Role for Endogenous BDNF in Endocannabinoid-Mediated Long-Term Depression at Neocortical Inhibitory Synapses,,

Endocannabinoids and neurotrophins, particularly brain-derived neurotrophic factor (BDNF), are potent neuromodulators that play critical roles in many behavioral and physiological processes. Disruption of either BDNF or endocannabinoid signaling is associated with an overlapping set of neurologic and psychiatric diseases, and both systems are currently major targets for the development of novel therapeutics, particularly in relation to depression, anxiety, autism, and schizophrenia.


Introduction
Endogenous cannabinoids (endocannabinoids, eCBs) are important regulators of synaptic function in the nervous system. In particular, endocannabinoids acutely modulate inhibitory and excitatory transmission throughout the forebrain, and mediate several forms of short-term plasticity at GABAergic and glutamatergic synapses, including depolarization-induced suppression of inhibition (DSI) and excitation (DSE). During DSI or DSE, endocannabinoids are released from postsynaptic sites in response to depolarization-induced calcium influx and act retrogradely via presynaptic type-1 cannabinoid receptors (CB1Rs) to suppress transmitter release. Endocannabinoids also mediate specific types of long-term depression at excitatory and inhibitory synapses (for review, see Chevaleyre et al., 2006;Mackie, 2008;Cachope, 2012). These forms of plasticity are dependent on CB1Rs, which are highly expressed throughout the forebrain, and are activated by both arachidonoylethanolamide (anandamide) and 2-arachidonoylglycerol (2-AG), the two bestcharacterized endogenous ligands of CB1R. It has been shown that the on-demand synthesis and release of endocannabinoids in the hippocampus, neocortex, cerebellum, and other areas can be stimulated by depolarizationinduced calcium influx, as well as by activation of phospholipase-C ␤ (PLC␤) triggered by G q -protein coupled receptors, particularly metabotropic glutamate receptors (mGluRs) (for review, see Hashimotodani et al., 2007;Castillo et al., 2012).
Recently it has been shown that brain-derived neurotrophic factor (BDNF) can also induce endocannabinoid release, part of a growing body of evidence for interactions between endocannabinoid signaling and BDNF. For example, there is evidence of eCB-BDNF interactions in visual cortex , hippocampus (Khaspekov et al., 2004;Roloff et al., 2010), and cerebellum (Maison et al., 2009). In support of this, there is strong colocalization of trkB receptors and CB1R throughout the forebrain. Within the neocortex, trkB is predominantly localized to layers 2/3 and 5 (Cabelli et al., 1996;Fryer et al., 1996;Miller and Pitts, 2000), and the highest levels of CB1R in the neocortex are also found in layer 2/3 (Matsuda et al., 1993;Tsou et al., 1998;Egertová et al., 2003). Recent studies in the neocortex have shown that endocannabinoid synthesis and release can be rapidly mobilized by BDNFϪtrkB signaling. In neocortical layer 2/3, acute application of BDNF rapidly suppresses GABAergic transmission via release of endocannabinoids from the postsynaptic pyramidal cell, which act in a retrograde manner to suppress presynaptic transmitter release (Lemtiri-Chlieh and Levine, 2010). This effect of BDNF is initiated by postsynaptic trkB signaling, requires downstream PLC signaling, and is independent of mGluR activation (Zhao and Levine, 2014).
Endogenous BDNF has been shown to play a critical role in LTP at excitatory synapses, but its role at inhibitory synapses and its potential role in endocannabinoidmediated synaptic plasticity are less clear. Because exogenous BDNF can trigger endocannabinoid mobilization that suppresses GABA release at cortical synapses, we explored whether endogenous BDNF plays a role in longterm depression at cortical inhibitory synapses. In particular, we hypothesized that stimulation-induced release of BDNF, acting through trkB and downstream diacylglycerol lipase (DGL) signaling, can trigger endocannabinoid release to cause long-term depression at cortical inhibitory synapses.
Cells were voltage clamped at Ϫ70 mV during recording. All electrical events were filtered at 2.9 kHz and digitized at Ͼ6 kHz using a HEKA EPC9 amplifier and ITC-16 digitizer (HEKA Elektronik). Series resistance (R s ) was compensated up to 40% at 100 s lag. Input resistance (R i ) was monitored with 10 mV (50 ms) hyperpolarizing voltage steps at the end of each sweep. Cells were rejected from analysis if R i fell below 50 M⍀ or the holding current dropped by 150 pA during the course of an experiment.

