GABA-Synthesizing Enzymes in Calbindin and Calretinin Neurons in Monkey Prefrontal Cortex

Abstract

Non-overlapping groups of cortical γ-aminobutyric acid-releasing (GABAergic) neurons are identifiable by the presence of calbindin (CB), calretinin (CR), or parvalbumin (PV). Boutons from PV neuron subtypes are also distinguishable by differences in protein levels of the GABA-synthesizing enzymes GAD65 and GAD67. Multilabel fluorescence microscopy was used to determine if this diversity extends to boutons of CB and CR neurons in monkey prefrontal cortex. CB and CR neurons gave rise to 3 subpopulations of GAD-containing boutons: GAD65+, GAD67+, and GAD65/GAD67+. Somatostatin and vasoactive intestinal peptide-expressing neurons, subtypes of CB and CR neurons, respectively, also gave rise to these distinct bouton subpopulations. At the transcript level, CB and CR neurons contained mRNA encoding GAD67-only or both GADs. Thus, the distinct subpopulations of CB/GAD+ and CR/GAD+ boutons arise from 2 unique subtypes of CB and CR neurons. The different CB and CR GAD-expressing neurons targeted the same projection neurons and neuronal structures immunoreactive for PV, CR, or CB. These findings suggest that GABA synthesis from CB/GAD67+ and CR/GAD67+ neurons would presumably be more vulnerable to disease-associated deficits in GAD67 expression, such as in schizophrenia, than neurons that also contain GAD65.

Introduction

Proper structure and function of cortical networks depend on the intricate interplay between inhibitory γ-aminobutyric acid (GABA)-releasing (GABAergic) neurons and excitatory pyramidal cells. GABAergic neurons constitute approximately 25% of all neurons in the monkey neocortex (Hendry et al. 1987). They are comprised of diverse groups of cells that can be differentiated by molecular, morphological, and physiological characteristics (Ascoli et al. 2008). The multiple subtypes of GABAergic neurons work in concert to regulate the integration of local and long-range synaptic inputs and thereby control the output of the neuronal network (Kepecs and Fishell 2014).

Most (80–90%) of GABAergic neurons in the primate prefrontal cortex (PFC) can be differentiated into non-overlapping subtypes based on the expression of 1 of 3 calcium-binding proteins—calbindin (CB; ∼20%), calretinin (CR; ∼45%), or parvalbumin (PV; ∼20%; Conde et al. 1994; del Rio and DeFelipe 1996; Gabbott and Bacon 1996; Barinka and Druga 2010). Axonal features and/or the expression of neuropeptides can be used to further divide these 3 subtypes. For example in rodent, approximately 50% of GABAergic CB neurons contain somatostatin (SST) and >85% of SST neurons contain CB (Rogers 1992; Kubota et al. 1994; Gonchar and Burkhalter 1997). Boutons from CB/SST neurons target dendritic compartments of other GABAergic, non-SST neuron subtypes and pyramidal cells (Melchitzky and Lewis 2008; Lovett-Barron et al. 2012; Pfeffer et al. 2013; Lovett-Barron et al. 2014). The overall function of these neurons is regulation of neuronal input–output transformations within local cortical circuits (Lovett-Barron et al. 2012) and from thalamic afferents (Xu et al. 2013), and to provide feedback inhibition in and between cortical layers (Wang et al. 2004).

Greater than 80% of CR neurons express vasoactive intestinal peptide (VIP) and approximately 80% of VIP neurons contain CR (Gabbott and Bacon 1997). Boutons from CR/VIP neurons mainly target dendrites of PV (Donato et al. 2013; Pi et al. 2013) and SST neurons (Pfeffer et al. 2013; Pi et al. 2013), and to a lesser extent the dendrites and soma of pyramidal cells (del Rio and DeFelipe 1997; Rajkowska et al. 2007; Melchitzky and Lewis 2008). Signaling from CR/VIP neurons plays a crucial role in synaptic plasticity (Donato et al. 2013). By contacting other GABAergic neurons, CR/VIP neurons mediate disinhibitory control of pyramidal cells, which lead to selective amplification of local signal processing (Pi et al. 2013).

Cortical PV neurons are comprised of basket and chandelier cells, which mainly target the perisomatic region of pyramidal cells. In adult monkey PFC, boutons from PV basket cells contain both GADs (GAD65/GAD67+), while those from PV chandelier cells contain only GAD67 (GAD67+; Fish et al. 2011). The physiological meaning of this difference is unclear because both PV subtypes exhibit similar short-term depression across species and cortical regions (Maccaferri et al. 2000; Gulyas et al. 2010; Dugladze et al. 2012), but PV chandelier cells would presumably be more vulnerable to disease-associated deficits in GAD67 expression. To extend our understanding of GABAergic neuron vulnerability to disease-associated changes in GAD expression, we examined GAD mRNA expression and bouton protein content of CB and CR neurons using quantitative fluorescence microscopy.

Materials and Methods

Animals and Tissue Preparation

For immunocytochemical studies, 3 young adult (37–46 months) male macaque (Macaca mulatta) monkeys were deeply anesthetized and perfused transcardially with ice-cold 1% paraformaldehyde in phosphate-buffered saline (PBS) followed by 4% paraformaldehyde as previously described (Oeth and Lewis 1993). Brains were immediately removed, and blocks (5–6 mm thick) were cut coronally and immersed in 4% paraformaldehyde for 6 h at 4 °C. Blocks were then immersed in a graded series of sucrose solutions at 4 °C and stored in a cryoprotectant solution at −30 °C. Sections (40 µm) were exhaustively cut from left hemispheric blocks containing the entire rostral–caudal extent of the principal sulcus (area 46).

