Structural Organization of Perisomatic Inhibition in the Mouse Medial Prefrontal Cortex

Abstract

Perisomatic inhibition profoundly controls neural function. However, the structural organization of inhibitory circuits giving rise to the perisomatic inhibition in the higher-order cortices is not completely known. Here, we performed a comprehensive analysis of those GABAergic cells in the medial prefrontal cortex (mPFC) that provide inputs onto the somata and proximal dendrites of pyramidal neurons. Our results show that most GABAergic axonal varicosities contacting the perisomatic region of superficial (layer 2/3) and deep (layer 5) pyramidal cells express parvalbumin (PV) or cannabinoid receptor type 1 (CB1). Further, we found that the ratio of PV/CB1 GABAergic inputs is larger on the somatic membrane surface of pyramidal tract neurons in comparison with those projecting to the contralateral hemisphere. Our morphologic analysis of in vitro labeled PV+ basket cells (PVBC) and CCK/CB1+ basket cells (CCKBC) revealed differences in many features. PVBC dendrites and axons arborized preferentially within the layer where their soma was located. In contrast, the axons of CCKBCs expanded throughout layers, although their dendrites were found preferentially either in superficial or deep layers. Finally, using anterograde trans-synaptic tracing we observed that PVBCs are preferentially innervated by thalamic and basal amygdala afferents in layers 5a and 5b, respectively. Thus, our results suggest that PVBCs can control the local circuit operation in a layer-specific manner via their characteristic arborization, whereas CCKBCs rather provide cross-layer inhibition in the mPFC.

SIGNIFICANCE STATEMENT Inhibitory cells in cortical circuits are crucial for the precise control of local network activity. Nevertheless, in higher-order cortical areas that are involved in cognitive functions like decision-making, working memory, and cognitive flexibility, the structural organization of inhibitory cell circuits is not completely understood. In this study we show that perisomatic inhibitory control of excitatory cells in the medial prefrontal cortex is performed by two types of basket cells endowed with different morphologic properties that provide inhibitory inputs with distinct layer specificity on cells projecting to disparate areas. Revealing this difference in innervation strategy of the two basket cell types is a key step toward understanding how they fulfill their distinct roles in cortical network operations.

Introduction

The medial part of the PFC (mPFC) has a critical role in controlling various higher-order cognitive functions (Goldman-Rakic, 1988; Miller, 2000; Fuster, 2006), such as working memory, attention, decision-making, or fear behavior (Maren and Quirk, 2004; Salzman and Fusi, 2010; Euston et al., 2012; Tovote et al., 2015; Lee and Seo, 2016; Uddin, 2021), yet our knowledge about the circuit organization within this cortical area is still limited. The mPFC networks are composed of excitatory glutamatergic pyramidal neurons and a variety of GABAergic inhibitory cells (DeFelipe and Fariñas, 1992; Tasic et al., 2016; Gouwens et al., 2020; Anastasiades and Carter, 2021). Although the inhibitory neurons constitute only 10–25% of the whole neuronal population in the mPFC (Markram et al., 2004; Le Merre et al., 2021), they are able to effectively modulate and control the activity of large pyramidal cell populations (Isaacson and Scanziani, 2011; Kepecs and Fishell, 2014; Pelkey et al., 2017), to synchronize pyramidal cell firing at different frequencies (Buzsáki and Chrobak, 1995; Jung and Carlen, 2021), to channel the information flow within the local circuits (Reyes et al., 1998; Pouille and Scanziani, 2004; Isaacson and Scanziani, 2011; Cummings et al., 2022; Binette et al., 2023), and to promote the storage of information (Frankland and Bontempi, 2005; Topolnik and Tamboli, 2022).

Previous studies have shown that similar to other cortical regions several different GABAergic cell types are present in the mPFC (Kawaguchi and Kubota, 1997; Lucas and Clem, 2018; Gouwens et al., 2020; Singh and Topolnik, 2023). Different inhibitory interneurons innervate distinct, often nonoverlapping, membrane domains of postsynaptic neurons, allowing them to fulfill specific roles in microcircuit function (Miles et al., 1996; Petilla Interneuron Nomenclature et al., 2008; Murayama et al., 2009; Pouille et al., 2013; Veres et al., 2017; Lourenço et al., 2020). Perisomatic-region-targeting interneurons innervate mostly the somata, the proximal dendrites, and axon initial segment of pyramidal cells. As the synaptic inhibition of these membrane surfaces controls the spike generation very efficiently (Cobb et al., 1995; Miles et al., 1996; Veres et al., 2014; Veres et al., 2017), inhibitory cells innervating the perisomatic region are well suited to regulate the discharge of neurons (Cobb et al., 1995; Miles et al., 1996; Somogyi et al., 1998; Kawaguchi and Kondo, 2002). In cortical structures two types of basket cells (BCs), expressing either parvalbumin (PV) or cholecystokinin cannabinoid receptor 1 (CCK/CB1), give rise to innervation of the somata and proximal dendrites of pyramidal cells (Somogyi et al., 1998; Freund and Katona, 2007; Takács et al., 2015; Vereczki et al., 2016). Although these basket cells target the same membrane surface, they are distinct in their morphologic features and electrophysiological characteristics (Kawaguchi and Kubota, 1998; Hefft and Jonas, 2005; Klausberger et al., 2005; Glickfeld and Scanziani, 2006; Szabo et al., 2010; Vereczki et al., 2016; Barsy et al., 2017) as they have been found in other cortical structures, suggesting they play distinct roles in network operation (Freund, 2003). At present, however, our knowledge is limited regarding the function of CCKBCs in comparison with PVBCs in the mPFC. By revealing how basket cell networks are structurally organized in the mPFC, we can predict their function at the microcircuit level.

In this study, we first defined the layers within the mouse mPFC, which helped us to investigate the organization of perisomatic inhibition in a layer-dependent manner. Next, we determined the source and the ratio of GABAergic inputs innervating the perisomatic region of pyramidal cells. Then, we characterized and compared the morphologic features of interneurons providing the vast majority of perisomatic inputs on pyramidal cells. We observed morphologically distinct subgroups of basket cells based on the differences in their input/output properties. Finally, we examined whether the extraprefrontal cortical afferents target PV-containing interneurons in the different layers with distinct likelihood. Based on the morphologic differences, PVBCs in the mPFC can accomplish layer-specific regulation of pyramidal cell operation, whereas CCKBCs may control pyramidal cell function more broadly.

Materials and Methods

Experimental design

Experimental animals

Experiments were approved by the Committee of the Scientific Ethics of Animal Research (22.1/360/3/2011), and all procedures involving animals were performed according to methods approved by Hungarian legislation (1998 XXVIII, section 243/1998, renewed in section 40/2013) and institutional guidelines of the ethical code. All procedures complied with the European convention for the protection of vertebrate animals used for experimental and other scientific purposes (Directive 86/609/CEE and modified according to the Directives 2010/63/EU). Every effort was taken to minimize animal suffering and the number of animals used. Adult mice postnatal day 50–140 from the following lines were used: C57BL/6J, FVB/Ant-Fx (wild type, Charles River Laboratories), BAC-PV-eGFP (Meyer et al., 2002), BAC-CCK-DsRed (Maté et al., 2013), Ai6 reporter line (CAG-LSL-ZsGreen1, strain #007906, The Jackson Laboratory), and Pvalb-IRES-Cre (PV-Cre, strain #017320, The Jackson Laboratory).

Surgical procedures

Retrograde cell labeling

Anesthesia was induced with 125 mg/kg ketamine and 5 mg/kg xylazine. Wild-type mice from both sexes [n = 2 female, 2 male for basal amygdala (BA) and periaqueductal gray (PAG) injection and n = 5 female, 5 male for contralateral PFC (cPFC) and dorsal striatum (DS) injection] were secured in a stereotaxic frame. Unilateral or bilateral injections of Fluoro-Gold (FG) or choleratoxin B (CTB) subunit tracers dissolved in glycerol (FG 2% iontophoresis by 2 µA pulses with 2/2 s on/off duty cycle for 5 min and CTB 0.5% iontophoresis by 5 µA pulses with 2/2 s on/off duty cycle for 7–10 min) or retrograde pAAVrg-CAG-GFP (catalog #37825-AAVrg, Addgene) or retrograde AAVrg-EF1a-mCherry-IRES-Flpo (catalog #55634-AAVrg, Addgene) viruses (200 nl with 3 nl/s flow rate) were aimed at the following coordinates from bregma (in cm): for BA injections, AP, −0.15; ML, 0.31; DV, 0.43; for DS injections, AP, 0.06–0.07; ML, 0.13; DV, 0.23; for PAG injections, AP, −0.46–0.49; ML, 0.05; DV, 0.11–0.12; for cPFC injections, AP, 0.15–0.18; ML, next to the sinus; DV, 0.1–0.15. After 4–8 d of tracer injection and after 4 weeks of virus injection, mice were transcardially perfused with 4% paraformaldehyde (PFA) in 0.1 m phosphate buffer (PB) for 30–40 min, and PFC sections of 50–100 μm thickness were prepared using a Vibratome (Leica) and stored in 0.1 m PB with 0.05% Na-azide until further processing.

