × 96 pixels,interpolated to a final image size of 256 × 256 pixels; repetition time,3360 milliseconds; echo time, 90 milliseconds; and interpolated DTI voxelsize, 0.86 × 0.64 × 5.00 mm³. Scanning time for thecomplete diffusion tensor sequences was 16.34 minutes. Total magnetic resonancescanning time including the localizer scan was about 30 minutes.

After reconstruction of the DTIs, eigenvalues and eigenvectors of thediffusion tensor were determined. Fractional anisotropy values were calculatedfor each image. The 2-dimensional diffusion maps were incorporated into the3-dimensional anatomy data sets through interpolation to the same resolution(voxel size, 1 mm³). This allowed us to superimpose 3-dimensionalstatistical maps onto the 3-dimensional anatomical data sets for visualizationof statistical differences. The 2-dimensional diffusion images and 3-dimensionalstructural measurement were aligned using position files of the magnetic resonancescanner. For confirmation of the coregistration accuracy, a structural 2-dimensionaldata set was obtained by reslicing the 3-dimensional data and compared visuallywith the DTI data in each individual. For each subject, the structural datasets were transformed into Talairach space, following the transformation procedurepublished elsewhere.¹³ On the basis of theparameters for importing the 2-dimensional DTI slices into the 3-dimensionaldata and on the parameters for transforming the 3-dimensional data sets intoTalairach space, the complete DTI data set of each subject was also transformedinto Talairach space. The study focused on alterations in WM fiber tracts.Therefore, after the individual anatomical data sets at the gray matter–WMboundary were segmented,⁴⁶ the analysis wasrestricted to voxels identified as WM. Because the segmentation in the regionof the basal ganglia was inconsistent across subjects, this region was excludedfrom analysis. To avoid spurious effects, group statistics were computed onlywhere there was a significant (P≤ .10) preponderanceof WM over gray matter as assessed by χ² statistics acrosssubjects (Figure 2).

In a first step, a voxel-based 1-way analysis of variance (ANOVA) restrictedto WM volume was performed with the factor "group" (C, H, and N) as a singlefactor. To avoid potential registration errors and disease-related misalignmentsin Talairach space, a region of interest (ROI) analysis based on the ANOVAwas computed. The WM tracts identified as significantly different betweengroups by the ANOVA were selected in each study subject by using the individualcoregistered T1 and DTIs. The mean FA values of each ROI were used for a 3-wayANOVA (factors were "group" [C, H, and N], "hemisphere" [left, right], and"ROI" [lateral and medial arcuate fasciculus]). Post hoc unpaired t tests were computed between the groups where appropriate.

Subsequently, for further descriptive statistics, 3-dimensional statisticalmaps (using a voxelwise, 2-tailed, unpaired t test)were computed, comparing the FA values between healthy control subjects andpatients, between the 2 patient groups (H vs N), and between each patientgroup and control subjects (H vs C and N vs C). To identify the most involvedregions and to reduce the problem of multiple testing, clusters containingmore than 99 neighboring voxels (100 mm³) with differences of P≤.05 were identified. For each cluster, FA values wereaveraged and tabulated (Table 3 and Table 4). Further, the Talairach⁴⁷ coordinates of the center of gravity were noted.Clusters were assigned to the underlying WM fiber tracts using 3-dimensionalanatomical data. Data analysis and visualization were realized with BrainVoyagerand in-house software.

In the voxel-based 1-way ANOVA, several significant clusters in longassociation and commissural fiber tracts were found. As we focused on speech-and language-related brain regions, the subsequent confirmatory ROI analysis(3-way ANOVA) included values for the medial and lateral arcuate fasciculusof both hemispheres (Figure 3A).The ANOVA demonstrated a significant main effect of hemisphere and a significantinteraction of group × hemisphere (Table 5). Post hoc t tests (P≤.05) showed that (1) FA values were higher in the right than inthe left hemisphere, (2) patients with AH had higher FA values in the leftlateral arcuate fascicle than did both patients without AH and healthy controlsubjects, and (3) control subjects had higher FA values in the medial arcuatefascicle than did patients with AH and patients without AH (Table 5) (Figure 3B).