For in situ hybridization studies, tissue sections from 2 unperfused young adult (37 and 44 months) male macaques (M. mulatta) were used. The monkeys were deeply anesthetized and the brain was extracted and blocked coronally. Blocks were quickly frozen in isopentane on dry ice and stored at −80 °C. Serial sections (12 µm) were cut from left hemisphere blocks containing the principal sulcus, thaw-mounted on Superfrost Plus Gold slides (Fisher Scientific, Pittsburgh, PA, USA), and stored at −80 °C until processed.

All experimental procedures were conducted in accordance with the NIH Guide for the Care of Animals and with the approval from the University of Pittsburgh's IACUC.

Immunocytochemistry

Nine different immunocytochemistry experiments were performed (Table 1). Two sections (separated by 3 mm with a random start) from each of 3 monkeys spanning the rostral–caudal axis of the principal sulcus (Fig. 1A) were used to assess GAD content in boutons arising from unique subtypes of GABAergic neurons (Table 1—Experiments A–D). One section per monkey was used to assess GAD content in boutons apposed to immunoreactive (IR) structures from specific neurons (Table 1—Experiments E–K). For each experiment, sections were permeabilized with 0.3% Triton X-100 in PBS for 30 min at room temperature (RT), incubated in 20% donkey serum in PBS for 2 h at RT, and then incubated for approximately 72 h at 4 °C in PBS containing 2% donkey serum and primary antibodies (Table 1). Sections were then rinsed for 2 h in PBS and incubated in secondary antibodies (Donkey) conjugated to Alexa 488, 568, and 647 (Invitrogen; 1 : 500 for all) for 24 h in PBS containing 2% donkey serum at 4 °C. For triple-labeled experiments (Table 1—Experiments C and D), the sections were rinsed in PBS and mounted. For quadruple-labeled experiments (Table 1—Experiments A, B, and E–K), biotin (Fitzgerald; 1 : 200) was included in the secondary antibody solution. The sections were then rinsed in PBS (2 h) followed by a tertiary incubation (24 h) with streptavidin Alexa 405 (1 : 200), and then rinsed in PBS (2 h) and mounted. Secondary antibody specificity was verified by omitting the primary antibody in control experiments. Multiple pilot studies were performed to determine if any primary/secondary combinations influenced the outcome; results from these studies indicated that the ability to detect IR puncta was not dependent on the secondary antibody spectra.

Table 1.

Antibodies and immunocytochemistry experiments

Antigen	Species	Dilution	Source	Experiment
CB	Rabbit	1 : 1000	Swant	A, E, F, G, J
CR	Rabbit	1 : 1000	Swant	B, I, K
CR	Mouse	1 : 1000	Swant	F
CR	Goat	1 : 1000	Swant	H
SST	Rabbit	1 : 1000	Dr Robert Benoit	C
VIP	Rabbit	1 : 1000	ImmunoStar	D
PV	Mouse	1 : 1000	Swant	E
PV	Rabbit	1 : 1000	Swant	H
SMI-32	Mouse	1 : 1000	Sternberger-Meyer Immunocytochemicals	G, I
MAP2	Mouse	1 : 500	BioLegend	J, K
vGAT	Mouse	1 : 500	Synaptic Systems	A, B
GAD65	Guinea pig	1 : 500	Synaptic Systems	A, B, C, D, E, F, G, H, I, J, K
GAD67	Goat	1 : 100	R&D Systems	A, B, E, F, G, I, J, K
GAD67	Mouse	1 : 1000	Millipore	C, D, H

Note: Rows with the same letter in the experiment column indicate the antigens labeled in each immunocytochemistry assay. The specificity of each antibody was verified by western blot in our laboratory (data not shown and Fish et al. 2011) or other laboratories [CR (Schwaller et al. 1993), vGAT (Guo et al. 2009), and Synaptic Systems data sheet; GADs (Gottlieb et al. 1986; Chang and Gottlieb 1988)]. In addition, the specificity of the SST28_1–12 antibody has previously been demonstrated by radioimmunoassay (Benoit et al. 1982, 1985) and immunohistochemical blocking experiments (Morrison et al. 1983; Bakst et al. 1985; Lewis et al. 1986).

Figure 1.

Principal sulcus and layer delineation within monkey PFC. (A) Schematic of lateral view of monkey cortex showing the approximate locations (dashed black lines) of PFC tissue sections used for this study. (B) Schematic view of left hemisphere PFC coronal tissue section designating the dorsal (46-D) and ventral (46-V) banks of the principal sulcus (PS; area 46). Dashed lines approximate the border of area 46. Image stacks were collected from 46-V. (C) Nissl-stained monkey tissue section from 46-V showing layer delineation by cytoarchitectonic criteria (roman numerals), and cortical zones L1, L2/3s, L3d/4, L5, and L6, which correspond to 0–10%, 10–35%, 35–60%, 60–80%, and 80–100%, respectively, of the total gray matter area spanning from pia to the gray/white matter border. Bar = 250 µm.

In Situ Hybridization

In situ hybridization probes were designed by Advanced Cell Diagnostics, Inc. (Hayward, CA, USA) to detect mRNA encoding GAD65 (GAD2), GAD67 (GAD1), CB (CALB1), and CR (CALB2). Tissue samples were processed using the RNAscope^® 2.0 Assay as previously described (Wang et al. 2012). Two experiments were performed to simultaneously assess GAD65 and GAD67 mRNAs in neurons containing CB mRNA or CR mRNA (Figs 3A and 7A, respectively). Briefly, tissue sections were incubated in a protease treatment, and then the probes were hybridized to their target mRNAs for 2 h at 40 °C. The sections were exposed to a series of incubations that amplified the target probes, and then counterstained with NeuroTrace blue-fluorescent Nissl stain (1 : 50; Molecular Probes). GAD65 and GAD67 mRNAs were detected with Alexa 488 and 647, respectively, and CB and CR mRNAs were detected with Alexa 550.

Figure 3.

