Density of Glutamic Acid Decarboxylase 67 Messenger RNA–ContainingNeurons That Express the N-Methyl-D-AspartateReceptor Subunit NR2A in the Anterior Cingulate Cortex in Schizophreniaand Bipolar Disorder FREE

The anterior cingulate cortex (ACCx) (Brodmann's area 24) is a key regionof the large-scale neural network that comprises the limbic system, the dorsolateralprefrontal region, and the motor and premotor cortices. This distributed neuralsystem mediates a wide range of functions, such as affective regulation, motivation,selective attention, separation calls, executive control, and the dynamicorchestration of motor programming. The ACCx is involved in mediating thesefunctions via postulated capabilities, such as error detection and conflictmonitoring.^1- 5 Becauseperturbations of many aspects of these functions are commonly seen in schizophreniaand bipolar disorder, it is perhaps not surprising that converging lines ofevidence from postmortem and neuroimaging studies^6- 11 haveconsistently demonstrated that the ACCx is structurally and functionally alteredin these disorders.

γ-Aminobutyric acid (GABA) interneurons play an important rolein information processing in the cerebral cortex. Disturbances of these neuronshave been strongly implicated in the pathophysiology of schizophrenia.^12- 19 Increasing,albeit still somewhat limited, evidence^7,20- 23 suggeststhat GABAergic function may also be perturbed in bipolar disorder. In fact,some of the parameters of GABA neurotransmission that have been examined seemto be even more severely altered in subjects with bipolar disorder than inthose with schizophrenia. For example, it has been shown in the ACCx thatthe density of cells with a nonpyramidal shape, putative GABA interneurons,and terminals is decreased in schizophrenia and bipolar disorder, but theextent of this reduction is considerably greater in the latter condition.^7,21 In addition, GAD₆₅-immunoreactiveterminals have also been found to be substantially reduced in the ACCx ofbipolar, but not schizophrenic, subjects.²⁴

Converging lines of clinical and preclinical observations^25- 29 stronglysuggest that disturbances of glutamatergic neurotransmission contribute tothe pathophysiology of schizophrenia. Furthermore, it has been postulated,largely based on animal studies, that such disturbances may involve hypofunctioningof N-methyl-D-aspartate (NMDA) receptorson GABA interneurons.^30- 31 TheNMDA receptor complex is a heteromeric structure composed of different subunits.Among them, the NR_2A subunit is abundantly present in the adultcerebral cortex. It has also been implicated in the pathophysiology of schizophrenia.For example, mice lacking the NR_2A subunit demonstrated an increasein the release of dopamine in the striatum.³² Behaviorally,these animals exhibited hyperlocomotion, which could be attenuated by treatmentwith antipsychotic agents. Furthermore, NMDA-mediated GABA release in theseanimals was markedly decreased. In this study, as a first step to explorethe question of whether glutamatergic innervation of GABA interneurons viathe NMDA receptor may be altered in schizophrenia and bipolar disorder, weused a double in situ hybridization procedure to simultaneously examine theexpression of messenger RNA (mRNA) for the NR_2Asubunit, labeledwith ³⁵sulfur ([³⁵S]), in cells containing GAD₆₇ mRNA, labeled with digoxigenin (DIG), in the ACCx from normal control,schizophrenic, and bipolar subjects.

A cohort of 51 human brains obtained from the Harvard Brain Tissue ResourceCenter at McLean Hospital was used in this study and included 17 normal controls,17 subjects with schizophrenia, and 17 subjects with bipolar disorder (Table 1). Each of the schizophrenic subjectswas matched to a subject with bipolar disorder and to a normal control subjecton the basis of age, postmortem interval, and, whenever possible, sex, hemisphere(ie, right vs left), and pH. The female-male ratio was 7:10 for the bipolardisorder group and 8:9 for the schizophrenia and normal control groups. Theright hemisphere–left hemisphere ratio was 10:7 for the schizophreniagroup and 8:9 for the bipolar disorder and normal control groups. The mean± SD freezer storage time of brains was not significantly differentamong the normal control (1391 ± 1012 days), schizophrenia (1766 ±958 days), and bipolar disorder (1847 ± 1084 days) groups (F_2,48 = 0.97; P = .39). Measurements of tissue pHwere available for 12 of 17 cases in the schizophrenia and bipolar disordergroups and for 15 of 17 cases in the normal control group. The mean ±SD pH was not different among the 3 groups (normal control group: 6.52 ±0.27; schizophrenia group: 6.54 ± 0.31; bipolar disorder group: 6.45± 0.24).

