All animal experiments were performed in accordance with institutional guidelines and were approved by the Animal Experiment Committee at RIKEN. Male wild-type mice (10–13 weeks old) (C57BL/6JJmsSlc; Japan SLC, Shizuoka, Japan) were used in this study. In all experiments, the mice were housed in a 12 h-light/12 h-dark light cycle environment with ad libitum access to food and water.

The following plasmids were obtained from Addgene (Addgene, Watertown, MA, USA): pAAV-CaMKII-ChrimsonR-tdTomato (cat. #9923; RRID:Addgene_99231) and pAAV-hDlx-Flex-green fluorescent protein (GFP)-Fishell_6 (cat. #83895; RRID:Addgene_83895). AAV was produced as previously described (Kobayashi et al., 2016; Kato et al., 2018). The following AAV vectors were obtained from Addgene: AAV-hSyn-Cre-WPRE-hGH (cat. #105553-AAV1; RRID:Addgene_105553) and AAV-hDlx-Flex-GFP-Fishell_6 (cat. #83895-AAV; RRID:Addgene_83895).

The animals were anesthetized with isoflurane (2–3% for induction and 1.2% for maintenance) using anesthesia equipment (AN-487-0T, Shinano, Japan). Stereotaxic injections were administered to deliver Cholera Toxin Subunit B (CTB) conjugate 647 (Cat#C34778, Molecular Probes, USA) or AAV to specific cortical areas. A craniotomy was performed above the injection site and CTB or AAV was injected (10 nL/min) via a pulled glass pipette (Cat#BR708744, BRAND, USA). To mark the injection of AAVrg-hSyn-Cre-WPRE-hGH, we co-injected Hoechst 33342 (Cat#19172-51, NACALAI TESQUE, Japan), because fluorescent probes were not present in the virus itself (Mukherjee et al., 2020). The coordinates, volumes, and titers were as follows: left M2 [anteroposterior from bregma (AP), + 2.19 mm; mediolateral from the midline (ML), 0.6 mm; dorsoventral from the cortical surface (DV), 3 depth in 0.2–0.8 mm, 600 nL, 5.50 × 10¹² GC/mL for AAV1-CaMKII-ChrimsonR injections, 100 nL, 1.00 × 10¹³ GC/mL for AAV1-hsyn-Cre-hGH-WPRE], left S1 (AP, −0.75 mm; ML, 1.70 mm; DV, 0.2–0.7 mm, 150 nL, 0.5%, for CTB injections, 200 nL, 3.33 × 10¹² GC/mL for AAV1-hdlx-flex-GFP). For the experiment using DISECT, the AAV1-hdlx-flex-GFP that was produced in our own laboratory was used in four of five mice and those obtained from Addgene was used in one of five mice. We confirmed that the VGAT-expression, the laminar distribution, and the expression of neurochemical markers were similar between neurons labeled by both of the two vectors.

Mice were deeply anesthetized using intraperitoneal injection of urethane and perfused transcardially with 20 mL of Hanks’ Balanced Salt Solution (HBSS; 14025076, Life Technologies, USA) supplemented with heparin (10 units/mL), followed by perfusion with 4% formaldehyde in 0.1 M phosphate buffer (PB; pH 7.4) and postfixation in the same fixative for 16–20 h at 4°C. For subsequent in situ hybridization experiments, the brain blocks were cut into 40-μm-thick coronal sections using a freezing microtome (ROM-380, Yamato, Japan) after cryoprotection with 30% sucrose in PB. For the other experiments, the brain blocks were cut into 80-μm-thick coronal sections on a vibratome (VTS1200S, LEICA, Germany).

