Postsynaptic diacylglycerol lipase α mediates retrograde endocannabinoid suppression of inhibition in mouse prefrontal cortex

Non-technical summary

In multiple brain regions, endogenous cannabinoids suppress inhibitory synaptic transmission; however, the biochemical/molecular pathways for endocannabinoid synthesis are poorly understood. Endocannabinoid signalling may be crucial for microcircuit function in the prefrontal cortex (PFC), a cortical region involved in complex behaviours. However, endocannabinoid signalling remains largely unexplored in the PFC. Using enzymatic inhibitors, we show that modulation of inhibitory synaptic transmission in PFC neurons is mediated by the endocannabinoid 2-arachidonoylglycerol synthesized postsynaptically. Interestingly, diacylglycerol lipase (DAGL), the 2-arachidonoylglycerol synthesis enzyme, has two isoforms: DAGLα and DAGLβ. Studying PFC neurons from DAGLα^−/−, DAGLβ^−/− and wild-type mice, we show that only DAGLα is involved in the suppression of inhibitory transmission in the PFC.

Abstract

Abstract

Depolarization-induced suppression of inhibition (DSI) is a prevailing form of endocannabinoid signalling. However, several discrepancies have arisen regarding the roles played by the two major brain endocannabinoids, 2-arachidonoylglycerol (2-AG) and anandamide, in mediating DSI. Here we studied endocannabinoid signalling in the prefrontal cortex (PFC), where several components of the endocannabinoid system have been identified, but endocannabinoid signalling remains largely unexplored. In voltage clamp recordings from mouse PFC pyramidal neurons, depolarizing steps significantly suppressed IPSCs induced by application of the cholinergic agonist carbachol. DSI in PFC neurons was abolished by extra- or intracellular application of tetrahydrolipstatin (THL), an inhibitor of the 2-AG synthesis enzyme diacylglycerol lipase (DAGL). Moreover, DSI was enhanced by inhibiting 2-AG degradation, but was unaffected by inhibiting anandamide degradation. THL, however, may affect other enzymes of lipid metabolism and does not selectively target the α (DAGLα) or β (DAGLβ) isoforms of DAGL. Therefore, we studied DSI in the PFC of DAGLα^−/− and DAGLβ^−/− mice generated via insertional mutagenesis by gene-trapping with retroviral vectors. Gene trapping strongly reduced DAGLα or DAGLβ mRNA levels in a locus-specific manner. In DAGLα^−/− mice cortical levels of 2-AG were significantly decreased and DSI was completely abolished, whereas DAGLβ deficiency did not alter cortical 2-AG levels or DSI. Importantly, cortical levels of anandamide were not significantly affected in DAGLα^−/− or DAGLβ^−/− mice. The chronic decrease of 2-AG levels in DAGLα^−/− mice did not globally alter inhibitory transmission or the response of cannabinoid-sensitive synapses to cannabinoid receptor stimulation, although it altered some intrinsic membrane properties. Finally, we found that repetitive action potential firing of PFC pyramidal neurons suppressed synaptic inhibition in a DAGLα-dependent manner. These results show that DSI is a prominent form of endocannabinoid signalling in PFC circuits. Moreover, the close agreement between our pharmacological and genetic studies indicates that 2-AG synthesized by postsynaptic DAGLα mediates DSI in PFC neurons.

Introduction

The two major endocannabinoids found in brain, 2-arachidonoylglycerol (2-AG) and anandamide, are effective agonists of the primary brain cannabinoid receptor, the cannabinoid receptor 1 (CB1R) (Kano et al. 2009). Endocannabinoids are released rapidly via non-vesicular mechanisms following stimulation of their synthesis, and retrogradely inhibit neurotransmitter release via presynaptic CB1Rs (Wilson & Nicoll, 2001). Among other stimuli, endocannabinoid synthesis is activated by postsynaptic depolarization, which produces a CB1R-dependent retrograde suppression of GABA release. Endocannabinoid-mediated depolarization-induced suppression of inhibition (DSI) is synapse-specific and short-lasting, decaying within seconds (Katona et al. 1999; Nyiri et al. 2005; Glickfeld & Scanziani, 2006; Galarreta et al. 2008).

In the hippocampus, cerebellum and striatum, multiple properties of DSI were previously studied, including the contribution of 2-AG and anandamide. Some studies using endocannabinoid synthesis inhibitors suggested that DSI requires 2-AG without a significant anandamide contribution (Kano et al. 2009). However, several experimental discrepancies have arisen (Di Marzo, 2011). For example, in some studies, 2-AG synthesis inhibitors failed to affect DSI even though they blocked other forms of endocannabinoid-mediated synaptic modulation (Chevaleyre & Castillo, 2003; Safo & Regehr, 2005; Min et al. 2010b). Such discrepancies may be due to the use of different enzymatic inhibitors, some of which could have non-specific effects, the use of cell cultures versus acute slices, or regional differences in the roles of 2-AG versus anandamide in DSI.

In the prefrontal cortex (PFC), a neocortical region with substantially different circuitry to hippocampus or cerebellum, endocannabinoid-mediated signalling remains largely unexplored, although the PFC contains molecular components of the endocannabinoid system, including CB1Rs (Eggan & Lewis, 2007; Lafourcade et al. 2007; Burston et al. 2010; Chiu et al. 2010), fatty acid amide hydrolase and monoacylglycerol lipase, the anandamide- and 2-AG-degrading enzymes, respectively, and diacylglycerol lipase (DAGL), the key enzyme for 2-AG synthesis (Hansson et al. 2007; Lafourcade et al. 2007; Volk et al. 2010).

Genes for two DAGL isoforms with very similar enzymatic activity, DAGLα and DAGLβ, have been cloned (Bisogno et al. 2003). Interestingly, in hippocampal (Katona et al. 2006; Yoshida et al. 2006; Ludanyi et al. 2011) and prefrontal (Lafourcade et al. 2007) pyramidal cells, DAGLα is highly expressed in dendritic spines, where it can retrogradely modulate glutamate release (Katona & Freund, 2008). However, DAGLα in dendritic spines is ultrastructurally distant from most GABA synapses and is thus unlikely to contribute to DSI, since the lipid-soluble nature of 2-AG severely limits its diffusion in the extracellular space.

Notably, DAGLα was reported to be undetectable at CB1R-containing GABA synapses in PFC (Lafourcade et al. 2007), suggesting that 2-AG synthesized by DAGLα mostly or exclusively modulates glutamate synapses. In fact, in PFC endocannabinoids modulate excitatory synaptic transmission (Lafourcade et al. 2007, 2011), but whether endocannabinoids produce DSI in PFC circuits has not been reported. The very low levels of DAGLα near CB1R-containing GABA synapses in PFC (Lafourcade et al. 2007) suggest that DSI in PFC might depend on DAGLβ, although the ultrastructural localization of DAGLβ has not been determined. Alternatively, DSI in PFC may depend on anandamide instead of 2-AG. Anandamide mediates some forms of suppression of inhibition onto pyramidal neurons (Lourenço et al. 2011) and, interestingly, plays a critical role in PFC circuit function. For example, application of exogenous anandamide to the PFC or manipulating the effects of endogenous anandamide in the PFC, both produce significant behavioural effects (Rubino et al. 2008; Aguiar et al. 2009; Lisboa et al. 2010).

Two recent studies showed that DSI is abolished in hippocampus and cerebellum of DAGLα-deficient mice, suggesting that DSI requires 2-AG synthesized via DAGLα (Gao et al. 2010; Tanimura et al. 2010). However, in such studies DAGLα deficiency decreased both 2-AG and anandamide levels in total brain tissue (Gao et al. 2010) or specifically in hippocampus and cerebellum (Tanimura et al. 2010). Thus, such studies could not definitively exclude a contribution of anandamide to DSI (Min et al. 2010a; Di Marzo, 2011). Moreover, those studies did not examine endocannabinoid signalling in neocortical circuits. To study endocannabinoid-mediated modulation of inhibitory synaptic transmission in PFC, we used electrophysiology, pharmacology and genetically modified mice. We show that in PFC pyramidal neurons DSI occurs independently of anandamide signalling or DAGLβ activity, but requires 2-AG synthesized by DAGLα localized in postsynaptic pyramidal cells.

