Alterations in Chandelier Neuron Axon Terminals in the Prefrontal Cortex of Schizophrenic Subjects

At autopsy, the brain was removed and the left hemisphere was cut into coronal blocks (1.0 cm thick). Tissue blocks were immersed in ice-cold 4% paraformaldehyde in phosphate buffer for 48 hours, washed in a graded series of sucrose solutions, and then stored in an antifreeze solution at –30°C. Tissue storage time did not differ across the three subject groups (table 2). Previous studies (13, 14) have shown that these storage conditions do not affect immunoreactivity.

Coronal sections (40 µm) were cut on a cryostat, and every tenth section was stained for Nissl substance with thionine. These sections were used to identify the location of dorsolateral prefrontal cortex area 46 by cytoarchitectonic criteria (15, 16). From a series of sections determined to contain area 46, we chose four sections separated by 400 µm, beginning with a random starting point. Tissue sections were treated with 1% hydrogen peroxide for 15 minutes to eliminate endogenous peroxidase activity; they were then processed with the use of a rabbit anti-GAT-1 antibody (1:1000 dilution; Chemicon, Temecula, Calif.) and the Vectastain Elite ABC kit as previously described (8). Sections from each triad of subjects were processed together through all steps and were mounted on coded slides. The specificity of the anti-GAT-1 antibody has been previously demonstrated (8, 17).

All quantification was performed without knowledge of subject number or diagnosis. GAT-1-immunoreactive cartridge density was measured in the superficial layers (2–3a), middle layers (3b–4), and deep layer (layer 6) of area 46. These layers were selected for analysis because 1) they contain the highest densities of GAT-1-immunoreactive cartridges in normal subjects, and 2) the pyramidal cells in these layers have different patterns of afferent and efferent connections (11). Laminar boundaries were determined by measuring the distance from the pial surface to the borders of each layer on Nissl sections and calculating the location of each border as a percentage of total cortical thickness. On average, the layer 1–2 border was located at 10%, the layer 3a–3b border at 35%, the layer 4–5 border at 60%, and the layer 5–6 border at 80% of the distance from the pial surface to the white matter. On each section, three separate counting regions (corresponding to the superficial, middle, and deep layers) were outlined with the Stereo Investigator Fractionator program (MicroBrightfield, Inc., Colchester, Vt.). A sampling grid (350 µm × 350 µm) was randomly placed over each counting region, and all cartridge-like profiles that fell within the sampling frames (225 µm × 225 µm) were counted at a final magnification of 400× by using rules for unbiased counting. The average Scheaffer coefficients of error (calculated by the Fractionator program) were 11.3%, 13.2%, and 13.0% for the normal, schizophrenic, and other psychiatric subjects, respectively. When corrected for sampling frames located along the exclusion boundaries for each of the counting regions, these values decreased by 15%–30%, resulting in coefficients of error of approximately 10% for each cortical layer.

The procedures described above were used for the 45 subjects (triads 16–30) that were new to this study. For triads 1–15, relative cartridge density was assessed by using a cortical traverse counting procedure as previously described (8). Direct comparison has shown that the two procedures produce similar density estimates (8). For triads 1–15, cartridge density values had previously been reported for the entire cortical thickness. For the present study, the subjects from triads 1–15 were reanalyzed by applying the laminar percentages described above to the cortical depth coordinates of the cartridges in order to obtain density measures of GAT-1-immunoreactive cartridges in layers 2–3a, 3b–4, and 6. Mean densities of GAT-1-immunoreactive cartridges did not differ between the subjects in triads 1–15 and triads 16–30 in layers 2–3a (mean=83.2/mm², SD=47.0, and mean=72.5/mm², SD=40.5, respectively), layers 3b–4 (mean=47.6/mm², SD=36.7, and mean=45.1/mm², SD=26.1), and layer 6 (mean=98.4/mm², SD=49.9, and mean=111.8/mm², SD=51.9) (F<1.55, df=1, 88, p>0.22).

To assess the influence of psychotropic medications on the density of GAT-1-immunoreactive cartridges, we studied four male cynomolgus (Macaca fascicularis) monkeys treated chronically with haloperidol decanoate. Each drug-treated animal was matched to an untreated comparison animal on sex, age, and weight. The animals initially received daily doses of haloperidol in order to determine the amount required to produce 8-hour trough serum levels of 4–8 ng/ml—levels shown to be associated with a therapeutic response in humans (18). As per standard clinical practice, that dose was multiplied by 15 to obtain the dose of haloperidol decanoate to be administered intramuscularly every 4 weeks. The mean dose of haloperidol decanoate was 16.0 mg/kg (SD=2.1), and the mean trough serum level obtained just before the next dose was 4.3 ng/ml (SD=1.1). Consistent with our attempt to mimic the clinical model of “neuroleptic threshold dosing” (19), all haloperidol-treated animals developed mild extrapyramidal symptoms that were controlled with maintenance benztropine mesylate (1 mg b.i.d.). After 9–12 months of treatment, both animals in a matched pair were euthanized with an overdose of pentobarbital, and the brains were removed. All procedures were approved by the University of Pittsburgh’s Institutional Animal Care and Use Committee.