Free-floating sections were incubated in a blocking solution [2% normal goat serum (NGS) and 0.3% Triton X-100 in phosphate-buffered saline (PBST)] at room temperature for 1 h, followed by incubation with 1/2000-diluted rabbit anti-Parvalbumin IgG (Cat#PV27, Swant, Switzerland, RRID:AB_2631173) and 1/500-diluted goat anti-Somatostatin IgG (Cat#sc-7819, Santa Cruz, RRID:AB_2302603) in 2% NGS overnight at 4°C. After washing for 10 min with PBST three times, the sections were incubated with donkey anti-rabbit secondary antibodies conjugated to 1/300-diluted Alexa Fluor Plus 568 (Cat#A10042, Molecular Probes, USA, RRID:AB_2534017) and donkey anti-goat secondary antibodies conjugated to 1/300-diluted Alexa Fluor Plus 647 (Cat#A32849, Molecular Probes, USA, RRID:AB_2762840) in 2% NGS for 2 h at room temperature. Subsequently, the sections were rinsed with PBS and mounted on a cover glass using Fluoromount (Cat#K048, Diagnostic BioSystems). For experiments requiring laminar identification, the sections were incubated with 1/500-diluted NeuroTrace™ 435/455 Blue Fluorescent Nissl Stain (Cat#21479; Thermo Fisher Scientific) overnight. To identify vasoactive intestinal peptide (VIP) and neuropeptide-Y (NPY) neurons, 1/300-diluted anti-VIP (Cat#20077, ImmunoStar, RRID:AB_572270) and 1/50-diluted anti-NPY (Cat#NMD-MSFR104610, Nittobo) were also used. However, we could not obtain results with a good signal-to-noise ratio sufficient for subsequent automatic detection of somata. Therefore, we analyzed the images obtained by the fluorescent in situ hybridization staining method.

Fluorescent in situ hybridization (FISH) was performed using the RNAscope system (Advanced Cell Diagnostic, Newark, CA, USA). Two 40 μm coronal sections of the S1 area were processed according to the RNAscope Multiplex Fluorescent Reagent Kit v2 User Manual (Cat#323100; ACD). RNAscope^® Target Probes against VGAT (Mm-Slc32a1-C2 Mouse, Cat#416631-c2), VIP (Mm-Vip-C2 Mouse, Cat#415961-c2), or NPY (Mm-Npy-C2 Mouse, Cat#313321-c2) were used. After the final wash buffer, the sections were washed with 0.3% PBST for 20 min. Subsequently, sections were incubated with 1/300-diluted chicken anti-GFP IgY (Cat#13970, Abcam) diluted in 2% NGS overnight at 4°C. After washing three times for 5 min with PBST, the sections were incubated with 1/250-diluted goat anti-chicken secondary antibodies conjugated to Alexa Fluor Plus 488 (Cat#A32931, Thermo Fisher Scientific) in 2% NGS for 2 h at room temperature. The sections were washed once with PBST for 5 min, treated with 1/1000 diluted 4′,6-diamidino-2-phenylindole (DAPI; Cat#D9542, Sigma-Aldrich) in PBS for 10 min, rinsed with PBS, then mounted on a cover glass using Fluoromount.

To confirm the injection sites or the labeled tissues of the brain blocks and fluorescence-labeled sections, the specimens were observed under an inverted microscope (IX83P2, Olympus) equipped with a 1.25 × /0.04 N.A. air objective (Cat#PLAPON1.25X, Olympus), as shown in Figures 2B, 3B, 4B, or a 4 × /0.16 N.A. air objective (Cat# UPLXAPO4X, Olympus), as shown in Figures 2C, 3C, 4C. To acquire images for subsequent analysis of coexpressing neurons labeled with CTB, GFP, or each neurochemical marker, the sections were observed under a confocal laser scanning microscope (FV3000RS, Olympus) with a 20 × /0.80 N.A. air objective (Cat # UPLXAPO20X, Olympus) (Figures 2D, E, 3D, E, 4D, E, 5A).

All analyses were performed using custom-built programs in Python or Fiji/ImageJ 1.53c (Schindelin et al., 2012).

Dlx enhancer-restricted Interneuron-SpECific transsynaptic Tracing is a simplified, two-step viral injection approach. In the first step, injection of AAV1-Cre into the presynaptic region enabled Cre-induced gene expression in recipient postsynaptic neurons. In the second step, injection of a Cre-inducible Dlx enhancer-specific AAV vector restricted the transduction of genes to the postsynaptic inhibitory neurons (Figure 1C). DISECT has two anatomical requirements. This example of an M2-S1 top-down circuit illustrates the need for these requirements. First, DISECT requires direct projection from the M2 to the S1 postsynaptic region because AAV1-Cre needs to spread beyond the first-order synapse at S1 (Figure 2). Second, it needs no retrograde projections from S1 inhibitory neurons (INs) to M2 neurons (Figure 3). If the circuit has retrograde inhibitory projections, it causes false labeling of M2 projecting S1 INs as well as M2 input-recipient S1 INs.

