Vasoactive Intestinal Polypeptide-Immunoreactive Interneurons within Circuits of the Mouse Basolateral Amygdala

In cortical structures, principal cell activity is tightly regulated by different GABAergic interneurons (INs). Among these INs are vasoactive intestinal polypeptide-expressing (VIP+) INs, which innervate preferentially other INs, providing a structural basis for temporal disinhibition of principal cells. However, relatively little is known about VIP+ INs in the amygdaloid basolateral complex (BLA). In this study, we report that VIP+ INs have a variable density in the distinct subdivisions of the mouse BLA. Based on different anatomical, neurochemical, and electrophysiological criteria, VIP+ INs could be identified as IN-selective INs (IS-INs) and basket cells expressing CB1 cannabinoid receptors. Whole-cell recordings of VIP+ IS-INs revealed three different spiking patterns, none of which was associated with the expression of calretinin. Genetic targeting combined with optogenetics and in vitro recordings enabled us to identify several types of BLA INs innervated by VIP+ INs, including other IS-INs, basket and neurogliaform cells. Moreover, light stimulation of VIP+ basket cell axon terminals, characterized by CB1 sensitivity, evoked IPSPs in ∼20% of principal neurons. Finally, we show that VIP+ INs receive a dense innervation from both GABAergic inputs (although only 10% from other VIP+ INs) and distinct glutamatergic inputs, identified by their expression of different vesicular glutamate transporters. In conclusion, our study provides a wide-range analysis of single-cell properties of VIP+ INs in the mouse BLA and of their intrinsic and extrinsic connectivity. Our results reinforce the evidence that VIP+ INs are structurally and functionally heterogeneous and that this heterogeneity could mediate different roles in amygdala-dependent functions. SIGNIFICANCE STATEMENT We provide the first comprehensive analysis of the distribution of vasoactive intestinal polypeptide-expressing (VIP+) interneurons (INs) across the entire mouse amygdaloid basolateral complex (BLA), as well as of their morphological and physiological properties. VIP+ INs in the neocortex preferentially target other INs to form a disinhibitory network that facilitates principal cell firing. Our study is the first to demonstrate the presence of such a disinhibitory circuitry in the BLA. We observed structural and functional heterogeneity of these INs and characterized their input/output connectivity. We also identified several types of BLA INs that, when inhibited, may provide a temporal window for principal cell firing and facilitate associative plasticity, e.g., in fear learning.


Introduction
The basolateral amygdaloid complex (BLA), an extension of the cortex deep in the temporal lobe, has been shown to contribute to a bewildering variety of behavioral functions. These include regulation of goal-directed and social behaviors as well as the forma-tion and storage of affective memories (Phelps and LeDoux, 2005). In this regard, the BLA has been found to be critical in the formation of long-lasting fear memories (LeDoux, 2000;Maren, 2001;Tovote et al., 2015). The BLA consists of the lateral (LA), basal (BA), and accessory basal nuclei, and is composed of ϳ85% pyramidal-like principal neurons and ϳ15% GABAergic interneurons (INs;McDonald, 1992). The LA and BA nuclei can be further divided into nuclear subdivisions based on distinct neurochemical and cytoarchitectonic features (Martínez-García et al., 2012), but also internuclear and intranuclear connections (Pitkänen and Amaral, 1991;Pitkänen et al., 1997).
Although the BLA has been intensely studied, its intrinsic neural circuits remain poorly understood. One characteristic of BLA INs that has not been fully explored is its diversity. BLA INs vary significantly in their firing, molecular, and morphological properties, including axonal projections (Spampanato et al., 2011;Capogna, 2014;Vereczki et al., 2016;Andrási et al., 2017). Similar to their neocortical and hippocampal counterparts, BLA IN subtypes may serve profoundly different roles in the spatiotemporal control of local network activity associated with different brain states and behavioral contexts (Tremblay et al., 2016).
Most BLA INs are positive for one of the calcium binding proteins, either calbindin or calretinin (CR) (McDonald and Mascagni, 2001;Spampanato et al., 2011). In rats, CRϩ INs also express to a certain extent the neuropeptides vasoactive intestinal polypeptide (VIP) and cholecystokinin (CCK; McDonald and Mascagni, 2002;Mascagni and McDonald, 2003) with some overlap between these two subgroups.
Despite the critical role of VIPϩ INs in regulating circuit operation, there have been no studies directly examining VIPϩ INs in the mouse BLA so far. The present study was undertaken to provide basic information about the distribution and connectivity of VIPϩ INs in the mouse BLA as well as some of their key neurochemical and electrophysiological properties. Because the afferent systems that can recruit VIPϩ INs are crucial determinants of the time course and specificity of the ensuing disinhibition, we have also investigated the inhibitory and excitatory inputs to these INs.

Materials and Methods
Animals. All procedures involving animals were performed according to methods approved by the Austrian Animal Experimentation Ethics Board (Bundesministerium für Wissenschaft und Verkehr, Kommission für Tierversuchsangelegenheiten) as well as by Hungarian legislation (1998. XX-VIII. section 243/1998, renewed in 40/2013 and institutional guidelines. All procedures complied with the European convention for the protection of vertebrate animals used for experimental and other scientific purposes (ETS number 123). Every effort was taken to minimize animal suffering and the number of animals used. For this study, the following mouse lines were used: C57BL/6J (Charles River), VIPtm1 Surgical procedures and viral vectors. Anesthesia was induced with 120 mg/kg ketamine and maintained with 1-2% vaporized sevoflurane. Mice were secured in a stereotaxic frame and injections were aimed at the following coordinates: 1.3 mm posterior to bregma; 3.4 mm lateral to the midline; and 4.5 mm deep from the cortical surface. A total of 500 nl of AAV2/6-CBA-FLEX-GFP (Murray et al., 2011; flow rate: 50 nl/min) was unilaterally injected into the BLA of 12-week-old homozygote VIP-IRES-cre mice. In addition, a total of 300 nl of AAV2/5-EF1-DIO-hChR2(H143R)-mCherry (University of North Carolina Vector Core Facility) was bilaterally injected into the BLA of 5-6-week old offspring produced by crossing VIP-IRES-cre and GAD65-EGFP mice. The cannula was slowly withdrawn 5 min after injection. GFP and ChR2 expression was allowed for 2 and 4 -5 weeks, respectively, before the animals were killed.
Tissue preparation for immunocytochemistry. After being anesthetized with thiopental (120 mg/kg; Sandoz), adult male mice were transcardially perfused with 0.9% NaCl followed by an ice-chilled fixative solution. For fluorescence microscopy, a fixative solution composed of 2% paraformaldehyde (PFA) and 15% saturated picric acid in 0.1 M phosphate buffer (PB) was used. On the other hand, for horseradish peroxidase (HRP) reactions, the fixative solution contained 4% PFA and 15% saturated picric acid in 0.1 M PB, pH 7.4. The same fixative was also used for pre-embedding electron microscopy (EM) with the addition of 0.05% glutaraldehyde. Brains were immediately removed from the skull, washed in 0.1 M PB, and sliced coronally in 50-m-thick sections for light microscopy or in 70-m-thick sections for EM on a Leica VT1000S vibratome (Leica Microsystems). Sections were stored in 0.1 M PB containing 0.05% sodium azide at 6°C until further use. Brains processed for EM were first cut in 6 -8-mm-thick blocks, cryoprotected in a solution of 20% sucrose in 0.1 M PB, and freeze-thawed once before sectioning.
Antibody characterization. All primary and secondary antibodies used in this study are listed in Tables 1 and 2. The rabbit polyclonal VIP antibody (ImmunoStar, catalog #20077, lot #1339001) was raised against a synthetic VIP peptide. The specificity of the antiserum was confirmed by soluble preadsorption test with VIP at a concentration of 10 Ϫ5 M, which abolished VIP immunolabeling (Sloviter and Nilaver, 1987; man-ufacturer's information). The other rabbit polyclonal VIP antibody used in this study has been characterized previously (Köves et al., 1990).
