Relative Glucose Metabolic Rate Higher in White Matter in Patients With Schizophrenia

The study group comprised 170 subjects on whom both PET and magnetic resonance imaging (MRI) were obtained: 103 normal volunteers and 67 patients with schizophrenia (N=61) or schizoaffective disorder (N=6). The patients (49 men, 18 women; mean age=34.3 [SD=17.9] years, range=13–73; mean education=12.9 [SD=2.5] years) were recruited from the Mount Sinai and Bronx Veterans Affairs Hospitals. The patients were evaluated with the Comprehensive Assessment of Symptoms and History (18) and diagnosed according to DSM-IV criteria. Patients were either never previously medicated (N=33) or had been neuroleptic-free for at least 2 weeks (N=34). Fifty-nine patients were right-handed, four were left-handed, three were ambidextrous, and one was undetermined based on the Edinburgh Handedness Inventory (19) . Administered rating scales included the Positive and Negative Syndrome Scale (20) (total psychopathology score=23.9 [SD=14.9]). Positive and Negative Syndrome Scale scores were not available for nine patients because of scheduling logistics.

The 103 normal volunteers (51 men, 52 women; mean age=43.4 [SD=17.7] years; range=20–80; mean education=16 [SD=2.4] years; right-handed [N=96]; left-handed [N=6]; ambidextrous [N=1]) were recruited by word-of-mouth or advertising. These volunteers were assessed using the Comprehensive Assessment of Symptoms and History and were found to have no history of psychiatric illness in themselves or their first-degree relatives.

All participants were screened by medical history, physical examination, and laboratory testing. Individuals with a history of substance abuse/dependence, neurological disorders, head trauma, clinical dementia, or a positive urine test for drugs of abuse on the day of the PET scan were excluded. After a complete description of the study, all participants provided written informed consent. Volumetric data from 40 of the 67 patients’ MRI scans and PET data for the thalamus and cortex (27 patients) have been reported previously (16, 21, 22) .

A 24-gauge angiocath was inserted (by Dr. M.S. Buchsbaum) in the antecubital vein, and a 0.9% saline drip was started. Patients were then moved to a sound-attenuated room and were read standard instructions about the serial verbal learning task (23), which was specifically developed for the 32-minute [ ¹⁸ F]fluorodeoxyglucose (FDG)-uptake period; the serial verbal learning task is analogous to and based on the principles of the California Verbal Learning Test (24) . As described elsewhere (23), the serial verbal learning task was selected because it requires both short-term memory (or “working memory”) (25) and encoding for long-term memory.

PET scans (20 slices, 6.5 -mm thickness) were obtained with a head-dedicated General Electric (GE, Milwaukee) scanner (model-PC2048B) with a measured resolution of 4.5 mm in plane (4.2 to 4.5 mm across 15 planes). We injected 5 mCi-FDG and processed the images as described elsewhere (16) . T1-weighted axial MRI GE scans were acquired with the Signa 5x system with the following parameters: repetition time=24 ms, echo time=5 ms, flip angle=40°, slice thickness=1.2 mm, pixel matrix=256×256, field of view=23 cm, total slices=128.

First, we used Automated Image Registration 5.2.5 (26) and a six-parameter rigid-body transformation to coregister each subject’s FDG-PET scan to his or her anatomical MRI. We used FSL FAST (27) to segment the MRI into gray and white matter and CSF and create a binary image for each of the three tissue types. These images were then smoothed with a 5-mm full-width at half-maximum filter to approximate the resolution of FDG-PET images. Smoothing resulted in three images for each subject, representing tissue concentration and ranging continuously in value from 0 to 1. To avoid potential artifacts arising from differences in the proportion of gray and white matter tissue, we computed the mean brain activity within gray (gray matter concentration >0.75) and white matter voxels (white matter concentration >0.75), respectively, and then computed the average of these tissue means. This yielded a mean brain metabolic rate equally weighted for gray and white matter tissue and was used to divide the raw FDG-PET values to compute the rGMR for significance probability mapping. A t test comparing the composite gray-white denominator did not reveal statistically significant differences between patients and volunteers.

We used Automated Image Registration 5.2.5 (http://bishopw.loni.ucla.edu/AIR5/index.html) to align each subject’s MRI to the ICBM-452 template with a 12-point affine transformation. On the basis of this transformation, average images were created for the schizophrenia patients and the normal volunteers. These two templates were then averaged to create a custom template unbiased with respect to between-group differences in brain shape. All MRI were then renormalized to this custom template using nonlinear registration with the Automated Image Registration program “align_warp” and a fourth-order 105-parameter model.