CB neurons are distinguishable by their GAD65 and GAD67 mRNA content. (A) Single-plane image of a monkey PFC tissue section labeled for CB, GAD65, and GAD67 mRNAs, and counterstained with NeuroTrace. Bar = 5 µm. (B) Scatterplot showing the number of GAD65 and GAD67 mRNA molecules from 50 randomly selected CB mRNA-containing neurons.

Figure 7.

CR neurons are distinguishable by their GAD65 and GAD67 mRNA content. (A) Single-plane image of monkey PFC tissue section labeled for CR, GAD65, and GAD67 mRNAs, and counterstained with NeuroTrace. Bar = 5 µm. (B) Scatterplot showing the number of GAD65 and GAD67 mRNA molecules from 50 randomly selected CR mRNA-containing neurons.

Microscopy

Data were collected on an Olympus (Center Valley, PA, USA) IX81 inverted microscope equipped with an Olympus spinning disk confocal unit, a Hamamatsu EM-CCD digital camera (Bridgewater, NJ, USA), and a high-precision BioPrecision2 XYZ motorized stage with linear XYZ encoders (Ludl Electronic Products Ltd, Hawthorne, NJ, USA) using a 60x 1.40 NA SC oil immersion objective. The equipment was controlled by SlideBook 5.0 (Intelligent Imaging Innovations, Inc., Denver, CO, USA), which was the same software used for post-image processing. Three-dimensional (3D) image stacks (2D images successively captured at intervals separated by 0.25 µm in the z-dimension) that were 512 × 512 pixels (∼137 µm × 137 µm) were acquired over 50% and 100% of the total thickness of the tissue section starting at the coverslip for the immunocytochemical and in situ hybridization studies, respectively. The stacks were collected using optimal exposure settings (i.e., those that yielded the greatest dynamic range with no saturated pixels), with differences in exposures normalized during image processing.

Sampling

The gray matter of the ventral bank of the principal sulcus, which contains PFC area 46 in monkey (Fig. 1B), has a six-layer (L) lamination pattern. As determined by measurements made in Nissl-stained sections, the boundaries of each layer can be estimated based on the distance from the pial surface to the white matter. For the present study, we have designated the superficial and middle cortical zones as L2/superficial L3 and deep L3/L4, respectively. Thus, the 6 cortical layers were divided into 5 zones consisting of: L1 (pia—10%), L2/superficial L3 (L2/3s; 10–35%), deep L3/L4 (L3d/4; 35–60%), L5 (60–80%), and L6 (80%–gray/white matter border; Fig. 1C). Ten, systematic randomly sampled image stacks were taken within each laminar zone using a sampling grid of 180 × 180 µm² for the studies assessing GAD content in unique GABAergic neuron subtypes and the in situ hybridization studies. For the studies assessing GAD content in boutons apposed to IR structures from specific neurons, 10 randomly sampled image stacks were taken within each laminar zone assessed using a sampling grid of 180 × 180 µm². Studies assessing boutons apposed to IR structures for SMI-32, PV, CR, and CB were performed in L3d/4, while those assessing boutons apposed to IR structures for MAP2 were performed in L1, L3d/4, and L6. For these studies, 100 randomly selected boutons apposed to each IR structure per laminar zone were assessed.

Image Processing

For immunocytochemical studies, each fluorescent channel was deconvolved using Autoquant's Blind Deconvolution algorithm. For data segmentation, a Gaussian channel was made for each deconvolved channel by calculating a difference of Gaussians using sigma values of 0.7 and 2, which improves the ability to discriminate puncta (Supplementary Fig. 1). Importantly, the Gaussian channel was used for data segmentation only. Data segmentation was performed as described (Fish et al. 2008), with a few exceptions. First, the Ridler–Calvard iterative thresholding algorithm (Ridler and Calvard 1978) was used to obtain an initial value for iterative segmentation for each channel within each image stack. Second, 100 iterations with subsequent threshold settings increasing by 50 gray levels were performed. Third, threshold segmentation was done in MATLAB (R2012). After each iteration, the object masks were size-gated within a range of 0.03–0.5 µm³. For analyses, the image stacks were virtually cropped in the x-, y-, and z-dimensions. This was done using x-, y-, and z-coordinates of the object masks, which represented each IR puncta. In the x- and y-dimensions, the center of each object mask had to be contained in the center 490 × 490 pixels of the image. To select the z-dimension used for analyses, the z-position of each object mask was normalized by the following equation:

$$Z_{\mathrm{coordinate}}\left(\frac{\text{no}.\mathrm{of}z\text{-}\mathrm{planes}\mathrm{for}\mathrm{image}\mathrm{stack}}{\text{4}0}\right).$$

Next, each object mask was placed in 1 of 40 z-bins based on its normalized z position. The mean object mask number and mean fluorescence intensity for the vesicular GABA transporter (vGAT) (Table 1—Experiments A and B), GAD65 and GAD67 (Table 1—Experiments A–D), and CB, CR, SST, and VIP (Table 1—Experiments A–D, respectively) was determined within each z-bin, which was then used for an analysis of variance with post hoc comparison via Tukey's honestly significant difference test. The maximum number of adjacent z-bins that were not significantly different for both intensity and object mask number across all channels was used for analyses. For the CB and CR experiments (Table 1—Experiments A and B), the exact same 17 bins were used, which corresponded to 8.5 µm of the cut tissue thickness. By taking this approach, we controlled for possible edge effects (i.e., all puncta assessed were fully represented in the virtual space), differences in antibody penetration, and differences in fluorochromes. The final object masks were then used to collect information on the deconvolved channels.

For the in situ hybridization studies, each channel was deconvolved using the No Neighbors deconvolution algorithm. After deconvolution, a 2D projection image was made and every Nissl-stained cell within a centered 365 × 365 pixel counting frame was manually masked. Data segmentation for the remaining channels was performed as described above.