Methods

Brain slice preparation

Brain slices were prepared from the frontal cortex of mice deeply anaesthetized with isoflurane and decapitated following procedures in accordance with NIH guidelines and approved by the University of Pittsburgh's Institutional Animal Care and Use Committee. The authors have read, and the experiments comply with the policies and regulations of The Journal of Physiology, given by Drummond (2009). In most experiments we used male C57BL6 mice (Charles River), 1–4 months of age. In the experiments conducted with tissue from DAGL-deficient mice and their wild-type littermates, we used 5- to 6-week-old male mice. The brain was quickly removed and immersed in ice-cold, choline-based artificial cerebrospinal fluid (ACSF) containing (mm): 110 choline chloride, 2.5 KCl, 7 MgCl₂, 0.5 CaCl₂, 25 NaHCO₃, 1.25 NaH₂PO₄, 11.6 sodium ascorbate, 3.1 sodium pyruvate and 25 d-glucose, pH 7.3–7.4, and continuously bubbled with 95% O₂ and 5% CO₂. The frontal cortex was sectioned into 300 μm-thick slices in the coronal plane, using a vibrating microtome (VT1000S, Leica Microsystems). Slices were immediately incubated for 30–60 min in a chamber maintained at 36°C and filled with standard ACSF (mm): 125 NaCl, 2.5 KCl, 1 MgCl₂, 2 CaCl₂, 25 NaHCO₃, 1.25 Na₂HPO₄ and 10 glucose, pH 7.3–7.4, and continuously bubbled with 95% O₂ and 5% CO₂. Following incubation, brain slices were stabilized at room temperature in the same solution for at least 50 min. Slices were then transferred to an electrophysiology recording chamber where the submerged slices were superfused at a flow rate of 3–4 ml min⁻¹ (to record carbachol-induced spontaneous (s)IPSCs or sIPSPs, see Hájos & Mody, 2009) or 2 ml min⁻¹ (to record agatoxin-resistant evoked IPSCs) with ACSF gassed with 95% O₂ and 5% CO₂ at 30–32°C. AMPA receptor-mediated transmission was blocked routinely by adding 10 μm 6-cyano-7-nitroquinoxalene-2,3-dione (CNQX). CNQX, carbachol, tetrahydrolipstatin (THL), N-arachidonoyl maleimide (NAM), WIN55212-2, and AM251 were purchased from Sigma Chemical Company, St Louis, MO, USA or Tocris Bioscience, Ellisville, MO, USA. SR141716A was provided by Bristol-Myers Squibb. Biocytin was purchased from Sigma or Invitrogen, Carlsbad, CA, USA. ω-Conotoxin-GVIA and ω-agatoxin-IVA were from Bachem, Torrance, CA, USA. URB597 was from Cayman Chemicals, Ann Arbor, MI, USA.

Electrophysiology

Patch-pipette, tight-seal whole-cell recordings were obtained from pyramidal cells in layers 2–3 of the infralimbic (IL), prelimbic (PL) or anterior cingulate (AC) regions of the mouse medial frontal cortex, here collectively referred to as mouse PFC (Fig. 1A). Cells were visualized using Olympus or Zeiss microscopes equipped with infrared illumination and differential interference contrast videomicroscopy. IPSCs were recorded using pipettes pulled from borosilicate glass having a resistance of 2–5 MΩ when filled with the following solution (mm): 120 CsCl, 2 KCl, 10 Hepes, 0.2 EGTA, 4 MgATP, 0.3 NaGTP and 0.5% biocytin, pH adjusted to 7.2–7.4 using CsOH. IPSPs were recorded using pipettes filled with (mm): 120 KCl, 10 NaCl, 10 Hepes, 0.2 EGTA, 4 MgATP, 0.3 NaGTP, 14 phosphocreatine and 0.5% biocytin. The intrinsic excitability of PFC pyramidal cells (Fig. 9) was determined using pipettes filled with a solution containing (mm): 120 potassium gluconate, 10 NaCl, 10 Hepes, 0.2 EGTA, 4.5 MgATP, 0.3 NaGTP and 14 sodium phosphocreatine. Recordings were obtained using Multiclamp 700A or 700B amplifiers (Axon Instruments). Signals were low-pass filtered at 4 kHz and digitized at 10 kHz using Power 1401 data acquisition interfaces (Cambridge Electronic Design, Cambridge, UK). Data acquisition and analysis were performed using Signal 4 software (Cambridge Electronic Design).

Figure 1. Depolarizing steps do not suppress sIPSCs recorded from layer 2/3 PFC neurons in the absence of carbachol.

A, diagram showing the location of recorded pyramidal neurons in layers 2/3 of the mouse medial PFC. AC: dorsal anterior cingulate cortex, PL: prelimbic cortex, IL: infralimbic cortex, CC: corpus callosum. B, example traces of sIPSCs recorded from layer 2/3 PFC pyramidal neurons. Note that GABA synapses elicited inward IPSCs due to the high chloride concentrations in the pipette recording solution. C, effect of depolarizing steps (–80 to 0 mV for 4 s) on sIPSCs recorded from layer 2/3 pyramidal neurons. The depolarizing steps did not alter the sIPSC frequency or amplitude, suggesting that the sIPSCs were mostly produced by cannabinoid-insensitive synapses. Here and in other figures, the current elicited by the depolarizing steps has been blanked. D, quantification of the effect of depolarizing steps on sIPSCs recorded from PFC neurons. As described in Methods, for each neuron the baseline sIPSCs (sIPSCs recorded prior to the depolarizing steps) were averaged and all IPSCs (baseline and post-depolarization) were normalized relative to the average baseline value. Here and in all other figures, a normalized sIPSC value of 1 means no change in sIPSC amplitude; a value of 0 means 100% sIPSC suppression.

Figure 9. DAGLα deficiency reduced the excitability of PFC pyramidal cells.

A, example traces illustrating the membrane potential response of wild-type and DAGLα^−/− PFC neurons to injection of depolarizing input current steps of increasing amplitude (50, 150 and 250 pA). Note the lower excitability of the PFC neurons from DAGLα^−/− mice revealed as smaller number of spikes for identical level of input current. To test the effects of input current and genotype on the excitability of PFC pyramidal neurons, we recorded from 7 layer 2/3 PFC neurons from wild-type mice and 8 neurons from DAGLα^−/− littermates. Input current was increased in 10 pA increments up to 400 pA above threshold. To monitor recording conditions over time, a small (–50 pA) hyperpolarizing current step was injected prior to the depolarizing stimulation. Note the weaker response to the hyperpolarizing step in neurons from DAGLα^−/− mice. Differences in the excitability of wild-type and DAGLα^−/− neurons were examined using current clamp recordings from n = 7 wild-type and n = 8 DAGLα^−/− neurons. B, firing frequency (prior to onset of spike-frequency adaptation) plotted as a function of input current. Note that PFC pyramidal neurons from DAGLα^−/− mice fired at lower frequency throughout a wide range of input current. Here and in panels C, D and E, input current is normalized relative to each neuron's current threshold, (the minimum level of current necessary to fire action potentials). Two-factor ANOVA showed significant effects of input current (F_(40,533) = 31.032, P < 0.00001) and genotype (F_(1,533) = 207.5, P < 0.000001) on the firing frequency, as well as a significant current–genotype interaction (F_(48,629) = 1.9234, P < 0.0005). C, voltage–current plot illustrating the membrane potential response of wild-type and DAGLα^−/− neurons to injection of hyperpolarizing current steps. Wild-type neurons displayed a steeper relationship, suggesting that DAGLα^−/− neurons had lower membrane resistance. Two-factor ANOVA indicated a significant effect of genotype on the membrane potential response to current injection (input current: F_(4,65) = 43.5, P < 0.0001; genotype: F_(1,65) = 4.9353, P = 0.0298). D, PFC neurons from DAGLα-deficient mice had higher voltage threshold for firing an action potential (first action potential fired in response to a current step) throughout a wide range of input currents. Two-factor ANOVA indicated a significant effect of genotype on firing threshold (F_(1,403) = 170.14, P < 0.0001). E, for input currents up to ∼100 pA above current threshold, DAGLα^−/− PFC neurons showed stronger spike-frequency adaptation, as revealed by a lower adaptation ratio, measured as the ratio between the first and last inter-spike interval (an adaptation ratio value of 1 means no adaptation and values below 1 mean a stronger spike-frequency adaptation). Two-factor ANOVA showed significant effects of input current (F_(48,629) = 14.115, P < 0.00001) and genotype (F_(1,629) = 48.566, P < 0.00001) on the adaptation ratio, as well as a significant current–genotype interaction (F_(48,629) = 1.9234, P < 0.0005).