The mouse monoclonal CCK-8 antibody (Frontier Institute, catalog #CCK8-MO-167-1) was raised against a synthetic CCK-8 peptide, whereas the rabbit polyclonal pro-CCK antibody (Frontier Institute, catalog #CCK-pro-Rb-Af350) was raised against the last 9 aa of the C terminus. These antibodies specifically react with CCK-8 and do not cross-react with gastrin (manufacturer's information). The guinea pig polyclonal VGluT1 antibody (Millipore, catalog #AB5905, lots #23050127 and #2484243) was raised against the synthetic rat VGluT1 protein. Preadsorption of the VGluT1 antiserum with the immunogen peptide eliminates the immunostaining (manufacturer's information).
Both the guinea pig polyclonal VGluT2 antibody from Millipore Bioscience Research Reagents (catalog #AB5907, lot #23041014) and from Synaptic Systems (catalog #135404) were raised against a synthetic rat VGluT2 protein. Preadsorption of the VGluT2 antiserum with the immunogen peptide eliminates all immunostaining (manufacturer's information).
The mouse monoclonal neuronal nitric oxide synthase (nNOS) antibody (Sigma-Aldrich, catalog #N2280, lot #082M4835) was raised against a recombinant nNOS fragment (amino acids 1-181) and on Western blot reacted specifically with brain NOS and did not react with NOS derived from macrophages and endothelial cells (manufacturer's technical information).
The rabbit polyclonal CB1 antibody (Cayman Chemicals, catalog #10006590, lot #04574771) was raised against a synthetic peptide from the C-terminal region of the human CB1 receptor (manufacturer's technical information). The antibody labels GABAergic axon terminals (e.g., in the amygdala; Vereczki et al., 2016). The goat polyclonal CB1 antibody (Frontier Institute, catalog #CB1-Go-Af450) was raised against the C-terminal region of the mouse protein CB1 receptor, and the specificity was tested in CB1 knock-out mice (manufacturer's technical information).
The guinea pig polyclonal VGAT antibody (Synaptic Systems, catalog #131004) was raised against the cytoplasmic part of the VGAT protein and the specificity of the antibody has been verified in knock-out mice (manufacturer's technical information).
The mouse monoclonal bassoon antibody (Abcam, catalog #SAP7F407) was raised against a recombinant rat bassoon protein (manufacturer's technical information).
The guinea pig polyclonal NPY antibody (Synaptic Systems, catalog #394 004) was raised against a recombinant mouse NPY (manufacturer's technical information).
The rabbit monoclonal CaMKII antibody (Abcam, catalog #ab52476) was raised against a synthetic peptide of the human CaMKII (manufacturer's technical information).
HRP-DAB immunocytochemistry. Immunocytochemistry experiments were performed according to previously published procedures with minor modifications (Sreepathi and Ferraguti, 2012). Primary antibodies were diluted in a solution containing 2% normal goat serum (NGS), 0.3% Triton X-100 in Tris-buffered saline (TBS), pH 7.4. Free-floating sections were incubated in primary antibodies (Table 1) for 2 d at 6°C and then in secondary antibodies (Table 2) overnight. After extensive washing, sections were incubated in an ABC complex solution (1:100; Vectastain ABC kit, Vector Laboratories) made up in TBS, followed by diaminobenzidine (DAB) as a chromogen (0.5 mg/ml in tris buffer) and 0.003% H 2 O 2 , as the electron donor, for 5 min. The sections were mounted onto gelatin-coated slides, dehydrated in an ascending ethanol series followed by an incubation in butylacetate, and coverslipped with Eukitt (Christine Gröpl, Tulln, Austria).
Double immunofluorescence labeling using tyramide signal amplification. Immunofluorescence experiments were performed according to previously published procedures with minor modifications (Sreepathi and Ferraguti, 2012). Primary antibodies (Table 1) were prepared in 2% NGS and 0.1%Triton X-100 in TBS (TBS-T). After incubation with primary and secondary antibodies (Table 2), the fluorescence signal for VIP was enhanced using a tyramide signal amplification kit (TSA Fluorescein System, PerkinElmer). For the enhancement, sections were first incubated in streptavidin-HRP (diluted 1:500 in 2% NGS in TBS-T) for 30 min at room temperature and then, after two washes in TBS, in a fluorophore tyramide solution for 6 min. Sections were then mounted onto gelatin-coated slides and coverslipped with Vectashield (Vector Laboratories). When mouse monoclonal antibodies were used, sections were pretreated with a Mouse-on-Mouse kit (Vector Laboratories) to reduce endogenous mouse Ig staining.
Three-dimensional reconstruction. Three-dimensional (3D) model reconstructions were made from HRP-DAB stained sections for VIP, covering the full length of the amygdala. Images were acquired through a BX51 microscope (Olympus) equipped with a motorized stage (MAC 6000, MBF Bioscience) and a digital camera (QImaging). Drawings and reconstructions were performed with the Neurolucida software (Version 11, MBF Bioscience, RRID:SCR_001775).
Volume correction for tissue shrinkage. To obtain the tissue shrinkage factor due to tissue fixation, the brain volumes of 13-16-week-old C57BL/6J mice (n ϭ 4; 25-30 g) were compared before and after transcardial perfusion with a 3 tesla whole-body MRI device. A resolution of 0.34 ϫ 0.34 ϫ 0.3 mm was obtained with a T2-weighted 3D turbo spinecho sequence. To guarantee imaging without movement artifacts, the animals were anesthetized with an intraperitoneal injection of ketamine and xylazine (80 mg/kg ketamine and 5 mg/kg xylazine dissolved in a 0.9% sodium chloride solution). Immediately after imaging, the animals were transcardially perfused with 4% paraformaldehyde and 15% picric acid. The brain was removed, placed in PBS-filled falcon tubes, and imaged again. To measure the volume, the image sequences were analyzed with the polygon selection tool of ImageJ (Version 1.48k, RRID: SCR_003070). The volume of the whole brain until the end of the cerebellum was measured and then the volume of the ventricles was subtracted. The volume shrinkage of brain tissue due to fixation was 19.9 Ϯ 3.0%. To determine the shrinkage factor due to the HRP-DAB processing, randomly selected unprocessed sections (n ϭ 4) were mounted on an object slide with PBS. Subsequently the section area was measured with the Neurolucida software using a 4ϫ objective. The area of these sections was measured again after HRP-DAB immunolabeling. The area shrinkage factor was 33.4 Ϯ 2.3%.
Pre-embedding immuno-EM of AAV2/6-CBA-FLEX-GFP-injected brains. EM was used to validate light microscopic observations because, while light microscopy gives an estimate of preferred targets, it can be inaccurate in the identification of synaptic contacts (Tamás et al., 1997). Pre-embedding immuno-EM experiments were performed according to previously published procedures with minor modifications (Sreepathi and Ferraguti, 2012). VGluT1 and VGluT2 were visualized by nanogoldsilver enhanced reaction. GFP-labeled profiles were revealed by an ABC-HRP reaction. Silver enhancement was always performed first. Fab fragment secondary antibodies coupled to nanogold (1.4 nm) were enhanced with a silver amplification kit (HQ Silver Enhancement Kit, Nanoprobes). Contrast was enhanced using 2% osmium tetroxide v/v (Agar Scientific) in 0.1 M PB for 40 min at room temperature and 1% uranyl acetate w/v (Agar Scientific) in 50% ethanol for 30 min at room temperature. The sections were then dehydrated in increasing gradients of ethanol, immersed in propylene oxide, and embedded in epoxy resin (Durcupan ACM, Sigma-Aldrich) on greased glass slides. Regions of interest were dissected under a stereomicroscope and re-embedded in Durcupan ACM. Ultrathin sections (70 nm) were cut with an ultramicrotome (Ultracut S, Leica Microsystems) and collected on Formvarcoated copper slot grids. The ultrastructural analysis of the specimens was performed using a Philips CM 120 electron microscope equipped with a Morada CCD transmission EM camera (Soft Imaging Systems).