Classification of Bouton GAD Protein Content

For immunocytochemical experiments including vGAT (Table 1—Experiments A and B), CB-IR and CR-IR puncta were classified as a bouton if they colocalized with vGAT and GAD65 and/or GAD67. For all other experiments, CB-IR, CR-IR, SST-IR, and VIP-IR puncta were classified as a bouton if they colocalized with GAD65 and/or GAD67. A multistep process was used to classify boutons as GAD65+, GAD67+, or GAD65/GAD67+. For example, all vGAT-IR puncta that contained GAD immunoreactivity were classified as GAD65+, GAD67+, and GAD65/GAD67+ boutons. To obtain the vGAT/GAD+ bouton subpopulations, mask operations, which assess the degree of overlap between voxels of different object masks, were used to identify GAD65 and vGAT object masks that overlapped each other's centers and did not overlap a GAD67 object mask (GAD65+ boutons). GAD67+ boutons were similarly defined. vGAT object masks that overlapped the center of both a GAD65 and GAD67 object mask were defined as GAD65/GAD67+ (see Supplementary Fig. 2A–I). Supplementary Figure 2J is a scatterplot of the mean GAD65 and GAD67 fluorescence intensities for every vGAT-IR puncta in a randomly selected image stack. Puncta that were classified as GAD65+, GAD67+, or GAD65/GAD67+ using the mask operations criteria stated above constituted approximately 70% of all vGAT/GAD+ boutons. Supplementary Figure 2K is a scatterplot of the mean GAD65 and GAD67 fluorescence intensities for every vGAT-IR puncta from Supplementary Figure 2J that were classified as GAD65+, GAD67+, and GAD65/GAD67+ using the mask operations criteria. Supplementary Figure 2L is every vGAT-IR puncta from Supplementary Figure 2J that contained partially overlapping GAD65-IR and GAD67-IR puncta, and therefore did not meet the mask operation criteria used to be classified as one of the GAD+ bouton subpopulations. Next, the mean fluorescence intensity of GAD65 for all GAD65+ and GAD65/GAD67+ boutons and the mean fluorescence intensity of GAD67 for all GAD67+ and GAD65/GAD67+ boutons that were classified using mask operation criteria (e.g., boutons from Supplementary Fig. 2K) per site were used as seeds in a K-means cluster analysis to classify all puncta within the site. Supplementary Figure 2M shows all vGAT-IR puncta from Supplementary Figure 2J that were classified as GAD65+, GAD67+, and GAD65/GAD67+. For experiments assessing GAD content in CB+ and CR+ boutons apposed to distinct IR neuronal structures, GAD65 and/or GAD67 content per bouton were qualitatively determined.

Classification of Somatic GAD mRNA Content

To assess the number of GAD65 and GAD67 mRNA molecules in CB-containing and CR-containing neurons, a total of 2338 NeuroTrace-labeled somas were masked between the 2 in situ hybridization experiments. An unbiased, highly conservative approach was used to classify somas as GABAergic because CB is also expressed by pyramidal neurons in primate PFC (DeFelipe, Hendry SHC, Jones 1989; Freund et al. 1990; Hof and Morrison 1991; Hayes and Lewis 1992). Specifically, to be defined as GABAergic individual somas had to contain ≥20 GAD mRNA molecules (GAD65 and/or GAD67), which was approximately 10× greater than the number of GAD mRNA molecules that were not contained within a NeuroTrace-labeled soma. Between both experiments, 407 somas were defined as GABAergic. To be defined as containing CB mRNA or CR mRNA, somas had to contain ≥5 CB or CR mRNA molecules, respectively, which was >2.5× the number of CB or CR mRNA molecules that were not contained within a NeuroTrace-labeled soma.

Statistical analysis

In all analyses, the statistics were performed on the mean values for individual monkeys. The density and percentage of each GAD+ bouton subpopulation arising from different GABAergic neuron subtypes were determined in the following way: (1) The statistics for each measure were averaged for each image stack; (2) the stack averages were averaged within layer; (3) layer averages were averaged within section; and (4) the section averages were used to generate the mean (± standard deviation [SD]) density and percentage of each GAD+ bouton subpopulation per monkey. In cases where statistics were compared within PFC gray matter, step 2 was omitted and stack averages were averaged within the section. The density of each GAD+ bouton subpopulation was assessed using the analysis of variance with post hoc comparison via Tukey's honestly significant difference test. For analyses with unequal variances between groups, post hoc comparison was performed via Dunnett's T3 test. A 2 × 3 chi-square analysis was performed to assess differences in the proportion of the GAD+ bouton subpopulations between CB and CR neurons.

Results

CB Neurons Give Rise to GAD65+, GAD67+, and GAD65/GAD67+ Boutons

Within PFC gray matter, 21.4 (±4.3)% of all GAD+ boutons were CB-IR (CB/GAD+; see Supplementary Materials and Methods for the densities of all GAD+ boutons), which is similar to the percentage of GABAergic neurons that express CB. This percentage differed across cortical layers (F_4,10 = 4.38, P < 0.05); the percentage of CB/GAD+ boutons was greater in L1 (32.2 ± 10.6) than in L5 (12.6 ± 3.6; P < 0.05), whereas the other layers did not differ from each other or from L1 and L5 (L2/3s: 22.5 ± 6.4; L3d/4: 17.4 ± 3.9; and L6: 19.3 ± 0.4). Qualitative assessment found that GABAergic CB neurons give rise to 3 distinct GAD+ bouton subpopulations: (1) CB/GAD65+, (2) CB/GAD67+, and (3) CB/GAD65/GAD67+ (Fig. 2A). Quantitative analysis found no differences between the density of CB/GAD+ boutons that were CB/GAD65+ (0.002 ± 0.0005 boutons/µm³), CB/GAD67+ (0.003 ± 0.0002 boutons/µm³), and CB/GAD65/GAD67+ (0.002 ± 0.0009 boutons/µm³); however, there were some laminar differences (Table 2), suggesting that unique CB/GAD+ neuron subtypes might exist.