We compared rGMR on a voxel-by-voxel basis using standard t tests and our own program written in the R statistical language. In addition, we used spatial mixture analysis (28) to identify significance thresholds that are automatically estimated from the computed statistical images by fitting a mixture of three Gaussian probability distributions (corresponding to activated, deactivated, and nonactivated voxels) to the locally weighted image histogram and thus providing a method for separation of signal from background noise without first selecting an arbitrary p value.

After reviewing the resulting maps, we noted that significant regions of gray matter reduction were confluent across the prefrontal and temporal areas (related in part to the cohort size [N=170]) and that reporting their centroid and highest t value would not be helpful to the reader. Raising the t-value threshold would exclude some regions already reported as hypoactive in the literature, preventing replicatory inspection.

In cortical gray matter regions, voxel-wise statistical comparison of rGMR in patients and the full set of normal volunteers (with age removed as a single covariate for both groups) revealed an activation pattern consistent with previous studies that found relatively decreased activity that was most pronounced in the anterior prefrontal, lateral temporal, and posterior parietal regions in schizophrenia patients ( Figure 1 ). In addition, however, the patients showed pervasive and widespread increases of rGMR in white matter tracts throughout the brain ( Figure 1 ), including the corpus callosum, superior and inferior longitudinal fasciculi, and uncinate fasciculus. Higher rGMR in white matter in patients relative to volunteers was evident in the sagittal sections ( Figure 2 ) in the white matter of the corpus callosum and superior longitudinal and uncinate fasciculi.

Because previous studies did not detect a generalized pattern of white matter hypermetabolism in schizophrenia, we carried out a series of supplementary analyses aimed at ruling out the possibility that the white matter differential resulted from a statistical or image-processing artifact or that task performance was the major source of variation. The following three plausible alternative hypotheses were established: 1) Because of known structural differences between the brains of patients and comparison subjects, differences in rGMR may be secondary to or confounded by systematic variation in gray/white/CSF tissue concentration in the two groups; 2) since patients performed more poorly relative to comparison subjects on the serial verbal learning task administered during glucose uptake, the appearance of globally heightened white matter activity may be entirely unrelated to the disease state, reflecting the physiology of poor performance instead—a novel but not implausible finding in its own right; and 3) patients moved more in the PET scanner relative to volunteers, blurring gray matter into white matter locations, spuriously raising white matter values.

We examined the normal-versus-schizophrenia t test values as a function of the average local (gray/white) tissue concentration score in spatially normalized gray and white matter templates. These templates were created by averaging the mean gray and white matter atlases for each group, yielding a measure of mean tissue concentration for every voxel in standard space. Throughout the entire brain, the greater the white matter tissue concentration, the higher the t value separating normal volunteers and patients ( Figure 3 ). Thus, the greatest contribution to the effect came from white matter voxels surrounded by other white matter voxels and with higher signal-intensity values. These are the voxels that are farthest from CSF voxels, in the centers of the larger white matter tracts or bundles, and therefore they are the voxels least likely to have rGMR changed by partial volume or resolution-limit effects.

For every pixel, we used a multivariate algorithm (“Matchit” implemented in R [http://GKing.Harvard.Edu/matchit]) to find the subset of normal volunteers that provided the best “match” for gray, white, and CSF tissue concentration. This procedure is based on optimizing a “propensity score,” which is defined as the probability that an individual is a member of one group given the variables on which to match. This is analogous to matching our normal volunteers and schizophrenia patients simultaneously on age, education, and marital status (analogous to gray, white, and CSF tissue concentration) and discarding some volunteers until the match is optimal. In this analysis, however, a different set of matched normal volunteers was used for each voxel in stereotaxic space. Thus, for each voxel, the group of schizophrenia patients was matched with a group of normal subjects with the same distribution of gray, white, and CSF tissue concentration. This was possible because our large cohort of normal healthy comparison subjects allowed for many potential matching subsets across the whole brain. Voxel-wise t statistics performed on the glucose values after tissue matching did not significantly alter the overall findings of higher rGMR in the white matter of the patients. Note that for each voxel, the groups of PET scans had relative metabolic rate determined in groups of individuals who were matched on the MRI tissue type. As shown in Figure 3, white matter concentration versus the difference in group means indicated that local gray and white matter concentration differences could not be the sole sources of the observed white matter hypermetabolic effect.