Anterograde trans-synaptic viral labeling

Anesthesia was induced with 125 mg/kg ketamine and 5 mg/kg xylazine. Male PV-Cre and both male and female Ai6 mice were secured in a stereotaxic frame, and bilateral injections of 200–300 nl adeno-associated virus (AAV)1 vectors (AAV1-EF1a-DIO-ChETA-eYFP (catalog #26968, Addgene) to PV-Cre mice and pENN-AAV1-hSyn-Cre-WPRE-hGH (catalog #105553, Addgene) to Ai6 mice) were aimed with a 3 nl/s flow rate at the following coordinates from bregma (in cm): for BA injections, AP, −0.15; ML, 0.32–0.35; DV, 0.44–0.50; for midline thalamus injections, AP, −0.12–0.13; ML, next to the sinus; DV, 0.30–0.33; for lateral entorhinal cortex (LEnt) injections, AP, −0.42; ML, 0.375; DV, 0.28. After 4 weeks of recovery, mice were transcardially perfused with 4% PFA in 0.1 m PB for 30–40 min, and PFC sections were prepared as described above.

Tissue processing and immunocytochemistry

All anatomic data, including those acquired with viral vectors and retrograde tracers, were obtained from immunostained brain slices. After perfusion, 50- to 100-μm-thick sections were prepared using a Vibratome (Leica VT100S). Slices were thoroughly washed in 0.1 m PB several times (4–5 times for 10–15 min). For fluorescent labeling, slices were blocked with 10% normal donkey serum (NDS; Vector Laboratories) or 10% normal goat serum (NGS; Vector Laboratories) and 0.5% Triton X-100 in 0.1 m PB for 30–60 min at room temperature. Then sections were incubated in a mixture of different primary antibodies diluted in PB containing 2% NDS or 2% NGS, 0.05% Na-azide and 0.5% Triton X-100 (except all PV immunolabeling where we applied 2% Triton X-100) overnight at room temperature for an additional 3–6 d at 4°C. The applied primary antibodies are listed in Table 1, grouped and numbered according to the following experiments: (1) determination of the borders of PFC layers, (2) visualization of GABAergic inputs on the perisomatic region, (3–6) visualization of inputs onto retrogradely labeled cells, (7, 8) validating the CCK or PV content in the transgenic mice, (9) examination of CB1 receptor content of visualized CCKBCs, (10) target distribution of CCKBCs and PVBCs on random PCs, and (11, 12) visualization of the monosynaptic input-receiving PV-containing cells. After washing out primary antibodies several times, slices were treated with secondary antibodies diluted in 0.1 m PB and 1% NDS or NGS for 2–4 h (Table 1). Following several washes in PB, sections were mounted on glass slides in VECTASHIELD (Vector Laboratories).

Table 1.

Antibodies used in anatomic experiments

Experiment no.	Primary antibody	Provider	Catalog no.	Concentration	Notes	Secondary antibody	Provider
1	Rabbit anti-WFS1	Proteintech	11558–1-AP	1:500	4 d 4°C	Alexa 647 DAR	Jackson
	Rat anti- Ctip2	Abcam	ab18465	1:1000		Alexa 488 DARat	Molecular Probes
	Mouse anti-FoxP2	Sigma-Aldrich	AMAB91362	1:5000		Cy3 DAM	Jackson
	Chicken anti-Calbindin	SYSY	214006	1:1000		Dyl405 DACh	Jackson
2	Mouse anti-Kv2.1	NeuroMab	75–014	1:1000	3 d 4°C	Alexa 647 DAM	Jackson
	Goat anti-CB1	Frontier Institute	CB1-Go-Af450	1:1000		Dyl405 DAG	Jackson
	Guinea pig anti-VGAT	SYSY	131004	1:1000		Alexa 488 DAGp	Jackson
		Frontier Institute	VGAT-GP-Af1000	1:1000
	Rabbit anti-PV	Swant	PV 25	1:5000		Cy3 DAR	Jackson
3	Rabbit anti-PV	Swant	PV 25	1:5000	6 d 4°C	Cy3 DAR	Jackson
	Goat anti-CB1	Frontier Institute	CB1-Go-Af450	1:1000		Alexa 488 DAG	Jackson
	Guinea pig anti-Fluorogold	Protos Biotech	NM-101 FluGgp	1:5000		Dyl405 DAGp	Jackson
	Mouse anti-Kv2.1	NeuroMab	75-014	1:1000		Alexa 647 DAM	Jackson
4	Rabbit anti-PV	Swant	PV 25	1:5000	6 d 4°C	Cy3 DAR	Jackson
	Goat anti-Coleratoxin B	List Biological Laboratories	703	1:20.000		Dyl405 DAG	Jackson
	Guinea pig anti-CB1	Frontier Institute	CB1-Go-Af530	1:1000		Alexa 488 DAGp	Jackson
	Mouse anti-Kv2.1	NeuroMab	75-014	1:1000		Alexa 647 DAM	Jackson
5	Guinea pig anti-PV	SYSY	195004	1:10.000	6 d 4°C	Cy3 DAGp	Jackson
	Goat anti-Coleratoxin B	List Biological Laboratories	703	1:20.000		Dyl405 DAG	Jackson
	Rabbit anti-CB1	Cayman	10006590	1:1000		Alexa 488 DAR	Jackson
	Mouse anti-Kv2.1	NeuroMab	75-014	1:1000		Alexa 647 DAM	Jackson
6	Guinea pig anti-PV	SYSY	195004	1:10.000	6 d 4°C	Cy5 DAG	Jackson
	Goat anti-CB1	Frontier Institute	CB1-Go-Af450	1:1000		Dyl405 DAG	Jackson
	Rat RFP	Chromotec	5F8	1:1000		Cy3 DARat	Jackson
7	Rabbit anti-proCCK	Frontier Institute	CCK-pro-Rb-Af350	1:1000	3 d 4°C	Alexa 488 DAR	Jackson
8	Rabbit anti-PV	Swant	PV 25	1:5000	3 d 4°C	Cy3 DAR	Jackson
	Goat anti-eGFP	Abcam	ab5450	1:5000		Alexa 488 DAG	Jackson
9	Goat anti-CB1	Frontier Institute	CB1-Go-Af450	1:1000	6 d 4°C	Dyl405 DAG	Jackson
10	Mouse anti-Kv2.1	NeuroMab	75-014	1:1000	6 d 4°C	Alexa 647 DAM	Jackson
11	Rabbit anti-PV	Swant	PV 25	1:5000	4 d 4°C	Cy3 DAR	Jackson
12	Goat anti-PV	Swant	PVG-214	1:5000	4 d 4°C	Cy3 DAG	Jackson

DAR, Donkey anti-rabbit; DARat, donkey anti-rat; DAM, donkey anti-mouse; DACh, donkey anti-chicken; DAG, donkey anti-goat; DAGp, donkey anti-guinea pig. All the secondary antibodies were used in a 1:500 dilution.

