Clinical, imaging, and postmortem evidence indicates that the neural circuits in which the ECx is embedded are involved in the pathogenesis of major psychiatric illnesses. Cortical regions such as the medial and lateral prefrontal cortex and the anterior cingulate gyrus are affected in schizophrenia (SZ) and bipolar 6 disorder (BD) (Bartha et al 1997; Benes et al 2000; Broadbelt et al 2002; Guidotti et al 2000; Rajkowska et al 2001; Thomas et al 2003; Woo et al 2004). Within the medial temporal lobe, several interconnected regions are thought to play a particularly important role in the pathophysiology of these disorders (Arnold 1997; Benes and Berretta 2000; Blumberg et al 2003; Cirillo and Seidman 2003; Goldsmith et al 1997; Heckers et al 1998; Kurachi 2003; Shenton et al 2001; Strakowski et al 2005; Torrey 2002). Of relevance is a large body of evidence pointing to anatomical, neurochemical, and functional abnormalities in the hippocampus of SZ and BD (Altshuler et al 1998; Benes and Berretta 2001; Benes et al 1997; Bogerts 1984; Heckers et al 2002; Konradi et al 2004; Pantazopoulos et al 2004; Zhang and Reynolds 2002).

Given its powerful interactions with these regions, it is not surprising that the ECx has also been found to be affected. In SZ, volume reductions of the parahippocampal gyrus (Bogerts et al 1985; Joyal et al 2002; McDonald et al 2000; Turetsky et al 2003) or selectively of the ECx (Falkai et al 1988) have been reported. Cytoarchitectural anomalies and abnormal distributions of glutamatergic fibers suggest abnormal development of the ECx (Akil and Lewis 1997; Arnold et al 1991; Falkai et al 2000; Jakob and Beckmann 1986, 1994; Kovalenko et al 2003; but see also Krimer et al 1997; Longson et al 1996). Decreases of synaptic proteins and abnormalities relative to specific neurotransmitter systems have also been detected in the SZ ECx (Bachus et al 1997; Benes 1995; Benes and Berretta 2001; Eastwood et al 1995, 1997; Hemby et al 2002; Mizukami et al 2002; Wolf et al 1995). Surprisingly, very few postmortem microscopic studies have thus far focused on the involvement of the ECx in BD (Law et al 2004; Webster et al 2002). Recently, results from imaging studies indicate that abnormalities relative to the volume of the temporal lobe might also be present in BD (Soares 2003; Wilke et al 2004).

Abnormalities of γ -aminobutyric acid (GABA)ergic transmission in SZ and BD seem to be among the most consistent findings in postmortem studies (Akbarian et al 1995; Bachus et al 1997; Beasley and Reynolds 1997; Benes 1995; Benes et al 1991, 1992, 1996, 1997, 1998; Danos et al 1998; Ishikawa et al 2004a, 2004b, 2005; Reynolds et al 1990; Simpson et al 1989; Todtenkopf and Benes 1998; Woo et al 1998, 2004), and strongly suggest the involvement of interneurons in multiple limbic regions. As a first step in testing the hypothesis that specific subpopulations of ECx GABAergic interneurons may be affected in BD and/or SZ, we chose to measure total numbers (Tn), numerical densities (Nd), and somata size of neurons expressing the calcium-binding protein parvalbumin (PVB) (Miettinen et al 1996; Wouterlood et al 1995; Wouterlood 2002). Abnormalities relative to PVB-immunoreactive (IR) neurons have been found in other regions, such as the hippocampus and the prefrontal cortex (Lewis et al 2001; Zhang and Reynolds 2002), and recent reports have highlighted the relevance of PVB as a marker for GABAergic abnormalities in major psychoses (Knable et al 2004; Reynolds et al 2004; Torrey et al 2005).

In the present study, PVB-IR neurons were counted in the ECx of normal control, BD, and SZ subjects. We hypothesized that SZ and BD would show reduced PVB-IR neuron Tn and Nd in specific ECx subdivisions.

Table 1 Sample Demographic and Descriptive Characteristics

Table 2 Lifetime Exposure to Psychotropic or Neurotropic Drugs

Several regions from each brain were examined by a neuropathologist. Brains with evidence for gross and/or macroscopic changes consistent with Alzheimer’s disease, cerebrovascular accident, ethanol and drug abuse, or other potentially confounding factors were excluded from the study.