To examine whether the overall white matter raw FDG uptake (in nanocuries) was associated with the gray matter uptake, we examined the correlation between the average uptake from white and gray matter tissue (with segmented tissue concentration values >0.75, which was the same fraction used in the computation of rGMR) in the spatially unnormalized (“native-space”) PET images. This correlation was 0.44 in normal volunteers and 0.41 in schizophrenia patients, and the difference was not significant (p=0.84). Not surprisingly, this indicated that individuals with higher uptake in gray matter also tended to have higher uptake in white matter. When whole brain mean nanocurie values were covaried out, the remaining partial correlations were 0.02 and 0.11, respectively, which was consistent with the lack of a global reciprocal relationship; this latter analysis suggests a very modest linkage between gray and white matter rGMR and argues against the white matter effect being a mere byproduct of gray matter differences. Last, we examined the raw nanocurie uptake values for gray and white matter tissue in the spatially unnormalized images. Here, white matter showed higher values in patients relative to normal volunteers, while gray matter values were not significantly different between the two groups (whole population [schizophrenia patients >normal volunteers]: gray matter: t=–0.348, p=0.73; white matter: t=3.711, p=0.0003; restricted to California verbal learning test-matched subjects [schizophrenia patients >normal volunteers]: gray matter: t=1.02, p=0.31; white matter: t=2.205, p=0.03; restricted to age-matched group [schizophrenia patients >normal volunteers]: gray matter t=–0.37, p=0.71; white matter: t=3.692, p=0.0004). Again, this is consistent with an effect in white matter, which appears despite differences in body weight, dose, and other factors that affect nanocurie uptake. In addition, the persistence of this effect on the unscaled PET data rules out an artifact introduced during the conversion to relative glucose values. Note that since this analysis relied only on native-space (not spatially normalized) images, nonlinear normalization artifacts cannot be the source of patient/normal volunteer differences.

Last, as an additional evaluation of the potential contribution of low gray matter metabolic rate to white matter increase in schizophrenia patients, we recomputed the group difference maps removing three nuisance variables, gray matter metabolic rate (segmentated gray matter value >0.75), an average value from the midbrain determined stereotaxically (chosen in the center of the cerebral peduncles and pons at z=–12 and –16), and age. This analysis ( Figure 4 ) showed that higher metabolic rates in the corpus callosum and frontal white matter still remained statistically significant.

Performance on the serial verbal learning task was better in normal volunteers than in schizophrenia patients (mean=13.1 [SD=1.84] versus mean=7.8 [SD=3.70], respectively, p<0.001). We carried out two analyses, first with the volunteers and patients matched on the serial verbal learning task total number correct (N=34) in each group, mean number correct (N=10.7), and second on the whole group with the serial verbal learning task as a covariate. As shown in Figure 4, both the subject-matching and covariance correction using all subjects had little effect on the tissue-concentration/mean-difference curve. Figure 4 also illustrates subanalyses of normal volunteers and patients matched on age and sex, demonstrating that scores continued to show the white matter hypermetabolic pattern.

Finally, patients might have had more head movement relative to normal volunteers during the PET scan, thereby blurring the image and contaminating white matter areas with actual gray matter activity levels. Although the subjects’ heads were restrained by a specially molded plastic headholder during scanning, the possibility of differential head movement between patients and volunteers cannot be dismissed out of hand. Therefore, we examined this potential artifact by outlining each FDG-PET with an automated edge-finding algorithm and each MRI by tracing the gray matter edge. If the patients had moved their heads more than volunteers, the consequently blurred FDG edge outline would tend to be larger than the MRI cortical edge; however, no significant group difference in PET/MRI size ratios was observed.

The data reported in the present study support earlier literature on frontal, temporal, striatal, and thalamic metabolic decreases and cerebellar increases in gray matter in schizophrenia patients as reviewed previously but also indicate white matter increases. White matter change has been little reported in FDG-PET research in schizophrenia, and for this reason we review competing hypotheses of white matter metabolic rate elevation and, briefly, known neurological conditions associated with increased metabolic rate in white matter to cover all potential pathophysiological mechanisms.