Voltage clamp

The pipette capacitance was compensated and series resistance was continuously monitored but was not compensated. Only recordings with a stable series resistance of less than 20 MΩ were used for analysis. To record IPSCs, pyramidal cells were held at –80 mV.

Current clamp

Series resistance and pipette capacitance were monitored and cancelled using bridge and capacitance neutralization. Membrane potential was not corrected for liquid junction potential. The membrane potential was maintained near –75 mV with current injection.

Extracellular stimulation

Agatoxin-resistant IPSCs were evoked by focal stimulation using electrodes fabricated with theta-type capillary glass pulled to an open tip diameter of ∼3–5 μm and filled with oxygenated ACSF. Silver wires inserted into the theta glass were connected to a stimulus isolation unit (World Precision Instruments) commanded by TTL pulses. Stimulation at 0.2 Hz (duration = 100 μs, amplitude = 10–100 μA) was delivered after placing the stimulation pipette near the soma of the recorded pyramidal cell.

Morphological analysis

Biocytin-filled neurons were visualized using the Vectastain Elite ABC kit (Vector Laboratories) and their axonal and dendritic trees reconstructed using the Neurolucida Tracing System (Microbrightfield Bioscience) as described previously (Gonzalez-Burgos et al. 2009).

Induction of DSI

In the majority of the experiments, we studied DSI of GABA_A receptor-mediated IPSCs induced in pyramidal cells by application of the cholinergic agonist carbachol. Previous studies showed that carbachol-induced IPSCs display strong DSI, consistent with the idea that carbachol activates cholecystokinin-containing interneurons which furnish cannabinoid-sensitive GABA synapses (however, see Gulyas et al. 2010; Szabóet al. 2010). We used 20 μm carbachol, a concentration that produces stable increases in sIPSC amplitude and frequency in submerged slices (Hájos & Mody, 2009). In PFC pyramidal neurons, carbachol application rapidly increased the IPSC frequency in a manner that was sensitive to blockade of N-type voltage-dependent Ca²⁺ channels (Fig. 2), which control GABA release from cannabinoid-sensitive synapses (Freund & Katona, 2007). In preliminary experiments we determined that, as reported previously (Hájos & Mody, 2009), the amplitude and frequency of carbachol-induced sIPSCs reached steady-state values within 1–5 min after the start of carbachol application (see Fig. 2B). Therefore, the depolarizing commands (–80 to 0 mV, 4 s) were applied to induce DSI starting at 5 min from the beginning of carbachol application. The effects of depolarizing steps on carbachol-induced IPSCs were tested repeating the depolarizing step protocol 3–6 times in each pyramidal neuron. Consistent with recent findings suggesting that some principal neurons receive cholecystokinin-positive DSI-sensitive inputs whereas others do not (Varga et al. 2010), in some layer 2/3 PFC pyramidal neurons DSI of carbachol-induced (evoked) IPSCs (eIPSCs) was absent. The results obtained from cells that were apparently DSI-negative were still included in the data analysis, thus somewhat underestimating the magnitude of IPSC suppression by endocannabinoids in DSI-positive cells.

Figure 2. Endocannabinoid-mediated suppression of sIPSCs recorded in the presence of carbachol.

A, sIPSCs recorded from a PFC pyramidal neuron in control conditions (top panel), after application of 20 μm carbachol (middle panel) and following additional application of the N-type calcium channel blocker ω-conotoxin GVIA (1 μm). Note that ω-conotoxin GVIA application reversed the increase in sIPSC amplitude and frequency produced by carbachol. B, a plot of sIPSC amplitude versus time, showing the increase in sIPSC amplitude by carbachol for the neuron in A. Following the increase in sIPSC amplitude and frequency by carbachol, application of a depolarizing step induced DSI. Once sIPSCs recovered from DSI, ω-conotoxin GVIA was applied, which reversed the effects of carbachol on sIPSC amplitude. C, bar graph summarizing the effects of carbachol application on sIPSC frequency (P < 0.05, Student's t test). D, depolarizing steps (–80 to 0 mV, 4 s) produced a transient suppression of carbachol-induced sIPSCs. Here and in E, the insets show baseline sIPSCs and post-depolarization sIPSCs on an expanded time scale. E, the CB1R antagonist SR141716A (10 μm) abolished suppression of carbachol-induced sIPSC by depolarizing steps. F, left panel: plot of normalized sIPSC amplitude versus time showing the time course of sIPSC suppression by depolarizing steps, in control conditions (vehicle DMSO 0.1%) and in the presence of the CB1R antagonist SR141716A (10 μm); right panel: bar graph summarizing the sIPSC suppression in control conditions versus CB1R blockade with SR141716A (10 μm), P < 0.01 compared with DMSO, Student's t test.

In other experiments (Fig. 9), we examined DSI of IPSCs evoked in PFC pyramidal cells by focal extracellular stimulation (eIPSCs) of perisomatic GABA synapses in the presence of ω-agatoxin-IVA (250 nm). This toxin blocks GABA release from the cannabinoid-insensitive GABA synapses in various brain regions, including those of parvalbumin-containing neurons in PFC (Zaitsev et al. 2007), thus producing agatoxin-resistant eIPSCs that display significant DSI (Wilson et al. 2001). Agatoxin-resistant IPSCs were evoked every 5 s. After five stimuli were delivered to obtain a baseline of IPSC amplitude, a depolarizing command (–80 to 0 mV, 4 s) was applied, followed by 20 additional stimuli delivered to determine recovery from DSI. Such a sequence of five baseline eIPSCs, followed by the voltage command and recovery eIPSCs, was repeated at least 6 times for each neuron and eIPSC amplitude measures were averaged for all repetitions for each cell.

To study DSI induced by action potential firing (Fig. 10), IPSPs were induced in pyramidal cells by application of carbachol. The experiment design was similar to that employed for carbachol-induced IPSCs, except that the depolarizing command was replaced by trains of short suprathreshold current steps evoked at 20 Hz at different durations.

Figure 10. DAGLα deficiency abolishes DSI induced by action potential-mediated stimulation of endocannabinoid synthesis.

