Differential surface density and modulatory effects of presynaptic GABAB receptors in hippocampal cholecystokinin and parvalbumin basket cells

The perisomatic domain of cortical neurons is under the control of two major GABAergic inhibitory interneuron types: regular-spiking cholecystokinin (CCK) basket cells (BCs) and fast-spiking parvalbumin (PV) BCs. CCK and PV BCs are different not only in their intrinsic physiological, anatomical and molecular characteristics, but also in their presynaptic modulation of their synaptic output. Most GABAergic terminals are known to contain GABAB receptors (GABABR), but their role in presynaptic inhibition and surface expression have not been comparatively characterized in the two BC types. To address this, we performed whole-cell recordings from CCK and PV BCs and postsynaptic pyramidal cells (PCs), as well as freeze-fracture replica-based quantitative immunogold electron microscopy of their synapses in the rat hippocampal CA1 area. Our results demonstrate that while both CCK and PV BCs contain functional presynaptic GABABRs, their modulatory effects and relative abundance are markedly different at these two synapses: GABA release is dramatically inhibited by the agonist baclofen at CCK BC synapses, whereas a moderate reduction in inhibitory transmission is observed at PV BC synapses. Furthermore, GABABR activation has divergent effects on synaptic dynamics: paired-pulse depression (PPD) is enhanced at CCK BC synapses, but abolished at PV BC synapses. Consistent with the quantitative differences in presynaptic inhibition, virtually all CCK BC terminals were found to contain GABABRs at high densities, but only 40% of PV BC axon terminals contain GABABRs at detectable levels. These findings add to an increasing list of differences between these two interneuron types, with implications for their network functions.


Introduction
The perisomatic domain of cortical pyramidal cells (PC) is under GABAergic control of a specific subset of inhibitory interneurons called basket cells (BCs) (Buhl et al. 1994;Freund and Buzsáki 1996;Klausberger and Somogyi 2008). BCs are heterogeneous, comprising two main types with distinct neurochemical and anatomical characteristics, which are believed to underlie their divergent roles in functional networks. The two BC types are primarily distinguished on the basis of their selective expression of either the neuropeptide cholecystokinin (CCK) or the calcium-binding protein parvalbumin (PV) (Nunzi et al. 1985;Kosaka et al. 1987;Freund and Katona 2007;Lee and Soltesz 2011;Booker et al. 2013Booker et al. , 2017. Furthermore, CCK and PV BCs also differ in their physiological properties, most prominently their regularversus fast-spiking discharge pattern, respectively. Action potential-evoked release of GABA from inhibitory terminals, including those of BCs, is dependent on reliable activation of high voltage-gated calcium channels (VGCCs) leading to vesicular GABA release, which is tightly controlled by presynaptic neuromodulatory receptors for multiple transmitter systems (Wu and Saggau 1995;Lüscher et al. 1997;Lee and Soltesz 2011). The two BC types show striking differences in presynaptic mechanisms as well: in axon terminals of CCK BCs Ca v 2.2 (N-type) VGCCs are localized to the active zone, while in PV BCs Ca v 2.1 (P/Q-type) channels are present (Hefft and Jonas 2005;Lee and Soltesz 2011). Moreover, the compliment of presynaptic neuromodulatory receptors also diverges: CCK BCs possess high levels of endocannabinoid 1 receptors (CB1Rs), whilst PV BCs possess muscarinic acetylcholine 2 (M2Rs) and l-opioid receptors (Hajos et al. 1997;Katona et al. 1999;Fukudome et al. 2004;Neu et al. 2007;Armstrong and Soltesz 2012;Lenkey et al. 2015). Despite these differences, both CCK and PV BCs appear to express GABA B autoreceptors at their output synapses (Lee and Soltesz 2011;Booker et al. 2013).
GABA B Rs are heterodimeric receptors composed of the GABA B1 and GABA B2 subunits (Kaupmann et al. 1998). At presynaptic membranes GABA B Rs comprise the Sushidomain containing GABA B1a splice variant (Guetg et al. 2009) and negatively couple to VGCCs via the G-protein (G i/o ) bc subunit. Accordingly, activation of presynaptic GABA B Rs by the canonical agonist baclofen has been shown to result in a reduction of unitary IPSC amplitudes at both CCK and PV BC synapses (Hefft et al. 2002;Lee and Soltesz 2011;Booker et al. 2013;Jappy et al. 2016). However, direct comparison is difficult due to large differences in the concentration of agonist applied and recording conditions in the individual studies. Furthermore, the subcellular localization and density of the GABA B Rs in synaptic and extrasynaptic membrane segments of CCK and PV BC axon terminals has not yet been investigated.