PVB was detected with standard immunocytochemical methods. Sections from a 40-μm compartment from each brain were randomly assigned to a six-well staining dish. Sections were washed in .1 mol/L PB .5% Triton-X, boiled for 3 min in PB with 1:100 Antigen Unmasking Solution (Vector Labs, Burlingame, California), and incubated with a monoclonal anti-PVB antibody for 48 hours at 4°C (1:10,000; P3088, lot #10K4846; clone PARV-19, ascites fluids; Sigma-Aldrich, St. Louis, Missouri). Immunoblot characterization showed a single band corresponding to 12 kD (information kindly provided by Sigma-Aldrich). This antibody has been found to react with PVB (12 kDa) from human. It does not react with other members of the EF-hand family of calcium binding proteins. Sections were incubated with biotynylated secondary serum horse anti mouse (1:500; Vector Labs), followed by streptavidin (1:5,000; Vector Labs). A solution of diaminobenzidine (.02%; Sigma-Aldrich), nickel sulphate (.08%), and hydrogen peroxide (.006%) was used for detection of the antigen/antibody complex. All steps were followed by rinses with PB-Triton. Immunostained sections were mounted on gelatin-coated glass slides, coverslipped, and coded for quantitative analysis blinded to diagnosis. Sections from all brains included in the study were processed simultaneously within the same ICC session to avoid procedural differences. Care was taken that each six-well staining dish contained sections from SZ, BD, and normal control subjects and was carried through each immuno-cytochemistry step for the same duration of time to avoid sequence effects. Identical protocols for PVB ICC with the omission of the first or secondary antibodies did not yield detectable signal. The patterns of PVB-IR in the ECx as well as the amygdala were identical to those described by other laboratories (Mikkonen et al 1997; Sorvari et al 1995).

For volume measurements, the ECx was traced in Nissl-stained serial sections encompassing the entire structure (10–12 sections/ subject) and sampled with the systematic random sampling criteria described (see Methods). Included in the region identified as ECx are the rostral (ECx-R), lateral (ECx-L), intermediate (ECx-I), caudal (ECx-C), olfactory, and medial intermediate subdivisions of the ECx. For measurement of PVB-IR neuron soma size, the surface area of IR neurons was measured on immunostained sections by outlining the perimeter of each IR somata.

For estimates of Tn and Nd (cells/mm³) of ECx PVB-IR neurons, the entire ECx was outlined in a complete set of serial PVB-immunostained sections from each subject (10–12 sections/ subject) with a 1.6× objective. Adjacent Nissl-stained sections, from which volume estimates were derived, were used for confirmation of the ECx borders. In each section, the ECx was thoroughly scanned within its full x-, y-, and z-axes with a 40× objective. All detectable PVB-IR neuron somata were counted.

As a second step, Nd of PVB-IR neurons were estimated within ECx subregions (ECx-R, ECx-L, ECx-I, and ECx-C; Figure 1). This step was intended to investigate whether potential differences between diagnosis groups might originate from distinct ECx subdivisions and/or layers. Nd of PVB-IR neurons were measured in a sub-sample of three sections/subject, chosen to represent the fully formed rostral, intermediate, and caudal portions of the ECx (Figure 1). The olfactory, medial intermediate, and caudal limiting ECx subdivisions were not considered for data collection because preliminary studies had shown that they contain only very sparse PVB-IR neurons. Again, the borders of each region were outlined with a 1.6× objective. Each subdivision and layer was systematically scanned through the full x-, y-, and z-axis with a 40× objective in order to determine the location of each PVB-IR neuron with respect to these subdivisions and layers.

Figure 1 Diagrams representing typical rostro-caudal levels chosen for analysis of parvalbumin-immunoreactive (PVB-IR) somata numerical densities (Nd) within entorhinal cortex (ECx) main subdivisions and layers. The following specific criteria were used to choose (more ...)

PVB-immunostained and adjacent Nissl-stained sections were used to identify ECx subdivisions according to specific cytoar-chitectonic criteria and nomenclature described by Insausti et al (1995; Mikkonen et al 1997; Tunon et al 1992) (Figure 1). Within these subdivisions, six cortical layers were identified, with a 4× objective, on the basis of the distribution of PVB immunoreac-tivity (Mikkonen et al 1997) and confirmed on adjacent Nissl-stained sections according to the criteria of Insausti et al (1995) (Figures 1 and 2). Particular care was taken to match the rostro-caudal level of each section across all subjects (see legend in Figure 1). The most rostral section contained the ECx-R and the rostral portion of the ECx-L. The middle section contained the ECx-I and the caudal portion of the ECx-L. The third and most caudal section contained the ECx-C and was located rostrally to the transitional zone that defines the distinct features of the caudal limiting subdivision of the ECx.