Chemicals
Unless otherwise stated, all drugs were from Tocris Biosciences and were delivered by bath perfusion. Drugs were first prepared as concentrated stock solution in solvents and stored at Ϫ20°C. The stock solutions of DNQX, AM251, ANA-12, K252a, SR 141716A, and tetrahydrolipstatin (THL) were dissolved in 100% dimethyl sulfoxide (DMSO). The final concentration of DMSO did not exceed 0.1%, which by itself had no effect on evoked inhibitory synaptic currents (Zhao and Levine, 2014) and had no effect on evoked excitatory currents (102.5 Ϯ 0.3% of baseline, n ϭ 3). CPP, E4CPG, and BDNF (Pep-roTech) were dissolved in 18 M⍀ water. Drug stock solutions were diluted into aCSF on the day of recording to the final concentrations. Bovine serum albumin (BSA; Sigma-Aldrich) was added to the BDNF solution at a concentration of 0.1 g/l to reduce nonspecific binding. BSA by itself had no effect on evoked IPSC amplitude (112.7 Ϯ 13.2% of baseline, n ϭ 3). Preincubation time for all drugs was 15-30 min prior to experiment.

Data analysis
Off-line analysis was carried out using Clampfit 10 (Molecular Devices) and Prism 6 (GraphPad Software). Statistical comparisons were made between average amplitudes of baseline responses and 27-35 min postin-duction using two-tailed Student's paired t test unless otherwise stated. p Ͻ 0.05 was taken as a statistically significant effect. In individual examples, sweeps of evoked responses were averaged traces of four consecutive evoked IPSCs around corresponding time windows. Group data are reported as mean Ϯ SEM.

Strong theta frequency burst stimulation induces eCB-dependent iLTD
Evoked IPSCs were recorded from layer 2/3 pyramidal neurons in response to intralaminar stimulation (0.05 Hz). In the absence of high frequency repetitive stimulation, these responses were stable over time (101.7 Ϯ 2.6% of baseline, n ϭ 4, after 40 min of stimulation). Using thetafrequency burst stimulation (TBS), we examined longterm depression at inhibitory synapses (iLTD) at inhibitory synapses onto layer 2/3 cortical pyramidal neurons. Stimulation consisted of seven trains of TBS (7ϫ TBS) delivered with a 5 s intertrain interval. Each TBS train contained 10 bursts (200 ms interburst interval), each burst consisted of five stimuli at 100 Hz. This protocol induced a stable long-lasting suppression of inhibitory transmission (Fig. 1A). As shown in Figure 1B This TBS-induced iLTD was independent of ionotropic glutamate receptor signaling because it was evoked in the presence of NMDA and non-NMDA receptor antagonists (see Materials and Methods, above). To confirm that these receptors were effectively blocked during the repetitive stimulation used to induce iLTD, we recorded evoked EPSCs in layer 5 pyramidal neurons. Intralaminar stimulation evoked a dual component EPSC that was almost completely abolished in the presence of CPP and DNQX (12.6 Ϯ 1.6% of predrug baseline, n ϭ 4). Following 7ϫ TBS stimulation, there was no change in the amplitude of the residual evoked response (1 min post-TBS; 94.5 Ϯ 4.7% of pre-TBS baseline, n ϭ 4), indicating effective antagonism of these receptors during repetitive stimulation.
We next examined whether this iLTD requires CB1R signaling. Because CB1Rs are predominantly expressed on presynaptic terminals, we hypothesized that there would be a change in the paired-pulse ratio before and after iLTD induction. We found that the paired-pulse ratio was significantly increased after iLTD induction (137.3 Ϯ 13.3% of baseline, p ϭ 0.025, n ϭ 7; Fig. 1C). We then examined the effect of 7ϫ TBS during bath application of the selective CB1R antagonist AM251 (5 M) or SR141716A (5 M). Identical results were obtained using these structurally similar drugs, so the results were combined. As shown in the group data in Figures 1D and 2C, 7ϫ TBS in the presence of either AM251 or SR141716A failed to induce iLTD at these synapses (94.1 Ϯ 5.4% of baseline, p ϭ 0.3201, n ϭ 9; baseline, 1.17 Ϯ 0.1 nA; 35 min-post, 1.10 Ϯ 0.1 nA). Application of CB1R antagonists had no effect on basal eIPSC amplitude of layer 2/3 pyramidal neurons under similar conditions (Trettel and Levine, 2002). These results suggest that TBS induces eCB-dependent iLTD at inhibitory synapses onto layer 2/3 pyramidal neurons.