Interneuron recording, labeling in vitro

In vitro biocytin labeling of interneurons was conducted as described previously in detail (Veres et al., 2014). Briefly, CCK-DsRed and PV-eGFP mice were deeply anesthetized with isoflurane and decapitated. The brain was quickly removed and placed into ice-cold cutting solution containing the following (in mm): 252 sucrose, 2.5 KCl, 26 NaHCO₃, 1 CaCl₂, 5 MgCl₂, 1.25 NaH₂PO₄, and 10 glucose, bubbled with 95% O₂ and 5% CO₂ (carbogen gas). Coronal slices of 200 μm thickness containing the PFC region were prepared with a Leica VT1200S Vibratome and kept in an interface-type holding chamber containing artificial CSF (ACSF) at 36°C, which gradually cooled down to room temperature. ACSF contained the following (in mm): 126 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 2 MgCl₂, 2 CaCl₂, 26 NaHCO₃, and 10 glucose, bubbled with carbogen gas. Interneurons were selected based on the presence of the fluorescent proteins (DsRed or eGFP) excited by a UV lamp and visualized by a charge-coupled device camera (Hamamatsu Photonics or Andor Zyla). Targeted cells were recorded under visual guidance using differential interference contrast microscopy (Olympus BX61W or Nikon FN1) at 50–100 μm below the surface of the acute slice. Interneurons were recorded in whole-cell mode using a K-gluconate-based intrapipette solution, with biocytin to label their processes, containing the following (in mm): 110 K-gluconate, 4 NaCl, 2 Mg-ATP, 20 HEPES, 0.1 EGTA, 0.3 GTP (sodium salt), 10 phosphocreatine, and 0.2% biocytin adjusted to pH 7.3 using KOH and with an osmolarity of 290 mOsm/L. Firing patterns of the cells were tested with 800-ms-long alternating depolarizing and hyperpolarizing current steps with increasing amplitudes (10–600 pA, 10–100 pA steps). After the recordings, slices were fixed in 4% PFA and Alexa 488–coupled streptavidin (1:10,000 in TBS; Invitrogen) or Alexa 647–coupled streptavidin (1:10,000 in TBS; Invitrogen) was used to visualize the fine details of the neurons in the entire slice.

Images and analysis

Fluorescent images were taken with a Nikon A1R or C2 confocal laser scanning microscope using the following distinct settings for different objectives: CFI Super Plan Fluor 20×, NA, 0.45; z step size, 1 μm; x, y, 0.62 μm/pixel for the cell reconstruction; CFI Plan Apo VC10X, NA, 0.30; single plane; x, y, 0.31 μm/pixel for the determination of the borders of layers within the mPFC and for the validation of reporter protein content of animals; CFI Plan Apo VC60X Oil objective, NA, 1.40; z step size, 0.13 μm; x, y, 0.08 μm/pixel for the visualization of GABAergic inputs on the perisomatic region and for the analysis of the density of CB1 and PV boutons on the retrogradely labeled cells; CFI Plan Apo VC60X Oil objective, NA, 1.40; z step size, 0.2; x, y, 0.21 μm/pixel for the analysis of target distribution of basket cells on random pyramidal cells; and CFI Plan Apo VC10X, NA, 0.30; z step size, 3 μm; x, y, 0.63 μm/pixel for the visualization of the monosynaptic input receiving PV-containing cells.

Reconstruction and analysis of the 3D confocal images were performed with Neurolucida 10.53 software (MBF Bioscience) and NIS Elements software (Nikon). The properties of axonal and dendritic arbors and surface analysis were performed with Neurolucida Explorer software (MBF Bioscience). Values were corrected for shrinkage and flattening of the tissue (x, y, and z axes correction on pictures taken by using CFI Plan Apo VC60X Oil objective, 1.08; x and y-axes correction was 1, and z-axis correction was 2.5 on pictures taken from the biocytin-labeled basket cells by using CFI Super Plan Fluor 20× objective). Schematic representation of brain slices from thalamus injection and PFC were achieved using the Inkscape software open access program.

Quantification of inputs on Kv2.1-immunolabeled somata

During quantification, different aspects were taken into account. Boutons were considered putative contacts if no apparent gap was visible between the labeled bouton and the surface of Kv2.1-stained somata when examined in 3D view of confocal images. Varicosities located at branch points were not counted as putative contacts, as it has been shown by electron microscopy that these boutons do not form synapses (Veres et al., 2014; Veres et al., 2017).

Statistical and cluster analysis

For comparison of data with normal distribution according to the Shapiro–Wilk test, the two-sample t test and ANOVA were used. For data with non-normal distribution, the Mann–Whitney (M–W) U test, Wilcoxon signed rank test, and Kruskal–Wallis (K–W) ANOVA were used. For post hoc analysis, Dunn's test or the M–W test was used. For the comparison of distributions, the two-sample Kolmogorov–Smirnov (K–S) test and chi-square homogeneity test were used. All statistics were performed using Origin 8.6 or 9.2 software or online LibreTexts statistical programs (https://stats.libretexts.org/). Asterisks on figures represent significant differences. Exact p values were indicated when p was higher than 0.001, considering the rounding rules. Data are presented as mean ± SEM, unless indicated otherwise. The cluster and principal component analysis (PCA) were performed using Origin 8.6 or 9.2 software. The PCA showed the main components from the five axonal and five dendritic distribution ratios between layers, and cluster analysis applying Ward's method was made based on these selected main components.

Results

Defining the layers in the mPFC

As the layers in the mPFC are not labeled consistently across studies (Clarkson et al., 2017; Lu et al., 2017; Anastasiades et al., 2018), we first defined the distinct layers by visualizing a combination of markers that have already proven to be good tools for differentiating layers in other cortical areas, like the primary somatosensory cortex (S1; Fig. 1A) that served as a well-defined layered cortical structure for comparison (Cruikshank et al., 2001; Arlotta et al., 2005; Luuk et al., 2008; Hisaoka et al., 2010). Our study was conducted primarily in the prelimbic (PrL) area of the mouse mPFC, but as the borders between this area and the neighboring cortical regions (the anterior cingulate cortex and the infralimbic cortex) are ill defined, we refer to the investigated area as the mPFC. To visualize the borders between the two superficial layers, layers 2 and 3, the transcription factor Wolfram syndrome 1 protein (WFS1) and the calcium-binding protein Calbindin (Calb) were used. WFS1 revealed the neurons in layer 2 (Luuk et al., 2008; Fig. 1B_1–2), whereas Calb was expressed at high levels in both layers 2 and 3 (Cruikshank et al., 2001; Fig. 1C_1–2). We surprisingly observed that in the dorsal part of the mPFC, these two layers fuse and form a narrow layer 2/3 (Fig. 1B₂–C₂, preventing the separation of these two superficial layers. To determine the borders of deep layers, two additional transcription factors were applied. COUP-TF (chicken ovalbumin upstream promoter transcription factor) interacting protein 2 (Ctip2) was found to be present in neurons located in layer 5b and layer 6 (Arlotta et al., 2005; Fig. 1D_1–2), whereas Forkhead box protein P2 (FoxP2) was expressed predominantly in layer 6 neurons (Hisaoka et al., 2010; Fig. 1E_1–2). As layer 4 cannot be defined in the rodent PFC (Uylings et al., 2003), the borders between layers in the mPFC are clearly determined by the visualization of these markers (Fig. 1F), allowing us to investigate the structural organization of perisomatic inhibition in a layer-dependent manner

Figure 1.

Defining the layers in the mPFC by using antibodies developed against Calb and transcription factors WFS1, Ctip2, and FoxP2. A, A low-magnification multicolor confocal image taken from an mPFC section immunostained against WFS1, Calb, Ctip2, and FoxP2. White empty rectangles indicate the PrL subregion of the mPFC and the S1 (shown at a higher magnification in B–E). Scale bar, 1 mm. B–E, Higher-magnification confocal images taken from the S1 and PrL region and immunolabeled for WFS1, Calb, Ctip2, and FoxP2. Dashed white lines indicate the boundaries of the layers (L) defined by the immunolabeling. F, A multicolor confocal image taken from a PrL region that was immunostained against WFS1 (green), Ctip2 (yellow), and FoxP2 (magenta), indicating the clear boundaries of the layers. G, H, Immunostaining against PV and CB1 in the PrL region of mPFC. Scale bar, 100 µm.