In the voxelwise t tests, FA values were lowerin patients with schizophrenia in many parts of the WM. The identificationof voxel clusters and the corresponding WM tracts yielded 17 significant regions(Table 3), 8 clusters in the leftand 9 in the right hemisphere. Significantly higher FA values in the voxel-basedanalysis for control subjects were observed in the long anterior-to-posteriorassociation fibers in large parts of the arcuate fasciculus (superior longitudinalfasciculus), the uncinate fasciculus, and the inferior longitudinal fasciculusin both hemispheres (Figure 4).Further, higher FA values were observed in control subjects in parts of thecorpus callosum (CC) (Figure 4).There were no clusters with higher values in patients with schizophrenia.

Patients with AH showed voxels with significant higher FA values inmany parts of the WM (Figure 5A). A total of 12 clusters were located in the left hemisphereand 5 in the right. There were 15 clusters with higher FA values in patientswith AH compared with patients without AH and only 2 clusters with higherFA values in patients without AH compared with patients with AH. The mostimportant differences for the 2 patient groups could be observed in the arcuatefasciculus (10 of 17 clusters), with a slight dominance of the left hemisphere(6 of 10 clusters). The significantly higher FA clusters in patients withAH were distributed from prefrontal WM passing the parietal WM down into thetemporal lobe part of the arcuate fasciculus. The largest cluster (402 voxels)was located in the left temporoparietal section of the arcuate fasciculus(Figure 5A). Furthermore in theleft cingulate bundle, part of the limbic system, patients with AH had higherFA values compared with patients without AH. Finally, there were higher FAvalues in the anterior and posterior parts of the interhemispherical commissuralfibers of the CC (Figure 5 A). Theclusters with higher FA values in patients without AH were located in theleft inferior longitudinal fasciculus.

There were 16 significant clusters in the comparison of patients withschizophrenia with AH vs healthy control subjects. Thirteen demonstrated lowervalues in patients with schizophrenia with AH whereas 3 showed higher valuesin patients with schizophrenia with AH (Table 4B). These regions in which patients with AH had significantlyhigher FA values compared with control subjects were located in the lateralleft and right temporoparietal section of the arcuate fasciculus close tothe posterior end of the Sylvian fissure—where the Wernicke area islocated—and close to the auditory regions (Figure 5B). Patients with AH showed generally higher FA values inthe lateral part of the arcuate fasciculus whereas control subjects showedgenerally higher FA values in the medial parts of the arcuate fasciculus.Further, patients with AH did have higher FA values in the left anterior CCcompared with control subjects (Figure 5B).

In 13 clusters (6 in the left, 7 in the right hemisphere), lower FAvalues were observed in patients with AH in comparison with control subjects.These clusters were distributed in the prefrontal to the temporal arcuatefasciculus (medial), in the uncinate fasciculus, and in the inferior longitudinalfasciculus. Further clusters were located in the more ventral part of theanterior CC (Figure 5B).

In all 21 clusters with significant differences, patients with schizophreniawithout AH had lower FA values compared with control subjects (Table 3). Eleven of the 21 clusters were located in the left hemisphere.Significant clusters were located in the frontotemporal arcuate fasciculus(Figure 5C), the uncinate fasciculus,and the inferior longitudinal fasciculus as well as in the anterior and medialsections of the CC (Figure 5C).

