White matter abnormalities in bipolar disorder and schizophrenia detected using diffusion tensor magnetic resonance imaging

Abstract

Objectives:  Strong qualitative and quantitative evidence exists of white matter abnormalities in both schizophrenia and bipolar disorder (BD). Diffusion tensor imaging (DTI) studies suggest altered connectivity in both disorders. We aim to address the diagnostic specificity of white matter abnormalities in these disorders.

Methods:  DTI was used to assess white matter integrity in clinically stable patients with familial BD (n = 42) and familial schizophrenia (n = 28), and in controls (n = 38). Differences in fractional anisotropy (FA) were measured using voxel-based morphometry and automated region of interest analysis.

Results:  Reduced FA was found in the anterior limb of the internal capsule (ALIC), anterior thalamic radiation (ATR), and in the region of the uncinate fasciculus in patients with BD and those with schizophrenia compared with controls. A direct comparison between patient groups found no significant differences in these regions. None of the findings were associated with psychotropic medication.

Conclusions:  Reduced integrity of the ALIC, uncinate fasciculus, and ATR regions is common to both schizophrenia and BD. These results imply an overlap in white matter pathology, possibly relating to risk factors common to both disorders.

Bipolar disorder (BD) is strongly associated with white matter abnormalities. Postmortem studies have revealed a reduction in glial cells in the subgenual prefrontal cortex (1) and there is evidence from T₂-weighted magnetic resonance imaging (MRI) that patients with BD have an increase in white matter hyperintensities compared to healthy controls (2, 3). T₁-weighted MRI has also revealed white matter reductions using both region of interest (ROI) (4, 5) and voxel-based morphometry (VBM) approaches (6).

Conventional T₁-weighted MRI is, however, limited in its ability to identify abnormalities of white matter, since the signal is largely dependent upon water content and is potentially confounded by non-neuronal tissue components. Diffusion tensor imaging (DTI) is a more recent technique that measures the diffusivity of water molecules within tissues in vivo. Yet there have been few DTI studies in BD and all of these have had relatively small sample sizes. Their findings have included reduced prefrontal (4, 5, 7) and posterior internal capsule (7) white matter integrity, and altered integrity in orbital frontal white matter (8). Despite this paucity of BD-related research, the available evidence suggests altered prefrontal-subcortical connectivity, in line with a growing body of functional imaging literature implicating dysfunction in the frontostriatal and frontothalamic circuits in the development of BD. These findings are supported by an expanding body of work including abnormalities of myelin-associated gene expression (9) and studies using conventional T₁-weighted MRI (10, 11).

Schizophrenia and BD show some overlap in relation to shared genetic aetiology (12), clinical symptoms, pharmacological interventions, and course (13). The most common positive DTI findings in schizophrenia have included decreased fractional anisotropy (FA) in temporal and prefrontal regions (14–16) and their connections: the uncinate fasciculus (17, 18), internal capsule (19), and genu of the corpus callosum (20). Indirect comparison of the literature in these disorders using T₁-weighted MRI and DTI therefore suggests that deficits in the internal capsule may be common to both disorders, whereas uncinate deficits have been found only in schizophrenia. However, conclusions about whether or not these represent common or unique white matter abnormalities are hampered by the lack of any direct DTI comparison, where both disorders have been examined in the same study. Our study sought to address this issue.

Despite its increasing use in clinical populations, there remains a general lack of agreement about how DTI data are best analysed. We and others have previously used an adaptation of VBM techniques to compare FA at specific points on prespecified tracts in schizophrenia (18). We therefore wished to apply the same methodology in this study to provide a direct comparison with our previous findings. The VBM approach is, however, potentially biased by registration differences between groups. We therefore conducted a complimentary automated ROI study of FA between the three groups in which the specific ROIs were hand traced, both to provide an additional test of internal replication and to deal with the potential issue of misregistration.