In this study therefore we comparatively assessed the function and localization of presynaptic GABA B Rs in identified BC axon terminals, using whole-cell patch-clamp recordings and quantitative SDS-digested freeze-fracture replica (SDS-FRL) immunoelectron microscopy.

Characterization of presynaptic GABA B R function
Following breakthrough into whole-cell configuration, intrinsic properties of recorded neurons were characterized in current-clamp mode from resting V M . Neurons were physiologically identified based on their response to hyperto depolarizing current steps (-500 to 500 pA, in 100 pA steps, 500 ms duration). CCK BCs were identified as responding to depolarizing stimulus with low to moderate frequency trains of action potentials (APs); PV BCs were identified as firing high-frequency non-accommodating trains ([100 Hz) of APs. PCs showed characteristic large amplitude and broad APs, and lower-frequency accommodating trains of APs. Neurons were rejected from further analysis if V M was more depolarized than -50 mV, APs did not overshoot 0 mV, or initial series resistance exceeded 30 MX or changed by [20% over the course of the experiment.
Minimally stimulated monosynaptic inhibitory post-synaptic currents (IPSC) in PCs were elicited in the presence of NBQX (10 lM) and APV (50 lM) by extracellular stimulation with a glass monopolar electrode (a patch pipette filled with 2 M NaCl) placed in str. pyramidale at a distance of 10-20 lm from the soma. IPSC were evoked every 10 s with stimulus amplitudes of 1-20 V, minimally exceeding threshold to evoke a response. A 5-min control baseline was collected after which 1 lM WIN-55,212 (WIN) was briefly applied to the bath (2 min) to test presynaptic CB1R sensitivity. Following washout of WIN and recovery of IPSC amplitude to control levels (*5 min) baclofen (10 lM) was applied to the bath and steady state inhibition was observed at 3-5 min following wash in. Following baclofen application, CGP-55,845 (CGP, 5 lM) was applied to the bath confirming that the observed responses were GABA B R mediated. In recordings where the IPSC was not sensitive to WIN application, we subsequently applied the selective M2R agonist arecaidine but-2-ynyl ester tosylate (ABET; 10 lM) (Chiang et al. 2010) to the bath to confirm that these WIN-insensitive IPSCs were mediated by M2R-containing putative PV axon terminals. In all recordings the whole-cell holding current (I WC ) in the CA1 PC was monitored to confirm successful activation of postsynaptic GABA B Rs, given that presynaptic receptors show higher baclofen affinity (Dugladze et al. 2013). Given the optimal voltage-clamp conditions for the recording of perisomatic synaptic events meditated by BCs, we did not block post-synaptic Kir3 potassium channels in any of these experiments.
To confirm that presynaptic GABA B R mediated effects were specific to identified BCs, we performed paired recordings between a CCK or PV BC and a CA1 PC. Conditions were identical to those of single cell recordings albeit with a lower 0.5 mM EGTA in the intracellular solution (compensated by a higher K-gluconate content of 110 mM). Following characterization of intrinsic properties (see above) of pre-and postsynaptic neurons, 10 trains of 10 APs (1-2 nA, 1 ms, 20 Hz) were elicited in the BC while recording the CA1 PC in voltage-clamp (as above). If a unitary synaptic connection was observed, seen as shortlatency (\4 ms) IPSCs in the PC following presynaptic APs in averages of ten traces, it was assumed that there was a direct synaptic connection. If synaptic connectivity was not observed in the CA1 PC, the recording was abandoned, an outside-out patch formed, and a neighboring CA1 PC recorded. Once a synaptic connection was found we recorded[50 traces under paired-pulse protocols (2 stimuli with 50 ms interval), following which baclofen (10 lM) was bath applied for 5 min, after which CGP (5 lM) was bath applied. The IPSC amplitude was measured from the preceding baseline as an average over a 0.4 ms window corresponding to the peak within 10 ms from the start of the AP. Onset latency of the IPSC was assessed as the duration from the time of maximal rise rate of the presynaptic AP to the onset of the IPSC. Mean IPSC responses shown and measured from at least 30 traces.