A, top panel: trains of supra-threshold current steps elicited at 20 Hz produced a transient suppression of IPSPs (arrow) induced by application of 20 μm carbachol. Note that together with IPSP suppression, the action potential trains produced a prolonged hyperpolarization consistent with the outward current observed following depolarizing steps in voltage clamp recordings (see Fig. 7). GABA synapses elicited depolarizing IPSPs due to the high chloride concentrations in the pipette recording solution (see Methods). The depolarizing IPSPs occasionally elicited action potentials (marked by *). Spikes were truncated for clarity. Bottom panel: sub-threshold depolarizing current steps delivered in an alternating manner with the supra-threshold steps to the same PFC pyramidal cells of the top panel failed to induce IPSP suppression. In A, C and E action potentials were truncated. B, bar graph summarizing the IPSP suppression produced by alternating supra-threshold versus sub-threshold depolarizing current steps. **P < 0.01 compared with supra-threshold steps, t = 3.788, P < 0.002, paired samples Student's t test, n = 10 cells. C, example of the effect of action potential firing (4 s at 20 Hz) on carbachol-induced IPSPs recorded from a wild-type pyramidal neuron. D, time course plots showing the effects of action potential firing at 20 Hz for different durations on carbachol-induced IPSPs recorded from wild-type neurons. The normalized IPSP values at each time point were compared with the baseline value right before the onset of 20 Hz stimulation. The open symbols indicate significant differences (P < 0.05) compared with last baseline value; post hoc comparisons with Dunnett's test, after single-factor repeated measures ANOVA (F and P values for the ANOVA are shown in the graphs). Note the progressive increase in the duration of statistically significant IPSP suppression with increasing duration of the action potential trains at 20 Hz. E, example of the effect of action potential firing (4 s at 20 Hz) on carbachol-induced IPSPs recorded from a DAGLα^−/− pyramidal neuron. F, time course plots showing the effects of action potential firing at 20 Hz for different durations on carbachol-induced IPSPs recorded from DAGLα^−/− neurons. Note the absence of significant IPSP suppression compared with the baseline, independent of stimulus duration. Single-factor repeated measures ANOVA indicated no significant IPSP suppression (F and P values for the ANOVA are shown in the graphs. G, graph summarizing the IPSP suppression versus duration of 20 Hz stimulation in wild-type and DAGLα^−/− neurons. Two-factor ANOVA indicated a statistically significant effect of genotype on IPSP suppression, F_(1,184) = 25.244, P < 0.001. *Significant IPSP suppression compared with baseline IPSP value, P < 0.05 Dunnett's test. #Significantly different compared with IPSP suppression value for the same stimulus duration in wild-type neurons, P < 0.05 Bonferroni or Fisher's LSD post hoc comparison (wild-type neurons, 0.1 s: –0.016 ± 0.042, n = 17; 0.2 s: 0.025 ± 0.045, n = 17; 0.5 s: 0.079 ± 0.044, n = 31; 1.0 s: 0.234 ± 0.060, n = 15; 2.0 s: 0.230 ± 0.044, n = 15; 4.0 s: 0.269 ± 0.064, n = 15; 6.0 s: 0.219 ± 0.069, n = 15; DAGLα^−/− neurons: 0.1 s: –0.118 ± 0.042, n = 6; 0.2 s: –0.101 ± 0.069, n = 6; 0.5 s: 0.005 ± 0.039, n = 17; 1.0 s: 0.076 ± 0.024, n = 11; 2.0 s: –0.025 ± 0.040, n = 11; 4.0 s: 0.095 ± 0.042, n = 11; 6.0 s: 0.006 ± 0.074, n = 11).

Electrophysiology data analysis

Data were analysed using Signal 4 (Cambridge Electronic Design) and Mini Analysis (Synaptosoft). To assess DSI of sIPSCs induced by carbachol, data were acquired in 120 s-long traces, with application of the 4 s depolarizing step (–80 to 0 mV) to stimulate endocannabinoid synthesis starting at 20 s from the trace onset (see Fig. 1C). For each neuron, the depolarizing stimulus protocol was repeated at least 3 times, producing at least three traces of 120 s each. For analysis, each trace (excluding the depolarizing step) was divided into 0.5 s bins, computing the peak current within each bin, using the peak detection option in Signal 4 software, which subtracts from the maximum value within the bin, a mean baseline value for each bin. Within each trace, the peak current values measured from bins prior to the depolarizing step (baseline bins) were averaged and their mean was used to normalize the peak current values for all bins in a trace. For each neuron, the normalized peak current values for each bin (same time window) were averaged across traces, obtaining an average current value. Changes in IPSPs by DSI were assessed similarly, except that traces were divided into 0.1 s bins to minimize the effects of summation of overlapping IPSPs, which have slower decay kinetics than IPSCs, during detection by the peak measurement option in Signal 4 software. Suppression of IPSC or IPSP amplitude by DSI would directly decrease the peak current or peak potential values measured using this method. In fact, previous studies have shown that DSI strongly reduces the sIPSC amplitude at cannabinoid-sensitive synapses studied in isolation in synaptically connected pairs (Glickfeld & Scanziani, 2006; Neu et al. 2007; Galarreta et al. 2008). In addition, other studies have shown that DSI decreases IPSC frequency (Pitler & Alger, 1992b). In some of our experiments, DSI transiently decreased sIPSC frequency to zero (Fig. 2B). Our DSI measurement method also detects decreases in IPSC frequency, because the absence of sIPSCs within a bin is measured as a peak current amplitude equal to zero. In a group of experiments, we compared the magnitude of DSI determined using the method described above versus directly measuring sIPSC frequency. These results are shown in Supplemental Fig. 1 (available online only), demonstrating that measurements of DSI by these two methods produced highly correlated estimations of DSI magnitude. For classification of DSI-positive or DSI-negative cells we calculated the standard deviation for the baseline responses recorded before the depolarizing step by using the values of 1 s bins. A cell was considered to be DSI-positive when the two consecutive normalized IPSC values right after the depolarizing step (values at 1 and 2 s after the depolarization ended) satisfied the following criteria: (1) both normalized IPSC values were below 1, and (2) either IPSC value was less than 1 – 1 × SD of the baseline.

To quantify the currents observed immediately following application of the depolarizing steps to induce DSI (Fig. 7), we divided each trace into 0.5 s bins and used the maximum value option in Signal 4 software to measure the holding current in each bin independently of the minimums represented by sIPSCs. For each trace, the initial holding current value was subtracted from the value measured in each bin.

Figure 7. Depolarizing steps produce an outward current that is reduced by DAGLα deficiency.

A, the holding current for a holding potential of –80 mV was measured every 0.5 s starting 5 s before the depolarizing step (–80 to 0 mV, 4 s) used to induce DSI. The initial holding current value in each trace was subtracted from all values measured in the trace. The plot shows the changes in holding current relative to the initial value averaged across at least three traces per neuron and then across neurons (wild-type, n = 17 cells; DAGLα^−/−, n = 14 cells; DAGLβ^−/−, n = 13 cells). The current evoked by the depolarizing step was blanked. Note the outward current observed in PFC neurons from wild-type and DAGLβ^−/− mice but not in DAGLα^−/− neurons. The plot shows the current values relative to initial baseline. The arrow marks the time of peak for the outward current recorded in neurons from wild-type mice. B, bar graph summarizing the data of peak outward current measured at the time indicated by the arrow in A (time of peak of outward current in wild-type neurons). The hyperpolarizing current was significantly smaller in neurons from DAGLα^−/− mice. Single-factor ANOVA, F_(2,41) = 9.575, P < 0.0005; Dunnett's test versus wild-type, DAGLα^−/−: P < 0.005, DAGLβ^−/−: P = 0.533). C, holding current measured as described in A, in PFC pyramidal neurons from C57BL6 mice in the presence of the vehicle DMSO, the CB1R antagonist SR141716A (10 μm), the DAGL inhibitor THL (bath-applied 10 μm, intracellularly applied 5 μm) or the inhibitor of 2-AG degradation NAM (1 μm). D, bar graph summarizing the effects of the manipulations described in C on the magnitude of the peak hyperpolarizing current. Single-factor ANOVA followed by post hoc tests revealed no significant differences between groups, F_(4,57) = 1.3432, P = 0.265.

Generation of mutant embryonic stem cells and mice

The generation of the OmniBank gene trap library has been described elsewhere (Zambrowicz et al. 1998, 2003). Mutant mice were generated by microinjection of embryonic stem cell clones into host blastocysts using standard methods (Joyner, 2000). The precise genomic insertion site of the retroviral gene trapping vectors was determined by inverse genomic PCR as described (Silver & Keerikatte, 1989).

Mouse genotyping

Genotyping of mice was carried out using a multiplex PCR strategy on genomic DNA as described (Schrick et al. 2006). Oligonucleotide primers (LTR-rev: 5′-ATAAACCCTCTTGCAGTTGCATC-3′, Fwd: 5′-GACATTTTTCAAGCAGTCAACTCCC-3′ and Rev: 5′-AATGAGACATT TCCTCCCTAGCCT-3′) were used to amplify wild-type and mutant DAGLα alleles. Oligonucleotide primers (LTR2: 5′-AAATGGCGTTACTTAAGCTAGCTTGC-3′, Fwd: 5′-AAGGAGGCAA AGACAGCAAAGTGC-3′ and Rev: 5′-TATCCTAGGTGCAGACAGATTGTGC-3′) were used to amplify wild-type and mutant DAGLβ alleles. Interbreeding of DAGLα^+/− or DAGLβ^+/− mice gave rise to progeny in the predicted Mendelian ratios (DAGLα: 881^+/+, 1826^+/− and 923^−/− mice; DAGLβ: 173^+/+, 332^+/− and 169^−/− mice).