Patients and Methods

Participants

Individuals with bipolar I disorder or schizophrenia were identified from the case loads of consultant psychiatrists across Edinburgh and the Lothian region of Scotland. Case note diagnoses of BD and schizophrenia were first established using the OPCRIT symptom checklist (21) and confirmed using the Structured Clinical Interview for DSM-IV (SCID) by a single psychiatrist (AMM). All recruited patients had at least one first- or second-degree relative with the same diagnosis in order to reduce heterogeneity and enrich the sample in terms of genetically determined effects. The diagnosis of affected family members was confirmed wherever possible using the SCID and OPCRIT (21) symptom checklist or the Family Interview for Genetic Studies (22). At the time of scanning, all patients were outpatients and clinically stable as subjectively defined by their treating physician. Unaffected controls were identified from the same geographical regions and communities as the patients themselves and did not have any first- or second-degree relatives with schizophrenia or BD. The diagnostic status of controls was also confirmed using the SCID. A pilot study with five candidates in each group demonstrated a requirement for a minimum of 28 participants per group to have 80% power to detect between-group differences at p < 0.05 in the uncinate and control-patient differences in the anterior internal capsule.

Clinical symptoms were assessed using the Young Mania Rating Scale (YMRS) (23), Hamilton Depression Rating Scale (HDRS) (24) and the Positive and Negative Symptoms Scale (PANSS) (25). The premorbid full-scale IQ of all subjects was estimated using the National Adult Reading Test (26). All participants provided written informed consent. Ethical approval was granted by the Lothian Research Ethics Committee.

MRI protocol

All MRI data were collected using a GE Signa LX 1.5 T (General Electric, Milwaukee, WI, USA) clinical scanner equipped with a self-shielding gradient set (22 mT/m maximum gradient strength) and manufacturer-supplied ‘birdcage’ quadrature head coil. The MRI examination consisted of standard fast spin-echo (FSE) T₂-weighted sequence, fast spoiled gradient-echo (FSPGR) T₁-weighted volume sequence, and a whole brain DTI protocol. In the latter, a single-shot pulsed gradient spin-echo echo-planar imaging (EPI) sequence was used to acquire sets of axial images (b = 0 and 1000 s/mm²) with diffusion gradients applied sequentially along 51 noncollinear directions arranged uniformly in space (27). A total of 48 contiguous axial slice locations were imaged and, in addition to the 51 diffusion-weighted EPI volumes, three T₂-weighted baseline EPI volumes were also acquired [echo time = 93.4 ms, repetition time = 17 sec, field of view = 220, slice thickness = 2.8 mm, and an imaging matrix of 96 × 96 (zero filled to 128 × 128)].

Image processing

All DICOM-format magnitude images were transferred from the scanner to a Sun Blade 2000 workstation (Sun Microsystems, Mountain View, CA, USA) and converted into Analyze (Mayo Foundation, Rochester, MN, USA) format. Using the FLIRT toolbox (http://www.fmrib.ox.ac.uk/fsl), a three-dimensional computational image alignment program (28), bulk patient motion and eddy current-induced artefacts were removed from the DTI data by registering the diffusion-weighted to the T₂-weighted EPI volumes. The apparent diffusion tensor of water (D) was calculated in each voxel from the signal intensities in the component EP images (29). Maps of FA for every subject were generated on a voxel-by-voxel basis from the sorted eigenvalues of D and converted into Analyze format, resulting in a series of skull-stripped FA and T₂-weighted volumes for further analysis.

VBM was performed using SPM2 (http://www.fil.ion.ucl.ac.uk/spm/). A study-specific template was constructed to minimise the amount of warping required in bringing the images into a common analysis space. This was achieved by coregistering the subject T₂-weighted images with the Montreal Neurological Institute (MNI) skull-stripped T₂-weighted template using only linear transformations to account for large differences in brain size and position [i.e., 9-point affine (3 translations, 3 rotations, and 3 shears)]. All subject images were then normalised to this target image using a reduced number of nonlinear warps, described by a set of 4 × 5 × 4 basis functions, to avoid distortion of our images. These were then averaged and the study-specific template was formed by smoothing the mean image at 8 mm full-width at half maximum (FWHM).

