Spatial Characteristics of White Matter Abnormalities in Schizophrenia

Abstract

There is considerable evidence implicating brain white matter (WM) abnormalities in the pathophysiology of schizophrenia; however, the spatial localization of WM abnormalities reported in the existing studies is heterogeneous. Thus, the goal of this study was to quantify the spatial characteristics of WM abnormalities in schizophrenia. One hundred and fourteen patients with schizophrenia and 138 matched controls participated in this multisite study involving the Universities of Iowa, Minnesota, and New Mexico, and the Massachusetts General Hospital. We measured fractional anisotropy (FA) in brain WM regions extracted using 3 different image-processing algorithms: regions of interest, tract-based spatial statistics, and the pothole approach. We found that FA was significantly lower in patients using each of the 3 image-processing algorithms. The region-of-interest approach showed multiple regions with lower FA in patients with schizophrenia, with overlap at all 4 sites in the corpus callosum and posterior thalamic radiation. The tract-based spatial statistic approach showed (1) global differences in 3 of the 4 cohorts and (2) lower frontal FA at the Iowa site. Finally, the pothole approach showed a significantly greater number of WM potholes in patients compared to controls at each of the 4 sites. In conclusion, the spatial characteristics of WM abnormalities in schizophrenia reflect a combination of a global low-level decrease in FA, suggesting a diffuse process, coupled with widely dispersed focal reductions in FA that vary spatially among individuals (ie, potholes).

Introduction

As the number of diffusion tensor imaging (DTI) studies in schizophrenia edges over the century mark, it is challenging to make sense of the considerable heterogeneity in the location of reported white matter (WM) abnormalities. ¹ Certain WM tracts are frequently reported as abnormal, viz, the corpus callosum, ² , ³ cingulate bundle, and frontal WM pathways; ^4–7 however, these tracts are larger and contain more coherent WM tracts and have been the target of more directed studies, ie, region of interest (ROI) analyses. ¹ , ² Studies have also found localized WM differences in multiple other regions, including the internal capsule, ⁸ , ⁹ cerebellum, ¹⁰ arcuate fasciculus, ⁹ , ¹¹ hippocampus, ¹² fornix, ¹³ and the temporal, ¹⁴ , ¹⁵ parietal, ¹⁰ , ¹⁶ and occipital lobes. ¹⁷ , ¹⁸ Because many of the studies utilize voxel-based analyses, a positive finding in only 1 region of the brain implies negative findings in a number of other regions. These negative findings in similar brain regions are not reported and do not enter into the meta-analytic studies of DTI in schizophrenia. Thus, although there is considerable evidence for WM abnormalities in schizophrenia, there is a lack of consistency in the location of these WM regions.

Understanding the spatial characteristics of WM abnormalities in schizophrenia is important because different analysis strategies require specific assumptions regarding the underlying spatial profile. Studies using ROI or voxel-based approaches assume that the underlying WM abnormalities are in the same general spatial location. In other words, if a specific ROI is traced or if the brains are all placed in the same standard space, the assumption is that the majority of WM abnormalities will have a regionally specific overlap, which can be statistically identified. Yet, primary WM disorders such as multiple sclerosis, although having a predilection for certain brain regions, are not associated with lesions in the same region in every individual. ¹⁹ Thus, it is possible that different patients can have spatially distinct WM abnormalities, perhaps even at different locations along the same WM tract (ie, frontal WM pathways), ²⁰ which may not be found using ROI or voxel-based approaches.

Thus, the goal of this article was to evaluate the spatial characteristics of WM abnormalities in schizophrenia. We utilized DTI data from the Mind Research Network Study, a large, multisite study of schizophrenia involving the Universities of Iowa, Minnesota, and New Mexico, and the Massachusetts General Hospital. WM abnormalities have been identified in this sample of patients with schizophrenia, ²¹ and thus it is well suited to apply several different distinct methodologies to assess the spatial characteristics of WM abnormalities in patients with schizophrenia. We utilized 3 distinct image-processing approaches to test 3 different spatial patterns of WM abnormalities in schizophrenia. One possibility is that the WM abnormalities are present in the same general location in the majority of patients. The second possibility is that there are focal abnormalities that are not spatially overlapping (“potholes”). ²⁰ The third possibility is that the WM deficits involve a diffuse process, affecting different areas equally.