Figure 2 Photomicrographs showing parvalbumin-immunoreactive (PVB-IR) immunostaining in the lateral entorhinal cortex (ECx-L) from a normal control subject. (A) Six cortical layers were identified, using a 4 × objective, on the basis of the distribution (more ...)

Quality of PVB-IR was tested in a blinded pilot study. Four sections from each ECx were subjectively rated on a scale from 1 to 5, with intensity of immunoreactivity, immunolabeling penetration along the z axis, and immunoreactivity distribution pattern as the main criteria. The quality of PVB-immunostaining was entirely consistent both within and between groups and optimal for readily and unambiguously recognizing IR neurons throughout the thickness of each section.

Volume of the ECx and Tn, Nd, and somata size of PVB-IR neurons were considered for analysis with a stepwise linear regression process. An initial baseline model that included diagnostic group as the main comparison and age, gender, interaction of age and gender, hemisphere, and interaction of gender and hemisphere as covariates was adopted in consideration of following data. In the cohort used for this study, women were older than men (p = .03), and the left hemisphere was more frequently represented in men than in women. The left ECx was larger than the right (p = .04). Differences in size between men and women and the interaction of hemisphere and gender on the volume of ECx were not significant. The PVB-IR neurons Tn and Nd were significantly higher in the left hemisphere (p = .01 and p = .03, respectively) and in men (p = .03 and p = .04, respectively).

The effects of postmortem time interval (PMI; hours), brain weight, age at illness onset, years of illness, cause of death, and lifetime as well as final 6 months’ (Table 3) exposure to antipsychotic drugs and lithium treatment were tested systematically within the framework of the baseline stepwise regression model just described for each of the outcome variables and included in the model if they significantly improved the model goodness-offit. Cause of death was categorized as acute (e.g., myocardial infarction) or chronic (e.g., cancer). In addition to drug exposure levels, we evaluated the potential effects of treatment-adherence to pharmacological treatment and exposure to drugs other than antipsychotic or lithium (Table 2) using analysis of variance in a separate analysis. The PVB-IR soma size was considered both as an outcome measure and as covariate because of its potential to affect Tn and Nd estimates (Guillery and Herrup 1997). The possible confounding effect of cell size on Tn were also tested by applying the same statistical analyses to Abercrombie corrected values (Abercrombie corrected Tn = Tn × (z/cs × z); where cs 3 average cell size measured) (Abercrombie 1946).

Table 3 Exposure to Psychotropic or Neurotropic Drugs During the Last 6 Months of Life

Statistical analyses were performed with JMP v5.0.1a (SAS Institute, Cary, North Carolina).

Figure 3 The volume of the entorhinal cortex (ECx) is not altered in bipolar disorder (BD) and schizophrenia (SZ). Scattergram showing volumetric values (mm3; logarithmic transformation) for the ECx of normal control, SZ, and BD subjects. No statistically significant (more ...)

Effect of Confounding Variables Differences between hemispheres (left > right) were retained in diagnostic groups, although they were not always significant. None of the covariates tested within the baseline stepwise regression model (i.e., PMI, brain weight, age of onset, cause of death, years of illness, and lifetime as well as last 6 months’ antipsychotic and lithium exposure) had significant effects on Tn and Nd of PVB-IR neurons in the entire ECx and Nd of PVB-IR neurons in the ECx subregions considered. Furthermore, correlations between CPZ-eq dose (within the last 6 months or throughout the illness) and the main outcome variables were not significant. Similarly, no significant correlations were detected between lithium treatment (within the last 6 months or throughout the illness) and the main outcome variables. Therefore, the effect of diagnostic group on Tn and Nd of PVB-IR neurons in the entire ECx and Nd of PVB-IR neurons in each of the ECx subregions considered was tested with the baseline model only.

In contrast to Tn and Nd, somata size was affected significantly by lifetime grams of antipsychotic drug exposure. For instance, within the baseline linear model used for this analysis, the effect of CPZ-eq dose on PVB-IR neuron size in the entire ECx was significant (p = .01). This covariate was included in the final model testing the effect of diagnostic group on somata size.