Psychiatric diagnoses were established using a retrospective reviewof medical records and an extensive family questionnaire that included themedical, psychiatric, and social history of the subjects. For the diagnosisof schizophrenia, the criteria of Feighner et al³³ wereused, and the diagnoses of schizoaffective and bipolar disorder were madeaccording to DSM-III-R criteria. Of the 17 schizophrenicsubjects, 3 (cases B2166, B4875, and B4907) had a diagnosis of schizoaffectivedisorder, whereas the remaining cases has a diagnosis of schizophrenia. Threeof the 17 schizophrenic subjects (cases B3146, B4875, and B4256) were nottaking antipsychotic medications at the time of death. In the bipolar disordergroup, 9 subjects were taking antipsychotic medications at the time of death.The dose of antipsychotic drugs that subjects with bipolar disorder (267.4± 383.5 mg) were receiving (expressed as chlorpromazine-equivalentdose) was less than half that of the schizophrenia group (618.3 ± 809.7mg). Some subjects in both disease groups were also taking concomitant psychotropicmedications, such as mood stabilizers, antidepressants, or anxiolytics (Table 1). No subject in the normal controlgroup was receiving any psychotropic agents at the time of death.

Tissue blocks (3 mm thick) from Brodmann's area 24 were removed fromfresh brain specimens at the level of the rostrum of the anterior cingulategyrus between the points at which the gyrus curves above and below the corpuscallosum.⁷ The blocks were immediately fixedin 0.1% paraformaldehyde in ice-cold 0.1M phosphate buffer (pH 7.4) for 90minutes, immersed in 30% sucrose in the same buffer overnight, and then frozenin Tissue-Tek OCT embedding meduim for frozen tissue (Sakura Finetek USA Inc,Torrance, Calif) on dry ice. Tissue blocks were then sectioned at a thicknessof 10 µm on a cryostat. Two sections per subject and therefore 6 sectionsper matched triplet were used for in situ hybridization. The 6 sections fromeach triplet were mounted on 3 slides as follows: (1) normal control + schizophrenia,(2) normal control + bipolar disorder, and (3) schizophrenia + bipolar disorder.This method of mounting sections controls for potential variability of hybridizationsignals between slides. All mounted sections were stored at –70°Cuntil riboprobe labeling was performed.

Radiolabeled Complementary RNA Probe for NR_2A mRNA. The complementary RNA (cRNA) probes for the NR_2A subunit(provided by Christine Konradi, PhD) were transcribed in vitro from linearizedcomplementary DNA (cDNA) subclones encoding the rat NMDA NR_2A subunit.The specificity of the probe was verified by Northern blot analysis (datanot shown). The probe was derived from a cDNA spanning nucleotides 1185 to2154 (GenBank Accession No. M91561) within the coding region of the subunit.A corresponding sense probe was used as a control. Radiolabeled cRNA probewas prepared by first drying down [³⁵S]UTP (500 µCi/mL ofprobe, PerkinElmer Life and Analytical Sciences Inc, Boston, Mass) in a DNAspeed vac (Savant, Farmingdale, NY); 100 ng/µL of the cDNA template,0.1M dithiothreitol, 3 U/µL of RNasin, 5mM NTPs, 0.8 U/µL of T3or T7 RNA polymerases (for antisense and sense probes, respectively), and5× transcription buffer were then added. The transcription mixture wassubsequently incubated at 37°C for 1 hour. The cDNA template was digestedby incubating the mixture with R1Q DNase at 37°C for 15 minutes. UnincorporatedNTPs were removed by running the mixture through a push column (NucTrap; Stratagene,La Jolla, Calif). The eluate was collected, and probe concentration was determinedby scintillation counting. The probe was stored at –20°C until use.