Quantification of VIPϩ neuron and bouton density. The density of VIPϩ cell bodies and boutons (n ϭ 8 amygdalae) was calculated on images of HRP-DAB sections immunolabeled for VIP by using the Neurolucida software. Borders of the LA and BA nuclei were outlined according to the pattern revealed by immunocytochemistry. The nuclear subdivisions were identified with the help of a mouse brain atlas (Franklin and Paxinos, 2007). The total number of VIPϩ neurons was determined by counting VIPϩ neurons in all sections containing the BLA. To measure the bouton density, we used a 100ϫ objective and a counting frame (50 ϫ 50 m). Bouton density was counted in every fourth section from three nonoverlapping counting frames per subdivision and per section using the Neurolucida software. To ensure that the neurons and boutons were counted as accurately as possible, the focus was maintained through the whole thickness of each section.
Colocalization of VIP with other neuronal markers. To quantify the colocalization of VIP and CR (n ϭ 6 amygdalae), images were taken on an epifluorescence microscope (AxioImager M1, Zeiss) using the Openlab software (Version 5.5.0; RRID:SCR_012158). Images were then imported and analyzed with the Neurolucida software. Each counting field was taken at three different z levels and for both channels in every third section. All images were analyzed using nonoverlapping counting frames (150 ϫ 150 m). For each BLA, a fixed number of counting frames were analyzed: 10 for the dorsolateral subdivision of the lateral nucleus (LAdl), 6 for the ventrolateral subdivision of the lateral nucleus (LAvl), 5 for the ventromedial subdivision of the lateral nucleus (LAvm), 30 for the anterior subdivision of the basal nucleus (BAa), and 14 for the posterior subdivision of the basal nucleus (BAp). The colocalization degree of VIP and CCK (n ϭ 6 amygdalae) was obtained by counting all neurons in every third section in the respective subdivision without using counting frames. Similarly, BLA sections at different rostrocaudal levels were tested using a combination of guinea pig anti-PV (Synaptic Systems; 1:3000) and rabbit anti-VIP (Immunostar; 1:1000) and visualized with donkey anti-rabbit Cy3 (1:400) and anti-guinea pig Alexa Fluor 488 (1: 1000; Jackson ImmunoResearch).
In a different set of experiments, we calculated the colocalization of CCK and CR in the BLA. Here, CCK-containing and CR-containing neurons were visualized with the rabbit anti-pro-CCK (Frontiers Institute; 1:1000) and guinea pig anti-CR (Synaptic Systems; 1:1000), respectively, using Cy3-conjugated donkey anti-rabbit and Alexa Fluor 488-conjugated donkey anti-guinea pig secondary antibodies. The potential colocalization between VIP and NPY in BLA neurons was analyzed in coronal sections obtained from NPY-GFP mice (Jackson Laboratory) stained with a rabbit anti-VIP (Immunostar; 1:1000) and visualized using a donkey anti-rabbit HRP (Invitrogen; 1:500) and TSAbiotin streptavidin-Dylight 649 (Vector Laboratories).
Digital image processing. Digital images were adjusted using Photoshop CS6 Extended (RRID:SCR_014199) on an Apple computer for contrast, brightness, and tonal values for the entire frame.
Estimation of the target distribution of VIPϩ varicosities in the amygdala. To estimate the ratio of the GABAergic and non-GABAergic postsynaptic targets of VIPϩ boutons, we used two offspring (P63 and P76) of a VGAT-Cre and Ai14 breeding pair. In these mice, the fluorescent protein tdTomato was selectively expressed in those cells expressing the vesicular GABA transporter (VGAT), i.e., in GABAergic cells. The mice were transcardially perfused with 4% PFA in 0.1 M PB. The brains were resectioned into 50-m-thick coronal sections, and those containing the amygdala were further processed for immunocytochemistry. Briefly, the amygdala-containing sections were incubated with a solution containing 0.2% Triton X-100 and 10% normal donkey serum (NDS) in 0.1 M PB for 1 h at room temperature. This was followed by a 5-d-long incubation at 4°C in a solution containing 0.2% Triton X-100, 2% NDS, 0.05% sodium azide, and a mixture of primary antibodies: rabbit anti-VIP [courtesy of Professor Tamás Görcs (Köves et al., 1990); 1:20,000], goat anti-CB1 (Frontier Institute; 1:1000), and rat anti-RFP (Chromotek; 1:1000) to enhance the endogenously expressed tdTomato signal. The primary antibodies were visualized using the following secondary antibodies: Alexa Fluor 488-conjugated donkey anti-rabbit (Invitrogen; 1:500), DyL405-conjugated donkey anti-goat (Jackson ImmunoResearch; 1:500), and Cy3-conjugated donkey anti-rat (Jackson Immu-noResearch; 1:500) antibodies. Images from the lateral and basal amygdala were obtained using a Nikon C2 confocal microscope (Plan Apo VC 60ϫ oil objective; numerical aperture, 1.40; xy, 0.10 m/pixel; z-step size, 0.25 m), and analyzed using ImageJ software.
For this analysis, we used the following criteria: (1) VIPϩ boutons in close apposition to a tdTomato-expressing profile were considered to establish an apposition onto GABAergic profiles; (2) in those cases in which the tdTomato-containing profile fully overlapped with the VIPϩ puncta, the bouton was considered to be a VGATϩ bouton and, therefore, not considered a putative postsynaptic target. The results obtained in the two animals were pooled together, and graphs were made using OriginPro 2015 (RRID:SCR_014212).
After Ն1 h of incubation, slices were transferred individually into a submerged-type recording chamber perfused with ACSF at 30 Ϯ 2°C with 2-3 ml/min flow rate. Recordings were performed under visual guidance using differential interference contrast microscopy (Olympus BX61W). tdTomato-containing VIPϩ cells in slices obtained from VIP-IREs-cre::Ai14 mice, and GFP-containing GAD65ϩ cells obtained from VIP-IREs-cre::GAD65-EGFP mice, respectively, were excited by a mercury lamp, and the fluorescent signal was detected by a CCD camera (Hamamatsu Photonics). Patch pipettes were pulled from borosilicate glass capillaries with inner filament (Hilgenberg) using a P-97 Micropipette puller (Sutter Instruments). For whole-cell recordings, pipettes with 0.188 mm wall thickness were used and had a resistance of ϳ6 -8 M⍀ when filled with the intrapipette solution. K-gluconate-based intrapipette solution used in all recordings contained the following (in mM): 110 K-gluconate, 4 NaCl, 2 Mg-ATP, 20 HEPES, 0,1 EGTA, 0.3 GTP (sodium salt), and 10 phosphocreatine adjusted to pH 7.3 using KOH and with an osmolarity of 290 mOsm/L. Biocytin in a concentration of 0.2% was added to the intrapipette solution. Recordings were made with a Multiclamp 700B amplifier (Molecular Devices), low-pass filtered at 3 kHz, digitized at 10 kHz, and recorded with in-house dataacquisition and stimulus software (Stimulog, courtesy of Professor Zoltán Nusser, Institute of Experimental Medicine, Hungarian Academy of Sciences, Budapest, Hungary). Recordings were not corrected for junction potential. To record the firing characteristics, cells were injected with 800-ms-long hyperpolarizing and depolarizing square current pulses with increasing amplitudes from 10 to 600 pA. These voltage responses were analyzed using in-house analysis software SPIN1.0.1 (courtesy of Professor Zoltán Nusser, Institute of Experimental Medicine, Hungarian Academy of Sciences) for the positive steps, and a custom-made program written in Matlab (RRID:SCR_001622) for the negative steps (courtesy of Dr. Szabolcs Káli, Institute of Experimental Medicine, Hungarian Academy of Sciences). Analysis of spiking properties was based on previously published procedures (Antal et al., 2006). Comparison of single-cell properties (including both active and passive membrane characteristics) has been done between groups of neurons categorized based on their neurochemical marker content. These data are reported in Tables 3, 4, and 5. In addition, we rearranged the dataset shown in Table 5 into three groups based on the spiking features of VIPϩ INs reported in Figure 5 and regardless of their CR content.