After a 45-minute postmortem interval, coronal tissue blocks from the left hemisphere were prepared and processed for GAT-1 immunocytochemistry in a manner identical to that described for the human subjects. The densities of GAT-1-immunoreactive cartridges in layers 2–3a and 3b–4 of area 46 were determined in a fashion identical to that described above with one exception. Because of the thinner cortical mantle in monkeys, the sampling frames (190 µm × 190 µm) were randomly placed at intervals of 225 µm in each counting region. The coefficient of error in each sampling zone of each animal was less than 10%.

For the human studies of GAT-1-immunoreactive cartridge density, regression analyses revealed a significant association between cartridge density and age, but not sex, postmortem interval, or tissue storage time. Because the last three measures also did not differ between subject groups (table 2), differences in cartridge density within each layer were analyzed with the use of a general linear model based on a complete randomized block design, with diagnostic group, triad (block), and age as independent variables (20). Analyses were done in PROC GLM in SAS software, and F ratios were interpreted with type III sum of squares. Post hoc pairwise comparisons with the schizophrenic group as the target were made by means of Dunnett’s adjustment for multiple comparisons, as this test controls type I experimentwise error. To examine the magnitude and variability of these comparisons, Dunnett’s t tests were used with an alpha level of 0.05 to calculate confidence intervals. For the monkey studies, analysis of covariance, with body weight as a covariate, was used to assess group differences in cartridge density for each layer.

We report relative densities as opposed to absolute numbers of GAT-1-immunoreactive cartridges in this study, since the latter measure was neither feasible nor necessary (24). To obtain an estimate of the absolute number of cartridges in area 46, the volume of this cyto­architectonic region for each subject must be accurately identified. However, because boundaries between cytoarchitectonic areas are not sharp and involve extensive transition zones in the human dorsolateral prefrontal cortex (15, 16), the total number of cartridges in area 46 cannot be reliably determined. In addition, the thickness of tissue sections may shrink by 70% during processing for immunocytochemistry, interfering with the three-dimensional sampling required for estimation of absolute numbers. Furthermore, two potential limitations of the use of profile-counting procedures to determine relative density (differences in cartridge size or cortical volume across diagnostic groups) do not appear to have been confounding factors in the present study. We previously demonstrated that the length of GAT-1-immunoreactive cartridges did not differ across the subject groups in triads 1–15 (8). In addition, the subjects with schizophrenia in the present study showed no increase and perhaps a slight decrease in cortical thickness. This decrease in cortical thickness, consistent with reports of decreased dorsolateral prefrontal cortex volume in schizophrenia (25–27), suggests that the decrease in GAT-1-immuno­reactive cartridges in schizophrenic subjects may actually be greater than that reported in the present study. Finally, the use of systematic random sampling techniques and assessments of sampling adequacy in this study ensured that the measures obtained were reliable estimates of relative cartridge density in each subject.

The three groups of subjects were also matched on variables critical to interpreting postmortem studies, such as postmortem interval and tissue storage time, neither of which was correlated with cartridge density. In addition, although the subjects with schizophrenia differed from the other psychiatric subjects on measures of age at illness onset and duration of illness, these variables did not correlate with GAT-1-immuno­reactive cartridge density independent of age, suggesting that length of illness does not account for the differences in GAT-1-immunoreactive cartridge density between these two groups. Finally, decreased GAT-1-immunoreactive cartridge density was not associated with sex, suicide, or diagnosis of an alcohol or substance use disorder within the group of subjects with schizophrenia, suggesting that these variables did not affect our findings.

Understanding the significance of decreased cartridge density depends on whether this decrease is integral to the pathophysiology of schizophrenia or a result of treatment of the illness. We previously reported that the regional densities of GAT-1-immunoreactive cartridges did not differ between normal comparison subjects and nonschizophrenic psychotic subjects treated with antipsychotic medications (8). These findings suggested that the observed decrease in cartridge density in schizophrenic subjects was not a consequence of medical treatment. Three observations from the present study confirm this interpretation. First, in the schizophrenic subjects who were free of antipsychotic medications at the time of death or had never received them, average GAT-1-immunoreactive cartridge density was less than that of their matched normal comparison subjects across all cortical layers. Second, cartridge density did not differ between schizophrenic subjects on or off antipsychotic medications at the time of death (figure 5). Third, the density of labeled cartridges was not decreased in the dorsolateral prefrontal cortex of monkeys treated for 9–12 months with haloperidol and benztropine.

Although the decreased density of GAT-1-labeled axon cartridges might be due to a decrease in the number of chandelier neurons, the majority of studies have failed to detect a decrease in the number or density of prefrontal neurons in schizophrenia (5, 15, 28, 29). Moreover, neither the density nor the laminar distribution of neurons containing parvalbumin, a protein present in chandelier neurons (30, 31), was altered in the dorsolateral prefrontal cortex of 15 schizophrenic subjects (triads 1–15) examined in the present study (32).