DIG-Labeled GAD₆₇ mRNA Probe. The DIG-UTP–labeled cRNA probes were transcribed using 100 ngof full-length, linearized human cDNA clones inserted in a bluescript vector(provided by Allan Tobin, PhD, and Niranjula Tillakarantne, PhD, Departmentof Physiological Sciences, University of California at Los Angeles) in thepresence of 0.1M dithiothreitol; 3 U/µL of RNasin; 0.8 U/µL ofT3 and T7 RNA polymerases; 10mM ATP, CTP, and GTP; 6.5mM UTP; and 3.5mM DIG-labeledUTP (Boehringer Mannheim, Indianapolis, Ind). The mixture was incubated at37°C for 1 hour. The cDNA template was digested with RQ1 DNase. Probeconcentration was determined using a standard with known concentrations.

To ensure adequate tissue penetration, the GAD₆₇ probe washydrolyzed to 0.8 kilobase (kb) with an equal volume of sodium bicarbonate–sodiumcarbonate buffer (pH 10.2; 40mM sodium bicarbonate and 60mM sodium carbonate)at 60°C for 6 to 10 minutes. The reaction was stopped by adding 0.08 volof 2M sodium acetate in 6.25% glacial acetic acid. Probes were then reconstitutedin a hybridization buffer consisting of 50% formamide, 0.1% yeast transferribonucleic acid, 10% dextran sulfate, 1× Denhardt solution, 0.5M EDTA,0.02% sodium dodecyl sulfate, 4× isotonic sodium chloride solution–sodiumcitrate buffer, 10mM dithiothreitol, and 0.1% single-stranded DNA at a finalconcentration of 0.4 ng probe/µL hybridization buffer. Before hybridization,mounted tissue sections were air-dried and warmed to room temperature. Theywere then postfixed in 4% paraformaldehyde for 10 minutes and incubated in0.1M tetraethylammonium for 5 minutes at room temperature before being dehydratedin a graded series of ethanol. Probes were then added to slides for hybridizationin a prewarmed, humidified dish. Sections were covered with coverslips andincubated at 55°C for 3 hours. At the end of hybridization, coverslipswere soaked off in 4× isotonic sodium chloride solution–sodiumcitrate in the presence of 100 µL of ßMer alcohol. Tissue wasthen incubated in 0.5M sodium chloride/0.05M phosphate buffer for 10 minutes,0.5M sodium chloride with 0.025 mg/mL of ribonuclease (pancreatic) at 37°Cfor 30 minutes, followed by a high-stringency wash with a solution containing50% formamide, 0.5M sodium chloride/0.05M phosphate buffer, and 100 µLof ßMer at 63°C for 30 minutes. Sections were finally washed overnightin 0.5× isotonic sodium chloride solution–sodium citrate with20mM ßMer alcohol at room temperature.

After incubation in blocking buffer (100mM Tris hydrochloride, 150mMsodium chloride [pH 7.5], 3% normal goat serum, and 0.3% Triton X-100), sectionswere incubated overnight at 4°C in buffer containing 1:200 dilution ofalkaline phosphatase–conjugated sheep α-DIG antibody (Roche Diagnostics,Indianapolis). Sections were then incubated in an alkaline phosphatase substrate(Vector Red; Vector Laboratories, Burlingame, Calif), at room temperaturefor 40 minutes in complete darkness.

It was determined that sufficient autoradiographic signal had developedafter the slides were apposed to x-ray film (Kodak BioMax MS; Kodak, Rochester,NY) for 10 days. The slides were then dipped in emulsion (Kodak NTB-2; Kodak),air-dried, and stored at 4°C in darkness for 5 weeks. After developmentin the dark with developer (Kodak D-19; Kodak), slides were counterstainedwith methyl green and coverslipped.