For recording IPSPs, postsynaptic principal neurons were held in current-clamp mode at ϳϪ50 mV [the reversal potential (Erev) of IPSPs using the intrapipette solution containing 4 mM Cl Ϫ was found earlier to be ϳϪ77 mV; . Bridge balance was adjusted throughout the recordings. Postsynaptic responses in the presence of 10 M CNQX [an AMPA/kainate (AMPA/KA) type of ionotropic glutamate receptor antagonist] were evoked by three 5-ms-long pluses of blue light at 10 Hz every 30 s using a Polygon 400 (Mightex Systems). After a 10-min-long baseline period, WIN 55,212-2 (1 M; a cannabinoid receptor agonist) together with CNQX (10 M) were washed in for 10 min to test the sensitivity of evoked responses for CB1. In four experiments, bath application of WIN 55,212-2 was followed by the wash-in of a mixture containing WIN 55,212-2 and AM 251 (both at 1 M; the latter is a CB1 antagonist) in the presence of 10 M CNQX. Recordings were analyzed with EVAN 1.3 (courtesy of Professor István Mó dy, Department of Neurology and Physiology, UCLA, Los Angeles, California) and Origin 9.2 (OriginLab, RRID:SCR_014212). CNQX disodium salt was added to the superfusate from a 10 mM stock solution dissolved in H 2 O, WIN 55,212-2 was added from a 20 mM stock solution dissolved in 0.1 M HCl, and AM 251 was added from a 10 mM stock solution dissolved in DMSO.
In vitro recordings obtained in slices prepared from VIP-IRES-cre:: GAD65-EGFP mice injected with AAV-DIO-ChR2-mCherry into the BLA were performed in ACSF and the postsynaptic responses in EGFPϩ neurons were evoked by 10-ms-long pulses of blue light applied every 20 s. In these recordings, intrapipette solutions containing 4 mM Cl Ϫ (as above) or 60 mM Cl Ϫ were used. The latter had a composition of 54 K-gluconate, 4 NaCl, 56 KCl, 2 Mg-ATP, 20 HEPES, 0,1 EGTA, 0.3 GTP (sodium salt), and 10 phosphocreatine in mM adjusted to pH 7.3 using KOH and with an osmolarity of 290 mOsm/L. Biocytin at a concentration of 0.2% was added to the intrapipette solution. To evaluate the magnitude of postsynaptic responses evoked upon light stimulation using both intrapipette solutions, we calculated the conductance of the inhibitory postsynaptic responses (IPSCs) by dividing the peak amplitude of events with the difference between the Erev of IPSCs and the holding potential (i.e., with the driving force). Erev of IPSCs using an intrapipette solution containing 4 mM Cl Ϫ was Ϫ77 mV (see above), while this value for an intrapipette solution containing 60 mM Cl Ϫ was determined experimentally as Ϫ31 mV. IPSCs were recorded at a holding potential of Ϫ45 and Ϫ65 mV having 4 and 60 mM Cl Ϫ in the intrapipette solution, respectively. When the light stimulation evoked action current(s) in a postsynaptic neuron [e.g., ChR2 was expressed in an EGFPϩ cell (n ϭ 7)], data were excluded from further analysis.
Immunostaining following in vitro recordings. After recording, slices were fixed overnight by immersion in a solution containing 4% PFA in 0.1 M PB. After fixation, slices were rinsed three times in PB and kept with sodium azide at 4°C for Յ1 week. Biocytin-filled recorded cells were visualized with Cy3-conjugated streptavidin. We found that the intensity of the Cy3-conjugated streptavidin labeling was strong enough to clearly distinguish the processes filled by biocytin from the endogenous tdTomato signal (Vereczki et al., 2016). All cells were imaged using a Nikon C2 confocal microscope (Plan Apo VC 20ϫ objective; numerical aperture, 0.75; z-step size, 0.5 m; xy, 0.62 m/pixel). Those INs (n ϭ 12) that had the most axon collaterals in the slice were further used for morphological reconstructions. The dendritic and axonal arbors of the intracellularly filled INs were reconstructed with Neurolucida 10.53 software (RRID:SCR_001775), using confocal stacks acquired from the cell before resectioning. The drawings of each cell were analyzed with Neurolucida Explorer (RRID:SCR_001775), and the values were corrected for shrinkage and flattening of the tissue (correction factor in the z-axis: 1.7; no correction in the x-axis and y-axis). Branched structure analysis was used to study the dendritic/axonal length and number of nodes. Sholl analysis was used to estimate the complexity of the dendritic arbor by determining the number of processes crossing concentric spheres centered on the cell soma with 50 m increments in their radius. In some cases, slices containing the biocytin-labeled INs were embedded in 1% agar and resectioned into 30-m-thick sections. The sections were then blocked with 10% NDS (Vector Laboratories) in TBS, pH 7.4. Sections from single slices were divided into two parts that were immunostained differentially. In one case, the following antibodies were used: rabbit anti-CR (Swant; 1:3000), mouse anti-bassoon (Abcam; 1:3000), and guinea pig anti-VGluT1 (Millipore; 1:3000) in TBS with 2% NDS and 0.05% Triton X-100. CR was visualized with Cy5-conjugated donkey anti-rabbit (Jackson ImmunoResearch; 1:500), bassoon was visualized with Alexa Fluor 488-conjugated donkey anti-mouse, and VGluT1 was visualized with DyL405-conjugated donkey anti-guinea pig (Jackson Im-munoResearch; 1:500). In the second case, the following antibodies were used: mouse anti-bassoon (Abcam; 1:3000) and guinea pig anti-VGluT2 (Synaptic Systems; 1:3000). Bassoon was visualized with Alexa Fluor 488-conjugated donkey anti-mouse (Jackson ImmunoResearch; 1:500), and VGluT2 was visualized with DyL405-conjugated donkey anti-guinea pig (Jackson ImmunoResearch; 1:500). Confocal images for investiga-   Investigation of the neurochemical marker content of EGFPϩ neurons in GAD65-EGFP mice. Some slices prepared from AAV-injected VIP-IREScre::GAD65-EGFP mice were fixed in 4% PFA in 0.1 M PB and after overnight incubation were resectioned at a 40 m thickness. Different sections were processed for single immunostaining with rabbit anti-pro-CCK (Frontiers Institute; 1:1000), guinea pig anti-NPY (Synaptic Systems; 1:1000), or rabbit anti-CaMKII (Abcam; 1:250). Visualization was performed with Cy5-conjugated secondary antibodies. In addition, double immunostaining was obtained using a mixture of rat anti-SOM (Millipore; 1:500) and guinea pig anti-PV (Synaptic Systems; 1:1000) antibodies or a mixture of rabbit anti-VIP (Immunostar; 1:4000) and guinea pig anti-CR (Synaptic Systems; 1:1000) antibodies. Here, DyL405-conjugated and Cy5-conjugated appropriate secondary antibodies were used to visualize the antigen-antibody complexes. Confocal images were obtained with a Nikon C2 confocal microscope (Plan Apo VC 20ϫ objective; numerical aperture, 0.75; z-step size, 2 m; xy, 0.62 m/pixel) and the analysis was conducted using Neurolucida Explorer (RRID:SCR_001775). Analysis of the colocalization between CCK and EGFP was limited to neurons with large soma, as these are known to be CB1ϩ basket cells (Szabó et al., 2014) and we did not find overlap between CCK and VIP in these cells in this mouse line. In addition, data for VIPϩ and CRϩ INs were combined.