It should be noted that we have included reanalyzed data (7 pairs) from a previous publication (Booker et al. 2013) where we had measured and reported only the unitary IPSC amplitudes for PV BC connections. We have assessed the presynaptic properties of those synaptic connections together with a set of newly recorded pairs (4 pairs) under identical conditions.
Neurons were imaged using a confocal laser scanning microscope (FluoView 1000, Olympus) with a 209 (N.A. 0.75) lens and z-axis stacks of images (4 Mpixel resolution, 1 lm z-steps) collected to allow identification of somatodendritic and axonal arborizations. Immunoreactivity of recorded neurons was tested at the soma with a siliconimmersion 609 (N.A. 1.3) objective lens. Neurons were reconstructed offline from digitally stitched image stacks, which were segmented and reconstructed using semi-automatic analysis software (Simple Neurite Tracer plug-in for the ImageJ/FIJI software package; http://fiji.org; Longair et al. 2011).

Antibody characterization for immunogold labeling
The specificity of the affinity-purified antibodies to CB1R (Fukudome et al. 2004), M2R (Alomone Labs, Israel), GABA B1 (B17, Kulik et al. 2002;B62, Kulik et al. 2006) were assessed with immunoblot analysis of crude membrane fractions derived from either mouse brains (CB1R and M2R) or rat brains (B17 and B62). Single protein bands were detected with M2R and CB1R antibodies, whereas GABA B1 antibodies gave rise to two immunoreactive products with estimated molecular masses of 130 and 100 kDa, as expected from the two splice variants of the GABA B1 subunit. In addition the band for M2R was not detected in M2R knock-out brains (data not shown). All antibodies used targeted the intracellular epitopes of the transmembrane proteins and therefore resulted in specific labelling on the protoplasmic face (Pface) of the replicas.

SDS-digested freeze-fracture replica immunogold labeling (SDS-FRL)
SDS-FRL was performed as previously described (Althof et al. 2015). Briefly, 60-day-old Wistar rats (2) were perfusion fixed with 2% PFA and 15% saturated picric acid in 0.1 M PB. Transverse 100 lm hippocampal sections were cut and cryoprotected O/N in PB with 30% glycerol at 4°C. Blocks containing the str. pyramidale and radiatum of the CA1 area were blocked-out and then frozen by a high-pressure freezing machine (HPM100, Leica, Austria). Frozen samples were freeze-fractured at -140°C and coated by deposition with carbon (5 nm), platinum (2 nm) and carbon (18 nm) in a freeze-fracture replica device (BAF 060, BAL-TEC, Lichtenstein). Replicas were digested for 18 h at 80°C in a solution containing 2.5% SDS and 20% sucrose in 15 mM Tris buffered saline (TBS) at pH 8.3. Replicas were then rinsed in washing buffer [0.05% bovine serum albumin, (BSA) and 0.1% Tween 20, in TBS] and blocked in a solution containing 5% BSA and 0.1% Tween 20 for 1 h at RT. Subsequently, replicas were incubated in one of the following mixtures of primary antibodies: (a) CB1R [Guinea pig (Gp); 1:3000, Fukudome et al. 2004] and GABA B1 [B17, Rabbit (Rb); 5 lg/ml, Kulik et al. 2002], (b) M2R (Rb; 10 lg/ml, Alomone Labs, Israel) and GABA B1 [B62, (Gp); 10 lg/ml, Kulik et al. 2006] made up in 50 mM TBS containing 1% BSA and 0.1% Tween 20 O/N at 4°C. Replicas were then rinsed in TBS, blocked for 30 min and reacted with a mixture of either (a) gold-coupled 10 nm goat anti-guinea pig IgG and 5 nm goat anti-rabbit IgG or (b) 10 nm goat anti-rabbit IgG and 5 nm goat anti-guinea pig IgG secondary antibodies (1:30; BioCell Research Laboratories, Cardiff, UK) diluted in TBS containing 1% BSA and 0.1% Tween 20 for 3 h at RT or at 4°C O/N. Replicas were subsequently rinsed in TBS, ultrapure water and then mounted on 100-mesh grids. Quantitative analysis was performed at 6000-10,0009 magnification under an electron microscope (Philips CM100). Images were taken with a CCD camera (Orius SC600, GATAN, Inc.) and acquired with GATAN software. Presynaptic axon terminals were collected in str. pyramidale and identified as perisomatic due to their direct apposition to a somatic or very proximal dendritic membranes. Putative CA1 PC spiny dendrites were collected from str. radiatum as positive control. Immunogold particle density was calculated by counting the number of particles on the 2D surface of identified axon terminals and dendrites using the FIJI software package.