Methods

Participants

A total of 252 subjects (114 patients with schizophrenia and 138 controls) were recruited from the Universities of New Mexico (NMex), Iowa, and Minnesota (Minn) and from the Massachusetts General Hospital (MGH). Table 1 contains the demographic information for the cohort. The study was a part of the consortium study of the Mind Research Network. Both the patients and controls received a semistructured diagnostic interview using the Structured Clinical Interview for the DSM-IV (SCID) ²² or the Comprehensive Assessment of Symptoms and History (CASH). ²³ An estimation of cognitive performance was obtained in both the patients and controls using the Wide Range Achievement Test, 3rd edition, reading test (WRAT3-RT). ²⁴

Table 1.

Demographics and Clinical Data of Patients With Schizophrenia and Controls

	Whole Group		Iowa		MGH		Minn		NMex
	Patients n = 114	Controlsn= 138	Patientsn = 19	Controlsn = 52	Patientsn = 27	Controlsn = 21	Patientsn = 27	Controlsn = 22	Patients n = 41	Controlsn = 43
Age, years (SD)	33.3 (11.2)	31.2 (10.9)	31.5 (10.2)	30.8 (10.4)	36.9 (10.4)	40.1 (8.0)	31.3 (10.1)	28.9 (8.9)	33.1 (12.6)	28.7 (11.6)
Sex (M/F)	84/30 *	81/57	15/4 **	23/29	19/8	12/9	18/9	12/10	32/9	34/9
Hand (R/L/both)	99/6/7	127/7/3	17/0/2	49/2/1	25/1/1	19/0/1	22/2/2	21/0/1	35/3/2	38/5/0
WRAT3-RT (SD)	46.9 (6.3)***	50.6 (4.3)	49.1 (4.8)	50.2 (4.2)	45.4 (7.3)***	53.2 (2.1)	46.0 (6.6)**	50.9 (3.7)	47.4 (5.8)	49.7 (5.1)
Father’s education (SD)	14.2 (3.7)	14.9 (3.3)	15.4 (3.2)	14.5 (2.8)	13.8 (3.9)	13.6 (3.4)	13.8 (3.5)	15.5 (2.5)	14.0 (3.9)	15.7 (3.9)
Mother’s education (SD)	13.5 (3.5)	13.9 (2.7)	14.9 (3.4)	13.7 (2.0)	14.2 (3.0)	13.3 (3.2)	12.8 (3.0)**	15.2 (2.5)	12.9 (4.0)	13.8 (3.2)
Clinical Measures
Years of illness	5.9 (9.9)		4.8 (7.5)		8.8 (11.3)		3.7 (7.1)		6.1 (11.3)
Antipsychotics a	35.4 (56.2)		46.5 (65.0)		44.4 (49.3)		33.3 (79.7)		24.1 (31.9)
Atypical antipsychotics	24.8 (34.9)		31.5 (42.6)		35.0 (40.2)		12.5 (22.0)		22.1 (32.1)
Typical antipsychotics	33.8 (64.0)		38.8 (42.7)		23.5 (49.1)		58.5 (107.9)		13.9 (19.9)
SANS / SAPS
Positive	4.7 (2.7)		3.6 (3.0)		5.1 (3.1)		4.9 (2.0)		4.8 (2.6)
Negative	7.6 (3.8)		8.1 (4.6)		7.2 (4.5)		7.1 (3.1)		7.9 (3.5)
Disorganized	1.6 (1.8)		1.9 (1.7)		1.3 (1.6)		1.8 (2.1)		1.6 (1.7)

Note: SANS / SAPS, Scale for the assessment of negative symptoms /Scale for the assessment of positive symptoms.

aAntipsychotics values are presented in dose years.

Patient-control differences: *P < 0.05; **P < 0.01; *** P < 0.001.

Healthy volunteers who were matched with the patients on age, sex, and handedness were recruited from the community. Controls were excluded from the study if they had any Axis I psychiatric disorder, including substance abuse/dependence or a history of a schizophrenia or bipolar spectrum disorder in a first-degree relative. Additional exclusion criteria for both patients and controls included a neurological disorder affecting brain function (ie, head injury with loss of consciousness and seizure disorder) or active substance abuse/dependence. Written informed consent was obtained from all subjects prior to participation, and the study was approved by the institutional review boards at each of the 4 sites.