The T₂-weighted images from each subject were then normalised to the study-specific template using the standard SPM default parameters (including 7 × 9 × 7 basis functions to describe the nonlinear components). Normalisation of the FA maps was achieved by applying the same transformation parameters (30) . Both normalised T₂-weighted images and FA maps were smoothed at 12 mm FWHM to make our statistical tests more sensitive to structures of approximately the same extent, to ensure normality of the data, and to compensate for any inaccuracy in the spatial normalization (31). Model residuals were subsequently checked to ensure that they approximated to a normal distribution.

Statistical parametric mapping analysis

FA was compared among the three groups using the General Linear Model (ANOVA) within the SPM2 package. As the groups were matched for both age and sex, nuisance covariates were not required in the analyses. Between-group differences in FA were then estimated using pairwise t-contrasts. Statistical parametric maps were thresholded with an uncorrected significance level of p = 0.001. To further investigate group differences in the hypothesised brain regions, small volume corrections (SVC) were applied to the prefrontal lobe and anterior limb of the internal capsule (ALIC). The prefrontal lobe SVC included the frontal portion of the uncinate fasciculus (UF). SVCs were defined by hand tracing the brain regions on the study-specific T1 template. Differences between groups were considered to be significant when the family-wise error corrected p-value was less than 0.05 or less than p = 0.01 (uncorrected) for the ALIC, a structure of less than one resel.

As described, the images were normalised to the study-specific template. In order to report these results in MNI space, it was necessary to warp the coordinates of the significant peak voxels (p-corrected), as identified from the study-specific template in ‘scanner’ space, into MNI space. The required transformation matrix was formed by performing a 12-point affine normalisation (comprising the nine linear transformations as detailed above and three nonlinear transformations) of the skull-stripped MNI to the study-specific T₂-weighted template. The MNI coordinates were then calculated by multiplying this matrix with the coordinates of interest.

Automated ROI analysis

Automated ROI analysis was also performed in SPM2. The right- and left-hand side of the ALIC and UF were traced separately onto the study-specific template to create masks from an MRI atlas of human white matter (32) . The boundaries of the ALIC were determined dorsally by bisecting the angle of the internal capsule on axial slices, medially by the caudate and laterally by the lentiform nucleus. A line connecting the lateral edge of the ventral caudate and the medial edge of the ventral lentiform nucleus determined the ventral boundary (33). To acquire subject-specific ALIC ROIs, the ‘automated ROI’ method described by Pagani et al. (34) was adopted. In brief, the normalised, T₂-weighted image (in template ‘space’) was warped back onto the un-normalised image (in the subject’s own ‘space’). The normalisation parameters from this transformation were then applied to the masks (created in template space) to produce a series of subject-specific ROIs. Using the SPM Image Calc function, the ROIs were multiplied with the un-normalised FA maps and a mean FA was acquired for each ROI for each subject. All ROIs were checked against the subject’s FA map to ensure correct coverage and registration.

The mean data from the left and right ALIC were imported into SPSS and, to replicate the contrasts investigated in SPM, independent t-tests were performed to identify any significant differences between groups.

Symptoms and medication effects

In order to examine the relationship of FA to psychiatric symptoms or medication, analyses of their associations were performed. For each significant finding reported in the between-groups analysis, the FA density at the peak voxel was extracted for all subjects and imported into SPSS. Multivariate backward-entry regression was then used to investigate the relationship between each medication class or symptom dimension (PANSS positive, PANSS negative, HDRS, YMRS), and FA values at each peak voxel result were reported. Analyses were conducted separately for each group. In addition, the effect of a lifetime presence or absence of psychotic symptoms (rated from the SCID) on FA in bipolar patients was investigated using an independent samples t-test.