RT-PCR analysis

DAGLα-deficient mice

RNA was extracted from brain and lung using a bead homogenizer and RNAzol (Ambion) according to the manufacturer's instructions. Reverse transcription was performed with SuperScript II (Invitrogen) and random hexamer primers, according to the manufacturer's instructions. PCR amplification (95°C, 30 s, 59°C, 45 s, 70°C, 60 s) was performed for 30 cycles using primers complementary to exons 4 and 7 of the DAGLα gene, flanking the insertion site of the vector (A: 5′- AATGACCTCACTGCTAAGAATGTCACC-3′ and B: 5′-GGCACGATGTCGAGGTC ACGGAAAA-3′). Control primers to the mouse α-actin gene (accession number NM_007393) were: 5′-GGCTGGCCGGGACCTGACGGACTACCTCAT-3′ and 5′- GCCTAGAAGCACTTGCGGTGCACGATGGAG-3′. DAGLα RT-PCR products were verified by sequencing.

DAGLβ-deficient mice

RNA was extracted from kidney and spleen using a bead homogenizer and RNAzol (Ambion) according to the manufacturer's instructions. Reverse transcription was performed with SuperScript II (Invitrogen) and random hexamer primers, according to the manufacturer's instructions. PCR amplification (95°C, 30 s, 59°C, 45 s, 70°C, 60 s) was performed for 35 cycles using primers complementary to exons 1 and 3–4 of the DAGLβ gene, flanking the insertion site of the vector (A: 5′-GTCTCTAGCCAGCGACGACTTGGTGTTCCC-3′ and B: 5′-CCAGCTGACAATGACGGTGGCGATGA-3′). Control primers to the mouse α-actin gene (accession number NM_007393) were: 5′-GGCTGGCCGGGACCTGACGGACTACCTCAT-3′ and 5′-GCCTAGAAGCACTTGCGGTGCACGATGGAG-3′. DAGLβ RT-PCR products were verified by sequencing.

Quantitative PCR

Total RNA from brain and liver was isolated from wild-type, heterozygous and homozygous DAGLα and DAGLβ mice using a bead homogenizer (BioSpec Products, Inc., Bartlesville, OK, USA) and TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's instructions. Total RNA (10 μg) was used for reverse transcription (RT) to generate cDNA using a High Capacity cDNA Archive Kit according to the manufacturer's instructions (Applied Biosystems, Foster City, CA, USA). The cDNA was diluted 1:10 and 5 μl was added to 25 μl master mix containing 250 nm probe and 900 nm primers (Applied Biosystems probe/primers Mm00813830_m1 and Mm00523381_m1 for DAGLα and DAGLβ, respectively), and 1x Universal TaqMan mix in a 96-well optical PCR plate. A wild-type brain sample was used to set up a 2-fold serial dilution standard curve. Sample reactions were run in triplicate on an ABI Prism 7900HT Sequence Detection System (Applied Biosystems) with programme conditions at 50°C 2 min, 95°C 10 min, followed by 40 cycles at 95°C 15 s and 60°C 1 min. SDS v2.2 software (Applied Biosystems) was used to analyse the data.

Measurement of brain 2-AG and anandamide

Tissue samples were flash-frozen and stored at –80°C. Frozen samples weighing over 40 mg were transferred to 15 ml falcon tubes and 10% (v/w) of 50 μm D8 2-AG (Cayman Chemicals) in methanol was added. Samples weighing less than 40 mg were transferred to 2 ml Eppendorf tubes and 20% of D8 2-AG was added as an internal standard. Fifteen volumes of ethyl acetate and 5 volumes of 100 mm Tris pH 8.5 (in this order) were added immediately following the addition of the D8 2-AG internal standard. The samples were homogenized using a Polytron (4 times, 5 s at 25,000 r.p.m.) with tubes placed on ice. Falcon tubes were spun in a Sorval Legend RT centrifuge at room temperature (10 min, 900–1000 g) and Eppendorf tubes were spun in an Eppendorf minicentrifuge (5 min, full speed). 7.5 volumes of the top, ethyl acetate fraction were recovered and dried in a speedvac. Samples were resuspended in acetonitrile containing 1.25 μm D8 2-AG (two times the original tissue volume for tissues weighing over 40 mg and 4 times the original volume for tissues weighing less than 40 mg), vortexed, spun down in a minicentrifuge for 5 min and the supernatant was transferred into a 96 well plate and analysed by liquid chromatography/mass spectrometry. In brief: 20 μl samples were injected into a Waters Symmetry C8 column and eluted with 35% A and 65% B from 0 to 0.7 min, 1% A and 99% B from 0.7 to 1.7 min and 35% A and 65% B from 1.7 to 2.5 min (A is 50% methanol, 10 mm ammonium formate, 0.1% formic acid; B is 100% acetonitrile, 10 mm ammonium formate and 0.1% formic acid). Anandamide was eluted at 1.1 min and 2-AG was eluted at 1.15–1.2 min. Anandamide and 2-AG were detected by a Micromass Ultima mass spectrometer using the MRM transition of 379.1 > 287.2 for 2-AG, 363 > 121 for anandamide and 387.2 > 295.2 for D8 2-AG. A standard curve of 125 nm to 8 μm 2-AG along with the appropriate dilution factor were used to determine the 2-AG concentration in the tissue samples. A standard curve for anandamide with a concentration range of 0.5–32 nm was used.

Statistical analysis

The statistical significance of differences between group means was determined using Student's t test or ANOVA followed by post hoc contrasts. Differences between group means were considered significant if P < 0.05. Results are expressed as mean ± standard error of the mean. For statistical comparisons and display of the data, in some cases the magnitude of DSI was expressed as IPSC suppression, where suppression = 1 – normalized IPSC. If normalized IPSC = 1, the suppression is null. Differences in the distribution of data between groups were assessed using χ² and Kolmogorov–Smirnov tests.

Results

DSI requires 2-AG synthesis by postsynaptic DAGL activity

To study modulation of inhibitory synaptic function by endocannabinoids, we obtained tight-seal whole-cell recordings. We targeted pyramidal neurons in layers 2/3 (Fig. 1A), since in the mouse PFC the density of CB1R-positive axons is highest in the superficial layers (Lafourcade et al. 2007). In hippocampus and somatosensory cortex, endocannabinoid modulation of inhibitory transmission is synapse-specific (Kano et al. 2009). For instance, DSI is very strong at synapses from perisomatic-targeting cholecystokinin-positive interneurons (Glickfeld & Scanziani, 2006; Galarreta et al. 2008; Lee et al. 2010), but DSI is very weak or absent in synapses from perisomatic-targeting parvalbumin-positive interneurons (Glickfeld & Scanziani, 2006; Galarreta et al. 2008) or from dendrite-targeting cholecystokinin-positive cells (Lee et al. 2010). As shown in Fig. 1B, layers 2/3 pyramidal cells displayed frequent spontaneous GABA_A receptor-mediated inhibitory postsynaptic currents (sIPSCs), probably produced by spontaneously active GABA neurons or spontaneous GABA release from single nerve terminals. Depolarizing steps (4 s, –80 mV to 0 mV) that typically produce DSI at cannabinoid-sensitive synapses failed to produce detectable sIPSC suppression (Fig. 1C and D), suggesting that such sIPSCs originated mostly in cannabinoid-insensitive GABA synapses.

In hippocampus and somatosensory cortex, endocannabinoid-mediated DSI is commonly studied in the presence of carbachol, a cholinergic receptor agonist that enhances DSI (Pitler & Alger, 1992a; Martin & Alger, 1999; Martin et al. 2001; Wilson et al. 2001; Trettel et al. 2004; Bodor et al. 2005). Carbachol may enhance DSI by stimulating the firing of cholecystokinin-positive interneurons (Kawaguchi, 1997; Cea-del Rio et al. 2010) that elicit cannabinoid-sensitive IPSCs onto pyramidal cells (Glickfeld & Scanziani, 2006; Galarreta et al. 2008; Lee et al. 2010). Furthermore, carbachol may directly stimulate endocannabinoid synthesis, thus potentiating DSI by synergism with membrane depolarization (Kim et al. 2002). As shown previously for hippocampal cells (Hájos & Mody, 2009), in PFC pyramidal neurons carbachol (20 μm) typically increased both the sIPSC amplitude and frequency within 1–5 min of application (Fig. 2A–C, see also Fig. 8 for similar results in a different group of neurons). Furthermore, the effect of carbachol was blocked (2 of 2 cells) by applying1 μmω-conotoxin GVIA (Fig. 2A and B), an inhibitor of the N-type Ca²⁺ channels that trigger endocannabinoid-sensitive action potential-dependent GABA release (Lenz et al. 1998; Wilson et al. 2001; Freund & Katona, 2007).