iLTD in layer 2/3 of somatosensory cortex is independent of mGluR signaling Because many forms of eCB-mediated LTD require mGluR signaling (for review, see Chevaleyre et al., 2006;, we examined the effects of 7ϫ TBS during bath application of the group I/group II mGluR antagonist E4CPG (500 M). Interestingly, blocking mGluR signaling did not prevent 7ϫ TBS from inducing iLTD at layer 2/3 inhibitory synapses ( Fig. 2A). At 35 min postinduction, eIPSC amplitude was significantly reduced compared to baseline (Fig. 2B,C, 53.88 Ϯ 8.7% of baseline, p ϭ 0.0001, n ϭ 6). The amount of suppression was comparable to that caused by 7X TBS alone (Fig. 2C, p ϭ 0.1447, unpaired t test). Previously it has been shown that this concentration of E4CPG is sufficient to block the effect of an mGluR agonist on synaptic transmission in layer 2/3 (Zhao and Levine, 2014). As an additional control to confirm the efficacy of E4CPG during repetitive stimulation, we examined its effects on LTD at layer 5 excitatory synapses. The same stimulation protocol (7ϫ TBS) induced significant LTD at these synapses (60.1 Ϯ 4.5% of baseline, p ϭ 0.0003, n ϭ 7), and this suppression was blocked by E4CPG (n ϭ 6, unpaired t test, p ϭ 0.0312).
It has been previously shown that BDNF induces eCB mobilization via DGL signaling (Lemtiri-Chlieh and Levine, 2010). We thus examined whether DGL plays a role in iLTD in layer 2/3. As seen in Figure 3, C and D, in the presence of the DGL inhibitor THL (5 M), iLTD was completely abolished (115.6 Ϯ 18.4% of baseline, p ϭ 0.5057, n ϭ 6; baseline, 1.13 Ϯ 0.2 nA; 35 min-post, 1.24 Ϯ 0.2 nA). Taken together, these findings suggest that endogenous BDNFϪtrkB signaling and downstream activation of DGL, but not mGluR activation, is necessary for eCBmediated iLTD at inhibitory synapses in layer 2/3 pyramidal neurons.

Exogenous BDNF facilitates iLTD induction
We next examined whether exogenous BDNF could facilitate iLTD induction by a subthreshold stimulus train. In contrast to the stable suppression caused by 7ϫ TBS, three trains of TBS (3ϫ TBS) did not induce significant iLTD, as shown in the group data in Figures 4, A and B, and 5B (102.6 Ϯ 10.9% of baseline, p ϭ 0.8943, n ϭ 7; baseline, 0.79 Ϯ 0.2 nA, 35 min-post, 0.78 Ϯ 0.2 nA). In fact, only one of seven cells showed suppression with 3ϫ TBS (Fig. 4B). We then examined the effects of BDNF alone on inhibitory synaptic transmission. It has previously been shown that BDNF at a concentration of 20 ng/ml (0.8 nM) significantly reduces inhibitory synaptic transmission (Lemtiri-Chlieh and Levine, 2010; Zhao and Levine, 2014). To determine a concentration of BDNF that does not by itself influence synaptic transmission, we examined the effect of 5 min bath-applied BDNF on eIPSC amplitude. Bath application of 0.2 nM (5 ng/ml) BDNF for 5 min caused a suppression of eIPSC amplitude in four of five cells, although this did not reach statistical significance (83.24 Ϯ 10.5% of baseline, p ϭ 0.2644, n ϭ 5). At a concentration of 0.1 nM (2.5 ng/ml), BDNF did not cause suppression (104.4 Ϯ 12.5% of baseline, n ϭ 5), nor did BDNF at 0.05 nM (104.5 Ϯ 8.8% of baseline; p ϭ 0.7086, n ϭ 6), as shown in Figure 4C. In addition, 0.05 nM BDNF for 5 min had no effect on eIPSC amplitude measured 35 min post-BDNF application (97.92 Ϯ 5.8% of baseline, p ϭ 0.8288, n ϭ 6), indicating that brief application of 0.05 nM BDNF does not induce iLTD. Interestingly, however, the combination of 3ϫ TBS with 0.05 nM BDNF (applied for 5 min around the time of TBS) produced significant iLTD (Figs. 4D,E, 5B, 70.95 Ϯ 9.1% of baseline, p ϭ 0.0305, n ϭ 7), with six of seven cells showing suppression. The average amount of suppression produced by BDNF ϩ 3ϫ TBS was not significantly different than that produced by 7ϫ TBS alone (Fig. 5B,  p ϭ 0.7978, unpaired t test). These data suggest that a subthreshold concentration of BDNF can synergize with weak TBS stimulation to induce iLTD at layer 2/3 inhibitory synapses.