Communication that is redundantly repeated, routed to multiple destinations, and/or poorly inhibited, synchronized, terminated or regulated by the frontostriatal-thalamic control circuit might demand greater metabolic activity in white matter pathways. Several sets of different but parallel routings are given in models (29) of these pathways, and minor misroutings, disturbances in the topographical organization, and the use of multiple rather than single pathways all provide potential explanations for inefficiency. The possibility of disorder in these pathways in schizophrenia has received theoretical, animal, and imaging support (16, 30 – 33) . While one might have expected higher rGMR in the gray matter areas associated with higher white matter traffic, it is also possible that longer interhemispheric or longitudinal circuits are used inappropriately in place of more local circuits. Analysis of interregional correlation patterns in individuals with high and low white matter rGMR, examination of electroencephalographic coherence patterns, and functional MRI studies contrasting tasks that require and do not require interhemispheric coordination might help to address this issue.

Tighter packing of axons, dehydration with reduced water between axons, and thinner myelin sheaths or reduced oligodendrocyte cellular volume might be associated with higher metabolic rates per unit area. Tangling, misrouting, brain edema, and thinner sheaths, or even the later development of axons in normal volunteers that are not myelinated in patients, might be associated with lower density. Indeed, recent neuropathological studies have demonstrated a significant loss of oligodendrocytes in the frontal white matter of schizophrenia patients, involving changes in myelin (9) as well as alterations in the spatial distribution and ultrastructural changes in these cells (10, 11) that may directly contribute to the differences in rGMR observed in this study.

Elevations in white matter metabolic rate have been reported in spinal cord radiotherapy for hypopharyngeal and thyroid cancer (34, 35), and this uptake has been suggested to be associated with the myelopathy known to be a consequence of radiation treatment. An autopsy study of one patient with elevated FDG uptake revealed demyelination (with axonal loss) and neuronal damage in the cervical spinal cord (36) . The investigators speculated that “functional recovery is supported by a process that probably differs from the restoration of the mechanism destroyed by the radiotherapy…an altered conduction mechanism of the action potential, involving an increased number of sodium channels along the demyelinated segments of the injured axons, which is fully congruent with the PET findings” (36) .

Lesions of the white matter in multiple sclerosis may also be metabolically hyperactive, which is consistent with inflammatory or reparative increases in metabolism (37) . In multiple sclerosis, the activation of astrocytes and microglia accompanies the destruction of myelin (38) . Other studies have not confirmed hypermetabolism in multiple sclerosis (39, 40), but some of them do not provide information on specific lesion areas or maps of comparison subjects versus patients (40) . Remyelination of damaged areas involves oligodendrocytes. In a recent review, Ludwin (41) suggests that this is more widespread in multiple sclerosis than often credited. Remyelination in schizophrenia has not received significant attention.

Hypoxic damage during childbirth has been noted as a risk factor in schizophrenia. However, oligodendrocytes appear more sensitive to hypoxia than astrocytes or microglia in in vivo studies of neonatal rat brain (42), suggesting that it might be patients with low white matter uptake rather than those with high white matter uptake who suffer this insult. Carbon monoxide toxicity has been associated with greater white matter than gray matter decreases in regional PET-FDG metabolism, which is again consistent with greater white matter sensitivity to hypoxia (43) . Patients with Wernicke-Korsakoff syndrome have relative hyper metabolism in white matter and hypo metabolism in subcortical gray matter (44) . The relatively confluent white matter mask appearance of the significance probability maps in the Wernicke-Korsakoff patients for Talairach z=28 is not entirely dissimilar to our findings in schizophrenia patients ( Figure 1 ).

An increased density of interstitial neurons in white matter might elevate the rGMR. Interstitial neurons have been found in the white matter of the frontal and temporal lobes, suggesting a defect during migration with a failure to reach their cortical targets (45, 46), and this effect may be more prominent in patients with the deficit form of schizophrenia (47) and in the more superficial regions of frontal white matter (48), although not all studies confirmed this finding (49) . Ectopic foci of gray matter in an adjacent region anterior to the caudate nucleus have been visualized on MRI (50) and, if sufficiently frequent, even if too small to be routinely observed (<1 mm), might elevate the rGMR in frontal white matter.