Statistical analysis
Statistical analysis was performed with Graphpad Prism 3.0 (GraphPad Software, CA, USA). Group data were compared with one-way ANOVA tests, combined with Bonferroni post-test to establish group differences. Analysis of unpaired and paired data was performed with Mann-Whitney or Wilcoxon matched-pairs tests, respectively. Data are shown as mean ± SEM throughout. Statistical significance was assumed if P \ 0.05.

Results
Presynaptic GABA B Rs differentially inhibit CB1R-and M2R-sensitive perisomatic inhibitory transmission To assess the role of presynaptic GABA B Rs in pharmacologically identified putative CCK versus PV BC axon terminals, we first recorded monosynaptic IPSCs evoked by minimal stimulation from CA1 PCs (15 cells) in the presence of ionotropic glutamate receptor blockers (NBQX: 10 lM and APV: 50 lM). All CA1 PCs were morphologically identified, as previously described (Degro et al. 2015), and had a soma in str. pyramidale with a single large caliber apical dendrite projecting through str. radiatum, with all dendrites decorated with dendritic spines. Stimulation via electrodes placed in the str. pyramidale, in close proximity to the recorded PCs (10-20 lm distal) consistently produced fast IPSCs. To identify CB1R-sensitive responses, corresponding to putative CCK BC afferents, brief (2 min) bath application of the highly potent receptor agonist WIN-55212 (WIN; 1 lM) was performed. In ten cells, this brief application of WIN reduced the IPSC amplitude to 58.3 ± 15.7% (P = 0.002, Wilcoxon matched-pairs test, Fig. 1a) confirming the presence of CB1R on the stimulated axons in this subset of recordings. Following complete washout of WIN and full recovery of the IPSCs to control levels, the selective GABA B R agonist baclofen was bath applied (10 lM, 5 min) resulting in a large decrease in IPSC amplitude to 29.9 ± 5.5% of control (P = 0.0001, Wilcoxon matchedpairs test, Fig. 1a, c, e). In addition, there was an outward I WC induced in the recorded PCs (95.6 ± 12.0 pA; P = 0.001) consistent with GABA B R-mediated activation of a postsynaptic potassium current (Booker et al. 2013(Booker et al. , 2017Fig. 1a, middle). Both the IPSC amplitude and I WC were fully recovered to control levels with the subsequent removal of baclofen and application of the potent and selective GABA B R antagonist CGP-55845 (CGP; 5 lM).
In five recorded PCs, WIN application had no inhibitory effect on the IPSC amplitude (Fig. 1b). In these WIN-insensitive afferents, bath application of baclofen reduced the IPSC amplitude only to 61.5 ± 11.3% of control (P = 0.04), indicating an approximately twofold weaker inhibition than for WIN-sensitive afferents (P = 0.024, Mann-Whitney test, Fig. 1b, d, e). Similar to the recordings for CB1R-sensitive afferents, baclofen application induced a postsynaptic outward I WC of 83.7 ± 18.3 pA in the postsynaptic PCs (compared to control: P = 0.02; CB1R-insensitive versus -sensitive: P = 0.75, Mann-Whitney test). To confirm that the CB1R-insensitive afferents corresponded to M2R-containing putative PV BC axons, we subsequently bath applied the selective agonist ABET (10 lM, 5 min). Indeed, in response to ABET application, IPSC amplitude was reduced in all five recorded PCs to 51.2 ± 14.4% of control. Neither WIN nor ABET application resulted in significant change in holding current of the recorded neurons, indicating a predominantly presynaptic locus of activity under these recording conditions (WIN: P = 0.89; ABET: P = 0.66; Wilcoxon signed rank tests). These data indicate that both populations of pharmacologically identified CA1 BC axon terminals contain functional GABA B Rs, but are differentially inhibited by these autoreceptors.

Strong presynaptic GABA B R-mediated inhibition at the synaptic output of CCK BCs
Our previous recordings of synaptically coupled PV BC and PC pairs were consistent with the results of the minimal stimulation experiments and indicated an approximately 40% (of control) reduction in the unitary IPSCs during application of baclofen (10 lM; Booker et al. 2013, Fig. 7). To directly assess the effects of GABA B R activation on GABA release from CCK BCs, we performed paired recordings between identified CCK BCs and CA1 PCs (Fig. 2a). Interneurons were anatomically identified following overnight fixation, visualization of biocytin and immunolabeling with CCK antibodies. CCK BCs were identified based on the presence of dense axon in and near the str. pyramidale, forming ring like ''baskets'' around the somata of putative CA1 PCs (Fig. 2).