Figure 2.

CB and SST boutons are distinguishable by their GAD65 and GAD67 protein content. (A) Single-plane image of a monkey PFC tissue section immunolabeled for CB, vGAT, GAD65, and GAD67. Gray-scale images are 1.5×-magnified single-channel images of the multichannel image. Arrows depict CB/GAD65+ (open arrowhead), CB/GAD67+ (solid arrowhead), and CB/GAD65/GAD67+ (arrow) boutons. Bar = 3 µm (multichannel images) and 2 µm (gray-scale images). (B–E) Single-plane images of a monkey PFC tissue section immunolabeled for SST, GAD65, and GAD67. (B–D) Single channel and (E) merged channels depicting SST/GAD65+ (B1–E1; open arrowhead), SST/GAD67+ (B2–E2; solid arrowheads), and SST/GAD65/GAD67+ (B3–E3; arrows) boutons. Bar = 1 µm.

Table 2.

Densities of CB/GAD+ bouton subpopulations across cortical layers

	CB/GAD65+	CB/GAD67+	CB/GAD65/GAD67+
L1	0.003 (0.002) [20.2 (7.1)]	0.005 (0.0004) [46.1 (13.9)]	0.004 (0.003) [33.6 (7.0)]
L2/3s	0.001 (0.0007) [15.7 (2.7)]	0.004 (0.0006) [49.0 (9.5)]	0.003 (0.002) [35.3 (6.9)]
L3d/4	0.001 (0.0003)A [19.7 (1.5)]	0.002 (0.0004)B [43.6 (6.0)]	0.002 (0.0005)AB [36.6 (4.8)]
L5	0.001 (0.0003) 32.1 (7.8)	0.001 (0.0003) 34.1 (2.5)	0.001 (0.0005) 33.8 (5.5)
L6	0.003 (0.0005)A 61.4 (5.5)	0.0007 (0.00004)B 13.5% (1.8)	0.001 (0.0003)B 25.1 (5.2)
All	0.002 (0.0005) [27.6 (2.1)]	0.003 (0.0002) [39.0 (6.3)]	0.002 (0.0009) [33.4 (5.5)]

Note: All density values are presented as mean (±SD) boutons/µm³. Bolded density values in the same row indicate significant differences. Values not sharing the same letter are different (P < 0.05 for all). Bracketed values are the mean (±SD) percentages. For the layers with significant differences: L3d/4—F_2,6 = 8.39, P < 0.05; L6 F_2,6 = 51.16, P < 0.001.

SST Neurons Give Rise to GAD65+, GAD67+, and GAD65/GAD67+ Boutons

GABAergic CB neurons can be subdivided into 2 groups based on the expression of SST. To assess if SST-expressing neurons account for one of the CB/GAD+ neuron subtypes, SST-IR bouton GAD content was assessed. All 3 distinct GAD+ bouton subpopulations were identified: (1) SST/GAD65+, (2) SST/GAD67+, and (3) SST/GAD65/GAD67+ (Fig. 2B–E). These subpopulations were present in all cortical layers (Supplementary Table 2). Thus, SST-expressing neurons do not appear to represent a CB/GAD+ bouton subpopulation.

CB Neurons Contain mRNA for Only GAD67 or Both GADs

To further assess if the distinct CB/GAD+ bouton subpopulations represent different subtypes of GAD-expressing CB neurons, we used the RNAscope^® 2.0 Assay (Wang et al. 2012) to examine GAD65 and GAD67 mRNA content in CB mRNA-containing neurons. Although some CB mRNA-containing neurons clearly contained only GAD67 mRNA, there was no quantitative or qualitative evidence of GAD65-only mRNA-containing CB neurons (Fig. 3). These findings suggest that CB/GAD65+ and CB/GAD65/GAD67+ boutons arise from CB neurons that contain both GAD65 and GAD67 mRNAs, whereas at least some CB/GAD67+ boutons arise from CB neurons containing only GAD67 mRNA.

CB/GAD+ Boutons Do Not Exhibit Target Specificity

To assess if the different GAD-expressing CB neuron subtypes targeted specific neuronal subtypes, GAD content in CB/GAD+ boutons that contacted neuronal IR structures for PV (Fig. 4A), CR (Fig. 4B), SMI-32, which recognizes a non-phosphorylated epitope of neurofilament proteins thought to be necessary for the maintenance of large subcortical projecting neurons with highly myelinated processes (Fig. 4C), or microtubule-associated protein 2 (MAP2; Fig. 4D) was assessed. CB/GAD67+ and CB/GAD65/GAD67+ boutons were apposed to IR structures for each of the neuronal subtypes (Fig. 5). In contrast, the CB/GAD65+ boutons assessed only overlapped with IR structures for SMI-32 and MAP2 (Fig. 5A and D, respectively). Moreover, a laminar assessment of GAD content in CB boutons apposed to MAP2-IR structures found a similar proportion of the different subpopulations of CB/GAD+ boutons in L1 and L3d/4 (Fig. 5D1 and D2, respectively); however, there was a greater percentage of CB/GAD65+ boutons apposed to MAP2-IR structures in L6 (Fig. 5D3), which is consistent with the finding that there is an overall higher percentage of CB/GAD65+ boutons in L6 than the other layers. Since the above mRNA findings suggest that CB/GAD65+ boutons arise from neurons expressing both GAD mRNAs, the different GAD-expressing CB neuron subtypes do not appear to have a preference for targeting the neuronal subtypes examined.

Figure 4.