To validate requirement 1, we performed anterograde axonal tracing from M2 to S1 (Figure 2A). We injected AAV1-CaMKII-ChrimsonR-tdTomato into the M2 (n = 3 mice). Three weeks after the injection, we confirmed tdTomato expression at the M2 injection site (Figures 2B, C) and M2 axons in S1 in coronal sections of the brain slices (Figures 2C, D). These axons were preferentially distributed in the superficial and deep layers of S1, which is the typical top-down cortical projection pattern (Felleman and Van Essen, 1991), as previously reported (Figures 2E, F; Manita et al., 2015).

To validate requirement 2, we investigated whether S1 INs project to M2. Although most INs in the neocortex have short-range projections, a few types of INs have long ranges (Tomioka et al., 2005). To identify whether S1 had INs with long-range projections to M2, we performed retrograde labeling of M2 projecting S1 neurons and calculated their fraction (Figure 3A). First, we injected the retrograde tracer CTB into the M2 and labeled M2 projecting S1 neurons (n = 2 mice). Five days after the injection, we observed the fluorescence of CTB at the M2 injection site (Figures 3B, C) and the CTB-labeled somata in S1 (Figures 3C, D) in brain slices. Next, we performed FISH on the slices to examine whether CTB-labeled S1 neurons expressed VGAT mRNA, a marker of INs. We found that only a small fraction of CTB-labeled S1 neurons expressed VGAT mRNA (Figures 3D, E). Quantitative analysis indicated that 1.7% (4 out of 240 neurons) of the CTB-labeled neurons expressed VGAT mRNA, and 98.3% (236/240) did not (Figure 3F). These results indicate that almost none of the M2 projecting S1 INs. Taken together, we conclude that DISECT is applicable to the M2-S1 top-down circuit.

We applied DISECT to study the properties of M2-input recipient S1 INs (Figure 4A). To transduce Cre recombinase into these neurons, we injected the AAV1-Cre into the M2 (Figure 4A). We performed a cocktail injection of Hoechst 33342 and AAV1-Cre to label the M2 injection sites. On the same day, we injected AAV1-hdlx-flex-GFP into the S1, which transduced GFP into the Cre + S1 inhibitory neurons (n = 5 mice). Four weeks after injection, we confirmed the fluorescence of Hoechst 33342 at the M2 injection site (Figures 4B, C), and GFP-labeled S1 neurons, which are M2-input recipient neurons, were located in all S1 layers (Figures 4C, D). To confirm whether the transduction of GFP by the hDlx enhancer was restricted to S1 INs, we performed FISH and confirmed that most GFP + neurons (95.4%, 83/87 neurons) expressed VGAT (Figures 4D–F). These results indicate that DISECT successfully labeled the M2 input recipient S1 INs.

We performed a quantitative analysis of the laminar distribution of the GFP + INs. To analyze the laminar specificity of the GFP + INs, we statistically compared the fractions of GFP + S1 INs (i.e., M2-recipient INs) and VGAT + S1 INs (the total population of GABAergic INs) using binomial tests. We calculated laminar specific ratios of VGAT + INs (number of VGAT + INs in each cortical layer to the total number of VGAT + INs in all the layers) as the expected fractions of INs (Figure 5A). We also calculated the ratios of the GFP + neurons (number of GFP + INs in each cortical layer to the total number of GFP + neurons in all the layers). Statistical analysis revealed that the ratio of GFP + S1 INs in L4 was significantly smaller than the ratio of VGAT + neurons (Figure 5B). This result suggests that M2 axons tend to avoid making synaptic connections with S1 INs in the middle layers. We also found that the GFP + S1 neurons were preferentially located in the upper and deeper layers (Figure 5C, green box). Furthermore, we found that the laminar specificity of GFP + S1 neurons aligned with the laminar density of M2 axons shown in Figure 2F (magenta line), which was also lower in the middle layers. These results support the hypothesis that DISECT-labeled neurons receive presynaptic inputs from M2 axons.