The vast majority of perisomatic GABAergic inputs originate from CB1- and PV-expressing axonal varicosities

The perisomatic membrane surface of neurons is a distinctive site, where the action potential generation can be controlled powerfully by inhibitory GABAergic synapses (Cobb et al., 1995; Miles et al., 1996; Veres et al., 2017). As prior studies elucidated, mostly PVBCs and CCKBCs provide these inhibitory axon terminals onto the perisomatic region of pyramidal cells in cortical areas studied so far (Bodor et al., 2005; Freund and Katona, 2007; Takács et al., 2015; Vereczki et al., 2016). However, at present it is unclear how the perisomatic inhibition originating from basket cells expressing PV or CB1 is organized in the mPFC at the structural level. Immunostaining against PV and CB1 revealed that the distribution of axonal boutons expressing these proteins varied among the layers in the mPFC (Fig. 1G,H), implying that pyramidal cells in different layers may receive a distinct ratio of perisomatic inputs from PVBCs and CCKBCs. To support this hypothesis, we first assessed the sources of perisomatic GABAergic inputs in the distinct layers of mPFC using multichannel high-resolution confocal microscopic investigations. The perisomatic region of neurons was visualized by immunostaining against the voltage-gated potassium channel subunit Kv2.1 (Vereczki et al., 2016), whereas the vesicular GABA transporter (VGAT) was used to reveal GABAergic axonal varicosities and their neurochemical content by immunostaining against PV and CB1 (Fig. 2A). Our analysis uncovered that the vast majority of the VGAT-immunolabeled boutons were immunopositive for PV or CB1 in layers 2/3, 5a, and 5b (Fig. 2B). Interestingly, we found that neurons in layer 5b receive a significantly larger PV+/CB1+ perisomatic input ratio than neurons in layer 2/3 and layer 5a (Fig. 2D,E). As pyramidal tract (PT) neurons projecting to subcortical targets are located predominantly in layer 5b in the mPFC (Gabbott et al., 2005; Kawaguchi, 2017; Anastasiades and Carter, 2021), our observations may indicate that depending on the target area of pyramidal cell axons, they could be innervated by various ratios of PV+/CB1+ boutons (Bodor et al., 2005; Varga et al., 2010) in a layer-selective manner (Fig. 1G,H). To directly test this hypothesis, we examined the CB1+ and PV+ axonal varicosities on the Kv2.1-labeled perisomatic region of pyramidal cells projecting to the BA, cPFC, DS, and PAG (Fig. 3). The soma location of pyramidal cells projecting to these remote areas was similar to that described earlier (Gabbott et al., 2005; Lu et al., 2017; Anastasiades and Carter, 2021; Fig. 3B,C). The analysis of the ratio of PV+/CB1+ boutons forming close appositions on the perisomatic membrane surface of pyramidal cells projecting to distinct regions revealed that the proportion of these two GABAergic inputs was similar in layers 2/3 and 5a regardless of where pyramidal neurons projected (Fig. 3D–G). This ratio, however, showed a difference in layer 5b and was higher on PAG-projecting pyramidal cells compared with cPFC-projecting pyramidal cells (Fig. 3G).

Figure 2.

The vast majority of perisomatic GABAergic inputs originate from CB1- and PV-expressing boutons. A, Multicolor maximum z-intensity projection confocal images taken of sections immunostained for Kv2.1, CB1, PV, and VGAT. Blue arrows indicate the CB1- and VGAT-coexpressing boutons on Kv2.1-labeled profiles, orange arrows point to PV- and VGAT-coexpressing terminals that are in close apposition with Kv2.1-immunostained soma membranes, whereas green arrows show VGAT-immunopositive boutons that lack PV or CB1 immunoreactivity on Kv2.1-labeled somata. Scale bar, 10 µm. B, The ratio of boutons expressing CB1, PV, and VGAT only on the Kv2.1-immunostained somata in three different layers (L) of the mPFC. (L2/3, only VGAT, 9.61 ± 1.88%; VGAT+ PV+, 50.02 ± 2.79%; VGAT+ CB1+, 40.37 ± 2.44%; L5a, only VGAT, 14.47 ± 2.6%; VGAT+ PV+, 46.69 ± 3.5%; VGAT+ CB1+, 38.84 ± 2.94%; L5b, VGAT, 7.69 ± 1.64%; VGAT+ PV+, 61.43 ± 3.46%; VGAT+ CB1+, 30.88 ± 2.92%). C, CB1 and PV bouton density on the surface of Kv2.1-labeled somata in different layers. Black lines represent the mean in each case; mean (bouton/µm²), L2/3 CB1, 0.0787; L2/3 PV, 0.0968; L5a CB1, 0.0731; L5a PV, 0.0971; L5b CB1, 0.0683; L5b PV, 0.1341. D, The ratio of PV- and CB1-expressing boutons on Kv2.1-labeled membrane profiles in distinct layers. Each dot represents the ratio that was determined on single pyramidal neurons, and black lines show the mean; L2/3, n = 26; L5a, n = 28; L5b, n = 19 somata were examined (1-way ANOVA, p = 0.0022; two sample t test, L2/3 vs L5a, p = 0.59; L2/3 vs L5b, p = 0.012; L5a vs L5b, p = 0.0061; mean, L2/3, 1.442; L5a, 1.327; L5b, 2.712.) n.s., not significant. E, Cumulative probability distributions of PV/CB1 ratios in different layers. Kolmogorov–Smirnov test confirmed that the distribution of the PV/CB1 ratio in L5b is different in comparison with other layers (L2/3 vs L5a, p= 0.73; L2/3 vs L5b, p < 0.001; L5a vs L5b, p < 0.001).

Figure 3.

Inhibitory inputs on the perisomatic region of pyramidal cells projecting to different brain regions. A, Schematic representation of the experimental procedure. Retrograde tracers were injected into the BA, cPFC, DS, and PAG, respectively. After 4–8 d of tracer injections, PFC-containing sections were prepared and immunolabeled. B, Multicolor confocal images taken from the mPFC showing the distribution of retrogradely labeled cells (magenta) in distinct layers. Scale bar, 100 µm. C, Layer-specific distribution of the somata of pyramidal neurons targeting distinct brain regions. Number of the quantified retrogradely labeled cells is as follows: BA, n = 240 in 6 slices from 2 mice; cPFC, n = 1249 in 9 slices from 3 mice; DS, n = 2158 in 9 slices from 3 mice; PAG, n = 70 in 6 slices from 2 mice. D, Multicolor maximum z-intensity projection confocal images taken from a section immunostained for Kv2.1, FG, CB1, and PV to visualize the perisomatic CB1 and PV inputs (arrows) on FG-labeled neurons. Scale bar, 5 µm. E, PV bouton density on the perisomatic surface of pyramidal neurons projecting to distinct brain regions. Red dots indicate cells located in layer (L)2/3, gray dots represent pyramidal cells located in L5a, and purple dots show pyramidal cells located in L5b. Each dot represents the PV bouton density on single pyramidal neurons. Black lines indicate means. F, CB1 bouton density on the perisomatic membrane surface of pyramidal cells projecting to distinct brain regions. Colors, lines, dots, and numbers are the same as in E. G, Perisomatic PV/CB1 ratio on pyramidal cells projecting to different brain areas and located in distinct layers. Red lines indicate means. Numbers of examined cells by layers and by projection brain areas are the following: L2/3, BA, n = 25; cPFC, n = 22; DS, n = 12; L5a, BA, n = 17; cPFC, n = 36; DS, n = 20; L5b, cPFC, n = 17; DS, n = 10; PAG, n = 24. BA-projecting cells were counted in 6 slices from 2 mice, cPFC-projecting cells were counted in 10 slices from 4 mice, DS-projecting cells were counted in 6 slices from 2 mice, and PAG-projecting cells were counted in 4 slices from 2 mice. K–W ANOVA, L2/3, p = 0.104; L5a, p = 0.633; and L5b, p = 0.026. In L5b M–W test showed the difference between cPFC- and PAG-projecting neurons; p = 0.012; DS-PAG, p = 0.121; cPFC-DS, p = 0.303. n.s, not significant.

In summary, our results showed that (1) pyramidal cells in layers 2/3, 5a, and 5b receive the vast majority of their perisomatic GABAergic inputs from two sources, and (2) the ratio of PV+/CB1+ GABAergic inputs in deep layers were higher on pyramidal cells projecting to the PAG, indicating that the primary source of perisomatic inhibition for PT neurons is PV+ cells.