All microscopic analyses were conducted under strictly blind conditions.[³⁵S] labeling of NR_2A mRNA appeared as clusters ofsilver grains after processing for emulsion autoradiography (Figure 1). After counterstaining with Vector Red, DIG labeling canbe visualized as a red-brown reaction product under a brightfield microscopeor as a fluorescent emission in the red range. Neurons that were single labeledwith DIG (Figure 1A) and those thatwere double labeled with DIG and [³⁵S] (Figure 1B) were identified on images captured on a computer screenusing a microscope (Laborlux; Leica Microsystems, Wetzlar, Germany), whichwas equipped with a solid-state charge-coupled device video camera connectedto an image analysis system (Bioquant Nova; R&M Biometrics, Memphis, Tenn).Using a 100× oil immersion objective lens at a final magnification of×1000, the distributions of single- and double-labeled neurons in a250-µm-wide column extending from the pial surface to the border betweenlayer 6 and the subcortical white matter were obtained for each section. Neighboringsections were stained with cresyl violet for determination of laminar boundaries.Densities of single- and double-labeled neurons for each cortical layer werethen obtained by dividing cell counts by laminar areas. Intrarater reliability,as assessed by counting and recounting profiles in the same column, was establishedto be 93% to 97% before the actual data collection process began.

To quantify the expression level of mRNA for the NR_2A subunitin individual GABA cells, the area occupied by each grain cluster was outlinedusing a cursor displayed on the computer monitor. For each cluster, this quantificationwas performed according to the principle of including the largest number ofgrains in the smallest possible area. The cluster area was measured by highlightingthe grains with a thresholding subroutine. This threshold was held constant,and the light intensity was adjusted to ensure that the size of the grainswas neither underrepresented nor overrepresented. This procedure was consistentlyfollowed throughout the entire study. The area covered by autoradiographicgrains in the cluster area was automatically computed based on the thresholdvalue and was represented as a pixel count for NR_2A transcriptexpression level. The pixel count was expressed as a function of cluster area(numerical density). By subtracting the background grain density (ie, pixelcount of the area covered by autoradiographic grains per unit area in squaremicrometers in the white matter), the corrected NR_2A expressionlevel was obtained. The average NR_2A expression level in GABA interneurons(ie, cells positive for GAD₆₇ mRNA) for each cortical layer foreach case was then computed. Intrarater reliability in grain density measurements,which was accessed by repeating the procedures described previously hereinon the same clusters, was determined to be consistently greater than 95% beforethe actual data collection process.

The numerical densities of single-labeled (GAD₆₇ mRNA only)and double-labeled (GAD₆₇ and NR_2A mRNA) neurons andthe amount of mRNA for the NR_2A subunit in GAD₆₇ mRNA–containingneurons were compared among groups across layers 2 through 6 using repeated-measuresanalysis of variance (ANOVA), with diagnosis (ie, schizophrenia vs controland bipolar disorder vs control) and layer as main effects. For post hoc analyses,2-tailed paired t tests were used. The Bonferroniprocedure was used to correct for type 1 error as a result of multiple comparisons(layers 2, 3, 5, and 6). Therefore, the α level for significance forall t tests was P = .01(ie, .05 ÷ 4). Layer 1 was not included in the analyses because therewere no GAD₆₇ mRNA–containing neurons with co-expressed NR_2A subunit mRNA in this lamina. To evaluate the potential effects ofconfounding variables, such as age, sex, postmortem interval, brain pH, freezerstorage time, hemispheric laterality, and exposure to antipsychotic medications(expressed as the chlorpromazine-equivalent dose), simple Pearson correlationswere obtained for the individual groups and for the control group combinedwith the schizophrenia and bipolar disorder groups. In addition, an analysisof covariance (ANCOVA) was performed to understand how these confounding variablesmight have affected our results. Because none of the conclusions derived fromour findings were affected by the ANCOVA analysis, only results from repeated-measuresANOVAs are reported. Effects of hemispheric laterality on our findings wereevaluated by using 2-tailed unpaired t tests to comparethe measures from the 2 hemispheres within individual groups and when casesfrom the control group were combined with those from the schizophrenia andbipolar disorder groups.

Neurons that express GAD₆₇ mRNA appeared to be distributedmore or less evenly across all layers in the ACCx, except for layer 1, wherethe density of these neurons was low. In the entire population of GAD₆₇ mRNA–expressing neurons, those that co-expressed NR_2A mRNA seemed to be most concentrated in layers 3 to 5, whereas the densityof these neurons seemed to be slightly lower in layers 2 and 6 (Figure 2).