In the mouse cortex, PV, SST, and VIP are exclusive neurochemical markers of the INs (Rudy et al., 2011), and NPY is a marker of neurogliaform neurons (Schuman et al., 2019). We examined the expression of PV and SST by immunohistochemistry and VIP and NPY by FISH to determine which subtypes of INs were most dominant in M2-recipient S1 INs (Figure 6A).

Before investigating the coexpression of GFP and the markers, we validated our neurochemical marker staining results. To this end, we calculated the fraction of all INs that expressed each subtype-specific marker (i.e., VGAT + neurons) located in all cortical layers or in individual layer; we then compared the results with those of a previous study (Gonchar et al., 2008). First, we obtained the ratio of neuron density between each subtype and VGAT + neurons (as background INs; for more details, see Section “2.8. Data analysis” in Section “2. Materials and methods”). Our analysis indicated that the dominant subtypes were PV, SST, and VIP INs, in that order (Table 1). Second, we also confirmed the distribution of each subtype among the different cortical layers. We obtained the neuron density ratio for each cortical layer (Table 1; Figure 6B) and found similar laminar profiles of INs, represented by the absence of PV + neurons in cortical layer 1 and laminar specificity of VIP + neurons in superficial layers. These results are consistent with those of Gonchar et al. (2008) (Table 2).

After the validation, to analyze the fraction of neurons expressing each neurochemical marker in the GFP + neurons identified by DISECT, we counted the number of GFP + neurons coexpressing each neurochemical marker and calculated their ratio to the total number of GFP + neurons (Figure 6C). We statistically compared the fractions of the neurons coexpressing GFP + with each marker neuron, and each subtype shown in Figure 4B using binomial tests. The results showed that the fractions of GFP+ PV+ and GFP+ SST+ neurons were not significantly different from their expected fractions (Figure 6C). In contrast, the proportions of GFP+ VIP+ and GFP+ NPY+ neurons were significantly smaller than those of their expected fractions. These results indicate that M2-input recipient S1 neurons preferentially express PV and SST rather than VIP or NPY. The laminar profile of each subtype was analyzed using the same method (Figure 6D; Supplementary Table 1). In all subtypes, we confirmed that the fraction of GFP + neurons in the S1 L4 was smaller than the expected fractions, which is consistent with the results shown in Figure 5B. However, the fraction of GFP+ PV+ coexpressing neurons was significantly smaller in L2/3 and significantly larger in L6 than in their expected fractions. We also discovered that the fraction of GFP+ SST+ neuron coexpression in L1 was significantly larger than its expected fraction. We did not find any significant differences in the laminar profiles of GFP+, VIP+, or GFP+ NPY+ neuron coexpression. These results suggest that the M2-input recipient S1 + INs preferentially expressed PV in the deep layers and SST in the superficial layers. Taken together, we identified the subtype specificity and laminar profiles of M2 top-down input-recipient S1 INs using DISECT (Figure 6E).

Dlx enhancer-restricted Interneuron-SpECific transsynaptic Tracing succeeded in greatly simplifying conventional AAV-mediated transsynaptic tracing. This simplification reduces the experimental procedures and costs, increases efficiency, and extends the range of applications, as discussed below.

First, DISECT simplifies the post hoc analysis for subtype determination of INs by not labeling excitatory neurons, which has two advantages. One advantage of not labeling excitatory neurons, including larger fractions of retrogradely-projecting neurons, is prevention of false labeling and disturbing the post hoc analysis. Another advantage is that the neurons labeled by DISECT can be clearly identified as GABAergic INs without any additional histological procedures and analyses. This observation allowed us to instantly quantify the subtype specificity of the postsynaptic INs (Figure 6D). For example, fractions of labeled neurons that are smaller or larger than the expected total fractions indicate lower or higher preferences for target INs, respectively.

Second, DISECT simply requires wild-type mice and commercially available viral vectors, but not the generation of transgenic animals (e.g., breeding) or viral vectors (e.g., vector purification). This can accelerate the experiments and contribute to reducing costs and the number of animals.