The severity of positive and negative symptoms over the past 2 weeks in the patient group was collected using the Scale for the Assessment of Positive Symptoms (SAPS) ²⁵ and the Scale for the Assessment of Negative Symptoms (SANS). ²⁶ Handedness was assessed using the Edinburgh Handedness Inventory. ²⁷ The majority of patients (n = 99) were on one or more antipsychotic medications at the time of testing. Of these patients, 89 were on only atypical medications, 8 were on a typical antipsychotic medication, and 3 were on both atypical and typical medications. Within the atypical antipsychotic medication group, 12 were on clozapine only and 5 were on clozapine in addition to another atypical medication. Fifteen patients were not on medication at the time of scanning. Antipsychotic dose years was calculated according to the formula 1 dose year = 100 chlorpromazine equivalents per day for 1 year. ²⁸ The clinical characteristics for the patients are presented in table 1.

Image Acquisition

High-Resolution Structural Magnetic Resonance Imaging All structural images were acquired with an in-plane resolution of 0.625 × 0.625mm, a slice thickness of 1.5mm, and a flip angle of 7^o. MGH and NMex utilized a Siemen’s 1.5-Tesla scanner with TR = 12, TE = 4.76, and NEX = 1, but 3 separate images were collected and averaged. Iowa utilized a GE 1.5-Tesla Genesis Signa scanner with TR = 20, TE = 6, and NEX = 3. Minn used a Siemen’s 3-Tesla scanner with repetition time (TR) = 2530, inverse time (TI) = 1100, echo time (TE) = 3.79, and number of excitations (NEX) = 1.

Diffusion Tensor Imaging: All DTI images were acquired on Siemens 3-Tesla scanners with 2-mm isotropic resolution. MGH utilized a Siemens Sonata with TR = 8900, TE = 80, B values of 0 and 700, NEX = 1, and NEX = 60 directions. Iowa utilized a Siemens TRIO with TR = 9500, TE = 90, B values of 0 and 1000, and NEX = 4 and 6 directions. NMex utilized a Siemens Sonata with TR = 9800, TE = 86, B values of 0 and 1000, and NEX = 4 and 12 directions. Minn utilized a Siemens TRIO with TR = 10 500, TE = 86, B values of 0 and 1000, and NEX = 2 and 12 directions.

Image Processing

The DTI images were analyzed using 3 different approaches to assess the spatial characteristics of WM differences. These approaches included an ROI approach, tract-based spatial statistics (TBSS), ²⁹ and the no-overlap-requirement spatial analysis (NORSA), also known as the pothole approach. ²⁰

ROI Approach: The anatomical images were processed using BRAINS2. ³⁰ This produced a skull-stripped T1-weighted image and a WM mask. The skull-stripped T1-weighted image was coregistered with an AC-PC-aligned atlas image using a rigid registration to a scaled version of the atlas image to account for linear stretching along each axis. The Talairach parameters were defined for the subject based on an affine registration of the atlas image into the raw subject space, allowing the Talairach atlas to be warped onto each subject. The T2-weighted image was then coregistered with the AC-PC-aligned T1-weighted image. Tissue classification was performed using a multimodal tissue classification. ³¹

The diffusion-weighted images were analyzed using the GTRACT program. ³² The diffusion-weighted images were first coregistered with the B0 image using a mutual-information image registration to correct for motion and distortions caused by eddy currents. The images were median filtered and the diffusion tensor was estimated. Scalar measures for fractional anisotropy (FA) were calculated on the DTI images for all subjects. ³³

The B0 image was then coregistered with the skull-stripped T1-weighted anatomical image from BRAINS2 using a rigid-body transformation and a mutual-information-similarity metric. Next, a B-spline transform was applied to remove distortion in the echo-planar images resulting from susceptibility changes at air-tissue interfaces. ³² The resulting transforms were applied to the scalar maps, placing them within the space of the anatomical image.

Measurements of FA were obtained in coronal Talairach sections ³⁴ from the anterior to the posterior region along the whole brain. The mean FA within each coronal Talairach slice was calculated from the intersection between the segmented WM obtained from the structural image and an FA threshold of 0.1 from the DTI image. This combination of structural and DTI images eliminated regions of signal loss resulting from magnetic susceptibility differences. The mean FA within each coronal slice was z-transformed within each site prior to pooling the data. In addition to obtaining coronal slices, regions from the Johns Hopkins University WM atlas ³⁵ were applied to the FA maps to extract mean FA values for each individual.

