MRI Brain Abnormalities in Chronic Schizophrenia: One Process or More? Bryan T. Woods, Deborah Yurgelun-Todd, Jill M. Goldstein, Larry J. Seidman, and Ming T. Tsuang Ht has been suggested that schizophrenia is primarily a prefrontal-temporallimbic circuitry disorder. Further, it has been argued that primary neurologic vulnerability 10 the illness is established only during early stages of brain development and is not progressive. We tested the Inypothesis of whether brain volume losses in prefrontal and temporal-limbic regions have occurred either before or after brain growth was hypothesized to be complete in schizophrenia, Nineteen chronic schizophrenic patients and 19 age- and sex-matched normal controls underwent magnetic resonance imaging (MRI). All cans were segmented into gray and white matter and cerebrospinal ‘fluid (CSF) compartments for the frontal and temporal lobes and pasterior cerebral hemispheres. Multivariate analysis of variance was used to analyze absolute intracranial cerebrum and subregion volumes, ie, gray, white and CSF, absolute tissue (i.e. gray plus white) volumes, and tissue 10 intracranial volume (ICV) ratios. Patients showed significant intracranial volume reductions only in the frontal lobes but highly significantly lower TCV ratios (ie, greater relative tissue loss) in all three major regions. It is suggested that the observed decreases in frontal intracranial volumes reflect a pathologic process in schizophrenia that impacted the frontal regions before brain growth was complete. We hypothesize that the generalized lower patient TCV ratios are attributable to a process that affected the whole cerebrum over a time period after brain volume had reached its maximum levels Key Words: Schizophrenia, MRI, brain imaging, frontal lobes, temporal lobes Biot PSycHiaTRY 1996;40-585-596 Introduction The evidence for cerebral anatomic abnormalities in schizophrenic patients is very strong (Weinberger 1984). ‘The occasional negative studies can reasonably be attrib- uted to smail sample size, unusual patient selection fac- rome Deprtnents of Neng an Py. Teas A eM Mac Schl and Tewology Secon, VA Medial Caner. Temp TX (BTW) Deptt of Peta, Harvant Medial Schools Br imaging Conrad Manan Rewach Coo, McLe Hosp, Belmont MA (DY 1 Hard lot yeh: Hdemblgy and Geeies UMC. LIS, MI an Hamad Depa iment of Pacha af Masachsets Meal Heath Cee, Bens MA ad Brockton Wes Roxbury VA Modis! Cert, reckon, MA OMG, LS MT Newepsyeiniog Labrty, Deparment of Psctiay, Haran Medial Scat © 1986 Sociey of Biologia Pysity tors, and possibly etiologic heterogeneity. Less unanimity exists for the localization and extent of the abnormalities, though the most frequently replicated findings are lateral ventricular enlargement and) medial temporal volume reduction. Even less settled are the issues of etiology and timing of the causative insults. The case for a developmental origin of brain abnormal- ities in schizophrenia comes from both clinical and neu- ‘Masschasts Mental Health Cntr, Boston, MA (JMG, LIS nd he Depa ‘of Epidemiology, Harvard Scho of Pubic Healy, Bostom, MA (TD. Ades rept requ o br. Bran T Woods, Neucogy Sexton, VA Mis Comer 1901 Sa St, Temple TX 6808. Received September 24 1598; reve Jy 24 1985 .006-32239615 00 pr ene 122505)00198 3586 BIOL psvcwarey Hedoses 6 ropathologic reports, Clinical imaging studies of new- fonset schizophrenic patients show some volume abnormalities are already present at this stage (DeLisi 19914; Lieberman 1992), Neuropathological studies show aan absence of gliosis in schizophrenic brains (Roberts 1991), and a loss of interneurons in layer Il of the cingulate gyrus and prefrontal cortex (Benes et al 1991), ‘These findings have been interpreted as indicating oceur- rence of the anatomical changes, pre-illness, and probably in the perinatal period (Weinberger 1987; Murray et al 1992). A recent neuropathological study of postmortem material from schizophrenic patients that stained for the neuron-specific enzyme nicotinamide-adenine dinucle- ‘tide phosphate-diaphorase (Akbarian et al 1993a,b), found evidence for decreased numbers of positive-staining cells in neocortex and increased numbers of positive- staining cells in subcortical white matter in both frontal ‘and temporal regions. It was concluded that both regions showed a defect in prenatal neuronal migration The question of whether there may also be brain changes that take place after onset of illness has long been disputed. There is some positive evidence for such con- tinuing pathological changes from both clinical and imag- ing studies. Recent clinical studies of schizophrenia indi- cate that many patients have a course in which they progressively deteriorate over the initial S~10 years of illness, and then plateau (MeGlashan 1988; Breier et al 1991), Several MRI cross-sectional studies have suggested. that generalized loss of brain volume in schizophrenia is related to duration of illness (Gur et al 1991) and age (O'Callaghan et al 1992). Gur et al (1991) found that, after ccovarying for age, schizophrenic