Experimental design and statistical analyses. Statistical tests were performed with the Prism (Version 7.0a for Mac OSX; RRID:SCR_002798) or the Origin software. For two-group comparisons, the Student's t test was used for normally distributed datasets and the Mann-Whitney test was used for nonparametric distributions. For multiple comparisons, one-way ANOVA followed by the Bonferroni post hoc test was used, whereas for nonparametric distributions, the Kruskal-Wallis test followed by the Dunn's multiple post hoc test was applied. Statistical significance was set at p Ͻ 0.05. Correlations were calculated by using the Pearson correlation test. Distributions (e.g., dendrite diameter targeted by VGluT1ϩ or VGluT2ϩ terminals) were analyzed using the twosample Kolmogorov-Smirnov test.

Density and distribution of VIP؉ neurons in the mouse BLA
In the mouse BLA, the general distribution and morphology of VIPϩ INs at the light microscopic level (Fig. 1A) was highly similar to those of previous studies in the rat (McDonald, 1985;Muller et al., 2003). VIP immunoreactivity was associated with the somata and primary dendrites of a discrete population of neurons as well as with their axon terminals. A dense plexus of VIPϩ axons was also observed in the central lateral amygdaloid nucleus (Fig. 1A). Neurons of the BLA containing VIP were mostly bipolar or bitufted having relatively small somata ( Fig.  1 B, C). Close immunofluorescence and EM examination of VIPϩ neurons showed that although the primary dendrites were mostly aspinous, the additional branches possessed a few stubby spines (Fig. 1D-F ).
To get the best approximation of the absolute number and density of VIPϩ INs within each subdivision of the LA and BA, we calculated the amygdala shrinkage due to both fixation and immunocytochemical staining procedures. Fixation shrinkage was obtained by comparing the volume of mouse brains before and after transcardial perfusion with a 3 tesla MRI device ( Fig.  2A). Shrinkage due to processing was obtained by comparing the area of sections before and after immunocytochemical procedures. A volume of 1.497 Ϯ 0.246 mm 3 (n ϭ 16 amygdalae) was obtained for the whole BLA (excluding the basomedial nucleus). The BAa contributed most of the volume (0.665 Ϯ 0.105 mm 3 ) followed by the BAp (0.420 Ϯ 0.072 mm 3 ), whereas the three LA subdivisions were substantially smaller (LAdl, 0.246 Ϯ 0.036 mm 3 ; LAvl, 0.096 Ϯ 0.020 mm 3 ; LAvm, 0.071 Ϯ 0.013 mm 3 ). To quantify the number of VIPϩ neurons, all the sections containing the BLA were immunostained. The BLA was found to contain 1583 Ϯ 89 VIPϩ neurons, without a significant difference between the left and right hemispheres ( p ϭ 0.35, t ϭ 1.109, df ϭ 3, t test). The density of VIPϩ neurons, however, varied signifi-cantly among the distinct subdivisions ( p Ͻ 0.0001, F ϭ 13.88, df ϭ 4, one-way ANOVA). The LAdl had the lowest density of VIPϩ cells (769 Ϯ 131 cells/mm 3 ), followed closely by the BAa (990 Ϯ 114 cells/mm 3 ). The VIPϩ cell density was slightly higher in the LAvm (1094 Ϯ 198 cells/mm 3 ) and LAvl (1102 Ϯ 99 cells/ mm 3 ), whereas the highest density was detected in the BAp (1347 Ϯ 216 cells/mm 3 ). We observed a rostrocaudal gradient with fewer VIPϩ neurons per mm 3 rostrally and a higher density caudally (Kruskal-Wallis test, H ϭ 147.7, p Ͻ 0.001; Fig. 2 B, C). However, the absolute number of VIPϩ cells was highest in the central sections of the BLA (rostral to caudal: Ϫ1.55 to Ϫ2.35; Kruskal-Wallis test, H ϭ 287.2, p Ͻ 0.001; Fig. 2D). The density of VIPϩ axon terminals in the BAa (325 Ϯ 12 terminals/100 m 2 ), BAp (389 Ϯ 12 terminals/100 m 2 ), and LAvl (322 Ϯ 14 terminals/100 m 2 ) was also analyzed and showed a significantly higher density ( p Ͻ 0.01, df ϭ 2, one-way ANOVA with Bonferroni post hoc test) in the BAp compared with the other two subdivisions. Surprisingly, when the density of VIPϩ somata was related to the density of axon terminals (Fig. 2E), no significant correlation could be observed (Pearson correlation coefficient: LAvm, r ϭ Ϫ0.54, p ϭ 0.17; BAa, r ϭ Ϫ0.04, p ϭ 0.92; BAp, r ϭ Ϫ0.39, p ϭ 0.34), which suggests a complex intrinsic and possibly extrinsic innervation pattern of VIPϩ axon terminals in the BLA.

Colocalization between VIP and other neurochemical markers
In another set of experiments, the colocalization of VIP with other interneuronal markers, such as the calcium-binding protein CR and the neuropeptide CCK, was investigated by double immunofluorescence labeling, as previous studies have shown their coexpression in the rat BLA (McDonald and Pearson, 1989;McDonald, 1994;Mascagni and McDonald, 2003). Our results showed a high degree of colocalization between VIP and CR (Fig.  3A), which, however, varied among distinct nuclear subdivisions ( p ϭ 0.008, df ϭ 4, one-way ANOVA). The highest rate of CR coexpression by VIPϩ cells was found in the LAvm (68.11 Ϯ 3.5%, 36 double-labeled cells out of 53) followed by the LAvl (61.04 Ϯ 7.37%, 50 double-labeled cells out of 80) and the LAdl (55.29 Ϯ 7.85%, 73 double-labeled cells out of 133). In the BAa, 47.59 Ϯ 6.97% (200 double-labeled cells out of 411) of the VIPϩ Figure 3. Colocalization of VIP with CR, CCK, or nNOS in the mouse BLA. A, Epifluoresence microscopic images of a neuron colocalizing VIP (magenta) and CR (green) in the BA nucleus. B, Epifluoresence microscopic images of a neuron colocalizing VIP (magenta) and CCK (green) in the BA nucleus. Arrows indicate neurons coexpressing VIP with another marker. C, No colocalization between VIP (magenta) and nNOS (green) was detected in the BLA. D, Percentage of colocalization between VIP and CR along the rostrocaudal axis of the mouse BLA. E, Percentage of colocalization between VIP and CCK along the rostrocaudal axis of the mouse BLA. For all datasets, one in every three sections (50 m) was analyzed. N ϭ 3 mice. Lineplots represent mean, filled areas represent SEM. Scale bar, 50 m. cells coexpressed CR, whereas the lowest amount of colocalization was detected in the BAp (41.62 Ϯ 4.77%, 67 double-labeled cells out of 161). The colocalization of VIP with CCK ( Fig. 3B) was relatively scarce, ranging from ϳ12 to ϳ3%, and did not statistically differ ( p ϭ 0.23, df ϭ 4, one-way ANOVA) across the distinct subdivisions of the BLA. The highest rate of CCK coexpression by VIPϩ neurons was found in the LAvm (12.01 Ϯ 5.66%, 5 double-labeled cells out of 61), followed by the LAdl (6.66 Ϯ 6.29%, 10 double-labeled cells out of 138), BAp (4.93 Ϯ 2.78%, 11 double-labeled cells out of 223), BAa (3.72 Ϯ 1.4%, 14 double-labeled cells out of 378), and LAvl (2.40 Ϯ 1.70%, 3 double-labeled cells out of 95). Plotting the percentage of VIPϩ neurons colocalized with CR or CCK along the rostrocaudal axis revealed a relatively uniform distribution with no significant difference (VIP/CR: one-way ANOVA, p Ͼ 0.05; VIP/CCK: Kruskall-Wallis ANOVA, p Ͼ 0.05) detected among the different rostrocaudal bregma levels (Fig. 3 D, E).
Because in the peripheral nervous system VIPϩ neurons are also known to coexpress NOS (Chino et al., 2002), we have examined the possible colocalization between VIP and this enzyme. However, no colabeled cells could be detected in the mouse BLA (Fig. 3C). No colocalization was also observed between VIP and PV or NPY (data not shown).