Tract-Based Spatial Statistics: The FA maps for the patient and control groups were preprocessed using TBSS ²⁹ from the Oxford Centre for Functional Magnetic Resonance Imaging of the Brain (FMRIB) Software Library. The TBSS automated voxel-based approach minimizes partial volume effects by its novel skeleton-based registration algorithm. No smoothing was applied to the FA maps, as there is evidence that the magnitude of spatial filtering can distort voxel-based results using DTI. ³⁶ TBSS was applied to the data at each of the 4 sites individually to assess patient-control FA differences along the skeletal tracts. Permutation testing was performed using a cluster-based thresholding to reduce type I error. A familywise error correction of q < 0.05 was used to correct for multiple testing. ³⁷

No-overlap requirement spatial analyses (NORSA): An overview of the NORSA, or the pothole approach, is described elsewhere. ²⁰ The method for the NORSA approach begins with FA maps from the control group, whichhave all been spatially normalized into MNI space using TBSS. ²⁹ Utilizing all the spatially normalized images from the controls, mean and SD images are created for each subject within each site. These mean and SD images are then used to transform the images from each patient and control into voxel-based z-images. The z-transformation is performed within each site, using the mean and SDs for the controls at that site. For each z-image generated, an algorithm searches for contiguous clusters of voxels that are below a z-score of z = –2. The total volume and spatial location of each cluster were used for further analyses.

Statistical Analyses

The evaluation of differences between patients and controls on demographic measures was performed using t test and χ² analyses. Site by demographic differences were evaluated with a one-way ANOVA. Within-site demographic variables of patients vs controls were evaluated using χ² for discrete data or a one-way ANOVA for continuous data. The clinical measures of length of illness and dose years of antipsychotic medications did not fit a normal distribution and were thus evaluated with the Kruskal-Wallis Test.

For the ROI analyses, the FA measures were converted to within-site z-scores and combined. A repeated-measures ANCOVA was performed to assess patient-control FA differences across the Talairach slices, with age and sex as covariates. Due to nonlinearity in duration of illness and dose years of medication, Spearman rank correlations were used to assess the relationship between duration of illness, dose years of medication, and FA measures within brain regions. Statistical analyses were performed utilizing the SAS statistical package).

Results

Demographic and Clinical Variables

There was a significantly higher proportion of males among patients compared to the control group (χ² = 6.2, df = 1, P = 0.01; 74% and 59%, respectively). Sex and age were used as covariates in all analyses. There were no differences between the patients and controls in age, handedness, or in the educational level attained by the father or the mother (table 1). The mean ages of the patients and controls were 33.3 (SD: 11.2) and 31.2 (SD: 10.9) years, respectively.

The mean positive, negative, and disorganized symptom scores in the patients were 4.7 (SD: 2.7), 7.5 (SD: 3.8), and 1.6 (SD: 1.8), respectively (table 1). The mean number of years between diagnosis and entry into the study was 10.7 (SD: 10.7) years. There were no significant differences in illness duration (log-transformed total number of years since diagnosis) between sites. A one-way ANOVA revealed no differences in total dose years of antipsychotic medications, atypical dose years, or typical dose years between the sites.

There were differences in cognitive performance between the patient and control groups. The mean WRAT3-RT score for patients was 46.9 (SD: 6.3) compared with 50.6 (SD: 5.0) in the control group (t = 5.46, df = 246, pP < 0.001). The highest level of education obtained was also significantly different between the patient and control groups (t = 7.1, df = 244, P < 0.0001). The patients completed on average 13.6 (SD: 2.6) years of education compared to 15.7 (SD: 2.2) years completed by the controls. Although there was a significant site-related difference in total displaced head movement between the different diffusion-weighted volumes (F _3,239 = 47.0, P < 0.001), there was no significant difference in total displaced head movement between the patients and controls when using site and age as covariates.