In 1919, Kraepelin⁴⁸ had already postulatedthat AH were a result of temporal lobe abnormalities. This hypothesis wassupported by severe abnormalities in the left temporal lobe in the brainsof patients with schizophrenia, found post mortem or in structural imagingstudies⁴⁹; these abnormalities were thoughtto be related to AH. In 1900, Wernicke⁵⁰ hypothesizedthat a pathologic activation of the primary acoustic cortex was the basisof the experience of external sensory stimulation during AH. In fact, a dysfunctionof temporal cortical areas, in particular of the primary auditory cortex,and of frontal speech areas was reported in AH, suggesting an associationwith impaired auditory and language-processing networks.¹⁹ However,it remained unclear to what degree WM alterations in fiber tracts constitutingparts of these functional networks were involved in the pathogenesis of AH.

In the present study, we investigated the integrity of WM fiber tractsin patients with schizophrenia with frequent AH, patients who had never experiencedhallucinations, and healthy control subjects. Using a line-scan DTI sequence,⁴³ we obtained FA data in a 3-dimensional volume coveringthe Sylvian fissure. The arcuate fasciculus contains—among others—fibersconnecting the frontal, parietal, and temporal language and auditory areas.Compared with healthy control subjects, patients with a history of AH demonstrated,especially in the left hemisphere, an imbalance in directionality of the arcuatefasciculus, with higher directionality in the lateral part of the arcuatefasciculus and decreased directionality of WM fibers in the medial part ofthe arcuate fasciculus. This pattern could not be demonstrated for patientswithout AH. The arcuate fasciculus is divided into (1) a medial part thatcontains longer fibers connecting the lateral frontal cortex with the dorsolateralparietal and temporal cortex and (2) a lateral part, with shorter U-shapedfibers connecting the frontoparietal, parietooccipital, and parietotemporalcortex⁵¹; fibers originate in the prefrontaland premotor gyri (mainly the Broca area) and project among others posteriorto the Wernicke area. Thus, the lateral part of the arcuate fasciculus providesa pathway by which frontal speech-production areas can influence auditoryand speech perception areas during overt and inner speech. The importanceof the arcuate fasciculus in language is underlined by results from neurologicalfindings in aphasia research. A disruption of the arcuate fasciculus leadsto a disturbance of the neuronal connections from the frontal Broca area tothe temporal Wernicke area, which results in a disturbance of the stream ofspeech.⁵² One link between AH and inner speechis the common clinical observation that the content of AH is often closelyrelated to the content of the patient's own thought and sometimes is evenreported as thoughts becoming loud. The exact neurobiological functional correlateof alterations of FA remains unclear; however, most in vivo human and animalstudies investigating FA values during neuronal development and functionaldisturbances in neuropsychiatric diseases suggest that an increase of FA valuesis related to an increase of connectivity in WM bundles.⁵³ Therefore,it can be hypothesized that high WM directionality in the lateral part ofthe arcuate fasciculus in AH is associated with high connectivity betweendistributed language and auditory areas. This may facilitate the dysfunctionalcoactivation of the primary auditory cortex and language-related areas thathas been previously described during AH.¹³ Inthe medial part of the arcuate fasciculus, both patient groups showed smallerdirectionality than did control subjects, which is in accordance with a recentreport observing reduced left hemispherical FA values in the arcuate fasciclein schizophrenia.³⁴ This might provide a structuralalteration associated with disrupted frontotemporal processing.^23- 25 Themore pronounced alterations in the left hemisphere, the same hemisphere reportedlymore affected in schizophrenia, may be related to language dominance.⁵⁴ The fiber tracts found to be affected here coincidewith those that show the strongest developmental changes during childhoodand adolescence.⁵⁵ This allows speculationabout a developmental origin of the aforementioned alterations of the arcuatefasciculus in patients with AH and suggests a relation between AH and WM aberrance.

The CC carries most of the commissural fibers in the cerebrum, interconnectingleft with the correspondent right hemispheric regions,³⁸ mediatinginterhemispheric communication. It was argued⁵⁶ thatthe deficit in information processing in schizophrenia might be related toalterations in the CC. Many studies on the size of the CC have shown smallervolumes in schizophrenia; however, these findings are not unequivocal.^49,57 Even if no concluding macrostructuralresults were obvious, behavioral experiments gave evidence for a specificexcessive callosal transfer in schizophrenia.⁵⁸ However,a recent study searching for a specific relation between AH and the CC failedto show clear volumetric differences.⁵⁹ Inthe present study, we report higher directionality in the anterior part ofthe CC, including fibers that connect the left with the right frontal speech-relatedareas. This gives—on a microstructural level—support to the earlierassumption of higher callosal transfer in patients with schizophrenia withAH.