In dual recordings from eight synaptically coupled CCK BC to CA1 PC pairs, APs (29 with 50 ms interval) evoked in the CCK BCs were reliably followed by short-latency unitary IPSCs in the CA1 PCs (Fig. 2b, upper), with an average amplitude of 29.2 ± 9.3 pA of first IPSC and failure rate of 35.1 ± 8.9%. The second IPSCs showed marked paired-pulse depression (PPD) with a ratio of 0.47 ± 0.05. Consistent with a relatively low temporal precision of release at these synapses (Hefft and Jonas 2005), the coefficient of variation (CV) of the onset latency was 0.52 ± 0.12 (Fig. 2c-h). Bath application of 10 lM baclofen significantly reduced the average amplitude of the first IPSCs to 17.6 ± 6.1% to 4.8 ± 2.5 pA (P = 0.02, repeated measures 1-way ANOVA). In fact, the reduction here was comparable to that observed in the minimal stimulation experiments for WIN-sensitive afferents (P = 0.19, Mann-Whitney test). Application of CGP (5 lM) resulted in a full recovery of the IPSC amplitude (Fig. 2b, c). Consistent with a presynaptic locus of GABA B R activity, baclofen application resulted in an increase in failure rate to 76.6 ± 8.6% (P \ 0.0001, repeated measures 1-way ANOVA) and enhanced PPD to 0.24 ± 0.08 (P = 0.02, repeated measures 1-way ANOVA). Furthermore, the CV of onset latency decreased to 0.32 ± 0.12 (P = 0.03, repeated measures 1-way ANOVA).
Given the apparent differences in the degree of presynaptic inhibition between CCK and PV BC-mediated transmission, we next performed paired recordings from PV BCs (4 cells) to further characterize these effects. PV BCs were all identified on the basis of PV immunoreactivity and axon collaterals localized to str. pyramidale ( Supplementary Fig. 1) as previously described (Booker  Fig. 1D). In all these PV BCs to CA1 PC paired recordings we observed a baclofen-produced inhibition of IPSC amplitude to 45.0 ± 5.1% of control levels, which recovered to control levels after the application of CGP (Fig. 2g). Thus, the presynaptic inhibitory effects in these pairs were markedly smaller than that observed for CCK BCs, but corresponded well to our previously published data (seven cells; 57.1 ± 0.4% of control; P = 0.49; Booker et al. 2013). As we had previously only analyzed the IPSC amplitude difference, we now reanalyzed these previously published cells for PPD and failure rate and included them with these four additional paired recordings (Fig. 2g-j). For the pooled PV BC data set (11 cells), GABA B R-mediated presynaptic inhibition reduced IPSC amplitude to 51.2 ± 6.0% of control levels (Fig. 2j), not different from that observed in M2R-sensitive synapses in Fig. 1 (P = 0.51, Mann-Whitney test), but reflecting a substantially lower inhibition than in CCK-BC pairs (P = 0.004, Mann-Whitney test). Analysis of IPSC failure rate showed that in baseline conditions PV BCs had less failures of transmission (22.4 ± 6.8%), which was not significantly different from that of CCK BCs (P = 0.39). However, during baclofen application PV BCs showed only a very modest increase in IPSC failure rate to 33.8 ± 9.4%, which was not different from control (P = 0.16, Wilcoxon signed rank test, Fig. 2h). Indeed, the failure rate of PV BC in baclofen was substantially lower than that of CCK BCs (P = 0.006, Mann-Whitney). Furthermore, as expected from the high release probability of PV BCs, they showed PPD (0.71 ± 0.05), which was removed by application of baclofen to 1.02 ± 0.1 (P = 0.02, 1-way ANOVA with multiple comparisons). Both the effect on IPSC amplitude and failure rate in PV BCs was reversed by subsequent application of CGP. Interestingly, unlike CCK BCs, CGP in PV BCs application resulted in a stronger PPD than under control conditions (0.54 ± 0.6, P = 0.025) suggesting that under control conditions there might be a level of tonic effect of GABA B R on presynaptic short-term plasticity. These data show that despite both populations of BCs express presynaptic GABA B Rs, the ability of these receptors to inhibit GABA release onto CA1 PC somata is substantially different. Furthermore, CCK and PV BC synapses onto CA1 PCs show divergence in the effects of GABA B R activation on short-term plasticity and the failures of transmission.