Validation of reporter protein expression in two transgenic mouse lines

Previous studies in other cortical regions have uncovered that CB1+ and PV+ axonal varicosities innervating the perisomatic region of pyramidal cells originate from two distinct kinds of basket cells (Hájos et al., 2000; Bodor et al., 2005; Hefft and Jonas, 2005; Glickfeld and Scanziani, 2006; Freund and Katona, 2007; Takács et al., 2015; Vereczki et al., 2016). To selectively target and study the morphologic characteristics of these interneurons, transgenic mouse strains can successfully be used. In our earlier studies, we used two transgenic mouse lines—BAC-CCK-DsRed (Maté et al., 2013) and BAC-PV-eGFP (Meyer et al., 2002)—to target CCKBCs and PVBCs in the hippocampus and basal amygdala (Gulyás et al., 2010; Szabó et al., 2014; Vereczki et al., 2016). Before examining the features of interneurons that can be targeted in acute slices containing the prefrontal cortical region, we validated the reporter protein expression in the mPFC in these two transgenic mouse lines (Fig. 4). Using immunostaining, we observed that both in layer 2/3 and layer 5 the majority of strong DsRed+ cells were immunoreactive for proCCK (74.1%, n = 80 proCCK⁺ of 108 DsRed+ cells), whereas neurons expressing weaker DsRed signals lacked detectable immunolabeling for this neuropeptide (Fig. 4B_1–2,D). As this proCCK antibody stains GABAergic interneurons in the hippocampus and basolateral amygdala (Kotzadimitriou et al., 2018; Rhomberg et al., 2018), we hypothesized that neurons displaying strong DsRed signals would also be interneurons in this cortical region, whereas neurons expressing weaker DsRed signals might belong to a subset of pyramidal cells. This assumption has been confirmed by subsequent in vitro whole-cell recordings, showing that strong DsRed+ cells were indeed interneurons shown by their spine-free dendrites and axon morphology (95%, n = 42 CCKBC of 44 biocytin-filled neurons). Although half of DsRed+ neurons in layer 1 showed immunoreactivity for proCCK (48.9%, n = 22 proCCK⁺ of 45 DsRed+ cells), they were excluded from the analysis as basket cells are not present in this layer (Markram et al., 2004; Jiang et al., 2013; Schuman et al., 2019). Our data imply that although not all strong DsRed+ cells are proCCK immunopositive both in layer 2/3 and layer 5, which can be because of the sensitivity of the antibody labeling, this mouse line can be used to target CCK⁺ interneurons with reasonable efficacy. Importantly, none of the strong DsRed+ cells showed immunoreactivity for PV (n = 0 PV+ of 66 DsRed+ cells from 2 mice) and no PV immunoreactive neurons expressed DsRed (n = 0 DsRed+ from 102 PV+ interneurons from 2 mice). In BAC-PV-eGFP mice, our analysis demonstrated a robust coexpression of PV and eGFP (93.7%, n = 686 PV+ of 732 eGFP+ cells; Fig. 4C_1–2,D).

Figure 4.

Validation of reporter protein expression in two transgenic mouse lines. A, Schematic representation of the experimental procedure. B₁, Multicolor confocal images taken from an mPFC-containing section prepared from a BAC-CCK-DsRed mouse; proCCK was revealed by immunostaining. Red arrows show the genetically labeled (L) DsRed+ cells, green arrows represent the proCCK-immunopositive cells, and yellow arrows indicate the colocalization of the two markers. Dashed white lines represent the boundaries of layers. Scale bar, 100 µm. B₂, An image of representative cells taken at a higher magnification, showing strongly labeled neurons, which are interneurons, and weakly labeled ones, which are pyramidal cells. Scale bar, 10 µm. C₁, Multicolor confocal images taken from the mPFC of a BAC-PV-EGFP mouse. The section was immunostained for PV. Green arrows show the genetically labeled EGFP cells, red arrows represent the PV-immunolabeled neurons, and yellow arrows indicate the colocalization of the two markers. Dashed white lines represent the boundaries of layers. Scale bar, 100 µm. C₂, An image of representative cells taken at a higher magnification. Scale bar, 10 µm. D, The ratio of DsRed/CCK (n = 130 in 3 slices from 2 mice) and EGFP/PV (n = 794 in 5 slices from 2 mice) cells within layers 2/3 and 5 of the mPFC. E₁, Maximum z-intensity projection image taken of an in vitro biocytin-filled CCKBC. Scale bar, 50 µm. E₂, Confocal images present the CB1 content of a biocytin-containing bouton of the same CCKBC. Scale bar, 1 µm. F, Maximum z-intensity projection image taken of an in vitro biocytin-filled PVBC. Scale bar, 50 µm. G, Maximum z-intensity projection image taken of an in vitro biocytin-filled chandelier cell (ChC). Inset, A typical bouton cartridge that characterizes ChCs. Scale bar, 50 µm; inset, 2 µm. Insets in E₁, F, and G show representative voltage responses of interneurons on current injections (−100 pA and +500 pA). Scale bars, x-axis, 200 ms; y-axis, 20 pA.

As we have revealed earlier, CB1 immunoreactivity could be detected in axon terminals of CCKBCs sampled in the hippocampus or basal amygdala of the BAC-CCK-DsRed mouse line (Szabó et al., 2014; Andrási et al., 2017; Veres et al., 2017). To test whether the strong DsRed+ neurons express CB1 on their axon terminals in the mPFC as well, supporting the adequateness of this mouse line to specifically target CCKBCs, we tested the CB1 immunoreactivity of biocytin-filled neurons (Fig. 4E₂). We found, indeed, that all biocytin-labeled interneurons tested expressed this type of cannabinoid receptor in their axon terminals (n = 20 CB1+ of 20 biocytin-filled neurons), that is, strong DsRed+ neurons in the mPFC are likely CB1/CCKBCs (Fig. 4E₁). Importantly, none of the DsRed+ interneurons displayed a fast-spiking phenotype, which is typical for PVBCs and chandelier cells (Povysheva et al., 2013; Miyamae et al., 2017). As PV is present both in basket and chandelier cells in cortical structures, we separated these interneurons based on the presence of cartridges formed by axonal boutons (Fig. 4G, inset), which is a characteristic and specific feature of chandelier cells (Somogyi, 1977), enabling the dissection of this GABAergic cell type from PVBCs based on the morphology of their axonal arbor. In this study, we distinguished 53 basket cells (Fig. 4F) and 10 chandelier cells (Fig. 4G). Importantly, chandelier cells expressing eGFP were sampled only in layer 2/3 but not in the deeper layers and were excluded from this study. Thus, in slices prepared from these transgenic mouse lines, basket cells expressing CCK and CB1 or PV can be readily sampled in the mPFC.

Morphologic characterization of basket cells in the mPFC

As the axonal boutons expressing PV+ and CB1+ are not evenly distributed in mPFC layers (Fig. 1G,H), we tested whether the dendritic and axonal tree arborization of basket cells had any layer preference. Therefore, in the next set of experiments, we aimed to investigate the morphologic characteristics of basket cells in the mPFC. As biocytin was added in the pipette solution for whole-cell recordings obtained in acute brain slices, we could reconstruct the dendritic and axonal arborizations of interneurons after visualizing the tracer molecules (Fig. 4A). Complete biocytin filling of the cells was confirmed by the presence of clear labeling in all the processes without a sign of a fading fluorescent signal, indicative for incomplete filling. Initial fluorescent microscopic examination suggested that morphologically distinct subpopulations of CCKBCs and PVBCs most probably exist in the mPFC. Therefore, we performed principal component and cluster analysis on the morphologic properties of dendrites and axons to uncover subgroups within the two types of basket cells (Fig. 5A_1,2). Based on the dendritic localization, CCKBCs were divided into superficial and deep morphologic subgroups; basket cells in the former group had dendrites predominantly in layer 1, whereas basket cells in the latter group arborized primarily in layer 5a (Fig. 5A₁,B₁–D₂). Regardless of the location of the somata and dendrites of the CCKBC (Fig. 5E), their axonal arbors were similar in size and were mainly restricted to layer 5a (Fig. 5F). Thus, based on the location of the dendritic tree, CCKBCs could be divided into two morphologic categories; however, both groups innervate pyramidal cells in the same layer of the mPFC.

Figure 5.