ROI Analyses

A repeated-measures ANCOVA with age and site as covariates found a highly significant difference in FA between the patient and control groups (F _1,223 = 20.8, P < 0.0001). In fact, post hoc ANCOVAs at every coronal slice showed significantly lower FA in the patients compared to the controls, except for the first (A1) and last (I2) coronal slices. The most anterior and posterior slices had the least WM volume and would have the greatest WM partial volume effects. The differences between patients and controls remained significant when controlling for age, sex, and performance on the WRAT3-RT.

Analyses of the coronal slices at each individual site gave mixed results, with no differences at the Iowa site and a number of differences at the other 3 sites (figure 1A). The site-specific differences were evaluated more closely using mean FA values extracted from each individual based on regions defined by the Johns Hopkins University WM atlas. ³⁵ The results of these analyses are presented in table 2 and show considerable heterogeneity in findings. The regions that are consistent across all 4 sites include the anterior and middle corpus callosum and the right and left thalamic radiation.

Fig. 1.

(A) Fractional anisotropy along coronal Talairach regions of interest between patients and controls at each of the 4 sites. Significant levels at coronal slices were determined using ANOVA with age as a covariate (*, P < 0.05 and **, P < 0.01). (B) Within-site differences in fractional anisotropy using tract-based spatial statistics.

Table 2.

Analyses of Fractional Anisotropy Within Regions of Interest Defined by the Johns Hopkins White Matter Atlas

	Right				Left
Region of Interest	Iowa	MGH	UMinn	NMex	Iowa	MGH	UMinn	NMex
Corticospinal tract	*
Medial lemniscus	*	*
Inferior cerebellar peduncle
Superior cerebellar peduncle		**				**
Cerebral peduncle		*	*			**	***	*
Internal capsule (anterior limb)	*	**	*	**		***		***
Internal capsule (posterior limb)			*	*			*	**
Internal capsule (retrolenticular)			*			**	*
Corona radiata (anterior)	*	***		**	*	***		***
Corona radiata (superior)			**	**		*	*	***
Corona radiata (posterior)		*	*	**		**		**
Thalamic radiation (posterior)	*	**	**	**	**	**	*	***
Inferior longitudinal fasciculus		*		**		**	*	**
External capsule			*	*		*	*	**
Cingulate bundle		*	*			**
Posterior cingulate / hippocampus		*	*		*	**	*
Stria terminalis	*	***	**		**	***		**
Superior longitudinal fasciculus			**	*		*	*
Uncinate fasciculus				*		*	*
Midline regions of interest
Corpus callosum (anterior)	**	***	*	***
Corpus callosum (mid)	**	***	*	***
Corpus callosum (posterior)		***		***
Fornix	***	*
Pontine crossing
Middle cerebellar peduncle	*

Note: * P < 0.05 ** P < 0.01 *** P < 0.001.

Tract-Based Spatial Statistics

Analyses utilizing TBSS resulted in widespread FA differences between patients and controls at 3 of the 4 sites (figure 1B). The FA differences at the Iowa site tended to be more focal and affected the frontal and anterior corpus callosum WM tracts. In fact, the anterior corpus callosum showed differences at all 4 sites. MGH, Minn, and NMex all showed patient-control FA differences along the body of the corpus callosum, the centrum semiovale, and posterior WM tracts.

No-Overlap Requirement Spatial Analyses (NORSA)

To assess for non-overlapping spatial WM abnormalities, we applied the no-overlap requirement spatial analyses (NORSA), or pothole approach, with the volume of the pothole being at least 50 voxels. ²⁰ A 4 (site) by 2 (diagnosis) ANCOVA of the number of potholes and age as a covariate resulted in a significant difference between patients and controls (F _1,238 = 74.3, P < 0.0001). There was also a significant effect of site (F _3,238 = 28.9, P < 0.0001). Evaluating differences within each the 4 sites using age as a covariate resulted in significant differences in Iowa (F _1,67 = 18.3, P < 0.0001), MGH (F _1,44 = 25.2, P < 0.0001), Minn (F _1,46 = 17.9, P < 0.0001), and NMex (F _1,75 = 20.1, P < 0.0001). Although the pothole volume was selected to contain at least 50 contiguous voxels, the patients had a significantly greater number of potholes irrespective of the pothole volume (figure 2).

Fig. 2.

The number of potholes compared to the volume of the potholes at each site between patients and controls. (A) the University of Iowa; (B) the University of Minnesota; (C) the Massachusetts General Hospital; and (D) the University of New Mexico.