In a subset of paired recordings, we found that the presynaptic CCK IN had axon collaterals in str. radiatum and oriens, targeting dendritic domains of CA1 PCs, consistent with the Schaffer-Collateral Associated (SCA) type (Booker et al. 2017). It is known that CCK SCA cells share many of the same presynaptic properties of CCK BCs, including CB1R sensitivity (Lee and Soltesz 2011), so we next asked if CCK SCA cells also showed presynaptic GABA B R-mediated inhibition. In the six synaptically coupled SCA to CA1 PC pairs (Fig. 3) we observed IPSCs in response to presynaptic APs with an average amplitude of 9.1 ± 7.0 pA (first IPSCs), which was reduced to 3.5 ± 1.5% of control following baclofen application (Fig. 3b, c, f), similar to that observed in CCK BCs (P = 0.13, Mann-Whitney test). IPSC amplitudes recovered to close to control levels with application of CGP (P = 0.06, Wilcoxon signed rank test). CCK SCA cells had a failure rate of 44.1 ± 13.2%, which increased to 88.8 ± 6.4% upon baclofen application (P = 0.03, Wilcoxon signed rank test, Fig. 3d). They showed a PPD of 0.56 ± 0.1, similar to that of CCK BCs (P = 0.28, Mann-Whitney test), but unlike CCK BCs, in SCA cells the PPD changed to facilitation with a ratio of 1.33 ± 0.32 upon application of baclofen (P = 0.02, Mann-Whitney test, Fig. 3e). However, in three SCA cells paired-pulse ratio could not be measured following baclofen application due to complete inhibition of the IPSC. Together, these data suggest that dendritic inhibitory CCK SCA cells possess similar presynaptic GABA B R inhibition to their BC counterparts, with potential divergence in its effect on short-term plasticity.

Different surface densities of presynaptic GABA B Rs at perisomatic synapses
To determine whether the differences in GABA B R-mediated presynaptic inhibition of BC terminals are due to differential distribution pattern and/or density of functional receptors we performed quantitative SDS-FRL electron microscopy. Immunolabeling for the two marker receptors, CB1R (Fig. 4a, b) and M2R (Fig. 4c, d), showed characteristic labeling of putative presynaptic axon terminals in str. pyramidale of CA1. Labeling for CB1R was consistently stronger on a subset of perisomatic terminals, while M2R-positive terminals showed generally weaker labeling. When GABA B1 subunit localization was examined on CB1R-positive terminals it was found that nearly all terminals contained immunogold particles for GABA B1 (95.2% of terminals, 67 terminals) (Fig. 4a, b), whereas only 40.6% of M2R-positive terminals (64 terminals) showed immunoreactivity for GABA B1 (Fig. 4c, d), which was significantly lower than on CB1R-positive terminals (P \ 0.0001, Fisher exact test, Fig. 4e). We then asked whether there was also a difference in the density of the GABA B1 receptor subunit on terminals. In a subset of CB1R-positive terminals, which all showed GABA B1 labeling (23 terminals), GABA B1 surface density was high at 20.6 ± 1.3 particles/lm 2 . This density was similarly high to that of CA1 PC dendrites in str. radiatum (17.0 ± 1.8 particles/lm 2 , 7 dendrites; P = 0.23, 1-way ANOVA with multiple comparisons). Interestingly, on M2R-containing axon terminals we observed a much weaker GABA B1 labeling density at 4.5 ± 0.5 particles/ lm 2 (38 axon terminals), which was substantially and significantly lower than that of both CB1R-containing terminals and CA1 PC dendrites (both P \ 0.0001, 1-way ANOVA with multiple comparisons, Fig. 4f).
These data confirm that there are IN type-specific differences in the presynaptic GABA B R expression in inhibitory axon terminals, as measured by the density of the constitutive subunit GABA B1 , between putative CCK and PV BC. Interestingly, this data shows that not only is the number of terminals that containing GABA B1 different, but also the receptor density on those terminals is different, plausibly explaining the differences we observed in GABA B R-mediated presynaptic inhibition between CB1Rand M2R-containing synapses.