In most parts of the CC, however, patients demonstrated smaller FA valuescompared with control subjects. Results on FA values in the CC of previousDTI studies are not fully consistent, even though most showed reduced FA valuesin schizophrenia.^30- 31 However,1 group failed to replicate their own findings in a second study.³⁶

The uncinate fasciculus connects the temporal pole with the orbitofrontalcortex. It includes projection fibers from the somatosensory cortex and auditorycortex.³⁸ In the present study, the directionalityin the uncinate fasciculus was significantly smaller in both groups with schizophreniacompared with healthy subjects, and there was no difference between patientswith and without hallucinations. Thus, compared with the arcuate fasciculus,the uncinate fasciculus appears less important for the generation of AH. Ingeneral, our results agree with previously found reduced FA values in theleft uncinate fascicle, although we found reduced values in both hemispheres.³⁴ Although we found reduced anisotropy in the uncinatefasciculus in schizophrenia, we could not confirm the previously reportedrelated loss of normal left-higher-than-right FA asymmetry,³² aphysiological asymmetry that might indirectly relate to specialized, lateralizedfunctions such as language to AH. Kubicki et al³² raisedthe issue of whether their findings were restricted to the uncinate fascicleor a general phenomenon in schizophrenia. Our results show various fasciculito be affected in schizophrenia and a differential involvement of the WM tractsdepending on psychopathologic features (in this study, AH).

The third important frontotemporal fiber tract besides the arcuate anduncinate fasciculus is the cingulate bundle. It is, in contrast to the arcuateand uncinate fasciculus, part of the limbic system. The cingulated gyrus isdiscussed as an "interface between emotion and cognition,"^60(p107) and impairment is found in several psychiatric disorders suchas schizophrenia,⁶¹ obsessive-compulsive disorder,⁶² and major depression.⁶³ Inthe cingulate bundle, we demonstrated a reduced anisotropy in patients withouthallucinations compared with patients who had hallucinations, but restrictedto the left hemisphere. This difference was not obvious in the comparisonof either patient group with control subjects. This is in accordance withthe earlier studies that also did not find differences of FA values in thecingulate bundle when comparing patients with schizophrenia without respectto psychopathologic features.³⁴ From functionalimaging studies, there is evidence of the involvement of the anterior^64- 66 and of the left-sidedposterior regions⁶⁷ of the cingulate gyrusin the generation of AH. Our finding of higher FA values in AH is locatedin the left medioposterior part of the cingulate bundle and may be relatedon a microstructural level to reports of pathologic, high activation shownin imaging studies.

In conclusion, those WM fibers that we found to be most altered in patientswho had AH constitute the most important connections between language-relatedfrontal and temporal regions. These alterations may have a developmental originand may contribute to an understanding of how internally generated languageis perceived to be generated externally. The aberrant connections may leadto abnormal activation in regions that normally process external acousticaland language stimuli. That accounts for these patients' inability to distinguishself-generated thoughts from external stimulation.

Correspondence: Daniela Hubl, MD, Department of Psychiatric Neurophysiology,University Hospital of Clinical Psychiatry, Bolligenstrasse 111, CH-3000 Bern60, Switzerland (hubl@puk.unibe.ch).

Submitted for publication September 29, 2002; final revision receivedJune 4, 2003; accepted February 6, 2004.

This study was supported by grants 3200-059077.99 (Dr Dierks) and 31-59082.99(Dr Kreis) from the Swiss National Science Foundation, Bern.

We thank Karin Zwygart for performing the magnetic resonance imagingmeasurements.