Distribution of dendritic and axonal arbor of different basket cell types within the layers of the mPFC. A_1–2, Cluster analysis of CCKBCs determined the superficial and deep categories based on the location of their dendrites, and analysis of PVBCs divided them into three subgroups based on the location of their dendrites and axons. B_1–5, Neurolucida reconstructions of dendritic and axonal arbor of different basket cells labeled in slice preparations. Black lines represent the axons, and colored lines show the dendritic trees, respectively. The dendrites of superficial CCKBCs are shown in light blue (B₁; n = 7 from 4 mice), and the dendrites of deep CCKBCs are shown in dark blue (B₂; n = 8 from 6 mice). The dendrites of layer (L)2/3 PVBCs are shown in yellow (B₃; n = 9 from 9 mice), the dendrites of L5a PVBCs are shown in light orange (B₄; n = 14 from 10 mice), and the dendrites of L5b PVBCs are shown in bright orange (B₅; n = 9 from 7 mice). Dashed dark gray lines represent the boundaries of the layers. Scale bar, 100 µm. C_1–5, The distribution of dendritic lengths in the different layers. D_1–5, The distribution of axonal lengths in the different layers. E, Comparison of the distribution of the dendritic lengths in different layers for the two CCKBC subgroups (M–W tests by layers, L1, p = 0.001; L2/3, p = 0.013; L5a, p = 0.001; L5b, p = 0.016). F, Comparison of the distribution of the axonal lengths of the two CCKBC subgroups in different layers (M–W tests by layers, L1, p = 0.013; L2/3, p = 0.003; L6, p = 0.019). G, Comparison of the distribution of dendritic lengths of the three PVBCs subgroups in different layers (K–W ANOVA by layers, L1, p < 0.001; L2/3, p < 0.001; L5a, p < 0.001; L5b, p < 0.001; L6, p = 0.007; post hoc Dunn's test by layers, L1, L2/3 PVBC vs L5a PVBC, p = 0.002; L2/3 PVBC vs L5b PVBC, p < 0.001; L5a PVBC vs L5b PVBC, n.s.; L2/3, L2/3 PVBC vs L5a PVBC, p = 0.005; L2/3 PVBC vs L5b PVBC, p < 0.001; L5a PVBC vs L5b PVBC, n.s.; L5a, L2/3 PVBC vs L5a PVBC, p = 0.006; L2/3 PVBC vs L5b PVBC, n.s.; L5a PVBC vs L5b PVBC, p < 0.001; L5b, L2/3 PVBC vs L5a PVBC, p = 0.03; L2/3 PVBC vs L5b PVBC, p < 0.001; L5a PVBC vs L5b PVBC, p = 0.008; L6, L2/3 PVBC vs L5a PVBC, n.s.; L2/3 PVBC vs L5b PVBC, p = 0.005; L5a PVBC vs L5b PVBC, n.s.). H, Comparison of the distribution of axons of the three PVBCs subgroups in different layers. (K–W ANOVA by layers, L1, p < 0.001; L2/3, p < 0.001; L5a, p < 0.001; L5b, p < 0.001; L6, p = 0.005; post hoc Dunn's test by layers, L1, L2/3 PVBC vs L5a PVBC, p = 0.005; L2/3 PVBC vs L5b PVBC, p < 0.001; L5a PVBC vs L5b PVBC, n.s.; L2/3, L2/3 PVBC vs L5a PVBC, p = 0.007; L2/3 PVBC vs L5b PVBC, p < 0.001; L5a PVBC vs L5b PVBC, p = 0.04; L5a, L2/3 PVBC vs L5a PVBC, p < 0.001; L2/3 PVBC vs L5b PVBC, n.s.; L5a PVBC vs L5b PVBC, p < 0.001; L5b, L2/3 PVBC vs L5a PVBC, p = 0.02; L2/3 PVBC vs L5b PVBC, p < 0.001; L5a PVBC vs L5b PVBC, p = 0.01; L6, L2/3 PVBC vs L5a PVBC, n.s.; L2/3 PVBC vs L5b PVBC, p = 0.004; L5a PVBC vs L5b PVBC, n.s.).

In comparison with CCKBCs, PVBCs showed a larger diversity. Based on the location of their dendrites and axons, cluster analysis separated PVBCs into three subgroups (Fig. 5A₂,B₃–D₅). The layer 2/3 PVBCs spread approximately half their axons within layer 2/3 (Fig. 5H), whereas their dendrites were located in layers 1, 2/3, and 5a with similar ratios (Fig. 5C₃,G). The layer 5a PVBCs and layer 5b PVBCs extended the majority of their dendrites and axons in layer 5a and layer 5b, respectively, with the vast majority of processes in the same layer where their soma was located (Fig. 5G,H). These observations indicate that PVBCs mostly innervate neurons in the same layers where they receive most of their inputs.

Next, we compared the characteristics of the dendritic and axonal arbors between the two basket cell types. First, when comparing the total length of the dendrites and axons we found no difference between CCKBCs and PVBCs (Fig. 6A,B), yet PVBCs spread their dendrites farther from the soma than CCKBCs (Fig. 6H₁). Although both cell types showed a similar axonal length, the axon of PVBCs had significantly more axonal nodes resulting in a lower length/nodes ratio in comparison with CCKBCs (Fig. 6C,D), indicating a more branchy axonal cloud. Second, for both basket cell types we correlated the total axonal length with the total dendritic length; a significant relationship was observed between these two parameters for PVBCs but not for CCKBCs (Fig. 6E_1–2). These results imply that those PVBCs that are collecting more inputs within the microcircuit may inhibit more pyramidal cells. Third, we examined the fine structure of their dendrites and axons. PVBCs emitted more primary, secondary, and tertiary dendrites, whereas CCKBCs had longer higher-order dendritic segments (Fig. 6F). Regarding the axons, we observed, using Sholl analysis (Fig. 6G_1–2), that the reconstructed CCKBCs had longer axons 250–400 µm from their somata than PVBCs, implying that the output structure of the two basket cells is distinct (Fig. 6H₂). Based on detailed morphologic analysis, these findings show that in comparison with CCKBCs, PVBCs have more complex dendritic and axonal trees, which arborize mainly within 200 µm from the somata (Fig. 6H_1–2). These structural features may be in line with their different roles played in mPFC operation.

Figure 6.

Comparison of the dendritic and axonal arborizations of CCKBCs and PVBCs. A–D, Box chart comparison of the total dendritic and axonal length, total number of axonal nodes, and average axonal length between nodes. The mean (small open square), the median (continuous line within the box), the interquartile range (box), and the 5–95% values (ends of whiskers bar) are plotted. (Statistical comparisons were performed with an M–W test, (C) p = 0.022; (D) p < 0.001.) E_1,2, Total axonal length as a function of the total dendritic length. Each dot corresponds to the values of a reconstructed BC. A significant relationship was found between these two values in the case of PVBCs (Pearson's r, r = 0.71, p < 0.001). F, Comparison of the dendritic length as a function of the dendritic order (M–W test, order 1, p < 0.001; order 2, p < 0.001; order 3, p = 0.019; order 4, p = 0.023; order 5, p < 0.001; order 6, p = 0.018). G₁,₂, Neurolucida reconstructions of dendritic and axonal tree of two example BCs labeled in slice preparations. Black lines represent the axons, colored lines show the dendritic trees. Concentric gray circles drawn on the reconstructions illustrate the 50 µm radii used for Sholl analysis. Dashed gray lines represent the boundaries of the layers (L) in the mPFC. Scale bar, 50 µm. H₁, The dendritic length as a function of the distance from the soma. (M–W test showed differences at 0–50 µm distance from the soma, p < 0.001, and at 100–150 µm, p = 0.001). H₂, The axonal length as a function of the distance from the soma. M–W test showed differences at 0–50 µm distance from the soma (p = 0.002), at 200–250 µm (p < 0.001), at 250–300 µm (p < 0.001), at 300–350 µm (p < 0.001), and at 350–400 µm (p = 0.002).

Postsynaptic target distribution of basket cells within the mPFC

To verify that interneurons sampled in the two transgenic mouse lines indeed provide the perisomatic innervation of neurons in the mPFC as in other cortical circuits (Glickfeld and Scanziani, 2006; Földy et al., 2007; Kohus et al., 2016; Veres et al., 2017), we investigated their postsynaptic target distribution. Therefore, we examined the ratio of axonal varicosities forming close appositions with the Kv2.1-labeled perisomatic region to those avoiding them (Fig. 7A,B). Previous studies in other neocortical regions have shown that on average, ∼30% of the axonal varicosities of a basket cell targets the somata of pyramidal cells, surrounding them in a basket-like manner (Kisvarday, 1992; Karube et al., 2004; Kubota et al., 2015). Our analysis revealed that ∼55% of the boutons of both CCK⁺ and PV+ interneurons targeted the somata (∼27%) and proximal dendrites (27–30%; Fig. 7D), an observation that proves that our recorded neurons are basket cells indeed. Importantly, although the ratio of axonal varicosities contacting the somata was fairly similar between the two basket cell types and between the morphologically distinct subpopulations (Fig. 7C–E), there was a substantial variance in the ratio of boutons innervating the proximal dendritic segments within both main groups (Fig. 7E). In addition, we addressed the question of whether there is a difference in the number of perisomatic contacts originating from CCKBCs and PVBCs at the single-cell level and how the number of contacts on the postsynaptic neurons changes as a function of distance measured from the interneuron. Therefore, we examined the bouton numbers on ∼20 Kv2.1-labeled neurons located inside or outside a circle with a radius of 150–200 µm from the in vitro filled basket cells. Interestingly, we observed that single Kv2.1-labeled neurons received more boutons from a PVBC than from a CCKBC regardless of the distance (Fig. 7F). Additionally, we found that the number of perisomatic contacts originating from both basket cell types decreased as a function of distance, and this reduction was significantly greater in the case of PVBCs (reduction of CCKBCs, 17.93% ± 5.51; reduction of PVBCs, 32.32% ± 3.9; M–W test, p = 0.045; Fig. 7F). Overall, our findings show that (1) CCK⁺ and PV+ interneurons (excluding chandelier cells) preferentially target the perisomatic region of mPFC neurons; that is, they are indeed basket cells, and (2) there are no interneurons expressing CCK or PV in the mPFC that innervate preferentially dendrites like in the hippocampus (Halasy et al., 1996; Cope et al., 2002; Somogyi and Klausberger, 2005; Szabó et al., 2014).