Discussion
Presynaptic neuromodulation is a central neuronal property, which defines cell types as well as our understanding of the role those cells play in local networks. Our results demonstrate that while both CCK and PV BCs contain presynaptic GABA B Rs at their axon terminals, acting as autoreceptors, their surface density and inhibitory potential at these two synapses show marked differences. We provide compelling evidence that virtually all CCK BC terminals contain GABA B Rs at high levels and GABA release is dramatically inhibited by their activation. In contrast, only 40% of PV BC axon terminals express GABA B Rs at detectable levels, leading to only a moderate reduction in inhibitory transmission at these output synapses upon GABA B R activation. These findings add to an increasing list of molecular and functional differences between these two major IN types with implications for their network function.

Presynaptic GABA B Rs as auto-and heteroreceptors
Presynaptic control of synaptic transmission by GABA B Rs has been established for various transmitter systems, including glutamatergic and GABAergic transmission (Bowery et al. 1981;Connors et al. 1988;Deisz and Prince 1989). Accordingly, the robust inhibitory effect of GABA B Rs as presynaptic autoreceptors at GABAergic synapses has been well documented in diverse brain regions including the neocortex and hippocampus (Misgeld et al. 1984;Peet and McLennan 1986;Oláh et al. 2009). However, while divergence in the sensitivity to GABA B Rmediated heterosynaptic inhibition at major glutamatergic pathways have been long recognized in the forebrain (Lanthorn and Cotman 1981;Ault and Nadler 1983;Molyneaux and Hasselmo 2002), little is known of GABA B Rs in diverse INs types, let alone the differential nature of this. Prior in situ and immunocytochemical studies suggested that GABA B Rs are present at high levels in somata of some, but not all interneuron types (Fritschy et al. 1999;Sloviter et al. 1999). In fact, for postsynaptic compartments our recent results demonstrated that morphological subtypes of PV and CCK INs show substantial differences in GABA B R surface density, as well as in the postsynaptic currents (Booker et al. 2013(Booker et al. , 2017. Furthermore, at the presynaptic level, dendritic-and perisomatictargeting PV IN subtypes were found to be sensitive to GABA B R-mediated inhibition at their output synapses, albeit differentially (Booker et al. 2013). For perisomatic synapses formed by putative BCs, earlier electrophysiological investigations using minimal stimulation indicated that GABA B Rs modulate GABA release at most, but not all b Fig. 2 Unitary IPSCs produced in CA1 PCs by identified CCK and PV BCs show differential presynaptic GABA B R-mediated inhibition. a Reconstruction of a synaptically coupled CCK BC (soma and dendrites in black, axon in red) and CA1 PC (soma and dendrites in blue) pair. Inset, top, voltage responses elicited in the CCK BCs and the CA1 PCs to hyper-to depolarizing current steps (-500 to 500 pA, 50 pA steps, 500 ms duration). Note the regular-spiking phenotype of the CCK BCs. Inset, right, immunolabeling for CCK (in green pseudocolor) in the biocytin-filled soma of the BC (in gray). Pyr., stratum pyramidale; Ori., stratum oriens; Rad., stratum radiatum; half being negative for GABA B Rs. In good agreement, our data obtained by minimally stimulation and paired recordings converges, demonstrating that GABA B R activation markedly reduces transmission at CCK output synapses (Neu et al. 2007;Lee and Soltesz 2011;Booker et al. 2013;Jappy et al. 2016), but has an approximately twofold weaker effect at PV synapses at the concentration of baclofen used in our experiments. In contrast to the findings of Lambert and Wilson (1993), however, all putative PV axons and identified unitary connections were affected to a similar degree and no baclofen insensitive connections were found. These electrophysiological findings suggest that the molecular dichotomy observed at the level of the individual axon terminals in the electron microscopic level (see Fig. 4) affects the multiple synaptic contacts formed between individual pre-and postsynaptic neurons (Buhl et al. 1994;Bartos et al. 2001  Insets in a-c show differences in particle size, with small 5 nm gold particles for GABA B1 (arrows) and the larger 10 nm particles for the interneuron-specific marker CB1R and M2R, respectively. e Summary bar chart of the proportion of GABA B1 on double-labeled CB1R-or M2R-containing axon terminals. The total number of examined CB1R-and M2R-immunopositive terminals is indicated in parenthesis. transmission is reduced by 70-100% in the presence of baclofen in a concentration range of 1-100 lM (Neu et al. 2007;Lee and Soltesz 2011;Jappy et al. 2016). This strong presynaptic autocrine modulation converges with the similarly strong retrograde control, exerted by the endocannabinoids system via CB1R (Katona et al. 1999;Neu et al. 2007;Lee and Soltesz 2011;Jappy et al. 2016). The effect of GABA B R activation at PV BC synapses, in contrast, is substantially smaller; however, our data is not consistent with a general lack of GABA B R at these synapses as assumed earlier (Freund and Katona 2007).