Soma Size of PVB-IR Neurons The BD subjects showed no significant changes of PVB-IR soma size in the entire ECx or in any of its subdivisions. In SZ subjects, somata size was not altered significantly when considered within the entire ECx (p = .1), but it was significantly smaller in the ECx-I (p = .009). No changes were detected in the other subdivisions (Figure 4). As previously noted, lifetime grams of CPZ-eq antipsychotic exposure was included in the model because it showed a significant effect on cell size. This variable was positively and significantly correlated with cell size throughout the ECx (p = .001) as well as in several of the ECx subdivisions (ECx-R superficial layers, p = .05; ECx-R deep layers, p = .03; ECx-L deep layers, p = .01; ECx-I superficial layers, p = .01; ECx-C superficial layers, p = .03; ECx-C deep layers, p = .03).

Figure 4 Size of parvalbumin-immunoreactive (PVB-IR) neurons somata in the entorhinal cortex (ECx). In bipolar disorder (BD), no changes relative to the size of PVB-IR neuron somata were detected in any of the ECx subdivisions investigated. In schizophrenia, a (more ...)

Tn of PVB-IR Neurons Within the Entire ECx The BD subjects showed a significant reduction of total numbers of PVB-IR neurons (p = .02). In contrast, no significant change was detected in the SZ group (Figure 5). Despite a lack of correlation of this measure with pharmacological treatment, the mean (± SEM) Tn of PVB-IR neurons ranked subjects with good adherence (5.71 ± .4) > subjects with poor adherence (5.09 ± .4) > non-treated patients (one SZ and three BD; 3.93 ± .5), suggesting a subtle effect of drug exposure (analysis of variance; p < .05).

Figure 5 Total number (Tn) and numerical density (Nd) of parvalbuminimmunoreactive (PVB-IR) neurons are significantly decreased in the entorhinal cortex (ECx) of bipolar disorder (BD). (A) A significant decrease of Tn of PVB-IR neurons (logarithmic transformation; (more ...)

Given the cell size decrease detected in SZ, Abercrombie correction for cell size was applied. The effect of cell size was also tested by including it as covariate in the stepwise model. Both of these strategies showed that results relative to Tn were not affected by cell size.

Nd of PVB-IR Neurons Within the Entire ECx In BD, PVB-IR neuron Nd is significantly reduced (p = .01) (Figure 5). Although the SZ group also seemed to show a reduction (Figure 5), the difference was not significant when tested for the effect of group only and became even less so when the other covariates included in the baseline model were considered (p = .9).

Nd of PVB-IR Neurons Within ECx Subdivisions and Layers In BD, significant decreases of PVB-IR neuron Nd were detected in the superficial layers of the ECx-L (p = .02), ECx-I (p = .04), and ECx-C (p = .01) (Figures 6 and 7; Table 4). Again, the SZ group did not show significant changes in Nd of PVB-IR neurons. Although most SZ subjects showed low PVB-IR neuron Nd in the superficial layers of ECx-L, three subjects exhibited high values (Figure 6). Review of the medical records did not provide distinguishing features that might define them as a distinct subpopulation. The standard confounding variables tested in this analysis, pharmacological treatment in particular, as well as other factors such as nicotine exposure and electroshock treatment did not seem likely to explain the segregation of these three subjects. Notably, the exclusion of these three subjects did not change the outcome of the linear model because of the strong effect of age and gender.

Figure 6 In bipolar disorder (BD), a significant decrease of numerical density (Nd) of parvalbumin-immunoreactive (PVB-IR) was detected in the superficial layers of the lateral entorhinal cortex (ECx-L). Within the ECx-L, reductions of PVB-IR neurons Nd were restricted (more ...)

Figure 7 Lower numerical densities (Nd) of parvalbumin-immunoreactive (PVB-IR) neurons in the superficial layers of the intermediate entorhinal cortex (ECx-I) and caudal ECx (ECx-C) of bipolar disorder (BD). In BD, significant reductions of PVB-IR neurons were (more ...)

Table 4 Numerical Densities of PVB-IR Neurons in the Superficial and Deep Layers of Distinct ECx Subregions in Normal Control, BD, and SZ Subjects

These results indicate that the Tn and Nd of PVB-IR neurons are decreased in the ECx of BD patients. To our knowledge, this is the first evidence for a disruption of inhibitory intrinsic PVB-IR neurons in the ECx of BD subjects. No significant changes were detected in SZ subjects, with the exception of a cell size decrease restricted to the ECx-I.