Results

Participants

A total of 108 participants completed a DTI scan of the brain. All groups were closely matched for both age and sex, and patients with BD had a similar mean age of onset to patients with schizophrenia (Table 1). Premorbid IQ differed between the groups, although the only statistically significant pairwise difference was between subjects with schizophrenia and controls. Patients with schizophrenia had higher positive symptom ratings (from the PANSS) than patients with BD. Depression and mania rating scores did not differ significantly between the clinical groups.

Table 1.   Characteristics of all participants undergoing diffusion tensor imaging scan

	Bipolar disorder (n = 42)	Schizophrenia (n = 28)	Control (n = 38)
NART = National Adult Reading Test; PANNS = Positive and Negative Symptoms Scale; YMRS = Young Mania Rating Scale; HDRS = Hamilton Depression Rating Scale.
Age, mean (SD)	">39.6 (10.1)	">38.0 (9.9)	">37.2 (11.9)
Male sex, n (%)	">22 (52.4)	">15 (53.6)	">19 (50.0)
Right handed, n (%)	">  39 (92.9)	">  26 (92.9)	">  36 (94.7)
NART IQ, mean (SD)	"> 111.6 (10.7)	"> 106.0 (10.3)	"> 113.5 (7.0)
Psychotic symptoms, n (%)	">  32 (76.2)	">  28 (100)	">–
PANSS positive, mean (SD)	"> 7.3 (0.6)	"> 10.5 (2.6)	">–
YMRS, mean (SD)	">0.6 (1.1)	">0.9 (2.6)	">–
HDRS, mean (SD)	">2.3 (5.8)	">3.4 (6.5)	">–
Duration of illness, mean (SD)	"> 18.3 (11.1)	"> 16.8 (9.6)	">–
Lithium, n (%)	">24 (57.1)	">0 (0)	">–
Antipsychotic, n (%)	">  19 (45.2)	">  28 (100)	">–
Antidepressant, n (%)	">  21 (51.2)	">  8 (28.6)	">–

Bipolar versus control subject differences

Voxel-based analysis. Patients with BD had significantly lower FA in the region of the superior thalamic radiation than controls (p = 0.018, Table 2, Fig. 1a). Using the frontal lobe SVC, significantly lower FA values were found in the regions of both the UF (p = 0.006) and anterior thalamic radiation (ATR; p = 0.037, Fig. 1b and 1c) in bipolar subjects compared with controls.

Table 2.   Regions of reduced fractional anisotropy in subjects compared with controls

Voxel (pFWE-cor)	Z-score	Cluster size	x, y, z (mm) MNI coordinates	Point of maximal change
pFWE-cor = family-wise error corrected p-value; MNI = Montreal Neurological Institute; ALIC = anterior limb of the internal capsule; SVC = small volume corrections.
Bipolar disorder
Whole brain analysis
0.018	5.04	13876	−19, −10, 49	Superior thalamic radiation
Frontal lobe SVC
0.006	4.90	502	"> 26, 19, −7	Uncinate fasciculus/inferior longitudinal fasciculus
0.037	4.35	1200	">−20, 22, 24	Anterior thalamic radiation/corpus callosum
ALIC SVC
< 0.001	3.48	19	"> 18, 12, 3	Anterior thalamic radiation
Schizophrenia
Frontal lobe SVC
0.007	4.60	850	">−24, 48, 8	Uncinate fasciculus/anterior thalamic radiation/inferior fronto-occipital fasciculus
ALIC SVC
< 0.001	3.95	40	"> 20, 12, 5	Anterior thalamic radiation

Figure 1.  Areas of reduced fractional anisotropy in bipolar disorder patients compared with controls. MNI=Montreal Neurological Institute.

In BD subjects, application of the ALIC SVC demonstrated a significant reduction in FA in the region of the ATR (p < 0.001) (Fig. 1d).