patients had a significant positive correlation of duration of illness and total CSF volume. O'Callaghan et al (1992) reported that cortical atrophy, but not ventricular size, showed a significantly {greater positive correlation with age for patients than for controls. In contrast another recent large cross-sectional MRI study that carefully controlled for age effects failed to find any significant relationship between volume loss and duration of illness (Zipursky et al 1992) The results of longitudinal imaging studies have also been mixed. Studies using CT, which cover a greater duration of illness because of longer availability of the technique, have not resolved the question of whether volume loss is progressive or not, since both negative (Nasrallah et al 1986; Iowsky et al 1988; Vita et al 1988; Abi-Dargham et al 1991) and positive (Kemali et al 1989; Woods et al 1990; Schwarzkopf et al 1990) studies have been reported Longitudinal studies using MRI are still in their early stages. Lieberman et al (1992) found a nonsignificant trend for chronie schizophrenics to have a higher preva- lence of morphologic brain abnormalities than acute pa- B.T. Woods et al tients; however, they raised the possibility that this trend might be accounted for by the poorer prognosis associated with the presence of brain abnormalities at initial diagno- sis and a subsequent “selection out” of the patients with fewer brain abnormalities. DeLisi et al (1991a), using a similar design, found that chronic, but not acute, schizo- phrenic patients showed a significant reduction in tempo- ral lobe volume as compared to age-matched controls. Interestingly, at a 2-year follow-up of the acute patients, in which both good-outcome and poor-outcome patients were restudied, there was an overall trend to increased ventricular enlargement, with several patients showing a ‘greater than 20% increase in ventricular size (DeLisi et al 19916). Itis evident that early and late changes are not mutually exclusive, and it is quite possible that both occur in the same patients by different mechanisms. There is a long standing ncuroradiological observation (Davidoft-Dyke- Masson syndrome) that offers the possibility of distin- ‘guishing brain volume loss beginning in early life from that occurring after brain growth is complete. Studies of patients with cerebral hemiatrophy occurring in early life show that the intracranial cavity is smaller on the side where the brain is atrophic, although the skull may be thicker (Taveras and Wood 1976; Zilkha 1980; Woods and Yurgelun-Todd 1991). This occurs because, barring the ‘uncommon event of a pathologic premature fusion of skull sutures, regional brain growth drives regional intracranial cavity growth during development, Thus, an early impair- ment of brain growth will result in a correspondingly smaller intracranial cavity; however, since intracranial cavity expansion is not reversible after skull sutures fase, regional loss of brain tissue volume after_growth is, complete will not cause a decrease in intracranial volume, but rather a compensatory increase in regional cerebrospi- nal fluid (CSF) space. It follows then that reductions in overall or regional intracranial volume in. schizophrenic patients relative to controls are likely to be the result of events that had their onset during the period of brain ‘growth, while decreases in the ratio of brain tissue volume to intracranial volume (TCV ratio) imply a tissue loss that ‘occurred after brain growth and development were essen- tially completed (ie., by the middle of the second decade of life), Finally, a combination of both reduced intracranial volume and a reduced TCV ratio is evidence for a process (or processes) that operated both during and after the period of brain growth, ‘The current study measured differences in intracranial volume and in brain tissue to cranial volume (TCV) ratio for the whole cerebrum and several subregions implicated in schizophrenia, ¢.g., frontal, temporal, ventricles, stria- tum, and pallidum. This study is a first step in beginning to determine whether patients showed evidence for gener-MRI in Schizophrenia alized or regional brain volume loss that occurred either before or after brain growth was complete, or both. ‘Methods Magnetic Resonance Imaging All MRI scans were carried out on a single General Electric Signa 1.5 Tesla system with shielded gradients and quadrature detection coils. Images were acquired using an inversion recovery sequence in the coronal plane perpendicular to the individual canthomeatal line. Slices were acquired from just behind the occipital pole to just ahead of the frontal pole, and were 5-6 mm thick (depending on individual head anteroposterior length) with no interslice interval. All images were analyzed on an auxiliary console using a method that isa variant of an approach originally described for computed tomography (CT) seans (Pfefferbaum etal 1986) and modified for MRI (Lim et al 1989) that segments a brain image into gray, white, and CSF components. The