Third, DISECT ensures higher applicability in a variety of experiments that require multiple intersectional systems. Combinations of orthogonal intersectional systems is necessary in various experimental procedures such as the conditional knockout of specific genes, optogenetic and chemogenetic manipulations, and activity-dependent labeling at the same time. However, the number of the available systems is currently limited to three (Fenno et al., 2020). The conventional tracing method which requires two major intersectional systems (Cre/cDIO and Flp/fDIO) would be difficult to be incorporated into the experiments because only one system (vCre/vcDIO) is left for the other procedures. Because DISECT requires only one of the systems, this will allow a variety of experimental methods to be used.

Using DISECT, we found the IN subtype specificity of projections from M2, which target PV+ or SST+ rather than VIP+ or NPY+ (Figure 6C). Previous studies have reported subtype-specific top-down projections (Schneider et al., 2014; Zhang et al., 2014; Ma et al., 2021). For example, projections from the cingulate cortex to V1 preferentially target VIP+ neurons, resulting in disinhibitory effects on V1 activity. Projections from M2 to A1 preferentially target PV+ neurons and suppress auditory responses. We also found the subtype-dependent laminar profiles of the INs, which are difficult to describe using conventional anterograde approaches (Figures 6D, E). Our results showed that the M2 input-recipient PV + INs were predominantly distributed in the deep layers and less so in the superficial layers, thereby avoiding the middle layers. The laminar specificity of M2 top-down projections is different from that of other projections, i.e., S2 top-down projections to S1 target L2/3 PV + INs (Naskar et al., 2021). Together, these differences in the top-down projection recipients of the IN subtype and their laminar specificity would reflect different forms of sensory processing, depending on the modality.

Although DISECT is a simple and efficient tracing tool, the following three points should be considered.

First, we considered the false labeling of retrogradely-M2-projecting S1 INs because retrograde transport of AAV1-Cre can also occur in INs. If the AAV1-Cre injected presynaptic site receives long-range inhibitory projections from a postsynaptic region, DISECT will result in false labeling of projecting INs as well as recipient INs in that region. Previous studies have reported a limited number of INs with projection distances longer than 2.0 mm in the cortical regions (Tomioka et al., 2005; Rock et al., 2018; Ruff et al., 2022). To avoid false labeling, it is necessary to confirm in advance whether the fraction of retrogradely-projecting INs in a targeted postsynaptic circuit is small enough to be ignored (Figure 3F).

Second, we considered subtype-dependent differences in trans-synaptic transport efficiency, which led to differences in comparisons between subtypes. A previous study confirmed that transsynaptic tracing methods can label both excitatory and inhibitory neurons (Zingg et al., 2020). However, the study also confirmed that postsynaptic neurons targeted by cholinergic projections cannot be labeled using AAV1-mediated methods. This result suggests potential differences in the labeling efficiency between IN subtypes. To clarify the anatomical results more precisely, comprehensive observations by multiple methods using retrograde or anterograde approaches, such as retrograde transsynaptic tracing using rabies viral vectors, channelrhodopsin-2 (ChR2)-assisted circuit mapping (CRACM), or anterograde transsynaptic tracing using other viral vectors (Li et al., 2021; Tsai et al., 2022), would be helpful.

Third, we examined the specificity of Dlx enhancers to IN subtypes. The enhancers used in DISECT—mDlx and hDlx—are gene sequences designed for gene transduction to universally target cortical INs. Previous studies have confirmed that there is little difference in the labeling efficiency between the major subtypes, including PV+, SST+, and VIP+ INs (Dimidschstein et al., 2016). However, a study using mGAD65 for gene transduction into INs have reported significantly higher transduction efficiency than mDlx in some subtypes, such as chandelier neurons (Hoshino et al., 2021). Therefore, the tracing results for some subtypes should be carefully interpreted in terms of enhancer specificity. Combinations of DISECT, other tracing methods, and functional recordings (e.g., optical or electrophysiology) with neural manipulations are crucial for further circuit-by-circuit clarification of the top-down-input-recipient inhibitory circuitry.