Estimating the ratio of GABAergic postsynaptic targets of VIP؉ INs
As VIPϩ INs in the hippocampus and dentate gyrus (Acsády et al., 1996;Hajos et al., 1996), as well as in the neocortex (He et al., 2016), form two major groups, i.e., IS-INs and basket cells expressing CCK and CB1 cannabinoid receptors, we first assessed the ratio of VIPϩ varicosities that originate from basket cells. To this end, we calculated the percentage of VIPϩ boutons that also showed immunoreactivity for CB1. We observed that only a minority of VIPϩ axon varicosities may belong to basket cells, as 18% (74 of 419) in the LA and 3% (13 of 424) of VIPϩ boutons in the BA, respectively, were found to be immunostained also for CB1 (Fig. 4). These data are overall in agreement with the colocalization results obtained at the soma level (see above) and sug-gest that the vast majority of VIPϩ INs in the BLA are putative IS-INs.
To provide further support for this assumption, we next estimated the ratio of VIPϩ boutons immunonegative for CB1 (i.e., that should originate from putative IS-INs), which form close appositions with GABAergic profiles. Through the selective expression of the fluorescent protein tdTomato in GABAergic cells, we could observe that VIPϩ/CB1Ϫ boutons preferentially formed close appositions with tdTomatoϩ somata and dendrites (Fig. 4). In the LA and BA, 61% (211 of 345) and 66% (272 of 411) of VIPϩ/CB1Ϫ boutons were found to be associated to tdTomatoϩ profiles, respectively. However, it should be noted that our estimate of the ratio of VIPϩ/CB1Ϫ boutons forming close appositions with GABAergic profiles may be an underestimation, as the tdTomato signal, even after enhancement by immunostaining, might be below the detection threshold in the distal dendrites of GABAergic cells. On the other hand, and in agreement with previous findings (Muller et al., 2003), only a minority of VIPϩ/CB1ϩ basket cell boutons contacted tdTomatoϩ profiles (in the LA, 20%, 13 of 64 boutons; in the BA, 23%, 13 of 57 boutons).
Together these results support the idea that in the BLA, as in other cortical structures, most VIPϩ INs preferentially target local GABAergic cells, but only few are basket cells.

Electrophysiological properties of VIP؉ INs
To characterize further BLA VIPϩ INs, we studied their active and passive membrane properties by whole-cell recordings in acute slices prepared from VIP-IRES-cre mice crossed with Ai14 reporter mice, in which tdTomato under the control of crerecombinase was selectively expressed only in VIPϩ neurons. All recorded cells, both in the LA nuclei (n ϭ 10) and BA nuclei (n ϭ 23), were filled with biocytin and visualized, following the recording, for anatomical characterization. VIPϩ INs revealed a relatively high input resistance (350.3 Ϯ 28.26 M⍀, n ϭ 30) and a small membrane capacitance (46.92 Ϯ 2.98 pF), consistent with their small cell-body size and few proximal dendrites. As shown in Table 3, the recorded VIPϩ INs showed very similar passive  (Table 3), strengthening the conclusion that no CCKϩ/CB1ϩ/VIPϩ basket cell was sampled from slices of VIP-IRES-cre::Ai14 mice. We could further assess the single-cell properties of CCKϩ/CB1ϩ/VIPϩ basket cells, as 4 of 16 neurons recorded in BAC-CCK-DsRed mice showed VIP immunoreactivity in their axonal varicosities. CCKϩ/CB1ϩ basket cells containing or lacking VIP immunoreactivity did not differ in any of the parameters measured (Table 4). We did not test the CR content of CCKϩ/CB1ϩ basket cells as CCK and CR are basically mutually exclusive in BLA neurons (1 CRϩ neuron was found among 162 CCKϩ neurons, 1 CCKϩ neuron was detected among 174 CRϩ cells, n ϭ 2 mice).
Upon current pulse injections, VIPϩ INs recorded from slices of VIP-IRES-cre::Ai14 mice diverged substantially in their firing characteristics (Fig. 5). Three spiking patterns could be recognized based on the maximum number of spikes emitted upon depolarization and the length of the spike train during the depolarizing current pulses ( p Ͻ 0.001, Kruskal-Wallis ANOVA). Eight cells fired Ͻ10 action potentials during the 800-ms-long current pulses (average number of spikes, 6.9 Ϯ 0.8; range, 3-10), a spiking that was restricted to the first 200 ms of the depolarizing steps (Fig. 5A1,B1,C). A second group of cells fired throughout the whole current pulses, but emitted Ͻ20 spikes at their maximal firing rate (average number of spikes, 15.4 Ϯ 1.6; range, 7-20; n ϭ 11; Fig. 5A2,B2,C). The third cell category fired throughout the depolarizing pulses (average number of spikes, 41.1 Ϯ 3.2; range, 31-63; n ϭ 14; Fig. 5A3,B3,C). Statistical comparison of the passive and active membrane properties, however, revealed no difference among these three cell categories ( p Ͼ 0.1, Kruskal-Wallis ANOVA). As nearly 50% of VIPϩ INs coexpress CR, we examined the presence of this calcium binding protein in the soma and/or axon varicosities of the recorded and biocytin-filled cells. Of 20 cells tested, 11 were found to be immunopositive for CR (Fig. 5D). The presence of CR, however, did not seem to make any distinction among VIPϩ cells with different firing patterns (three of six in the short spiking category, three of five among cells with a low spiking rate, and five of nine among cells with a high spiking rate). We also assessed whether the single-cell properties differed between VIPϩ/CRϩ and VIPϩ/CRϪ IS-INs. The comparison revealed no significant difference in any of the parameters tested (Table 5). Therefore, VIPϩ IS-INs have similar characteristics regardless of their CR content.
Furthermore, the visualization of the biocytin content of the recorded cells enabled us to study in more detail the morphological properties of VIPϩ INs. Their reconstruction using the Neurolucida software did not reveal any obvious difference among VIPϩ cells belonging to the three spiking categories (Fig. 6A-C). Overall, we confirmed that VIPϩ INs are rather small cells. The total length of the dendrites and the axon collaterals were found to be 2312.5 Ϯ 229.4 m (n ϭ 12) and 7364.5 Ϯ 1607.3 m (n ϭ 8), respectively. The number of the dendrite nodes (27.7 Ϯ 1.7) and axon nodes (94.6 Ϯ 24.4) was also modest. Sholl analysis revealed that the vast majority of dendrites (81.5%) and axons (78.1%) could be found locally near the soma (Յ200 m; Fig. 6C). Remarkably, the individual dendritic and axon segments were similar in average length (Fig. 6C Muller et al., 2003), and neocortex (Staiger et al., 2004;Pfeffer et al., 2013), our data indicate that a fraction of VIPϩ INs belongs to basket cells expressing CCK and CB1. To test whether principal neurons in the amygdala indeed receive functional innervation from CCKϩ/CB1ϩ/VIPϩ basket cells, we performed in vitro recordings combined with optogenetics. In acute slices prepared from offspring of VIP-IRES-cre and Ai32 mouse lines, we recorded principal neurons using whole-cell mode (Fig. 7A-C). In the presence of an AMPA/KA ionotropic glutamate receptor antagonist, CNQX, we observed IPSPs evoked by whole-field bluelight illumination in ϳ20% (Fig. 7D) of the recorded cells (n ϭ 78; IPSP amplitude, 1.104 Ϯ 0.275 mV, n ϭ 16). Bath application of the CB1 agonist WIN 55,212-2 (1 M) abolished the IPSPs (6.6 Ϯ 1.6% of control amplitude, in control: 0.808 Ϯ 0.185 mV; in WIN: 0.047 Ϯ 0.011 mV, n ϭ 9, p ϭ 0.002, paired t test). In all cases tested, the coapplication of WIN 55,212-2 (1 M) with the CB1 antagonist AM251 (1 M) reversed the effects of WIN 55,212-2 on the IPSP amplitudes [control: 1.085 Ϯ 0.38 mV; WIN: 0.072 Ϯ 0.015 mV (7.5 Ϯ 1.4% of control amplitude); WINϩAM251: 0.737 Ϯ 0.212 mV (73.8 Ϯ 11.5% of control amplitude); n ϭ 4; control vs WIN effect, p ϭ 0.035; WIN vs WINϩAM251, p ϭ 0.046; control vs WINϩAM251, p ϭ 0.17; Fig. 7 E, F]. To further confirm the GABAergic nature of boutons expressing VIP and CB1, we performed immunostainings against CB1 and VGAT in VIP-IRES-cre::Ai14 mice. Indeed, in all but one tdTomato-expressing VIPϩ varicosities immunoreactive for CB1 (n ϭ 30), we could detect the presence of VGAT (Fig. 7G).