Since the typical chandelier neuron furnishes 200–300 cartridges (9), it is possible that the axon arbor of these neurons is altered in schizophrenia such that each neuron gives rise to fewer cartridges. Such a reduction could occur in response to a decrease in the density of pyramidal neurons, the targets of chandelier neuron axon cartridges. However, consistent with other reports (29), the density of Nissl-stained pyramidal neurons in deep layer 3 was not decreased in the same 30 schizophrenic subjects examined in the present report (33). Furthermore, the size of parvalbumin-positive neurons, which includes the chandelier class, does not appear to be decreased in schizophrenia (32). Since a neuron’s somal size tends to be correlated with the total length of its axon (34, 35), these findings are not consistent with the notion that the axon arbor of chandelier neurons is reduced in schizophrenia.

Finally, a decrease in GAT-1 mRNA expression in chandelier cells could result in low (or absent) levels of the protein in axon cartridges so that they are un­detectable by immunocytochemical techniques. For example, during postnatal development of primate dorsolateral prefrontal cortex, other protein markers for axon cartridges decline substantially (36), without an apparent loss of terminals (37) or a change in the number of inhibitory synapses (38). In addition, expression of the mRNA for GAD₆₇, the synthesizing enzyme for GABA, is reduced in schizophrenia (5), and this change appears to be restricted to a subpopulation of GABA neurons, especially those located in the middle cortical layers (6). Together, these findings support the interpretation that the decrease in density of GAT-1-immunoreactive axon cartridges in schizophrenia represents a diminished amount of this protein in chandelier neurons. However, further testing of this hypothesis is necessary.

If the findings of the present study reflect a decrease in GAT-1 protein in axon cartridges, then this abnormality may represent a defect in chandelier cells such that they either underexpress GAT-1 mRNA or do not produce the GAT-1 protein or direct it to the axon terminal in a normal manner. Similar to other neurotransmitter membrane transporters, GAT-1 plays a central role in regulating the magnitude of effect and duration of action of GABA at postsynaptic sites (39, 40). Although less is known about the precise function of GAT-1, the dopamine membrane transporter appears to be the critical determinant of the availability of dopamine in the extracellular space (41). If GAT-1 functions similarly, then decreased GAT-1 protein in axon cartridges would be associated with increased GABA concentrations and enhanced inhibition at the axon initial segment of pyramidal cells. Indeed, in the dopamine membrane transporter knockout mouse, the absence of dopamine membrane transporter is associated with excess dopamine activity and a compensatory down-regulation of tyrosine hydroxylase, the rate-limiting enzyme in the synthesis of dopamine (41). As noted above, the expression of GAD₆₇ mRNA is decreased in the dorsolateral prefrontal cortex of schizophrenic subjects (5), and this decrease appears to be selective for a subpopulation of GABA neurons (6). By analogy to the dopamine system, these findings suggest that GAD₆₇ mRNA expression, and consequently GABA synthesis, may be down-regulated in chandelier neurons in order to compensate for excessive GABA activity at the axon initial segment of pyramidal cells. The notion that decreased GAT-1 represents a “primary” defect of chandelier neurons may be supported by the observation that both schizophrenic subjects on medication at the time of death and haloperidol-treated monkeys showed a trend toward an increase in the density of GAT-1-immunoreactive axon cartridges in the dorsolateral prefrontal cortex (figure 5). These observations suggest that some of the beneficial effects of antipsychotic medications may occur by reversing a GAT-1 deficiency.

Alternatively, decreased density of GAT-1-immuno­reactive axon cartridges in schizophrenia may reflect a “secondary” response to altered afferent input to the dorsolateral prefrontal cortex. For example, the number of neurons in the mediodorsal thalamic nucleus has been reported to be decreased in schizophrenia (42–44). Projections from this nucleus, the principal source of excitatory thalamic afferents to the dorsolateral prefrontal cortex, terminate primarily on the dendritic spines of pyramidal neurons in layers deep 3 and 4 (21, 22). In addition, the density of dendritic spines, markers of excitatory inputs, is decreased on the basilar dendrites of dorsolateral prefrontal cortex pyramidal neurons (45, 46), especially those located in layers deep 3 and 4 (46). Together, these findings support the hypothesis that the excitatory thalamic drive to dorsolateral prefrontal cortex pyramidal cells is diminished in schizophrenia and raise the possibility that the reported decrease in dorsolateral prefrontal cortex GAD₆₇ mRNA expression represents a compensatory down-regulation. For example, in primary visual cortex, decreased activity in thalamic inputs produced a decrease in GAD₆₇ expression that was most marked in the layers receiving these inputs (47). Although we are not aware of any data directly addressing this issue, if GAT-1 expression is co-regulated with GAD₆₇, our findings of a more prominent decrease in the density of GAT-1-immunoreactive cartridges, and in the density of neurons that express GAD₆₇ mRNA (6), in the middle layers of the dorsolateral prefrontal cortex in subjects with schizophrenia may represent a compensation for diminished input from the mediodorsal thalamic nucleus. Differentiating between these possible explanations for decreased GAT-1-immunoreactive axon cartridges in schizophrenia may provide ­further insight into the pathophysiology of the debilitating cognitive symptoms of this disorder.