The spatial characteristics of the potholes for the entire patient group can be seen in figure 3. Figure 3A shows that the majority of WM potholes are located along the corpus callosum and regions of the centrum semiovale. To examine the spatial overlap in potholes, the image masks containing each of the subject’s potholes were summed. The summation of the masks was performed voxel by voxel in standard space. This resulted in an image with a number of overlapping potholes at each voxel for all subjects. Thus, if a region in the corpus callosum had significantly lower FA in all subjects, the overlap for that region would be 114 (the total number of patients). Using this overlap image to evaluate the number of overlapping potholes within the patients (figure 3B), once the number of overlapping potholes exceeds 15, there is very little spatial overlap.

Fig. 3.

The spatial distribution of potholes in patients with schizophrenia. (A) The spatial location of at least 6 overlapping potholes; and (B) The number of overlapping potholes at the axial slice z = –25.

Relationship Between FA and Antipsychotic Treatment

The Spearman rank correlation between the dose years of antipsychotic medications (DY) and age in the patients was 0.64 (P < 0.0001). The correlation between FA and DY were significant at Talairach coronal slices C1 (r = –0.21, P = 0.04), C2 (r = –0.22, P = 0.03), D1 (r = –0.32, P = 0.002), D2 (r = –0.20, P = 0.05), and F2 (r = –0.24, P = 0.02). None of these measures remained significant after controlling for the effects of age.

Spearman rank-order correlations between age and the number of potholes found significant relationships with age at Minn (r = 0.56, P = 0.002) and NMex (r = 0.50, P = 0.001), but not at Iowa or MGH. The significant differences at Minn and NMex were present irrespective of the number of contiguous voxels, and the results for >100 contiguous voxels are presented above. Interestingly, there were no significant correlations between the number of potholes and DY within each site with or without age as a covariate.

Discussion and Conclusions

There is considerable evidence that WM abnormalities are involved in the underlying pathobiology of schizophrenia. ^38–40 Evidence for these abnormalities stem from postmortem, ⁴¹ , ⁴² genetic, ⁴³ and DTI studies. ¹ , ³⁸ , ⁴⁴ The WM abnormalities found in patients with schizophrenia suggest a disruption in the connectivity between different brain regions. The hypothesis is that this disruption results in inefficient or altered transmission of neuronal signals, which leads to the characteristic symptoms of schizophrenia, including a mixture of hallucinations, delusions, thought disorders, cognitive deficits, and negative symptoms. Based on the current literature, the WM abnormalities do not appear to involve only one WM region, but rather affect the WM either globally or in a distributed pattern. Thus, the goal of this study was to evaluate the spatial profile of WM abnormalities in schizophrenia.

We found that all 3 image-processing algorithms yielded significant differences between patients and controls. An ROI approach using subjects from all 4 sites found that the patients had significantly lower FA in nearly all coronal slices from the anterior to the posterior regions of the brain. However, when evaluating coronal slices separately at each site, the results become less clear. The data from Iowa had no significant WM differences, whereas the Minnesota and New Mexico sites showed considerable differences. Using the JHU WM atlas also resulted in variable findings between sites. The anterior and middle section of the corpus callosum shows a statistically significant decrease in FA at all 4 sites, consistent with the literature. ¹ The right and left thalamic radiations also show significantly lower FA in patients compared to controls at all 4 sites.

The TBSS approach resulted in significant patient-control differences in the genu of the corpus callosum and forceps minor in all 4 sites (figure 1B). In addition, 3 of the 4 sites had multiple overlapping differences involving the corpus callosum, the anterior and superior regions of the corona radiata, and the cingulate bundle. Finally, the NORSA approach using at least 6 overlapping potholes resulted in a very similar spatial profile as that seen in the TBSS approach (figure 3A). However, as the requirement for the number of subjects with overlapping potholes exceeded 6, the extent of spatial overlap quickly dropped. In fact, we found very little overlap with as few as 15 of the 114 patients (figure 3B).