Since most, if not all, principal neurons in the BA receive perisomatic GABAergic inputs from CCKϩ/CB1ϩ basket cells (Vereczki et al., 2016), it was surprising to observe that light stimulation of ChR2-expressing VIPϩ INs that include basket cells gave rise to evoked responses in only ϳ20% of principal neurons. This discrepancy may have two, mutually nonexclusive explanations: first, our recording conditions were not optimal to detect postsynaptic responses in principal neurons originated from CCKϩ/CB1ϩ/VIPϩ basket cells; or second, not all principal neurons receive perisomatic GABAergic input from CCKϩ/ CB1ϩ/VIPϩ basket cells. To test the latter scenario, we examined the percentage of BLA principal neurons, identified by their immunoreactivity for Kv2.1 channels (Vereczki et al., 2016), that received close perisomatic appositions by VIPϩ/CB1ϩ boutons. CB1ϩ varicosities contacted the somata of all examined principal neurons, as previously reported (Vereczki et al., 2016), but not all principal cells were targeted by VIPϩ/CB1ϩ boutons. Principal cell somata in LA and BA were found to be contacted by Ն1 VIPϩ/CB1ϩ varicosity in 59.5 and 74.4% of the cases, respectively. However, there was a marked difference in the ratio of  VIPϩ/CB1ϩ boutons per cell in these two amygdala nuclei. In LA, nearly half of the CB1ϩ varicosities (51.8 Ϯ 5.2%, n ϭ 240 varicosities in 39 cell somata) that contacted the soma surface of principal neurons showed immunoreactivity for VIP, whereas this ratio was much lower in BA (15.3 Ϯ 1.7%, n ϭ 600 varicosities in 38 cell somata).
Together these data indicate that VIPϩ basket cells innervate a selected subset of BLA principal neurons, and that they contribute to a different extent to the perisomatic inhibition of principal neurons in the LA and BA.

VIP؉ INs give rise to inhibitory inputs onto different types of BLA INs
In a new set of experiments, we examined whether VIPϩ INs in the mouse BLA innervate other INs. Previous studies in the neocortex and hippocampus have shown that VIPϩ INs innervate GABAergic cells that express PV or SOM (Dávid et al., 2007;Pfeffer et al., 2013;Pi et al., 2013), a finding that was recently reported also in the BLA . Therefore, we took advantage of a GAD65-EGFP knock-in mouse line, in which INs that contain PV or SOM do not express EGFP (Ló pez-Bendito et al., 2004). As the expression of EGFP in this mouse line has never been tested in the BLA, we first examined the neurochemical content of neurons that express this fluorescent protein (Fig. 8A). In the BLA, comparable fractions of EGFPϩ neurons were found to be immunoreactive for CCK (9.6%, 20 of 208), NPY (13.9%, 28 of 201), and VIP and/or CR (9.2%, 25 of 271, n ϭ 2 mice; Fig. 8 B, C). PVϩ INs did not contain EGFP (0 of 171 for PV), and only a few EGFPϩ cells showed immunoreactivity for SOM (1.7%, 3 of 171 for SOM, n ϭ 2 mice; Fig. 8 B, C), in agreement with previous results (Ló pez-Bendito et al., 2004). However, we noticed that a considerable fraction of EGFPϩ neurons with large somata could not be labeled for any of the markers that have been examined. When these neurons were assayed for the expression of CaMKII, a reliable marker for principal neurons in the BLA (McDonald and Mascagni, 2002), to our surprise, 34.3% of green cells were immunopositive for CaMKII (122 of 356, n ϭ 3 mice; Fig. 8 B, C). In addition, a subpopulation of EGFPϩ neurons was immunonegative for all neurochemical markers tested. However, these neurons could be false negative, due to low levels of any of the antigens. Thus, in the BLA of GAD65-EGFP mice, in addition to principal neurons, Ն3 largely nonoverlapping populations of INs can be investigated, namely CCKϩ basket cells (Rovira-Esteban et al., 2017), NPY-expressing neurogliaform cells (Mań ko et al., 2012), and VIP/CR-containing IS-INs (present study).
To test whether these three IN types receive innervation from VIPϩ INs, we crossed VIP-cre mice with GAD65-EGFP knock-in mice. Then, into the BLAs of the offspring, we injected a credependent viral construct-transducing ChR2. In agreement with our neurochemical results, the vast majority of recorded neurons could be divided into four groups based on their firing pattern: principal neurons (n ϭ 24), CCKϩ basket cells (n ϭ 8), neurogliaform cells (n ϭ 37), and IS-INs (n ϭ 31; Fig. 8D). The morphological features of recorded neurons, when the visualization of the dendritic and axon arbor was successful (in 58 of 103 cells), confirmed the categorization based on single-cell properties (Fig.  8D). In addition, we also recorded from three neurons whose firing pattern was different from that of the other three IN types. Whole-field blue-light stimulation evoked postsynaptic responses in all cell types, but the likelihood, as well as the magnitude, of light-evoked events varied among the cell types ( Fig.  8D-F ). In all three unidentified cells (unINs), light-evoked postsynaptic responses could be also detected (IPSG: 0.12 Ϯ 0.07 nS, n ϭ 3). The light-evoked responses were mediated via GABA A receptors, as bath application of gabazine, a selective GABA A receptor antagonist, abolished them (n ϭ 9, 4 CCKϩ basket cells, 4 IS- INs, and 1 unIN).
Therefore, VIPϩ INs in the BLA give rise to GABA A receptormediated postsynaptic responses in several INs, including CCKϩ basket cells, neurogliaform cells, and other VIP/CRϩ IS-INs, although the strength or efficacy of the innervation appeared largely different among postsynaptic neurons.

GABAergic innervation of VIP؉ INs
Taking advantage of the recorded slices from VIP-IRES-cre::Ai14 mice, we have estimated the density of GABAergic inputs received by VIPϩ INs and the ratio of VIPϩ and/or CB1ϩ boutons among GABAergic varicosities forming close appositions with tdTomatoϩ profiles (Fig. 9A). We found that on the somata of VIPϩ INs (n ϭ 9), the density of VGATϩ axon terminals was 10.5 Ϯ 1.4/100 m 2 , while on their dendrites an average of 32.4 Ϯ 2.1 boutons could be detected along a 100-m-long segment (Fig. 9B). Analysis of multicolor stainings revealed that on the somata of VIPϩ INs, 2.3% of the total VGATϩ inputs (ϳ0.25 bouton/100 m 2 ) was also VIPϩ, and 24.2% (ϳ2.5 bouton/100 m 2 ) was immunoreactive for CB1 (Fig. 9C). No bouton contacting the soma was immunopositive for both tdTomato and CB1. On the dendrites, sampled Յ150 m from the soma, the ratio of VIPϩ GABAergic boutons was higher (9.7%; ϳ3 boutons/100 m), while the proportion of CB1ϩ GABAergic varicosities was smaller (9.3%, ϳ3 boutons/100 m) compared with that observed on the soma. We also found a small fraction of VGATϩ boutons (3.8%; ϳ1 bouton/100 m) forming apposi-tions to dendrites that were coimmunolabeled for tdTomato and CB1. As each of the nine recorded and labeled VIPϩ INs had a small soma, characteristic for IS-INs, these data indicate that Ͻ10% of GABAergic inputs onto VIPϩ INs preferentially targeting GABAergic cells originate from other VIPϩ INs and that ϳ15% of inhibitory axon terminals contain CB1.