Examples of CB/GAD+ boutons apposed to distinct subtypes of neurons. (A1–7) Projection images (3 z-planes separated by 0.25 µm) from a monkey PFC tissue section immunolabeled for CB, PV, GAD65, and GAD67. (A1) Merged CB and PV channels. The asterisk indicates a PV-IR soma. (A2–7) 2× magnified images from the boxed region in A1. (A2–5) Gray-scale images of single GAD65 (A2), GAD67 (A3), CB (A4), and PV (A5) channels. (A6) Merged GAD65, GAD67, and CB channels, and an outline of the PV-IR structure from A5 (white lines). (A7) Corresponding GAD65 (red), GAD67 (green), and CB (blue) object masks, and an outline of the PV-IR structure from A5. (B) Projection image (2 z-planes separated by 0.25 µm) from a monkey PFC tissue section immunolabeled for CB, CR, GAD65, and GAD67. (C) Projection image (5 z-planes separated by 0.25 µm) from a monkey PFC tissue section immunolabeled for CB, SMI-32, GAD65, and GAD67. (D) Projection image (3 z-planes separated by 0.25 µm) from a monkey PFC tissue section immunolabeled for CB, MAP2, GAD65, and GAD67. Arrows indicate CB/GAD67+ (solid arrowheads) and CB/GAD65/GAD67+ (arrows) boutons. Bar = 10 µm (A1) and 5 µm (A2–7, B–D).

Figure 5.

Different GAD-containing CB neurons do not preferentially target distinct subtypes of neurons. GAD content was assessed in 100 randomly selected CB/GAD+ boutons apposed to IR structures for each specific neuronal subtype per laminar zone from PFC tissue sections of 3 monkeys (2 sections per monkey). (A–D) Pie charts showing the proportion of the subpopulations of CB/GAD+ boutons apposed to neuronal structure IR for (A) SMI-32 in L3d/4, (B) PV in L3d/4, (C) CR in L3d/4, (D1) MAP2 in L1, (D2) MAP2 in L3d/4, and (D3) MAP2 in L6.

CR Neurons Give Rise to GAD65+, GAD67+, and GAD65/GAD67+ Boutons

Within PFC gray matter, 15.8 (±2.5)% of all GABAergic boutons were CR-IR (CR/GAD+). This finding was surprising considering that CR neurons constitute roughly half of all GABAergic neurons in primate PFC. The percentage of GABAergic boutons that were CR-IR was similar across cortical layers (L1: 12.4 ± 3.8; L2/3s: 16.3 ± 2.4; L3d/4: 15.4 ± 3.2; L5: 19.1 ± 3.3; and L6: 16.7 ± 1.9). Similar to CB neurons, 3 distinct CR/GAD+ bouton subpopulations were identified: (1) CR/GAD65+, (2) CR/GAD67+, and (3) CR/GAD65/GAD67+ (Fig. 6A). Within PFC gray matter, there was no difference between the density of CR/GAD+ boutons that were GAD65+ (0.001 ± 0.0002 boutons/µm³), GAD67+ (0.001 ± 0.0002 boutons/µm³), and GAD65/GAD67+ (0.002 ± 0.0005 boutons/µm³). Similarly, there were no differences across cortical layers except L6 (Table 3). The identification of all 3 CR/GAD+ bouton subpopulations suggested that CR/GAD+ neuron subtypes might exist.

Figure 6.

CR and VIP boutons are distinguishable by their GAD65 and GAD67 protein content. (A) Projection images (5 z-planes separated by 0.25 μm) of a monkey PFC tissue section immunolabeled for CR, vGAT, GAD65, and GAD67. Gray-scale images are 1.5×-magnified single-channel images of the multichannel image. Arrows depict CR/GAD65+ (open arrowhead), CR/GAD67+ (solid arrowhead), and CR/GAD65/GAD67+ (arrow) boutons. Bar = 7.5 µm (multichannel images) and 5 µm (gray-scale images). (B–E) Projection images (5 z-planes separated by 0.25 µm) of a monkey PFC tissue section immunolabeled for VIP, GAD65, and GAD67. (B–D) Single-channel and (E) merged channels depicting VIP/GAD65+ (B1–E1; open arrowhead), VIP/GAD67+ (B2–E2; solid arrowhead), and VIP/GAD65/GAD67+ (B3–E3; arrow) boutons. Bar = 5 µm.

Table 3.

Densities of CR/GAD+ bouton subpopulations across cortical layers

	CR/GAD65+	CR/GAD67+	CR/GAD65/GAD67+
L1	0.001 (0.0004) [24.5 (1.7)]	0.001 (0.0003) [27.5 (3.7)]	0.002 (0.0009) [48.0 (4.9)]
L2/3s	0.001 (0.0003) [25.0 (1.3)]	0.002 (0.0001) [32.6 (4.2)]	0.002 (0.0007) [42.4 (4.7)]
L3d/4	0.001 (0.0002) [27.4 (2.6)]	0.001 (0.0003) [30.1 (2.7)]	0.002 (0.0006) [42.4 (5.0)]
L5	0.002 (0.0003) [27.6 (0.9)]	0.002 (0.0004) [31.0 (3.5)]	0.002 (0.0005) [41.4 (2.6)]
L6	0.001 (0.00009)AB [31.4 (2.6)]	0.001 (0.0003)A [25.0 (4.3)]	0.002 (0.0004)B [43.6 (3.6)]
All	0.001 (0.0002) [27.1 (0.8)]	0.001 (0.0002) [29.4 (3.3)]	0.002 (0.0005) [43.5 (4.1)]

Note: All density values are presented as mean (±SD) boutons/µm³. Bolded density values in the same row indicate significant differences. Values not sharing the same letter are different (P < 0.05 for all). Bracketed values are the mean (±SD) percentages. For the layers with significant differences: L6—F_2,6 = 7.31, P < 0.05.