Higher white matter rGMR could stem from reduced rGMR coupled with normal white matter activity. Since imaging studies of schizophrenia have found frontal and temporal metabolic rate decreases, paralleled by reduced frontal and temporal gray matter on anatomical MRI, the possibility merits careful consideration. Brains with proportionately less gray matter might have lower total brain uptake of tracer and a resulting higher white matter pixel/whole brain average ratio. To mitigate this bias for each subject, we divided the raw activity in each pixel by one-half the mean activity in all gray matter (>0.75 probability on MRI segmentation) plus one-half the mean activity in all white matter (similarly selected voxels). In this way, volume changes in gray and white matter in schizophrenia would not bias whole brain metabolic rate. We then performed several analyses. First, we compared rGMR of schizophrenia patients with rGMR in sets of normal volunteers that were selected for each voxel as representing the best match in gray, white, and CSF tissue concentration as determined by an MRI segmentation algorithm; these maps equally confirmed the white matter elevation. Second, we evaluated spatial mixture modeling (28), which uses a threshold based on a separate population of t values; this analysis confirmed the white matter effect (and yielded a somewhat more confluent indication of white matter change), suggesting separate populations of white matter rGMR values. Third, we examined the plots of tissue concentration/t values to assess whether the white matter elevation arose solely from gray matter decrease. The relationship between the t values and the white matter tissue concentration value could not arise from the rGMR divisor being biased low in patients since the same rGMR divisor is used for white matter voxels with concentrations ranging from 0.1 to 1.0 (had the effect come from the divisor, the increasing t values illustrated in Figure 3 from 0.4 to 1.0 would not have been expected). Moreover, this effect, as illustrated in Figure 3, was consistent across various matched subsets of the full group. We also examined global differences in gray and white matter activity in native-space images using unscaled nanocurie values. First, this analysis confirmed that the finding of white matter hypermetabolism was not an artifact of the spatial normalization process. We then showed that even when using raw nanocurie values, global white matter, but not gray matter, differences between groups were significant. Next, we found no evidence for differences in the correlation between gray and white matter nanocurie values between the two groups. Thus, this native-space analysis also shows that differences observed in the statistical parametric mapping analysis were not driven by an artifact incurred during the transformation to relative glucose values. Additionally, we examined the statistical images in detail. Effects were greatest in the central regions of large white matter areas, inconsistent with a head movement or whole brain divisor artifact. Finally, we used analysis of covariance to remove the mean gray matter value, mean cerebellar peduncles and pons values, and age. This analysis still confirmed statistically significant white matter values in schizophrenia. It should be noted that this covariance analysis included brain gray matter activity that is generally found to be low in schizophrenia, and thus this analysis might well remove some schizophrenia-related variance (just as using yearly income or the number of close personal friends as a covariate might attenuate gray matter metabolic rate reduction in the temporal and frontal lobes). Despite this possible overcompensation, both white matter increases and frontal metabolic decreases remained significant, even when the areas of cortical decrease—especially in the temporal lobes—were minimized.

The relative reduction of gray matter rGMR in the current statistical maps has been observed in earlier studies that included subgroups of this same population using regions of interest for the frontal lobe (16, 51), temporal lobe (51), caudate nucleus (52), dorsal cingulate gyrus (53), and mediodorsal nucleus of the thalamus (54) . Similar findings also have been widely reported in the FDG-PET literature (55, 56) .

In this large cohort of unmedicated patients, artifacts of neuroleptic medication are minimized, cognitive state is controlled, and statistical power to detect differences is enhanced. Some small cohort size studies have found increases in frontal and thalamic FDG uptake in unmedicated patients (N=10 [57] and N=4 [58] ); the lack of significance probability maps, uptake tasks, or white matter regions of interest makes comparison with the current white and gray matter results uncertain. The next level of scientific examination of individual differences in hypo- hyperfrontality and symptom constellations, medication responsiveness, genetic markers, and other features will clearly be necessary to understand this variation. Absolute glucose metabolic rates were not obtained, and thus the possibility that gray matter change did not affect the measurement of white matter activity cannot be fully excluded. Obtaining full glucose quantification on never-medicated schizophrenia patients will be valuable in replication and extension of this finding, although such large cohorts will be difficult to obtain.

In their important study, Weber et al. trenchantly suggest that “monitoring the energy supply of white matter may yield important information on the traffic on the highway and with it on cerebral communication” (14) . We feel that traffic that is inefficient, duplicative, erroneous, and unselected is the likely explanation for these relative metabolic increases and speculate that this may be related to the extensive literature on cognitive deficits in schizophrenia.