Figure 8. DAGLα deficiency abolishes DSI but does not generally alter GABA synaptic transmission.

A, left panel: perisomatic IPSCs evoked by focal extracellular stimulation in the presence of 250 nmω-agatoxin-IVA (ω-agatoxin-resistant eIPSCs) recorded from a wild-type PFC pyramidal neuron before (black arrow) and after (grey arrow) application of a depolarizing step (–80 mV to 0 mV, 4 s) to induce DSI. Right panel: ω-agatoxin-resistant eIPSCs recorded from a DAGLα^−/− PFC pyramidal neuron were not significantly altered by depolarizing steps. B, bar graph summarizing the magnitude of DSI of agatoxin-resistant eIPSCs in PFC neurons of wild-type mice, DAGLα^−/− mice or wild-type neurons in the presence of the CB1R antagonist AM251 (10 μm). ANOVA followed by Fisher's LSD post hoc test. C, effect of the CB1R agonist WIN55212-2 (1 μm) on ω-agatoxin-resistant eIPSCs. Note the significant reduction of IPSC strength which was reversed by the CB1R antagonist SR141716A (10 μm). D, plots of ω-agatoxin-resistant eIPSC amplitude versus time for an experiment with a wild-type neuron (top graph) or a DAGLα^−/− neuron (bottom graph). The horizontal bars indicate time of application of the drugs. E, bar graph summarizing the SR141716A-reversible effects of WIN55212-2 on ω-agatoxin-resistant eIPSCs recorded from layer 2/3 PFC pyramidal cells of wild-type versus DAGLα^−/− mice. Student's t test revealed no significant effect of genotype (t = 0.676, P = 0.509, n = 8 cells for each genotype) on IPSC suppression by WIN55212-2. F, examples of sIPSC recordings from PFC pyramidal neurons of wild-type and DAGLα^−/− mice in control conditions and after application of 20 μm carbachol. G, bar graphs summarizing the properties of IPSCs and carbachol-induced IPSCs.Wild-type neurons were recorded in control conditions (n = 17 cells) and in the presence of carbachol (n = 16 cells). Similarly, DAGLα^−/− neurons were recorded in control conditions (n = 15 cells) and in the presence of carbachol (n = 16 cells), whereas n = 12 DAGLβ^−/− cells were recorded in control and carbachol conditions. Top left: sIPSC frequency. Two-factor ANOVA indicated no significant effect of genotype (F_(2,83) = 0.581, P = 0.561) and a significant effect of carbachol (F_(1,83) = 20.2, P < 0.00005) on IPSC frequency. Top right: sIPSC amplitude. Two-factor ANOVA indicated absence of a significant effect of genotype (F_(2,83) = 0.696, P = 0.501) and a significant effect of carbachol (F_(1,83) = 4.919, P < 0.05). Bottom left: sIPSC decay time. Two-factor ANOVA indicated no significant effect of genotype (F_(2,83) = 1.42, P = 0.246) or carbachol (F_(1,83) = 0.61597, P = 0.434). Bottom right: sIPSC rise time. Two-factor ANOVA indicated no significant effect of genotype (F_(1,83) = 0.074, P = 0.928) or carbachol (F_(1,83) = 0.087, P = 0.768).

In contrast to sIPSCs recorded in the absence of carbachol (Fig. 1C and D), depolarizing steps produced a significant suppression of carbachol-induced sIPSCs, which was strongest initially (initial sIPSC suppression, non-DMSO control: 43.02 ± 5.26%, n = 13) and decayed within 5–30 s (Fig. 2D). These results are consistent with the idea that, as in hippocampus, carbachol enables or facilitates endocannabinoid-mediated DSI in PFC. Our method to quantify the magnitude of DSI was based on measurements of the sIPSC amplitude (see Methods), in part because DSI strongly reduces the amplitude of IPSCs purely originated at cannabinoid-sensitive synapses (Glickfeld & Scanziani, 2006; Neu et al. 2007; Galarreta et al. 2008; Lee et al. 2010). Because carbachol also increases the sIPSC frequency and not merely the sIPSC amplitude, an important question is whether in PFC neurons DSI reduces the frequency of carbachol-induced sIPSCs and if so, whether quantification of DSI via changes in sIPSC frequency is more efficient than via changes in sIPSC amplitude. As shown in Supplemental Fig. 1, in a subset of neurons we assessed DSI by measuring changes in either sIPSC amplitude or frequency. We found that DSI transiently reduced sIPSC frequency and that the magnitude of DSI measured in the two different ways was very similar and significantly correlated.

In hippocampus and somatosensory cortex, DSI may involve endocannabinoid-independent mechanisms, particularly in the presence of carbachol (Zilberter, 2000; Harkany et al. 2004; Makara et al. 2007). We therefore tested whether DSI of carbachol-induced sIPSCs in mouse PFC was mediated by endocannabinoid activation of CB1Rs. We first found that DSI was not significantly affected by the vehicle DMSO (Fig. 2E; initial sIPSC suppression: 32.92 ± 5.16%, n = 26, P > 0.05 versus non-DMSO control). In contrast, application of the CB1R antagonist SR141716A (10 μm) abolished DSI (Fig. 2E and F; initial sIPSC suppression: 7.95 ± 5.52%, n = 8, P < 0.01 versus DMSO group), showing that DSI in PFC neurons requires CB1R activation by endocannabinoids. All experiments described below, unless otherwise indicated, examined DSI of carbachol-induced sIPSCs. Although we have not determined the cholinergic receptor subtype mediating the effects of carbachol in PFC slices, previous studies have shown that the enhancement of DSI and endocannabinoid mobilization (Kano et al. 2009) and the stimulation of interneuron firing (Kawaguchi, 1997) are mostly mediated by carbachol activation of muscarinic receptors.

In mouse PFC, the 2-AG synthesis enzyme DAGLα is present at high levels in dendritic spines (Lafourcade et al. 2007), but is undetectable pre- or postsynaptically at CB1R-containing GABA synapses (Lafourcade et al. 2007), thus raising the question of whether DAGLα synthesis of 2-AG mediates DSI in PFC circuits. To determine if 2-AG is involved in producing DSI, we first used tetrahydrolipstatin (THL), an inhibitor of DAGL (Bisogno et al. 2006; Ortar et al. 2008). We found that THL (10 μm) bath-applied to PFC slices strongly inhibited DSI (Fig. 3B and G; initial sIPSC suppression: 7.36 ± 7.13%, n = 28, P < 0.05 versus DMSO group). Next, we tested if DSI is regulated by monoacylglycerol lipase, the key enzyme for 2-AG degradation (Dinh et al. 2002), which was localized in glutamatergic axon terminals (Straiker et al. 2009; Ludanyi et al. 2011). Incubation of slices with N-arachidonoylmaleimide (NAM, 1 μm), a monoacylglycerol lipase inhibitor (Saario et al. 2005; Matuszak et al. 2009), significantly prolonged DSI (Fig. 3C) as revealed by fitting exponential functions to the DSI decay time course (single exponential decay time constant ± SD of the regression, DMSO: 8.6 ± 1.5 s; NAM: 22.1 ± 1.9 s). Moreover, NAM application enhanced the DSI magnitude (Fig. 3C and G; initial IPSC suppression 56 ± 3.52%, n = 26, P < 0.05 versus DMSO group). In contrast, URB597, an inhibitor of anandamide degradation by fatty acid amide hydrolase (Fegley et al. 2005), applied at a concentration (1 μm) that enhances anandamide-mediated modulation of synaptic transmission in hippocampus (Chavez et al. 2010), did not alter DSI in PFC (Fig. 3D and G; exponential decay time constant: 11.6 ± 2.6 s; initial sIPSC suppression: 38.83 ± 6.03%, n = 13, P > 0.05 versus DMSO group).