Figure 7.

Postsynaptic target distribution of basket cells within the mPFC. A, High-magnification multicolor maximum z-intensity projection confocal image shows biocytin-filled boutons around Kv2.1-labeled neurons, dark blue arrows point to boutons forming contacts on the somata, light blue arrows indicate the proximal dendrite-targeting boutons, and gray arrows show varicosities that presumably target distal dendrites (i.e., avoid Kv2.1-labeled profiles). Scale bar, 10 µm. Inset, The in vitro biocytin-filled superficial CCKBC, whose axons are shown at the higher magnification. Scale bar, 100 µm. B, High-magnification multicolor maximum z-intensity projection confocal image shows biocytin-filled boutons around Kv2.1-labeled neurons, bright orange arrows point to boutons forming close contacts on a soma, light orange arrows indicate the proximal dendrite-targeting boutons, and gray arrows show varicosities that likely target distal dendrites (i.e., avoid Kv2.1-labeled profiles). Scale bar, 10 µm. Inset, The in vitro biocytin-filled L5b PVBC, whose axon collaterals are shown at the higher magnification. Scale bar, 100 µm. C, Ratio of the boutons of CCKBCs and PVBCs forming close contacts on Kv2.1-immunostained perisomatic regions of neurons. Each column represents a single BC (n = 20 CCKBCs and n = 24 PVBCs were examined), M–W test showed no difference in the ratio of boutons on Kv2.1-immunolabeled soma between CCKBC and PVBC (p = 0.99). D, Average ratio of biocytin-filled boutons forming contacts on Kv2.1-immunolabeled profiles or on unlabeled targets (pooled data from C). E, Ratio of the boutons of the basket cell subgroups. Each column represents a single basket cell (n = 3560 boutons of 10 superficial and n = 3518 boutons of 10 deep CCKBCs were examined; M–W test showed no difference in the ratio of boutons on Kv2.1-immunostained somas between distinct groups of CCKBCs, p = 0.97; n = 2838 boutons of 10 L2/3, n = 1986 boutons of 6 L5a, and n = 1995 boutons of 8 L5b PVBC were examined; K–W ANOVA showed no difference in the ratio of boutons on Kv2.1-immunostained somas between distinct groups of PVBC, p = 0.217). F, Comparison of the number of the perisomatic contacts received by single Kv2.1-labeled neurons from individual CCKBCs and PVBCs. Filled circles present the number of perisomatic contacts within 200 µm (near), and open diamonds show number of perisomatic contacts outside the circle with 200 µm radius (distant)., The median (continuous line in the box, CCKBC near, 2; CCKBC distant, 2; PVBC near, 3; PVBC distant, 2), the interquartile range (box, CCKBC near, 2; CCKBC distant, 2; PVBC near, 3; PVBC distant, 2), and the 5/95% values (ends of whiskers bar) are plotted. Two hundred eleven and 172 Kv2.1-labeled neurons innervated by 520 and 350 boutons, respectively, from 10 CCKBCs and 199 and 182 Kv2.1-labeled neurons innervated by 761 and 457 boutons, respectively, from 10 PVBCs were examined (K–W ANOVA, p < 0.001, post hoc Dunn's test, CCKBC near vs CCKBC distant, p = 0.007; CCKBC near vs PVBC near, p < 0.001; CCKBC distant vs PVBC distant, p = 0.009; PVBC near vs PVBC distant, p < 0.001).

PVBCs are preferentially targeted by thalamic and amygdalar inputs in layers 5a and 5b, respectively

We have shown that PVBCs preferentially innervate the same layer in the mPFC where their somata and most of their dendrites are located (Fig. 5). As different mPFC afferents typically target some layers, but not others (Anastasiades and Carter, 2021), this raises the possibility that PVBCs are capable of mediating feedforward inhibition in a layer-specific manner if they are preferentially targeted by extra-mPFC inputs in a given layer. To test this hypothesis, we investigated the innervation of PV+ interneurons by afferents originating either from the BA, midline thalamus (Thal) or LEnt. We combined anterograde trans-synaptic viral labeling with immunostaining and determined the ratio of PV+ interneurons that did and did not receive monosynaptic inputs from distinct projections in each layer. To obtain anterograde trans-synaptic labeling, which can be effectively achieved by using AAV1 (i.e., when AAV1 injected to a given brain region is transported anterogradely along the axon of the projecting cells, gets released, and infects those cells that are monosynaptically innervated by the neurons at the injection area; Zingg et al., 2017) two approaches were used. AAV1-hSyn-Cre was injected into the Ai6 reporter line resulting in ZsGreen1 expression in all neurons (both pyramidal cells and interneurons) that receive monosynaptic innervation from the neurons of the injected areas (BA, midline thalamus, lateral entorhinal cortex, respectively; Fig. 8A). In these experiments, PV+ interneurons were revealed by immunostaining. The other approach involved PV-Cre mice, in which AAV1-EF1a-DIO-ChETA-eYFP was injected into the same afferent areas as above, yet expressed only in those monosynaptically innervated cells that contain Cre, for example, directly revealing those PV-containing cells that are innervated by the afferent regions. By using these approaches, we were able to visualize those neurons in the mPFC that receive monosynaptic innervation from the injected area (Zingg et al., 2017; Fig. 8B). As both strategies gave rise to similar ratio of trans-synaptically labeled PV+ cells, this confirms that the labeling success was not dependent on the AAV1 or mouse strains used in these investigations (Fig. 8C). Therefore, we pooled the data from the two approaches. Next, we compared the distribution of PV+ interneurons in the different layers that received inputs from the BA, Thal, or LEnt input in the distinct layers with the distribution of all PV-immunolabeled interneurons in the mPFC (Fig. 8D). We observed that afferents from the BA and Thal innervated PV+ interneurons in a layer-specific manner but not the input arriving from LEnt. In case of BA inputs, a higher ratio of PV+ interneurons was labeled with AAVs in layer 5b compared with the relative distribution of all PV-immunostained interneurons, and the same was true in the case of Thal inputs in layer 5a (Fig. 8E). As we found no PV+ chandelier cells in layers 5a and 5b of the mPFC, in line with a previous study (Wang et al., 2019), these data imply that excitation from the amygdala and thalamus can drive PVBCs-mediated feedforward inhibition in the mPFC in a layer-specific manner.

Figure 8.

PV-expressing interneurons in the mPFC are preferentially innervated by Thal and BA afferents in layers 5a and 5b, respectively. A, Top, Schematic representation of AAV1-hSyn-Cre virus injection into Thal. Bottom, Localization of trans-synaptically labeled cells (trans-syn. lb. cells) within the mPFC. Scale bar, 1 mm. B, Multicolor maximum z-intensity projection confocal images taken from the PrL area of Ai6 mice injected with AAVs into the BA, Thal, and LEnt. Immunostaining was used to visualize PV interneurons. Yellow arrows point to ZsGreen1 (ZsGr) and PV double-positive cells. Scale bar, 50 µm. C, Comparison of PV-immunolabeled ZsGr- or GFP-positive cell distribution in each layer of Ai6 and PV-Cre mouse lines on AAV injections into different areas. Chi-square homogeneity test showed no differences between mouse lines. (BA, χ² = 0.09, p = 0.99; Thal, χ2 = 1.86, p = 0.6; LEnt, χ² = 0.2, p = 0.98). D, Box chart comparison of PV cell ratios in different layers. PV-immunolabeled interneurons are shown in black (n = 1283 PV immunopositive cells in 12 slices from 7 mice), the BA input receiving PV cells in orange (n = 570 cells in 45 slices from 10 mice), Thal input receiving PV cells in blue (n = 731 cells in 38 slices from 7 mice), and LEnt input receiving PV cells in green (n = 187 cells in 29 slices from 6 mice). The mean (small open square), the median (continuous line within the box), the interquartile range (box) and the 5/95% values (ends of whiskers bar) are plotted. E, The p values of an M–W test comparison between the ratios of PV cells receiving a given extra-mPFC input versus PV-immunolabeled cells in each layer (black vs colored bars in D). background indicates significant differences. L, Layer.