Decreases of PVB-IR neurons could result from several, non-reciprocally exclusive factors, such as downregulation of protein expression, developmental abnormalities, or actual cell death. Temporary changes in expression of calcium-binding proteins have been reported in experimental animal models (e.g., Johansen et al 1990; Magloczky and Freund 1995). Given the pivotal role that PVB plays in regulating several Ca^{+ +} -dependent aspects of neuronal physiology, it is reasonable to posit that its downregulation would result in profound functional abnormalities of the neurons expressing it as well as of the neural networks in which these neurons are embedded (Caillard et al 2000; Chard et al 1993; Lee et al 2000; Schwaller et al 2002; Vreugdenhil et al 2003). Developmental abnormalities affecting the ECx (Arnold et al 1991; Falkai et al 2000; Jakob and Beckmann 1986, 1994; Kovalenko et al 2003; Longson et al 1996) might contribute to lower PVB-IR neuron Tn and Nd. The late postnatal maturation of PVB-IR neurons in several brain regions supports this hypothesis (Berdel and Morys 2000; Lopez-Tellez et al 2004; Reynolds and Beasley 2001). Notably, the human ECx follows a protracted postnatal development, and ECx PVB-IR neurons are not detectable until the 5th month after birth (Grateron et al 2002, 2003). Finally, the possibility that cell death might be a contributing factor cannot be excluded. Existing evidence pointing to neuronal loss in BD (Manaye et al 2005; Rajkowska 2002) is consistent with this possibility. Regardless of its causes, a decrease of PVB-IR somata might be reasonably interpreted as reflecting altered functioning of this particular interneuronal population (Eyles et al 2002). Thus, these findings will be, from here on, referred to as a “decrease” of PVB-IR neurons to denote a decrease of function and/or number of cells.

The discrepancy between findings in BD and SZ might point at intriguing differences between these two diseases. We speculate that overlapping genetic vulnerabilities, possibly affecting PVB expression, might result in divergent phenotypes, depending on the timing, duration, and nature of concurrent developmental factors impacting on similar neural circuits as well as on other gene clusters concurrently affected.

Parvalbumin-immunoreactive neurons correspond to fast spiking inhibitory neurons (Jones and Buhl 1993) and form basket- or cartridge-like clusters of synaptic terminals around the somata and proximal axonal segments of projection neurons (Mikkonen et al 1997; Wouterlood et al 1995). These features enable them to exert a powerful control over the activity of output neurons. Furthermore, their dendritic arborization spans across ECx layers (Mikkonen et al 1997), suggesting that these neurons might be involved in the integration of input and output streams of information processed within the ECx. Such integration has been suggested to mediate functions such as memory processing and novelty detection (Witter et al 2000). Together, these considerations suggest that a decrease of PVB-IR neurons might result in altered information processing within ECx intrinsic networks and disrupted outflow of information to the hippocampal formation.

Finally, it is important to consider that the cingulate cortex (area 25) and the orbito-frontal cortex send heavy projections primarily to the superficial layers of the ECx-L and have been implicated in BD (Bakchine et al 1989; Benes et al 2001; Blumberg et al 1999; Cotter et al 2005; Drevets 1999; Eastwood and Harrison 2001; Starkstein et al 1991). We suggest that altered outflow of emotional information from these cortical regions might be further distorted by abnormal activity in this region. These considerations are consistent with the hypothesis that the pathophysiology of BD might emerge from interactions of abnormalities affecting interconnected networks of cortico-limbic regions, including several prefrontal cortex regions, the amygdala, ECx ,and hippocampal formation (Benes and Berretta 2001; Blumberg et al 2002; Phillips et al 2003; Rajkowska 2002). It follows that changes related to PVB-IR neurons in the ECx are not in themselves likely to account for the full clinical manifestation of this disease. The spread of values for Tn and Nd of these neurons in normal control subjects supports this idea. As part of pathological changes affecting cortico-limbic networks, a disruption of ECx intrinsic networks and outflow of information to the hippocampal formation might contribute to particular clinical features attributed to BD, including impairment of sensory gating, visuo-spatial processing, visually dependent sustained attention, and verbal memory (Baker et al 1990; Chang et al 2004; Clark et al 2005; El-Badri et al 2001; Kieseppa et al 2005; Olincy and Martin 2005; Saccuzzo and Braff 1986).

We thank the Harvard Brain Tissue Resource Center, funded by NIH R24MH068855 and directed by Francine M. Benes, M.D., Ph.D., for providing the tissue used in these investigations.