Discussion
In the present study, we investigated the interaction between BDNF and eCB signaling in iLTD. We found that seven trains of TBS induce stable eCB-mediated iLTD at cortical layer 2/3 pyramidal neurons, as inhibiting CB1R activation diminished this suppression. Surprisingly, this form of iLTD is independent of mGluR activation, in contrast to the most prevalent forms of eCB-mediated LTD (Maejima et al., 2001;Chevaleyre and Castillo, 2003;Bender et al., 2006;Lafourcade et al., 2007;Chiu et al., 2010;Jiang et al., 2010). In contrast, BDNF, which has also been shown to mobilize eCBs (Lemtiri-Chlieh and Levine, 2010; Zhao and Levine, 2014) is required for this iLTD, as inhibiting trkB tyrosine kinase activity or blocking trkB receptor activation prevented TBS-induced iLTD. Consistent with previous findings that DGL is required for BDNF-induced eCB mobilization (Lemtiri-Chlieh and Levine, 2010), the present results indicate that endogenous BDNF regulates eCB-mediated iLTD via DGL, as the DGL inhibitor THL diminished iLTD. Furthermore, we also found that a low concentration of exogenous BDNF, which has no effect on its own, synergizes with three trains of TBS to induce CB1R-dependent iLTD.
The seven-train TBS protocol induced a form of iLTD that requires BDNF/trkB signaling. The source of endogenous BDNF in this form of iLTD is not clear, but potential sources would include release from presynaptic terminals or dendritic release from neighboring neurons. Previous studies have identified BDNF-containing vesicles in presynaptic glutamatergic terminals (Hartmann et al., 2001;Brigadski et al., 2005;Dean et al., 2012;Dieni et al., 2012) and stimulation has been shown to induce the release of endogenous BDNF from glutamatergic terminals (Aloyz    Hartmann et al., 2001;Kohara et al., 2001;Dean et al., 2012). BDNF release from inhibitory terminals is not likely as BDNF mRNA is not expressed in interneurons (Ernfors et al., 1990;Gorba and Wahle, 1999).
BDNF has also been localized to vesicles in postsynaptic dendrites (Aoki et al., 2000;Avwenagha et al., 2006) and has been shown to be released from dendrites (Hartmann et al., 2001;Kuczewski et al., 2008;Matsuda et al., 2009;Dean et al., 2012). Dendritic release of BDNF can be triggered by the following: (1) Ca 2ϩ influx through ionotropic glutamate receptors or voltage-gated channels (Hartmann et al., 2001); (2) activation of Group I mGluR receptors, which subsequently triggers IP3-mediated Ca 2ϩ release from intracellular calcium store (Canossa et al., 2001); and (3) Ca 2ϩ -dependent Ca 2ϩ release from intracellular stores by ryanodine receptors (Balkowiec and Katz, 2002). However, in the present studies, iLTD was induced in the presence of both ionotropic and metabotropic glutamate receptor antagonists, and postsynaptic cells were voltage-clamped at a hyperpolarized membrane potential, arguing against postsynaptic release via these pathways. Astrocytes can also release BDNF, however this also requires mGluR signaling (Jean et al., 2008). Thus presynaptic release is the most likely source of the endogenous BDNF mediating the observed iLTD.