Overall, the repeated-measures ANOVA models revealed a significant diagnosiseffect in the schizophrenia group (F_1,32 = 10.19; P = .003). Furthermore, this effect seemed to be layer specific (diagnosis× layer interaction, F = 3.27; P = .03). Thus,the density of GAD₆₇ mRNA–expressing neurons showed the mostprominent change in layer 2 (Figure 3A),with a 53% reduction in the density of these neurons compared with controlsubjects (t = 4.41; P<.001).Besides layer 2, the density of these neurons was also significantly decreasedin layer 5, although the magnitude of reduction (28%) was more modest (t = 2.45; P = .01). The observeddecreases in the density of GAD₆₇ mRNA–expressing neuronsdid not seem to be artifactually related to differences in cortical thicknessbetween the 2 groups because the mean ± SD thickness of the ACCx inthe schizophrenia (1753.45 ± 269.1 µm) and control (1745.5 ±267.4 µm) groups was not significantly different (t = 0.009; P = .93). In subjects with bipolardisorder, the repeated-measures ANOVA initially did not demonstrate a significantdiagnosis effect (F_1,32 = 3.44; P = .07).On closer inspection of the data, it was noticed that the numerical densityof the GAD₆₇ mRNA–expressing cells in layer 2 in a subjectwith bipolar disorder (patient B3817) was 3 SD above the mean density forthat layer. This case and its matched control (case B5122) were, therefore,excluded from all of the numerical density comparisons between controls andsubjects with bipolar disorder reported herein. Case B3817 was the only casein which the mean neuronal density in any layer was beyond 3 SD. After theremoval of these 2 cases, the diagnosis effect in the repeated-measures ANOVAmodel was statistically significant (F_1,30 = 4.22; P = .04). When individual layers were examined, the density of GAD₆₇ mRNA–expressing cells in layer 2 was found to be significantlydecreased by 35% in the bipolar disorder group (t =4.12; P<.001). This reduction was smaller in magnitudethan the 53% reduction observed in layer 2 in subjects with schizophrenia(Figure 3A). This apparent differencein the magnitude of reduction in the numerical density of neurons betweenthe 2 subject groups was not statistically significant (t = 1.97, P = .06). Besides the reductionin layer 2, the numerical density of GAD₆₇ mRNA–expressingneurons was essentially unchanged in all other layers in subjects with bipolardisorder. As in the schizophrenia group, there was no statistically significantdifference in mean ± SD cortical thickness between subjects with bipolardisorder (1755.3 ± 318.2 µm) and control subjects (1745.5 ±267.4 µm) (t = 0.097; P =.92).

The effect of diagnosis on NR_2A-expressing GAD₆₇-positivecells was statistically significant in the schizophrenia (F_1,32 =8.97; P = .005) and bipolar disorder (F_1,32 = 4.42; P = .04) groups. In the schizophreniagroup, neuronal density was significantly decreased by 73% (t = 3.08; P = .007) and 52% (t = 2.95; P = .009) in layers 2 and 5, respectively,whereas 37% and 40% reductions in layers 3 and 6, respectively, did not achievestatistical significance under the stringent Bonferroni correction (t = 2.62; P = .02 and t = 2.27; P = .04, respectively) (Figure 3B). In the bipolar disorder group,the numerical density of GAD₆₇-positive and NR_2A-positivecells was significantly decreased by 60% in layer 2 (t =2.8; P = .01), but the decreases of 31%, 37%, and29% in layers 3, 5, and 6, respectively, did not achieve statistical significance(t = 1.63; P = .12, t = 2.24; P = .04, and t = 1.54; P = .14, respectively) (Figure 3B).

According to the repeated-measures ANOVA models, the effect of diagnosiswas not statistically significant in subjects with either schizophrenia (F_1,32 = 1.38; P = .25) or bipolar disorder (F_1,30 = 0.038; P = .85). However, in the schizophreniagroup, there was a significant diagnosis × layer effect (F_3,96 = 6.42; P<.001) that seemed to reflectthe 42% decrease in neuronal density in layer 2 (Figure 3C), and this reduction was statistically significant (t = 2.99; P = .005).