Discussion

In this study, GABAergic inputs innervating the perisomatic region of neurons in the PrL subregion of the mPFC were examined. We identified the sources of these inputs and compared the morphologic features of interneurons that provide the vast majority of perisomatic inhibitory innervation on the soma and proximal dendrites. In addition to the layer definition in the mPFC, our main findings are the following: (1) The perisomatic region of pyramidal cells in the mPFC is innervated mostly by PV+ or CB1+ GABAergic inputs; (2) these inhibitory inputs originate from PVBCs and CCKBCs; (3) at the population level, >50% of axonal boutons of both basket cell types contacted the perisomatic region, however, at the single-cell level, PVBCs innervated the perisomatic region via more boutons than CCKBCs; and (4) most importantly, we found that PVBCs are innervated by afferents originating from the BA and Thal in a layer-specific manner.

Although several studies have examined the cytoarchitectural and connectivity properties of the mPFC, the layers of this brain region are still ill defined and show inconsistency in previous work (Clarkson et al., 2017; Lu et al., 2017; Wang et al., 2019; Anastasiades and Carter, 2021). Other previous studies have established probably the most precise method to define a cortical layer by using single-cell transcriptomics (Tasic et al., 2018; Wang et al., 2018; Ortiz et al., 2020); unfortunately, it is hard to adapt this methodology to acute brain slices used in electrophysiological recordings. Additionally, as the thickness of layers in the mPFC changes continuously in dorsoventral and anterior–posterior directions, the measurements of the distance from the pia are not the most reliable way to define the layers. Therefore, we introduced an approach and defined the layers of the mPFC by immunostaining using a combination of antibodies developed against markers that have been used in other cortical areas as well (Cruikshank et al., 2001; Arlotta et al., 2005; Luuk et al., 2008; Hisaoka et al., 2010).

Similar to other cortical studies, we observed that the perisomatic region of neurons is innervated mostly by GABAergic inputs from two sources (Takács et al., 2015; Vereczki et al., 2016). Our findings revealed that ∼90% of inhibitory inputs on the perisomatic membrane surface is originating from PV+ or CB1+ axonal varicosities. Interestingly, there was a layer-specific difference in the ratio of perisomatic PV+/CB1+ inputs on randomly sampled neurons (Fig. 2D) and when examined on defined pyramidal cell populations projecting to specific areas (Fig. 3F). Thus, additionally regarding the observations that the two types of basket cells contribute mainly to perisomatic inhibition in the mPFC, it seems that pyramidal cells located in deeper layers are innervated by more inputs from PVBCs. A previous study using a combination of optogenetics with slice physiology proposed that PVBCs preferentially innervate PT neurons in comparison to intratelencephalic (IT) neurons (Lee et al., 2014), which is in line with our anatomic findings. However, these data contradict those obtained with paired recordings in the frontal cortex, suggesting that IT and PT neurons receive similar strength of inhibition from putative PV interneurons (Morishima et al., 2017). PT neurons in the mPFC are also innervated by CCKBCs as we found in the present study. This finding is in contract to our observations obtained in the somatosensory cortex, where we detected only negligible perisomatic inputs from CCKBCs onto the PT neurons (Bodor et al., 2005). Altogether, these results may support the hypothesis that in primary sensory cortices, PT neurons are controlled solely by PVBCs, whereas in higher-order cortical areas, the pyramidal cells are under the regulation of both basket cell types (Chiu et al., 2010; Toyoda, 2020), an idea that needs to be specifically tested.

Several studies have examined the features of these two basket cell types separately in rodent frontal and prefrontal cortices (Kawaguchi and Kubota, 1998; Lagler et al., 2016; Miyamae et al., 2017), yet the comparison of their morphologic characteristics has not been conducted in the mouse mPFC, although both interneuron types have been shown to be abundant in this area (Whissell et al., 2015). Our morphologic analysis and comparison revealed structural differences between the two basket cell types, similar to that revealed in mouse visual cortex (Gouwens et al., 2020; Koukouli et al., 2022). Although the total dendritic and axonal length of the two basket cell types was similar, we found differences in the structure of their dendritic and axonal arborizations. Interestingly, similar differences were observed in the basal amygdala too, suggesting that basket cells might not possess brain-region-specific structural features (Vereczki et al., 2016) but have a more uniform morphologic structure. We have also found that distinct basket cell types could be divided further into subgroups based on the location of their somata, dendrites, and axons. Such morphologic subgroups of CCKBCs have not been proposed so far in a given neocortical area, yet between neighboring cortical regions their arborization can differ (Koukouli et al., 2022). In previous studies (Jiang et al., 2015; Lagler et al., 2016; Miyamae et al., 2017; Gouwens et al., 2020) distinct morphologic types of PVBCs have been already identified but without any systematic comparison. As PVBCs play a pivotal role in the generation of gamma oscillations (Sohal et al., 2009; Gulyás et al., 2010), the layer-specific arborization of these GABAergic interneurons may ensure the independent oscillatory activities in different layers, a hypothesis that is in accord with previous in vivo findings (Senzai et al., 2019). Thus, the structural features of PVBCs may allow the neural computation to be performed synchronously across cortical layers or restricted to a given layer, depending on the primary excitatory inputs driving PVBC spiking. In addition, the restricted inhibitory output of PVBCs may also play a role in layer-specific plastic processes (Lourenço et al., 2014; Allene et al., 2015), altering the contribution of distinct pyramidal cell populations to cortical function.

Multicolor labeling allowed us to analyze the target distribution of distinct basket cells. We found that on average >50% of the biocytin-filled boutons of both basket cell types innervated the perisomatic region of pyramidal cells. Although this ratio of boutons may question the concept of perisomatic inhibition, as the other half of the boutons target the dendrites, our previous study performed in the BA demonstrated that basket cell boutons innervating the perisomatic region were primarily responsible for controlling principal cell spiking (Veres et al., 2017). These results argue for the functional separation of perisomatic and dendritic inhibition having distinct roles in microcircuit operation (Miles et al., 1996). Additionally, a large variability was observed in the ratio of contacts forming close appositions with the perisomatic membrane surface in both cases. Interestingly, this variance was because of the difference in the innervation of the proximal dendrites, as the ratio of boutons innervating the somata was similar within and between the basket cell types. These findings are different from those observed in the BA, where the ratio of boutons contacting the somata showed a large variance, whereas the proportion of boutons on proximal dendrites was rather similar (Veres et al., 2017). Furthermore, when we counted the number of boutons innervating single pyramidal cells in the vicinity of basket cells, we found that PVBCs targeted pyramidal cells with significantly more boutons than CCKBCs, similar to that found in the BA (Vereczki et al., 2016), an observation that did not depend on the distance (Fig. 7F).

Our morphologic analyses revealed that PVBCs extended mostly their dendrites and axons in the same layer where their somata were located (Fig. 5A3–C5). In addition, previous data demonstrated that extra-mPFC afferents from distinct brain regions show a layer-specific distribution within the mPFC (Parnaudeau et al., 2013; Oh et al., 2014; McGarry and Carter, 2016; Anastasiades and Carter, 2021). Therefore, based on these observations, we hypothesized that PV+ interneurons (which are predominantly basket cells in the mPFC based on our study) can be innervated preferentially by inputs from the BA, Thal, and LEnt cortex in a layer-specific manner. In line with this idea, we found that thalamic and amygdalar inputs favored PVBCs in layers 5a and 5b, respectively (Fig. 8D,E). Thus, our anatomic findings suggest that excitatory inputs from the BA and Thal, but not from the LEnt cortex, can mediate feedforward inhibition via PVBCs in the mPFC in a layer-specific manner. Revealing the long-range connectivity for CCKBCs, however, cannot be achieved using Cck-Cre mice even in a combination with intersectional strategy, as this approach would target a substantial fraction of PV-expressing interneurons and neurogliaform cells in addition to CCKBCs precluding to draw a solid conclusion (Rovira-Esteban et al., 2019).

Together, our results demonstrated the presence of morphologically different basket cells in the mPFC microcircuits. Furthermore, PVBCs in different layers are preferentially contacted by distinct extra-mPFC afferents. Thus, PVBCs innervating pyramidal cells in their vicinity have the possibility to control mPFC function in a layer-specific manner. In contrast, CCKBCs with extended axonal arborization can provide cross-layer inhibition in the mPFC. The different innervation strategy of the two basket cell types may play a key role in controlling neural operation within the individual layers and across the layers of the mPFC.