The location of the trkB receptors that are activated by endogenously-released BDNF in the present study is most likely on dendrites of postsynaptic pyramidal neurons in layer 2/3. There are high levels of trkB expression in postsynaptic dendrites in hippocampus and in cortex, particularly in layer 2/3 (Drake et al., 1999;Aoki et al., 2000), although there is also evidence for trkB expression at presynaptic sites (Xu et al., 2000;Kohara et al., 2001;Inagaki et al., 2008). Importantly, however, postsynaptic trkB receptor activation has been shown to be specifically required for exogenous BDNF-induced eCB release (Lemtiri-Chlieh and Levine, 2010). Postsynaptic calcium has also been shown to be required for BDNF-induced eCB release (Lemtiri-Chlieh and Levine, 2010); thus, release of calcium from intracellular stores may also be involved in the eCB-mediated LTD reported in the present studies.
Activation of postsynaptic trkB receptors induces the synthesis and release of eCBs that act retrogradely on presynaptic CB1Rs to trigger iLTD. The endogenous CB1R ligand that mediates iLTD in the present study is not completely clear, but is likely to be 2-AG. Activation of PLC␥ and PLC␤, downstream effectors of BDNF/trkB and mGluR, respectively, leads to cleavage of phosphatidylinositol 4,5-bisphosphate into the second messengers inositoltrisphosphate and DAG (Hashimotodani et al., 2007;Minichiello, 2009), and DAG is converted to the endogenous cannabinoid 2-AG by the enzyme DGL. In the present study, we found that inhibiting DGL blocked eCB-mediated iLTD, and exogenous BDNF-induced eCB release is also blocked by inhibiting DGL (Lemtiri-Chlieh and Levine, 2010). In addition, 2-AG has been shown to be the downstream effector of mGluR-dependent eCB release in many studies (for review, see Mackie, 2008; Katona and Freund, 2012; Ohno-Shosaku and Kano, 2014). However, a contributing role for anandamide cannot be ruled out, as several studies have shown a link between mGluR signaling and anandamide generation (Azad et al., 2004;Puente et al., 2011;Edwards et al., 2012;Huang and Woolley, 2012), presumably through PLC signaling (Liu et al., 2008).
In the current study, eCB-mediated iLTD was induced using TBS protocols that are similar to protocols that induce LTP at excitatory synapses, in which endogenous BDNF also plays a critical role (Aicardi et al., 2004;Abidin et al., 2006;Lu et al., 2010). For example, Lu and colleagues (2010) have shown that BDNF facilitates LTP by suppressing GABAergic inhibition and enhancing pyramidal neuron excitability. Thus, endogenous BDNF may enhance LTP and/or postsynaptic excitability at least in part by triggering eCB release which then induces long-term depression of inhibitory synapses, in addition to the direct effects of BDNF, as well as eCBs, at excitatory synapses. Interestingly, endogenous BDNF has also been shown to play a role in the induction of GABAergic LTP in the developing rat hippocampus (Gubellini et al., 2005). In order to fully understand the net effect of a given stimulation protocol on cortical circuits, it will be important to consider simultaneous changes at both excitatory and inhibitory synapses.
The present study focused on the role of BDNF in long-term eCB-mediated plasticity at inhibitory synapses. BDNF-induced endocannabinoid release may also play a role in eCB-dependent short-term synaptic plasticity at inhibitory synapses (DSI) or excitatory synapses (DSE). In fact, exogenous BDNF has been shown to enhance DSE via activation of the immediate-early gene homer1a (Roloff et al., 2010). It will therefore be interesting to explore whether endogenous BDNF contributes to eCB mobilization during DSI and DSE and whether it has synergistic effects with depolarization-induced eCB release.
There is increasing evidence for multiple types of interactions between BDNF and the eCB system. BDNF, for example, regulates CB1R expression as well as its sensitivity to endogenous agonists in cultured cerebellar granule neurons (CGNs) (Maison et al., 2009). De Chiara and colleagues (2010 found that in striatum, BDNF selectively antagonizes CB1R receptor function at inhibitory synapses in a cholesterol metabolism-mediated mechanism It has also been found that the neuroprotective effects of eCBs, particularly in depression and epileptic seizures, are mediated by changes in BDNF expression (Marsicano et al., 2003;Khaspekov et al., 2004;Aguado et al., 2007;Aso et al., 2008;Vinod et al., 2012). During interneuron differentiation, eCBs have additive effects with BDNF in regulating interneuron migra-  (Berghuis et al., 2005). A full understanding of the physiological roles of each of these systems will require a greater mechanistic understanding of their synaptic interactions. The current study, describing a novel form of eCB-mediated LTD at inhibitory synapses that requires endogenous BDNF/trk signaling, may contribute to this understanding.