process began with an operator using a cursor to outline the cerebrum and surrounding CSF on each coronal image of a complete front-to-back contiguous series, by tracing along the inner surface of the skull or dura mater, the upper surface of the tentorium, and the medial surface of the temporal lobes: the brainstem was transected at the mesencephalic-dien- cephalic junction. These outlined regions of interest (ROIs) summed over all images constituted the cerebral intracranial volume. For each image the outlined cerebral ROL was then duplicated on a splitscreen, with one image displayed with subjectively optimal gray-scale windowing and leveling, and the other image windowed to one (i.c., windowed so that all pixels were black or white). The level of the black-white image was then varied by the operator so as to ‘optimally maich the shape of the resulting silhouette to the appropriate tissue compartment on the adjacent gray-scale image. This level is the intensity threshold for the bound- ary separating two tissue components. The same process ‘was used for both the CSF-gray boundary and the gray- white boundary, and all pixels in the outlined ROI were categorized into one of the three components depending con their intensity values vis-t-vis the thresholds. The thresholds established in this way for each individual slice were used for all subsequent measurements on that slic. ‘The whole process was carried out independently by two raters blind to diagnosis and to one another's results ‘The next step was outlining smaller ROIs for subse- {quent segmentation. The independently outlined structures ‘were the separate left and right cerebral hemispheres, anterior (predominantly prefrontal) frontal lobes, major frontal subdivisions (dorsolateral, dorsomedial, and or- bitofrontal regions), temporal lobes and temporal hos, corpus striatum, ventral pallidum, and lateral ventricies. ‘The primary anatomic guides for this process were the coronal brain sections in the atlas of Matsui and Hirano (1978). The posterior cerebral region volumes were not ‘measured independently but were derived from the cere- bral hemispheric volumes less the sum of the ipsilateral frontal and temporal volumes. ‘The structure boundaries were traced as follows. First, all external borders were traced along the inner border of the skullidura mater. Second, the inferior and medial borders of the temporal lobes and posterior cerebrum followed the tentorium and medial brain border so as to exclude the cranial nerves, and the brainstem was transected at the mesencephalic-diencephalic junction. ‘Third, the posterior boundary of the frontal lobes was the rearmost slice passing fully through the genu of the corpus callosum. Fourth, the posterior boundary for the temporal lobes was the rearmost slice passing fully through the splenium of the corpus callosum, the superolateral bound- ary was the Sylvian fissure, and the superomedial bound- ary was a line drawn from the medialmost point of the Sylvian fissure straight down to the hippocampal fissure. Fifth, the ventral pallidum (Atheid and Heimer 1988) was ‘bounded by the triangle formed by the anterior commis- sure superiorly, the third ventricle medially, and the inferior margin of the cerebrum. Finally, the corpus striatum included the head of the caudate, the putamen, and that part of the body of the caudate that was included ‘on the putminal slices. It should be noted that the posterior boundaries we utilized to delineate the frontal and temporal lobes are not the true posterior anatomical boundaries of either region. In particular the frontal regions are largely prefrontal, and exclude the precentral and most of the premotor regions. With one exception, all of these structures were mea- sured on more than one slice. For reasons of anatomical landmark reliability, the subdivisions of the frontal lobes were outlined on the single slice that formed the posterior boundary of the frontal lobe. On this slice the anatomic division between dorsomedial and dorsolateral regions was the superior frontal sulcus, and the boundary between dorsolateral and orbitofrontal regions was the anterior ramus of the lateral sulcus (Nieuwenhuys et al 1988). In order to assess measurement reliability across both rater and subjects, nine control subjects were scanned twice on separate days, and two raters independently thresholded and outlined all slices of both studies for each subject. The intraclass coefficients (ICC) (Fleiss 1983) between raters for the same scans were: 0.99 for gray matter, white matter, and total intracranial volume; 0.94 for CSF; 0.80 for corpus striatum; and 0.46 for ventral588 OL PSYCHIATRY Tobe-85" 300, pallidum, Between scans, the ICCs were as follows: gray ‘matter, 0.92 for raters | and 2; white matter, 0.79 for raters 1 and 2; CSF, 0.87 for rater 1 and 0.89 for rater 2; and total intracranial volume, 0.99 for both raters. Subjects A total of 17 patients (14 men, 3 women) meeting DSM-III-R (American Psychiatric Association 1987) cri- teria for schizophrenia (SCZ), two women meeting criteria, for schizoaffective disorder, and 19 healthy controls (14 men, 5 women) were quantitatively evaluated with MRI. ‘The schizophrenic patients were a subsample of consecu- tive admissions of 50 schizophrenic and schizoaffective patients from four area hospitals who were recruited for a study of the heterogeneity of schizophrenia (M. T. Tsuang, principal investigator). Twenty consecutive patients who had agreed to participate in the larger study were asked to hhave an MRI study as well, and all agreed. One patient exceeded the weight limit of the scanner and was replaced by the next consecutive patient, who also agreed to participate. Ultimately, 19 patients were included in the MBI scan group, since one scan was rendered unusable by severe movement artifact. Patients had to be proficient in English, not have any lifetime history of DSM-ILR substance dependence nor any history of DSM-IIER substance abuse in the last 6 months, have no history of neurologic disease or injury ‘with persistent central nervous system (CNS) damage, not have undergone electroconvulsive therapy in the last 6 ‘months, and not have any systemic medical illness affect- ing brain function. DSM-II-R diagnoses were made on the basis of a consensus by three senior investigators, all of whom evaluated material from a structured psychiatric interview, the Schedule for Affective Disorder and Schizo- phrenia (SADS) (Spitzer and Endicott 1978), from an interview with the currently treating clinician, and from systematic review of past and present medical records. Reliability of diagnoses was high, with a x of 0.80 reflecting 11 agreements and only one disagreement be- tween the consensus diagnosis and two expert diagnosti- cians who made independent clinical assessments. Fifteen of the 17 SCZ patients met DSM-IILR criteria for non- paranoid subtype, while two were paranoid subtype. The patients who received MRI scans did not differ from those who did not receive MRI on demographic, clinical, or neuropsychologic criteria (Seidman et al 1994) Subjects and controls were between the ages of 20 and ‘55 years. Mean ages in years (with standard deviations in parentheses) were as follows: schizophrenic (SCZ) men, 34.4 (9.2); control (CTL) men, 34.6 (9.7); SCZischizoaf- fective women, 35.2 (6.7); and CTL women, 34.6 (6.7) B. T, Woods et al ‘Age did not differ significantly for patients versus con- trols, nor when stratified by sex. Subjects were all white non-Hispanics and had a mean education level of 12.6 years. More complete demographic and neuropsychologic {data on these subjects has been previously reported (Seid- ‘man et al 1994). Control subject scans were selected from fa larger panel of available normal control studies con- ducted on the same MRI system. Controls were selected ‘on the basis of sex and age by someone who was blind to sean results. The control pool consisted of hospital main- tenance workers, clerical staff, health professions students, and professional staff members. Their estimated mean education level was 15 years Data Analysis The independent variables for all analyses were diagnosis (SCZ, CTL) and hemisphere (eft, right). Dependent vari- ales were either the absolute volumes or the TCV ratios of the ROIs. Absolute volumes comprised total intracra- nial volumes (.e., gray + white matter + CSF) for the frontal, temporal, and posterior cerebral regions and the frontal subregions; brain tissue (i.e. gray + white) forthe cerebrum; the CSF segment of the ROI for the temporal horns and lateral ventricles; and the gray matter segment of the ROI for the corpus striatum and ventral pallidum. TCV ratios were the sum of the gray and white matter segments for cach ROI divided by the sum ofthe gray and White matter and CSF volumes for the same ROI. In the cease of frontal subregions, where measurements Were on & single slice, volumes rather than areas were nevertheless used as the unit of measurement because slice thickness was 5 mm for some patients and 6 mm for others. Due to partial volume effects, different slice thicknesses will result in slightly different measurements forthe same scan, even if the center of the slice is identical, although phantom measurements on our system indicated that these Jifferences would be less than 1%. All analyses employed either analyses of variance (ANOVA) or multivariate analysis of variance (MANOVA) (Abacus Concepts 1989). Separate MANO- VAs were utilized for the dependent variables that were measurements of absolute volumes and those that were ratios of measured volumes. The major cerebral regions (frontal, temporal, and posterior) were analyzed separately from the specific anatomic structures (frontal subregions, lateral ventricles, temporal hors, corpus striatum, and ventral pallidum) and from the cerebral segments (gray, White, and CSF). Moreover, because two of the female patients were diagnosed as schizoaifective rather than schizophrenic, all of the analyses were also run excluding these two patients and their matched controls; however,