Glial Cell Loss in the Anterior Cingulate Cortex, a Subregion of the Prefrontal Cortex, in Subjects With Schizophrenia

The brains from the comparison subjects were part of another repository collected from 1987 to 1991 from autopsied individuals following the Danish laws on autopsied human tissue. The Danish ethical committee for Copenhagen and Frederiksberg approved the study. Table 1 shows demographic and clinical data for the schizophrenia and comparison subjects, and Table 2 shows postmortem data. No brains had evidence of degenerative brain disorders. The comparison brains came from 14 individuals who were free of neurological or psychiatric disorders. None of the subjects with schizophrenia or the comparison subjects had any history of head injury or substance abuse. The schizophrenia subjects fulfilled DSM-III criteria and belonged to a subpopulation of severely ill patients hospitalized for a substantial part of their lives and treated with neuroleptic treatment, insulin comas, or ECT. Furthermore, for both groups information about the patient’s prehospital life was obtained from either close relatives or the general practitioner, thus excluding all patients with any CNS symptoms. The tissue was fixed within 8–72 hours after death and kept in 4% formaldehyde buffered with phosphate.

The meninges were removed from the brains, the right or left hemisphere chosen according to level of artifacts, and the hemispheres cut frontally into slabs that were 2.5 cm thick. The slabs were dehydrated in a series of alcohol solutions of 50%, 70%, and 80% (12 hours in each) and in 99% alcohol, xylene, and regular paraffin solutions (24 hours in each). The paraffin-embedded slabs were sectioned exhaustively into 40-μm-thick sections and were then systematically subsampled at random, providing about 10 sections per case (ranging from 9 to 15) containing Brodmann’s area 24 and 32. These were adhered on object glasses pretreated with a gamma-aminopropyl-trithoxysilane and stained with a modified Giemsa stain (25).

Brodmann’s areas 24 and 32 were delineated by two independent investigators on coded slides, according to the cytoarchitectonic criteria specified in the following section. The interrater reliability of delineations was 95%–98%. An example of the transitional zone between areas 24 and 32 is shown in Figure 1.

Brodmann’s area 24 occupies practically the entire extension of the cingulate gyrus, stretching from the adjacent area 33 to its dorsal limit. The lack of granular layer IV and the salience of the pyramidal layer V are characteristic features of the region (26). Layer V is composed of a population of well-stained big pyramids distributed in two levels: the upper sublayer Va, with a homogeneous distribution and high cell density in contrast to the underlying sublayer Vb, characterized by pyramids aggregating into distinct cellular clusters with an overall low cellular density and the presence (often in clusters) of spindle (fusiform) neurons (27, 28). These attributes influence the global aspect of its structural pattern, giving it a bistratified appearance clearly visible already at low magnification. The overall bilaminar aspect and the absence of layer IV help to differentiate area 24 from the neighboring dorsal and rostral areas. Ventral to the corpus callosum, the entire region shares the bilaminar layout in the lower layers described for area 24. Here the granularity criterion prevails, offering a reliable reference to the limit between the agranular area 24 and the granular Brodmann’s area 12.

Brodmann’s area 32 can be considered as a transitional region between peri- and isocortex and has been roughly associated to the paracingulate gyrus, also called the superior cingulate or paracingulate cortex. In this case, however, the relationship between microstructure and gross anatomical features proves to be unreliable, since the gyrus shows a high degree of morphological variability and is often fragmented. It is positioned basically parallel to the adjacent cingulate gyrus, connecting to the cingulate gyrus and the superior rostral sulcus at the level of the genu of the corpus callosum (29). The main properties of area 32 are the progressive predominance of supragranular layers II and III over the infragranular layers V and VI and the development of a discontinuous internal, granular layer IV. Area 32 is composed of pyramidal cells and shows a moderate but continuous increase in cell size from surface toward the white matter. Sublayer IIIc is particularly developed, and its cells intermingle with those belonging to the upper limit of the pyramidal layer V. Furthermore, layer V is more homogeneous than in area 24, with a diminishing sublayer Va, while simultaneously increasing cellular density and width of sublayer Vb. The appearance of layer IV and a wide layer III marks the ventral limit of area 32. Rostrally, area 32 is distinguishable from area 10 by the rich layer IV of the latter. The transition between the end of area 32 and the beginning of the supplementary motor area or cingulate magnoganglionic core of Braak (27, 30) is revealed by the sudden appearance of the pyramidal cells of Betz in layer Vb, a distinctive feature of motor cortices (31). No extension of area 32 ventral to the corpus callosum has been detected (31).