Automated ROI analysis.  Using the ALIC automated ROI analysis, participants with BD showed significantly lower FA in the left ALIC compared with controls (t = 3.159, p = 0.002). The UF automated ROI demonstrated significantly lower mean FA on the right side (t = −3.311, p = 0.001) and on the left side (t = −2.554, p = 0.013).

Schizophrenic versus control subject differences

Voxel-based analysis.  Patients with schizophrenia showed a significant reduction in FA in orbitofrontal white matter (p = 0.007) using the prefrontal SVC compared with controls. This was in the region of the frontal lobe portion of the UF/inferior fronto-occipital fasciculus (Fig. 2a). Reduced FA was found in patients with schizophrenia compared with controls in the ALIC (ALIC SVC: p < 0.001, Fig. 2b). This was in the region of the ATR.

Figure 2.  Areas of reduced fractional anisotropy in schizophrenia patients compared with controls. MNI=Montreal Neurological Institute.

Automated ROI analysis.  Using the automated ROI analysis, patients with schizophrenia showed significant reductions in right ALIC FA compared with controls (p = 0.002).

Schizophrenic versus bipolar differences

There were no significant differences in FA between patients with schizophrenia and those with BD using either voxel-based or automated ROI analyses. Patients with BD but not schizophrenia showed reductions in superior thalamic radiation FA compared with controls. Since the larger numbers of subjects with BD compared to schizophrenia provided more power to detect a patient-versus-control difference in FA, the mean FA at the peak voxel was extracted and plotted graphically for each group separately. The mean FA was similarly reduced in patients with both disorders and did not significantly differ between the patient groups (mean FA difference = 3.2 × 10⁻³, 95% confidence interval: −0.15 to 0.15).

Effects of symptoms and medication

BD patients with a lifetime history of psychotic symptoms showed no significant differences in FA from those who had never experienced psychosis (no p < 0.1). Within patients with schizophrenia, there was a significant positive relationship between FA and HDRS scores in the region of the UF (t = 2.4, p = 0.023). In patients with BD, there was a significant negative relationship between HDRS scores and FA in the region of the anterior thalamic radiation/corpus callosum (t = −2.23, p = 0.031). No other symptom-FA association was reported for any of the other peak voxel results (no p < 0.08).

There were no significant associations between FA and any class of psychotropic medication in either patient group (no p < 0.08), although a positive association between lithium and FA in the ALIC showed a nonsignificant trend (t = 1.79, p = 0.082) in bipolar subjects.

Discussion

Main findings

In this DTI study comparing patients with BD, patients with schizophrenia, and controls, deficits in white matter integrity were revealed in the ALIC, UF, and ATR across both conditions. These deficits are in keeping with our prediction that ALIC reductions would be common to both disorders, but do not support our hypothesis that uncinate deficits are unique to schizophrenia. This suggests that abnormal structural connectivity may underpin the pathology of both BD and schizophrenia. These findings replicate and extend earlier studies of people with schizophrenia, but are novel in BD.

Reductions in ALIC and ATR FA implicate abnormal structural connectivity between subcortical nuclei and the prefrontal cortex. The activity of striatum and thalamus is modulated by the prefrontal cortex, and this process is understood to be key to the appreciation of reward (35), emotional processing (36, 37), and mood state (38). Abnormal fronto-thalamic-striatal connectivity may therefore contribute to sustained functional deficits in these domains of cognitive function associated with both schizophrenia and affective disorders.