This dropping of the number of overlapping potholes in as few as 15 patients suggests that the WM differences are not present in the same location in the majority of patients. Had this been the pattern, the majority of patients would have spatially overlapping potholes. Rather, our results are more consistent with a pattern of diffuse, slight decreases of FA with interspersed potholes containing more dramatic decreases in FA. The spatial extent of these potholes is variable, but there does appear to be a predilection for the corpus callosum and frontal WM regions (figure 3A). This heterogeneous pattern of WM findings can help explain the heterogeneous pattern of WM abnormalities identified in the DTI studies of schizophrenia.

Understanding the spatial characteristics of WM abnormalities is important because a number of the DTI analysis approaches make underlying assumptions regarding spatial characteristics of the WM abnormalities. If these assumptions are not valid, then the use of specific statistical approaches may not provide a valid representation of the actual findings. Voxel-based approaches are well suited to evaluate abnormalities that generally occur in spatially select regions. Although the WM abnormalities found in patients with schizophrenia do have a predilection for specific regions (ie, frontal WM regions, cingulate bundle, internal capsule, superior longitudinal fasciculus, and corpus callosum), ¹ , ³⁸ , ⁴⁵ these regions did not show considerable overlap of potholes in our study. Thus, we question whether voxel-based approaches are well suited to evaluate the spatial profile of WM abnormalities in schizophrenia. Perhaps better approaches are either one or a combination of global ROI, NORSA, or tractography approaches. Tractography approaches would also be able to detect abnormalities at spatially different points along WM tracts.

Although the potential neuropathological underpinnings of spatially distinct WM abnormalities in schizophrenia are unclear, there are several possibilities. Disruptions in WM that are related to inflammation, ^46–48 vascular changes, ⁴⁸ or oligodendrocyte-related stochastic events could show a heterogeneous pattern of abnormalities.⁴²⁹ In addition, differential effects on cortical gray matter, possibility related to glutamatertic hypertoxicity, ^49–52 could result in secondary effects of cerebral WM. If the cortical gray matter is differentially affected, this could also differentially affect the cerebral WM. Postmortem studies utilizing the whole brain are required to assess whether there are spatially distinct patterns of WM abnormalities.

There are several limitations to our study. The subjects were recruited from 4 different sites, each using scanners with different field strengths (1.5 or 3 T) and diffusion-weighted sequence acquisition parameters that were optimized per site to yield highest-quality data. These differences, especially the number of directions encoded (6, 12, or 60), may increase the variance of the FA measures. To address this limitation, we did not attempt to pool the image data for the TBSS and NORSA approaches; rather, we performed the postprocessing of the imaging data separately for each site using the same algorithms. The TBSS and NORSA approaches were relatively consistent within each site, with 3 of the 4 sites having very consistent findings using the TBSS approach. The differences at the Iowa site may be related to a cohort effect, a smaller sample size of patients, a more rural setting, or greater variability due to the limitations imposed by the 6-direction acquisition sequence.

The differences in the field strengths, magnetic resonance imaging scanners, and sequences used can also be considered a strength. It would be optimal to have an analysis strategy that is not dependent on a specific field strength or scanner type, as this would result in improved generalizability. In addition, the NORSA approach at the MGH site appears to have a much better separation between patients and controls (figure 2). This sequence utilized 60 directions and b values of 0 and 700, which may provide reduced noise and be better suited for identifying WM potholes.

In summary, we evaluated the spatial characteristics of WM abnormalities using 3 different image analysis approaches in a group of patients with schizophrenia compared to a group of healthy controls. We did not find evidence for spatially localized WM abnormalities in the majority of patients, but rather we found a pattern consistent with more generalized decrease in WM with distributed WM potholes. The heterogeneous pattern in the spatial characteristics of the WM tracts in schizophrenia may not only explain the heterogeneous nature of WM findings reported in the literature, but may also underlie the considerable heterogeneity present in the illness. Studies that utilize only voxel-based or small ROI approaches may result in missing some of the WM abnormalities in schizophrenia. The NORSA approach or tractography are both well suited to detect spatially heterogeneous WM abnormalities. Future studies aimed at associating the individual spatial localization of the WM with specific cognitive and behavioral deficits are needed to assess whether the heterogeneity is associated with specific functional or cognitive deficits.

Funding

National Institute of Mental Health (K08 MH068540, MH060662); the National Institute on Drug Abuse (P2ODA024196); the National Association for Research in Schizophrenia and Affective Disorders (NARSAD); and the Mind Research Network.