Figure 3. Effects of inhibitors of endocannabinoid synthesis and degradation on DSI of carbachol-induced sIPSCs.

A, DSI produced in control conditions in the presence of the vehicle DMSO (0.1%). The grey line shows the time course of a single exponential decay function fitted to the DSI decay phase (decay time constant ± SD of the regression: 8.6 ± 1.5 s). B, pre-incubation (30 min) of the slices with the DAGL inhibitor THL (20 μm) followed by continuous THL application (10 μm) to the recording chamber, abolished DSI. C, pre-incubation (30 min) of PFC slices with 1 μm NAM, an inhibitor of 2-AG degradation by monoacyl glycerol lipase, followed by continuous NAM application (1 μm) to the recording chamber, significantly enhanced DSI. The black curve shows a single exponential decay function fitted to the data (time constant: 22.1 ± 1.9 s). D, pre-incubation (30 min) with URB597 (1 μm), an inhibitor of anandamide degradation by fatty acid amide hydrolase, followed by continuous application of URB597 (1 μm) did not affect DSI. The grey line shows an exponential decay function fitted to the data (decay time constant: 11.6 ± 2.6 s). E, time course of IPSC suppression showing the effects on DSI of intracellular application of the vehicle DMSO (i-DMSO). F, time course of sIPSC suppression showing the effects on DSI of intracellular application of 5 μm of the DAGL inhibitor THL (i-THL). G, bar graph summarizing the effects of pharmacological inhibitors of endocannabinoid synthesis and degradation on the magnitude of DSI. Single-factor ANOVA revealed a statistically significant difference (P < 0.05) of DSI magnitude between groups. P < 0.05 compared with DMSO, post hoc comparison with Fisher's LSD test. H, bar graph summarizing the effect of intracellular application of THL (i-THL) versus DMSO (i-DMSO) on the magnitude of DSI. P < 0.05, Student's t test.

Previous studies of hippocampal neurons showed that DSI was inhibited by bath-applied THL but not by intracellularly applied THL (i-THL), leading to the speculation that THL blocks DSI independently of DAGL activity (Edwards et al. 2008; Min et al. 2010b). To compare the effects of bath-applied versus intracellularly applied THL on DSI induced in PFC neurons, we obtained recordings from pyramidal cells with an intracellular solution containing THL (5 μm) or the vehicle DMSO. We found that following 20 min of whole-cell recordings, 5 μm i-THL, a THL concentration 50% lower than that used when bath-applied, strongly inhibited DSI (Fig. 3E, F and H; initial sIPSC suppression, i-THL: 5.04 ± 10.57%, n = 16; i-DMSO: 30.98 ± 5.28%n = 15; Student's t test, P < 0.005). Our pharmacological experiments therefore suggest that 2-AG synthesized by postsynaptic DAGL activity mediates DSI in PFC in pyramidal neurons and that anandamide is not involved in the mechanisms producing DSI.

Although carbachol-induced sIPSCs overall displayed significant DSI (Figs 2 and 3) the DSI magnitude varied significantly between cells, and in some individual neurons DSI was undetectable (Fig. 4A). Despite such variability, the distribution of initial IPSC suppression values (Fig. 4B) showed that DSI was significantly weaker in the THL group (P < 0.02 versus DMSO, Kolmogorov–Smirnov test), significantly potentiated by NAM (P < 0.01 versus DMSO, Kolmogorov–Smirnov test) and not changed by URB597 (P = 0.985 versus DMSO, Kolmogorov–Smirnov test). Our finding that DSI was undetectable in some PFC neurons is consistent with previous reports showing that some principal neurons in cortical circuits lack DSI (Bodor et al. 2005; Varga et al. 2010). DSI may be absent in some neurons if they lack either DAGL activity or, as in the entorhinal cortex (Varga et al. 2010), they lack cannabinoid-sensitive inputs. In that case, DSI potentiation by NAM should not change the percentage of DSI-negative neurons. Alternatively, endocannabinoid mobilization by carbachol (Kim et al. 2002) may tonically silence cannabinoid-sensitive synapses (Neu et al. 2007; Gulyas et al. 2010; Szabóet al. 2010) and occlude DSI in some neurons, which would therefore be classified as DSI-negative. If so, NAM should increase the fraction of DSI-negative neurons by increasing tonic CB1R activation. Finally, it is possible that all PFC pyramidal cells have the machinery necessary to express DSI but that in some neurons DSI was weak and undetectable. In this case, DSI potentiation by NAM should decrease the percentage of DSI-negative cells.

Figure 4. Variability of sIPSC suppression by endocannabinoid-mediated DSI.

A, four examples illustrating the variability between experiments in the magnitude of initial sIPSC suppression and in the baseline sIPSCs (sIPSCs recorded before application of the depolarizing step). The examples in the left panels illustrate cases of neurons that were classified as DSI-positive, with strong DSI (top) and weak DSI (bottom). The panels to the right illustrate cases of neurons classified as DSI-negative, with low baseline variability (top) and high baseline variability (bottom) (see Methods section for details on the criteria used to classify neurons as DSI-positive and DSI-negative). B, cumulative distribution histogram for the initial sIPSC suppression values observed for experiments in control conditions (DMSO), in the presence of 10 μm THL, 1 μm NAM and 1 μm URB597. Analysis using the Kolmogorov–Smirnov test revealed that THL significantly decreased sIPSC suppression (P < 0.05 versus DMSO), whereas NAM significantly enhanced IPSC suppression (P < 0.05 versus DMSO) and URB597 did not have statistically significant effects (P > 0.05). C, percentage of DSI-positive and DSI-negative neurons observed in each experimental group. Note that THL significantly decreased the percentage of DSI-positive cells, whereas NAM significantly increased the percentage of DSI-positive neurons. Statistical significance of the differences in percentage of DSI-positive and DSI-negative neurons was assessed using the χ² test.

To distinguish among these possibilities, we determined the effects of NAM and of other pharmacological inhibitors of endocannabinoid synthesis and inactivation (Fig. 3) on the proportion of DSI-negative and DSI-positive cells. Because the baseline IPSCs (IPSCs recorded before inducing DSI) showed significant variability between cells (Fig. 4A), DSI-positive cells were identified by comparing the initial IPSC suppression with the IPSC baseline on a cell-by-cell basis (see Methods). As shown in Fig. 4C, the percentage of DSI-negative neurons differed between groups: DMSO, 23.1%; THL, 67.9%; NAM, 0% and URB597, 30.8%. The proportion of DSI-negative cells (Fig. 4C) was significantly decreased by NAM (NAM versus DMSO, P < 0.05, χ² test), significantly increased by THL (THL versus DMSO, P < 0.005, χ² test), and not changed by URB597 (URB597 versus DMSO, P = 0.704, χ² test). Our finding that ∼100% of the PFC pyramidal neurons displayed DSI in the presence of NAM is consistent with the idea that most pyramidal neurons in layers 2/3 of PFC receive cannabinoid-sensitive GABA synapses and express DAGL, with some cells expressing weak DSI that was undetectable before NAM potentiation.

Generation of DAGLα-deficient and DAGLβ-deficient mice

The results of our pharmacological experiments indicate that DAGL activity in postsynaptic PFC pyramidal cells is crucial for retrograde endocannabinoid suppression of IPSCs. THL, however, does not selectively target the α or β isoform of DAGL (Bisogno et al. 2003). Moreover, THL and other DAGL inhibitors such as RHC80267 may also inhibit other enzymes of lipid biosynthesis and catabolism (Hoover et al. 2008; Di Marzo, 2011). To confirm the role of DAGL activity and determine if a particular DAGL isoform mediates DSI in PFC neurons, we generated mice lacking DAGLα or DAGLβ.