Previous studies using conventional T₁-weighted MRI have found reduced ALIC white matter density in patients with schizophrenia (6, 10, 39) (VBM) (40), including those with first psychotic episodes (41). Rose et al. (42) found reduced integrity of prefrontal-thalamic circuitry using DTI. Although there are other studies of prefrontal and subcortical white matter in patients with BD (4, 6, 7, 11, 43), the results have been somewhat inconsistent. We previously reported (6, 11) ALIC reductions in white matter density across patients with schizophrenia and BD using conventional T₁-weighted MRI, supporting the current finding. In contrast, Haznedar et al. (7), in a DTI study of patients with bipolar spectrum illness, found reduced FA in the posterior limb of the internal capsule, a region understood to include motor and somatosensory tracts. Adler et al. (4), in a study of nine patients and nine controls with BD, found reductions in prefrontal FA using an ROI approach. None of the ROIs were placed in the ALIC and it is unclear whether the ATR was implicated in their investigation.

The reduction demonstrated in the region of the UF replicates earlier work in schizophrenia and is an entirely new finding in patients with bipolar disorder. The UF fans out from the anterior three temporal convolutions and amygdala to terminate in the medial orbital cortex, gyrus rectus, and subcallosal area (44). Its integrity has been implicated in studies of general intelligence, visual and verbal memory (45, 46), and executive function (47, 48). These results replicate and extend previous DTI findings in schizophrenia (10, 17, 18, 47, 49, 50) and may provide a structural substrate for reduced fronto-temporal connectivity associated with the syndrome (51–53). Our findings suggest, in contrast to earlier work (54), that reduced UF FA is not simply related to psychotic symptoms. It is less clear how reduced white matter integrity in the superior thalamic radiation, common to both disorders (but significant only in BD), relates to the psychopathology of either syndrome, since this region is understood to carry somatosensory fibres. Nevertheless, it is potentially consistent with earlier findings suggesting involvement of the posterior internal capsule (7).

No association was found between white matter integrity and any class of medication, measures of current or lifetime psychotic symptoms, or mania rating scale scores in either patient group. In contrast, depressive symptoms were negatively associated with ATR/callosal FA in the BD group, providing further evidence that reductions in the ALIC and its associated tracts may be involved in the pathophysiology of affective symptoms. In contrast, a significant positive relationship was found between depressive symptoms and FA in the region of the uncinate in the group with schizophrenia. A preponderance of affective symptoms has previously been shown to be associated with a better outcome in patients with schizophrenia (55) and it is possible that those with good prognostic factors have less disruption in these pathways.

Strengths and limitations

Although our results are in need of replication, as far as we are aware, this is the largest DTI analysis to examine white matter integrity in patients with BD and the first to compare schizophrenia and BD directly within the same study. Nevertheless, it remains possible that a larger sample may detect smaller differences in FA.

All patients were clinically stable and had at least one first- or second-degree relative with the same diagnosis. Whilst this may affect the generalisability of our findings, it is likely to have reduced heterogeneity and enriched for genetically determined effects on white matter integrity. Although this study concerns medicated patients, the results presented were not simply related to the confounding effects of current medication or residual symptoms; however, a larger sample could assess these aspects further.

Whilst the analysis of DTI data remains a controversial area, this study utilised several methodological refinements to our previous work. First, VBM analyses were conducted using a study-specific template, and this is likely to have improved image registration. Furthermore, we used two complementary techniques: voxel-based analysis and automated ROI analysis. Since the latter technique is based on the analysis of unsmoothed images in native space, the reductions in ALIC FA are unlikely to be secondary to increases in lateral ventricular enlargement or other possible artefacts. Notwithstanding the benefits of these techniques, uncertainty remains about the precise location of reported white matter deficits and the effects of potentially imperfect image registration. Developing technologies, such as tractography, could improve this.

Acknowledgements

We thank all the participants who took part in the study, the radiographers who acquired the DTI scans, and other staff who contributed to the project at the Scottish Funding Council Brain Imaging Centre. This study was supported by a project grant from the Chief Scientist Office (Ref CZB/4/434) to AMM, GKSL, JH, MEB, ECJ, and SML. SML, GKSL, and SMM are supported by the Sackler Foundation. AMM is currently supported by the Health Foundation.