There were no differences in the density of silver grains in eitherthe schizophrenia (F_1,20 = 1.79; P = .20)or the bipolar disorder (F_1,24 = 0.21; P =.65) group, suggesting that for GABA interneurons that contained a detectableamount of NR_2A mRNA, the level of expression of the transcriptwas unaltered in either disease group (Figure4).

We examined the potential confounding effects of variables such as age,postmortem interval, brain pH, hemispheric laterality, and antipsychotic drugexposure on our findings. None of these factors seem to have affected ourresults (data not shown). Among these variables, pH was particularly importantbecause the integrity of mRNA is known to be sensitive to this variable.^34- 36 There was no statisticallysignificant difference in pH in the 3 study groups. In the statistical analyses,we also found no correlation between pH and any of our measurements eitherin individual diagnostic groups or when subjects from the disease groups andthose from the control group were combined. An ANCOVA incorporating pH asa covariate also did not statistically significantly alter the effect of diagnosison the cell density and grain density measurements. Similar analyses withchlorpromazine-equivalent dose also revealed that exposure to antipsychoticmedications was not statistically significantly correlated with any of ourmeasurements, and neither did it contribute to the observed differences inthe neuronal density measurements in the diagnostic groups. Therefore, thesefindings do not seem to be the result of antipsychotic medication treatmentor any other measurable potential confounds but may in fact reflect the underlyingdisease processes.

Multiple lines of evidence suggest that disturbances of GABA interneuronsrepresent a key feature of the pathophysiology of schizophrenia and bipolardisorder.^{10,12- 17,19,37- 40} Theactivities of GABA interneurons are subject to feedback and feedforward modulationby glutamatergic inputs from pyramidal neurons located locally and in distantcortical or subcortical regions. Together, these mechanisms regulate the flowof information in the cerebral cortex by adjusting the spatial and temporalarchitecture of GABA neurotransmission.^41- 43 Inthis study, we extended previous findings of altered GABAergic neurotransmissionin schizophrenia and bipolar disorder to demonstrate that alteration in glutamatergicinputs onto GABA interneurons via the NMDA receptor may contribute to disturbancesof GABA neurotransmission in both of these conditions. We cannot, however,exclude the possibility that there may be a primary dysregulation in the expressionof the NR_2A subunit in a subgroup of GABA neurons.

In situ hybridization labeling with [³⁵S] is more sensitivein detecting transcript signals than nonradioactive DIG-labeled probes.⁴⁴ In the present study, because DIG was used to labelthe GAD₆₇ transcript, it is possible that we may have underestimatedthe true density of GABA interneurons in all 3 study groups. If the amountof GAD₆₇ transcript per GABA interneuron is equivalent in the 3groups, our conclusions would not have been affected because they were basedon the "relative" changes in neuronal density among the groups. On the otherhand, it is possible that a subpopulation of GABA interneurons may in factexpress a lower level of GAD₆₇ mRNA in subjects with schizophreniaand bipolar disorder and that these cells fell below the detection thresholdfor DIG labeling of GAD₆₇. This scenario could have contributedto the observed reduction in neuronal density in the schizophrenia and bipolardisorder groups compared with controls. Although we cannot rule out thesepossibilities, they seem to be insufficient to account for the magnitude ofneuronal density reduction observed because DIG labeling of GAD₆₇ hasbeen estimated to be only slightly (<7%) less sensitive than similar insitu hybridization labeling with [³⁵S].⁴⁴ Analternative possibility is that the number of GABA interneurons may be inherentlydifferent in the 2 disease groups and that this was reflected in the differencesin neuronal density observed in this study (see the following paragraphs).

Our findings indicate that the decrease in the density of GAD₆₇ mRNA–expressing neurons may be more prominent in layer 2 thanin other layers in the ACCx in schizophrenia and bipolar disorder. Thus, therewas a 53% and a 35% reduction in the density of neurons that express GAD₆₇ mRNA in layer 2 in the schizophrenia and bipolar disorder groups,respectively, whereas there was no statistically significant reduction inthe density of these neurons in other layers in either subject group, exceptfor a 28% decrease in density in layer 5 in subjects with schizophrenia. Theseobservations are consistent with findings from previous studies⁴⁵ suggestingthat neural circuits in layer 2 in the ACCx may be a major site of diseasevulnerability in schizophrenia and bipolar disorder. Because neurons in thislayer receive extensive corticocortical projections from other regions ofthe cerebral cortex, such as the prefrontal cortex,⁴⁶ andthey also receive specific inputs from subcortical and limbic structures,such as the amygdala,⁴⁷ they may play a criticalrole in integrating diverse streams of information derived from the cognitiveand emotive domains. Therefore, disturbances of information processing inthe neural circuits in this layer could contribute to the multitude of symptomsobserved in schizophrenia and bipolar disorder.