VIP Neurons Give Rise to GAD65+, GAD67+, and GAD65/GAD67+ Boutons

CR neurons can be subdivided based on the expression of VIP. To determine if VIP-expressing neurons represent one of the CR/GAD+ bouton subpopulations, VIP-IR bouton GAD content was assessed. Three distinct subpopulations of VIP/GAD+ boutons were identified: (1) VIP/GAD65+, (2) VIP/GAD67+, and (3) VIP/GAD65/GAD67+ (Fig. 6B–E). These bouton subtypes were present in all cortical layers (Supplementary Table 3). These findings suggest that VIP-containing boutons do not represent one of the CR/GAD+ bouton subpopulations.

CR Neurons Contain mRNA for Only GAD67 or Both GADs

To further assess if distinct subpopulations of CR/GAD+ boutons represent different subtypes of GAD-expressing CR neurons, we used the RNAscope^® 2.0 Assay (Wang et al. 2012) to examine GAD65 and GAD67 mRNA content in CR mRNA-containing neurons. Similar to CB neurons, some CR mRNA-containing neurons clearly contained only GAD67 mRNA, whereas there was no quantitative or qualitative evidence of GAD65-only mRNA-containing CR neurons (Fig. 7). This suggests that CR/GAD65+ and CR/GAD65/GAD67+ boutons arise from CR neurons that contain both GAD65 and GAD67 mRNAs, and at least some CR/GAD67+ boutons arise from CR neurons containing only GAD67 mRNA.

CR/GAD+ Boutons Do Not Exhibit Target Specificity

Next, we examined if boutons from the different GAD-expressing CR neuron subtypes contacted specific neuronal subtypes. GAD content in CR/GAD+ boutons that contacted neuronal IR structures for SMI-32 (Fig. 8A), PV (Fig. 8B), CB (Fig. 8C), or MAP2 (Fig. 8D) was assessed. CR/GAD67+ and CR/GAD65/GAD67+ boutons were apposed to IR structures for each of these neuronal subtypes (Fig. 9). In contrast, CR/GAD65+ boutons were apposed to only IR structures for SMI-32 (Fig. 9A), PV (Fig. 9B), and MAP2 (Fig. 9D). Moreover, a laminar assessment of GAD content in CR boutons apposed to MAP2-IR structures found a similar proportion of the different subpopulations of CR/GAD+ boutons in L1, L3d/4, and L6 (Fig. 9D1–3). Similar to CB neurons, the different GAD-expressing CR neuron subtypes do not appear to have a preference for targeting the neuronal subtypes examined.

Figure 8.

Examples of CR/GAD+ boutons apposed to distinct subtypes of neurons. (A1–7) Projection images (3 z-planes separated by 0.25 µm) from a monkey PFC tissue section immunolabeled for CR, SMI-32, GAD65, and GAD67. (A1) Merged CR and SMI-32 channels. The asterisk indicates an SMI-32-IR soma. (A2–7) 2×-magnified images from the boxed region in A1. Gray-scale images of single GAD65 (A2), GAD67 (A3), CR (A4), and SMI-32 (A5) channels. (A6) Merged GAD65, GAD67, and CR channels, and an outline of the SMI-32-IR structure from A5 (white lines). (A7) Corresponding GAD65 (red), GAD67 (green), and CR (blue) object masks, and an outline of the SMI-32-IR structure from A5. (B) Projection image (3 z-planes separated by 0.25 µm) from a monkey PFC tissue section immunolabeled for CR, PV, GAD65, and GAD67. (C) Projection image (3 z-planes separated by 0.25 µm) from a monkey PFC tissue section immunolabeled for CR, CB, GAD65, and GAD67. (D) Projection image (3 z-planes separated by 0.25 µm) from a monkey PFC tissue section immunolabeled for CR, MAP2, GAD65, and GAD67. Arrows indicate CR/GAD65+ (open arrowheads), CR/GAD67+ (solid arrowheads), and CR/GAD65/GAD67+ (arrows) boutons. Bar = 10 µm (A1) and 5 µm (A2–7, B–D).

Figure 9.

Different GAD-containing CR neurons do not preferentially target distinct subtypes of neurons. GAD content was assessed in 100 randomly selected CR/GAD+ boutons apposed to IR structures for each specific neuronal subtype per laminar zone from PFC tissue sections from 3 monkeys (2 sections per monkey). (A–D) Pie charts showing the proportion of the subpopulations of CR/GAD+ boutons apposed to neuronal structure IR for (A) SMI-32 in L3d/4, (B) PV in L3d/4, (C) CB in L3d/4, (D1) MAP2 in L1, (D2) MAP2 in L3d/4, and (D3) MAP2 in L6.

The Proportions of the GAD+ Bouton Subpopulations Differ Between CB and CR Neurons

A chi-square test was performed to determine if the proportions of the GAD+ bouton subpopulations arising from CB and CR neurons were different. Within PFC gray matter, the proportion of CB+ boutons that were GAD65+ (28%), GAD67+ (38%), and GAD65+/GAD67+ (34%) was different (χ²= 2928, P < 0.0005) than that of CR+ boutons (28%, 30%, and 42%, respectively). These findings raise the possibility that CB neurons containing only GAD67 mRNA give rise to more boutons than GAD67-only mRNA-containing CR neurons.

Discussion

CB and CR neurons are non-overlapping subtypes that together constitute approximately 65% of all GABAergic neurons in the primate PFC (Conde et al. 1994; del Rio and DeFelipe 1996; Gabbott and Bacon 1996). Both subtypes can be further subdivided by molecular properties (i.e., SST- and VIP-containing, respectively). The findings presented here show that both CB and CR neurons give rise to GAD65+, GAD67+, and GAD65/GAD67+ boutons, and suggest that the GAD65+ and GAD65/GAD67+ boutons arise from neurons-expressing GAD65 and GAD67 mRNA, while the GAD67+ boutons arise from GAD67-only mRNA-expressing neurons. Thus, GAD65 and GAD67 expression by CB and CR neurons is another distinguishing characteristic that should be taken into account in both functional and disease-related studies.