Glutamatergic innervation of VIP؉INs
Likewise, to assess the density of glutamatergic inputs onto VIPϩ INs, sections from recorded and biocytin-filled cells (n ϭ 6) were stained for the VGluT1 or VGluT2 and for the scaffolding protein bassoon to visualize the presynaptic active zone (Fig. 10 A, B). Because of their complementary distribution in the brain, VGluT1 and VGluT2 can be used to discriminate at least in part the origin of glutamatergic inputs (Fremeau et al., 2004). VGluT1 predominates in cortical pyramidal neurons, whereas thalamic and hypothalamic projection neurons mostly have VGluT2 at their efferents (Fremeau et al., 2004). VGluT1ϩ axon terminals had a density of appositions on VIPϩ INs of 7.3 Ϯ 2.5/100 m 2 on the somata, an average of 16.0 Ϯ 1.6 m 2 on the proximal dendrite, and of 18.0 Ϯ 1.6 m 2 on the distal dendrites, as measured along a 100-m-long segment (Fig. 10C). VGluT2ϩ axon terminals had a similar density of appositions on VIPϩ INs with 18.0 Ϯ 1.7 boutons on the proximal and 17.0 Ϯ 1.8 on the distal dendrites per 100 m (Fig. 10D). As we could detect only two VGluT2ϩ appositions on the somata of VIPϩ INs in the analyzed sample, no mean density could be calculated. Based on their morphological and single-cell properties (see above), these biocytin-filled cells likely belong to IS-INs. Hence, our results suggest that VIPϩ IS-INs receive inputs from both cortical and subcortical structures.
To confirm that VGluT1ϩ and VGluT2ϩ axon terminals indeed form synapses with VIPϩ INs and to evaluate quantitatively postsynaptic target preference, we performed double immunopre-embedding EM experiments. To reveal the full dendritic domain of VIPϩ INs, we injected into the BLA of VIP-IRES-cre mice a viral vector to express the reporter protein GFP in a credependent fashion (Fig. 11 A, B). The presence of VGluT1 (Fig.  11C) or VGluT2 was visualized by silver-intensified immunogold and the GFP content by immunoperoxidase reactions, respectively (Fig. 11D). In agreement with the light microscopy results, both VGluT1ϩ and VGluT2ϩ axon terminals formed synapses preferentially with dendritic shafts, but also formed sporadic synapses on both somata and spines (Figs. 1 D, E, 10 E, F ). The dendrites targeted by VGluT1ϩ and VGluT2ϩ boutons showed a similar distribution ( p ϭ 0.08, two-sample Kolmogorov-Smirnov test) and range in terms of diameter ( p ϭ 0.08, Mann-Whitney test; Fig. 11 E, F ), in agreement with our light microscopy observations.

Discussion
In this study, we provide the first comprehensive analysis of the distribution of VIPϩ INs across the entire mouse BLA, as well as of their morphological and physiological properties. We also show that BLA VIPϩ INs are heterogeneous. Based on different criteria, in agreement with the general consensus on IN classification (DeFelipe et al., 2013), we have identified VIPϩ INs as belonging to IS-INs and CCKϩ basket cells (Fig.  12). Through a combination of electrophysiology and optogenetics in different transgenic mouse lines, we revealed that proach did not accommodate tests for PVϩ and SOMϩ INs, as they do not express EGFP in the reporter line we used. On the other hand, we have identified additional classes of INs contacted by VIPϩ IS-INs, which include CCKϩ basket cells, neurogliaform cells, and other VIP/CRϩ IS-INs. The number of postsynaptic INs within each of these different subtypes targeted by VIPϩ IS-INs, and/or the strength of the innervation, appeared quite diverse, suggesting a complex network regulation, which requires further functional characterization.
Our work substantially extends previous findings and shows for the first time that VIPϩ INs have a broad range of intrinsic postsynaptic targets, which include CCK basket and neurogliaform cells and IS-INs, in addition to PVϩ and SOMϩ INs. Particularly interesting is the innervation of neurogliaform cells as they evoke slow phasic inhibition of principal neurons by extrasynaptic release of GABA around their dendritic domains (Mań ko et al., 2012).

Synaptic inputs onto BLA VIP؉ INs
The BLA receives extensive cortical and thalamic glutamatergic inputs (Sah et al., 2003). We demonstrate that VIPϩ INs are direct targets of different excitatory afferents identified by their expression of VGluT1 and VGluT2. The synaptic inputs onto VIPϩ INs identified by light microscopy were confirmed by EM. The dendritic shafts of VIPϩ INs were the main subcellular domain targeted by glutamatergic inputs, with similar density on both the proximal and distal compartments. However, the density of excitatory inputs received by VIPϩ INs appears much lower compared with other INs, such as those expressing PV (Smith et al., 1998;Gulyás et al., 1999;Andrási et al., 2017). One EM study of CRϩ INs in the rat hippocampus revealed that they share a similar density of excitatory inputs to VIPϩ INs (Gulyás et al., 1999), thus suggesting that the glutamatergic innervation of VIPϩ and CRϩ INs significantly differs at least from PVϩ INs. Afferents to the BLA containing VGluT2 arise primarily from the thalamus and the ventromedial hypothalamus (Pitkänen, 2000;Fremeau et al., 2004). However, previous studies suggested that thalamic projections to the BA innervate dendritic spines of principal neurons almost exclusively (Carlsen and Heimer, 1988;Le-Doux et al., 1991). Considering that a selective innervation of VIPϩ INs by thalamic inputs would not be in conflict with a proportionally higher targeting of principal neuron spines, also in view of the relatively low density of these INs, tracing studies are needed to reveal the precise source of glutamatergic inputs containing VGluT2. The origin of VGluT1ϩ axon terminals, meanwhile, can arise from the prefrontal or other cortices as well as from the hippocampus, in addition to nearby principal neurons, since these are the primary areas projecting to the BLA and expressing VGluT1 (Pitkänen, 2000;Fremeau et al., 2004). In addition to excitatory inputs, VIPϩ INs displayed a high density of GABAergic synapses, nearly twice that of other BLA INs (Smith et al., 1998;Klenowski et al., 2015). Indeed, most of the synaptic inputs to their somata were GABAergic, similar to VIPϩ INs in the visual cortex (Hajó s and Zilles, 1988). Our findings indicate that the interconnectivity among VIPϩ INs accounts for only ϳ10% of their GABAergic synapses, and suggest that the remaining 90% come from other BLA INs. In the rat BLA, Muller and colleagues (2003)

Functional considerations
Recent data suggest that VIPϩ INs in cortical networks are critical for a functional disinhibition of principal cells (Kepecs and Fishell, 2014), as originally suggested by anatomical observations in the hippocampus (Acsády et al., 1996;Hajos et al., 1996). Disinhibition is emerging as a general mechanism for aversive learning that is not limited to the neocortex (Letzkus et al., 2015). Our findings suggest that VIPϩ INs can be recruited by excitatory inputs, which relay sensory stimuli from both cortical and subcortical areas. The disinhibitory role of VIPϩ INs may, therefore, provide a temporal window for firing of BLA principal neurons (Pouille and Scanziani, 2001;Wilent and Contreras, 2005) that may facilitate associative plasticity, e.g., the encoding of conditioned and unconditioned stimuli in fear conditioning. This requires a spatially and temporally coordinated regulation of the INs postsynaptic to VIPϩ IS-INs and their cooperative activity to release principal neurons from inhibition. Our findings also indicate that a minority of VIPϩ INs in the BLA are basket cells, whose specific role in circuit function remains to be elucidated. Future work using intersectional approaches combined with optogenetic manipulation of VIPϩ INs in behaving animals will be an important next step in determining their role in BLA computation.