In addition to the changes in the amplitude and reliability of ISPCs, baclofen has differential effects on the dynamic behavior of CCK and PV synapses: PPD observed at PV output synapses was fully abolished by the drug and recovered by subsequent application of the selective antagonist CGP. This finding is in good agreement with previous observations that GABA B R activation eliminates short-term depression at both excitatory and inhibitory synapses in the hippocampus and can thereby maintain synaptic transmission during repeated activation albeit at reduced amplitudes (Lei and McBain 2003). Surprisingly, PPD observed at CCK BC synapses was not reduced but enhanced in the presence of baclofen, which also recovered to control levels in CGP. These results are incompatible with a tonic control of perisomatic inhibitory transmission or a direct involvement of these receptors in frequency dependent behavior at the time scales investigated. Rather, our observations plausibly reflect indirect effects due to the modulation of release machinery and the probability of GABA release (Dittman et al. 2000;Lei and McBain 2003).
Finally, perisomatic CCK synapses differ in their longterm plasticity: while theta-burst stimulation induces typically LTP at inhibitory synapses, the output synapses of CCK interneurons display LTD. Although the induction of LTD requires activation of postsynaptic GABA B Rs with an inhibitory effect on adenyl cyclase and PKA activity (Jappy et al. 2016), it remains open what is the contribution of the highly expressed presynaptic receptors.

Functional implications of the dichotomy of perisomatic synapses
While the two BC types converge on the perisomatic compartment of their postsynaptic targets, their distinct properties and activity patterns in vitro and in vivo reflect distinct functions in the network. A differential contribution of the two types to network activity patterns is well established (Klausberger and Somogyi 2008): CCK BCs integrate incoming inputs over longer time scales and discharge and release GABA with low temporal precision, mediating a tonic form of inhibition to their targets (Hefft and Jonas 2005;Daw et al. 2009;Cea-del Rio et al. 2011). Activation of presynaptic GABA B Rs could therefore substantially suppress CCK BC output, resulting in dynamic disinhibition of principal cells, enhanced network activity and facilitation of long-term potentiation among principal cells (Davies et al. 1991;Mott and Lewis 1991). In contrast, PV BCs are proposed to act as fast-signaling devices and contribute to the generation of fast oscillations at gamma and ripple frequencies (Jonas et al. 2004;Bartos et al. 2007;Sohal et al. 2009). Enhanced GABA B R-mediated presynaptic inhibition at PV BC synapses during network activity (Scanziani 2000) may dampen, but will not abolish the entrainment of target cell populations, due to its moderate effect on IPSC amplitudes and the removal of PPD.
Beyond these physiological network functions, evidence is mounting for significant role for presynaptic GABA B Rs in the aetiology of neuropathological disorders, notably epilepsy and intellectual disabilities (i.e. Fragile X syndrome). In epilepsy, while a selective decrease of CCK BC terminals have been observed in the CA1 (Wyeth et al. 2010), GABA B Rs expression at CCK axon terminals appears to be increased leading to stronger disinhibitory effects of principal cells potentially promoting seizure generation and the progression of epilepsy (Dugladze et al. 2013). Interestingly, a similar hypersensitivity of GABA B R-mediated presynaptic inhibition of feedforward inhibition was observed in a mouse model of Fragile X syndrome (Wahlstrom-Helgren and Klyachko 2015). While, the IN subtype involved was not identified in these experiments, neurogliaform and CCK interneurons are likely candidates due to their high expression of the receptors. The increased presynaptic sensitivity leads to an altered excitation/inhibition balance in the network, which is thought to be an underlying circuit mechanism of cognitive impairments observed both in patients and animal models (Silverman et al. 2015;Wahlstrom-Helgren and Klyachko 2015).
In conclusion, we provide the first direct comparative and quantitative data showing that presynaptic GABA B Rs exert differential inhibition at CCK and PV BC synapses, resulting from a higher surface expression of the receptor at all CCK BC terminals and a lower protein density present in a subpopulation of boutons of PV BCs. These data provide further evidence that distinguishes these two major IN types with respect to their presynaptic molecular composition and physiological functions.
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://crea tivecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.