A combination of systematic uniformly random sampling, volume estimation, cell counting, and disectors was used to obtain the total number of neurons and glial cells.

For systematic uniformly random sampling, some of the most remarkable features are the efficiency in terms of time and precision that may be obtained by counting only a few hundred nerve cells in, for example, the human brain. The coefficient of error (CE) of the estimated total cell number was determined according to the formula given by Braendgaard et al. (25) and modified according to equations 20–22 of Gundersen et al. (32). In the estimation of total cell number, we considered the “ideal” coefficient of error to be about half of the observed coefficient of variation (CV) because of the relationship: observed CV²=biological CV² + estimator CE² (32). In this study, the values of the coefficient of variation were about 20%–30%, and the values of the coefficient of error were about 10%, which means that we met our general criteria for optimal precision of the estimations.

The total volume of the brain region of interest, i.e., the volume of the reference space [V(ref)] was estimated by using the Cavalieri principle (33). The total number of points [ΣP_i] is multiplied by the unit area per point [a(p)] to give the total area of interest: A_i=a(p) × ΣP. The area [A_i] of each region was estimated using a Leica DMLB microscope with a 2.5× objective and a final magnification of 172× and an area per point of 1517 mm². A camera lucida setup was used to superimpose the cross-sectional area of the region of interest onto a point counting grid, and the investigator counted only the points that were within the image of the area. The average number of points counted per brain was 122 (range=90–208) for area 24 and 87 (range=32–187) for area 32.

The third dimension of the reference volume [V(ref)] was obtained by multiplying the sum of the measured areas [ΣA_i] by the inverse sampling frequency of the sections [k] and the section thickness [t]: V(ref)=ΣA_i × k × t.

The cell counting was performed by using the optical disector method to estimate the number of cells divided into two groups (neurons and glial cells) in Brodmann’s areas 24 and 32. Bilateral cell number was estimated as the unilateral number multiplied by two.

The disector is a probe that samples isolated particles with a uniform probability in three-dimensional space in which the z axis (the height of the disector) is employed using 40-μm-thick sections in which the plane of focus is moved up or down, and the x and y axis are defined by a square (the counting frame) superimposed on the magnified image of the tissue on the computer screen (34). The area of the counting frame [a(frame)], the height of the disector [h], and the sampling fraction in the x-y axis were estimated in a pilot study and optimized to obtain a count of approximately 150–200 particles (cells) [ΣQ^–] per specimen. The dimension of the optical disector probe was a rectangular frame area of 2171 μm² and a disector height of 20 μm. The volume of the disector [v(dis)] is calculated as follows: v(dis)=h × a(frame). The equipment for optical disector counting consisted of a BX-50 Olympus microscope with a 100× oil-immersion objective with a high numerical aperture (1.35), a motorized stage, and an electronic Heidenhain microcator with a digital readout for measuring movements in the z direction with a precision of 0.5 μm. Counting took place at a final magnification of 3250×. Per brain, an average of 105 disector probes (range=32–208) were sampled with an average number of 1.5 neurons (range=0.9–2.8) and 3.0 glial cells (range=0.5–5.1) counted per disector, respectively. The thickness of the stained sections was measured at every second disector, showing a mean section thickness of 37.4 μm (range=31.9–45.3). The distinction between the largest glial cells and the smallest neurons is not a trivial problem. In this study, the neurons were identified on a combination of morphological criteria such as the chromatin pattern, size, and shape of the cell nucleus, a clearly visible nucleolus, and surrounding cytoplasm. Figure 1 shows examples of neuron and glial cells using the modified Giemsa stain applied in this study. In 1.3% of the cases, the cells were identified as nonclassifiable and were not included in the final estimate. One investigator (A.K.S.) performed all cell counting. Several brains were counted more than once in a coded design and showed an intrarater reliability of >95% in total cell number.

The total number of particles (cells) [N(part)] was obtained from the equation N(part)=[ΣQ^–/Σv(dis)] × V(ref), where ΣQ^–/Σv(dis) is equal to the numerical cell density [N_v]. Since the estimation of both the numerical cell density and the reference volume was obtained at the final, processed histological tissue level, the shrinkage does not need to be calibrated.

The differences between means were analyzed using an unpaired two-tailed Student’s t test. The p values of differences between means are presented as two-tailed probabilities. Variability is shown in brackets as coefficient of variability (CV=SD/mean) following the respective means.