The reduction in the density of neurons that express the mRNA for the67-kDa isoform of GAD may represent a loss of neurons. Alternatively,the amount of GAD₆₇ mRNA in a subpopulation of GABA interneuronsmay be decreased to an experimentally undetectable level. Consideration offindings from previous studies of quantification of nonpyramidal, presumablyGABA interneurons may provide some insights into the possible interpretationsof the current findings. Data from these studies^7,48- 49 demonstratethat the density of nonpyramidal neurons in layer 2 in the ACCx in bipolardisorder was decreased by 27% to 30%, whereas the magnitude of decrease inschizophrenia was only 16% to 17%. In bipolar disorder, the 35% reductionin the density of GAD₆₇ mRNA–containing neurons in layer2 observed in the present study is similar to the degree of reduction in thedensity of nonpyramidal neurons previously reported.⁷ Arecent study¹⁰ using immunohistochemical techniquesto examine the expression of various calcium-binding proteins, which are differentiallyexpressed by subpopulations of GABA interneurons,^50- 53 hasshown that the density of neurons that express calbindin was decreased by33% and 34% in bipolar disorder and schizophrenia, respectively. In addition,the density of neurons that expressed parvalbumin also seemed to be reducedin the 2 disorders, although the differences did not achieve statistical significance.Furthermore, the magnitude of reduction in the densities of the calbindin-and parvalbumin-expressing neurons seems to be similar in bipolar disorder,and it is quantitatively almost identical to the decrease in the density ofGAD₆₇mRNA–containing neurons reported in this study. Takentogether, these findings raise the possibility that a subpopulation of GABAinterneurons in layer 2, especially those that contain calbindin or parvalbumin,may indeed be lost in bipolar disorder. In schizophrenia, we observed a 53%reduction in the density of GAD₆₇ mRNA-containing neurons in layer2, which is far greater in magnitude than the 16% to 17% decrease in the densityof nonpyramidal neurons shown in previous cell counting studies⁷ andthe 34% decrease in the density of calbindin-containing neurons.¹⁰ Therefore,although cell loss may still occur to some degree in schizophrenia, it seemsto be insufficient to account for the degree of reduction in the GAD₆₇ mRNA–containing neurons. This conclusion is consistent withthe results of a recent study⁵⁴ demonstratinga paradoxical decrease in apoptosis markers in the ACCx of subjects with schizophrenia.

Because GABA interneurons are anatomically and functionally heterogeneous,^55- 57 characterizing theidentities of the GABA interneurons that show reduced expression of the NR_2A subunit will shed critical light on the nature of neural circuitrydisturbances and their functional consequences in schizophrenia and bipolardisorder. Because subpopulations of GABA interneurons can be characterizedby the presence of calcium-binding proteins and other neuropeptides, suchas cholecystokinin and vasoactive intestinal peptide,^50- 53 futurestudies will examine the co-expression of the NR_2A subunit andthese proteins or peptides. The information obtained from these studies willhelp define how glutamatergic modulation of specific GABAergic elements inlayer 2 in the ACCx may be differentially altered in schizophrenia and bipolardisorder. A novel treatment strategy for these conditions could potentiallyinvolve fine-tuning the relative levels of NMDA-mediated glutamatergic activityimpinging on GABA interneurons.

Correspondence: Francine M. Benes, MD, PhD, Program in Structuraland Molecular Neuroscience, McLean Hospital, 115 Mill St, Belmont, MA 02478(benesf@mclean.harvard.edu).

Submitted for publication October 2, 2003; final revision received February11, 2003; accepted February 13, 2004.

This study was supported by grants MH/NS31862, MH00423, and MH42261from the National Institutes of Health, Bethesda, Md.