The finding that some CB and CR neurons only expressed mRNA for GAD67 suggests that at least some of the CB/GAD67+ and CR/GAD67+ boutons arise from these unique subtypes. In contrast, no CB or CR neurons only expressed mRNA for GAD65, suggesting that CB/GAD65+ and CR/GAD65+ boutons arise from CB and CR neurons that express mRNA for both GADs. This finding raises the question—why are there CB/GAD65+ and CR/GAD65+ boutons? It is possible that the postsynaptic target influences bouton GAD content. For example, boutons from PV basket cells, which target the soma and proximal dendrites of pyramidal cells (Melchitzky et al. 1999), contain both GADs. In contrast, boutons from PV chandelier cells, which exclusively target the axon initial segment of pyramidal cells (DeFelipe, Hendry SH, Jones 1989; Lewis and Lund 1990), only contain GAD67 (Fish et al. 2011). In contrast, our findings indicate that GAD65+ and GAD65/GAD67+ boutons arising from CB and CR neurons did not preferentially contact specific neuronal subtypes (Figs 5 and 9). However, appositions are only an estimate of synapses. In addition, the subset of boutons innervating spines was not assessed. Thus, future electron microscopy studies are needed to fully address the question of synaptic specificity of the different CB and CR bouton subpopulations.

Alternatively, our findings might reflect differences in GAD65 and GAD67 trafficking and half-life. For example, because GAD65 has a very long half-life (>24 h) and is efficiently trafficked to axonal boutons (Kanaani et al. 2002), every bouton from CB and CR neurons expressing mRNA for both GADs would be expected to have detectable levels of GAD65. In contrast, GAD67's short half-life (∼2 h), along with a trafficking mechanism that is partially dependent on GAD65 (Kanaani et al. 2010), might result in undetectable levels of GAD67 in some boutons. Since changes in network activity drastically alter GAD67 protein levels (Lau and Murthy 2012), increased neuronal activity may lead to an increase in the proportion of boutons containing GAD67 in these neurons. Furthermore, the difference in half-life between GAD65 and GAD67 protein suggests that much less mRNA is needed to maintain normal GAD65 bouton levels compared with what is needed to maintain normal GAD67 levels. Thus, it is possible that CB and CR neurons expressing only GAD65 exist, but were not captured in our analysis because they did not have detectable, above background levels of GAD mRNA.

The physiological consequences of boutons containing both GADs compared with only GAD67 are unclear. For example, among PV neurons, basket cells give rise to GAD65/GAD67+ boutons, whereas chandelier cells give rise to GAD67+ boutons (Fish et al. 2011; Glausier et al. 2014), but both subtypes exhibit similar short-term depression across species and cortical regions (Maccaferri et al. 2000; Gulyas et al. 2010; Dugladze et al. 2012). Thus, the release properties of PV basket and chandelier neurons do not seem to differ as a result of differences in the presence of GAD65. Whether this is the case for CB and CR neurons that only express GAD67 mRNA is not known.

The function of CB and CR neurons that use only GAD67 to synthesize GABA would presumably be more affected by disease-driven reductions in GAD67 expression than neurons that also contain GAD65. Alterations in GABAergic neuron subtypes are implicated in the pathologies of several neurological disorders including schizophrenia (Tu et al. 1999) and epilepsy (DeFelipe 1999; Bernard et al. 2000). For example, lower GAD67 mRNA expression is perhaps the most widely reported finding in postmortem PFC tissue from schizophrenia subjects (Akbarian et al. 1995; Guidotti et al. 2000; Mirnics et al. 2000; Volk et al. 2000; Vawter et al. 2002; Hashimoto et al. 2003; Straub et al. 2007; Hashimoto et al. 2008; Duncan et al. 2010; Curley et al. 2011; Hyde et al. 2011; Hoftman et al. 2015; Kimoto et al. 2014), and is accompanied by lower levels of GAD67 protein levels (Impagnatiello et al. 1998; Guidotti et al. 2000; Curley et al. 2011). At the cellular level, the density of GABAergic neurons with detectable levels of GAD67 mRNA is approximately 30% lower across PFC layers 2–5 neurons, whereas the other neurons express GAD67 mRNA at normal levels (Akbarian et al. 1995; Guidotti et al. 2000; Volk et al. 2000). This deficit occurs without a change in total neuron density (Akbarian et al. 1995) or number (Thune et al. 2001) in the PFC, suggesting that a normal complement of GABAergic neurons are present in individuals with schizophrenia, but that a subset of GABAergic neurons have a markedly reduced capacity to synthesize GABA. Although GAD levels have not been assessed in CB or CR neurons in schizophrenia, a deficit in GAD67 mRNA expression in CB and CR neurons that only express GAD67 mRNA would presumably markedly decrease GABA synthesis in those cells. In addition, the finding that a higher percentage of CB boutons are GAD67+ than CR boutons suggests that CB neuron synaptic inhibition would be more impaired by GAD67 expression-level deficits than CR neurons. Considering that pyramidal neuron distal dendrites are a main target of CB neurons, and a minor target of CR neurons (del Rio and DeFelipe 1997; Rajkowska et al. 2007; Melchitzky and Lewis 2008), a deficit in GABA release from these boutons would presumably alter input–output transformations within local cortical circuits (Lovett-Barron et al. 2012) and from thalamic afferents (Xu et al. 2013). Since many CR neurons contact other GABAergic neurons, a disease-associated decrease in GABA release from their boutons would presumably alter disinhibitory control of pyramidal cells (Pi et al. 2013).

Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/.

Funding

This work was supported by the NSF (DGE-0549352 to B.R.R.) and NIMH (MH071533 to R.A.S.; MH051234 to D.A.L.; MH096985 to K.N.F.).