Mouse embryonic stem (ES) cells carrying a mutation in the DAGLα gene (accession number NM_198114) were obtained from OmniBank, a library of gene-trapped ES cell clones (Zambrowicz et al. 1998, 2003). The clone identified by the OmniBank Sequence Tag (OST) matching the mouse DAGLα sequences (OST-288027) was thawed and expanded. Inverse genomic PCR analysis (Silver & Keerikatte, 1989) of OST-288027 cell DNA showed the insertion of the gene-trapping retroviral vector in intron 4 of the DAGLα gene on mouse chromosome 19, downstream of the initiation codon in exon 1 (Fig. 5A, upper panel). OST-288027 cells were used to generate mice heterozygous for the DAGLα mutation using standard methods (Joyner, 2000) and DAGLα^−/− mice were produced by the interbreeding of DAGLα^+/− animals. Reverse transcription-PCR (RT-PCR) with primers complementary to exons 4 and 7, flanking the genomic integration site of the vector, showed that the wild-type DAGLα gene transcript was undetectable in tissue from DAGLα^−/− mice (Fig. 5A, middle panel), confirming the disruption of DAGLα transcription by the gene-trapping vector. Quantitative PCR (QPCR), showed the absence of the DAGLα transcript in DAGLα^−/− mice and intermediate levels in DAGLα^+/− compared with wild-type mice, whereas DAGLβ transcript levels were normal (Fig. 5A, lower panel).

Figure 5. Generation and characterization of DAGLα^−/− and DAGLβ^−/− mice.

A, top panel: gene trap mutation of the DAGLα gene. Analysis of genomic DNA from cells of the Omnibank ES cell clone identified by the OmniBank Sequence Tag (OST) 288027 showed the insertion of the gene-trapping retroviral vector in intron 4 of the DAGLα gene on mouse chromosome 19, downstream of the initiation codon. Middle panel: reverse transcription (RT)-PCR using primers complementary to exons 4 and 7 shows absence of endogenous DAGLα transcript in the brain and lung of DAGLα^−/− animals. Bottom panel: quantitative PCR (QPCR) shows absence of DAGLα transcript in brain from homozygous mutant mice for DAGLα, without any effect on the expression of the reciprocal DAGLβ locus. DAGLα expression in the heterozygous mutant animals is reduced by approximately 50%. B, top panel: gene trap mutation of the DAGLβ gene. ES cells carrying a mutation in the DAGLβ gene were obtained from the OmniBank clone OST-195261. In OST-195261 cells, the gene-trapping retroviral vector was inserted in exon 1 of the DAGLβ gene on mouse chromosome 5, downstream of the initiation codon. Middle panel: DAGLβ RT-PCR using primers complementary to exons 1 and 3–4 shows absence of endogenous DAGLβ transcript in the brain and lung of DAGLβ^−/− animals. Bottom panel: QPCR shows absence of DAGLβ transcript in brain from homozygous mutant mice for each line, without any effect on the expression of the reciprocal DAGLα locus. DAGLβ expression in the heterozygous mutant animals is reduced by approximately 50%. C, basal 2-AG tissue content measured with liquid chromatography/mass spectroscopy in cortex, cerebellum, hypothalamus and hippocampus of wild-type (wt), DAGLα^−/−, DAGLβ^−/−, and DAGLα^−/−β^−/− mice (double knock-outs). Note the reduction of 2-AG content in tissue from all brain regions of DAGLα^−/− but not DAGLβ^−/− mice. DAGLα^−/−β^−/− mice showed a tendency for lower 2-AG levels compared with DAGLα^−/− mice, but this tendency was not statistically significant. The numbers in or above each bar indicate the number of mice for which 2-AG concentration was determined in each group. Multiple-factor ANOVA revealed statistically significant effects of genotype (F_(3,65) = 88.451, P < 0.0001) and brain region (F_(3,65) = 14.687, P < 0.00001). P values displayed above the bars indicate significant differences compared with tissue content in the same brain region for the wild-type group, post hoc comparison with Fisher's LSD test. Post hoc comparison of 2-AG concentration in DAGLα- versus DAGαβ-deficient mice showed no significant differences (cortex: P = 0.1275, cerebellum: P = 0.3725, hypothalamus: P = 0.0723, hippocampus: P = 0.4552). Single-factor ANOVA performed for 2-AG levels in neocortex independently of other brain regions revealed a significant effect of genotype (F_(3,14) = 6.239, P = 0.0065) and post hoc comparisons (Fisher's LSD test) revealed a significant decrease of 2-AG in DAGLα- or DAGLαβ-deficient mice (P = 0.023 and P = 0.0023, respectively) compared with wild-type mice, but no significant decrease of 2-AG in DAGLβ-deficient animals (P = 0.715). Furthermore, post hoc comparisons following single-factor ANOVA of cortical 2-AG levels showed no difference between DAGLα^−/− and DAGLα^−/−β^−/− animals (LSD test: P = 0.293). D, cortical anandamide levels were not altered in DAGLα^−/− or DAGLα^−/−β^−/− mice, whereas, as described in a recent study (Tanimura et al. 2010), hippocampal levels of anandamide were significantly reduced by ∼50% in DAGLα^−/− or DAGLα^−/−β^−/− mice, although DAGLβ^−/− mice had normal hippocampal anandamide levels. The numbers inside each bar indicate the number of mice for which anandamide concentration was determined in each group. Multiple factor ANOVA revealed statistically significant effects of genotype (F_(3,62) = 6.6713, P < 0.001) and brain region (F_(3,62) = 19.018, P < 0.00001) on anandamide concentrations. P values displayed above the bars indicate significant differences compared with tissue content in the same brain region for the wild-type group, post hoc comparison using Fisher's LSD test. Single-factor ANOVA performed for anandamide levels in neocortex independent of other regions revealed absence of significant effect of genotype (F_(3,13) = 2.2135, P = 0.1352). Moreover, post hoc comparisons following single-factor ANOVA showed no significant decrease of anandamide in DAGLα- or DAGαβ-deficient mice compared with wild-type mice (P = 0.1015 and P = 0.1117, respectively).

ES cells carrying a mutation in the DAGLβ gene (accession number NM_144915) were obtained from the OmniBank clone OST-195261. In OST-195261 cells, the gene-trapping retroviral vector was inserted in exon 1 of the DAGLβ gene on mouse chromosome 5, downstream of the initiation codon (Fig. 5B, upper panel). Mice heterozygous for the DAGLβ mutation were generated using OST-195261 cells and DAGLβ^−/− mice were produced by interbreeding of DAGLβ^+/− mice. RT-PCR with primers complementary to exons 1 and 3–4, flanking the vector integration site, showed that DAGLβ gene transcription was disrupted, as wild-type DAGLβ transcript was undetectable in tissue from DAGLβ^−/− mice (Fig. 5B, middle panel). QPCR showed a complete reduction of DAGLβ message levels in DAGLβ^−/− mice and an intermediate reduction in DAGLβ^+/− mice, with normal levels of DAGLα transcript (Fig. 5B, lower panel).

To test whether disrupting DAGLα or DAGLβ gene transcription affected 2-AG synthesis, we determined the 2-AG content in brain tissue using liquid chromatography–mass spectroscopy. In either wild-type or mutant mice, we found higher 2-AG levels in hypothalamus than in neocortex, cerebellum or hippocampus (Fig. 5C), revealing a significant effect of brain region (F_(3,65) = 14.687, P < 0.0001). Moreover, 2-AG levels were significantly affected by genotype (F_(3,65) = 88.451, P < 0.00001), post hoc comparisons showing a significant reduction of 2-AG levels in DAGLα^−/− mice (Fisher's LSD test, DAGLα^−/−versus wild-type, P < 0.0001), but not in DAGLβ^−/− mice (Fisher's LSD test, DAGLβ^−/−versus wild-type, P = 0.2983). The reduction of 2-AG levels by DAGLα deficiency was significant in each of the brain regions examined (Fig. 5C). Interestingly (Fig. 5C), 2-AG content was lower in tissue from mice homozygous for both DAGLα and DAGLβ deficiency (DAGLα^−/−β^−/− mice) compared with DAGLα^−/− mice, although the difference did not reach statistical significance in each region separately (cortex: P = 0.127; cerebellum: P = 0.372; hypothalamus: P = 0.072; hippocampus: P = 0.455), nor when all regions were pooled (Fisher's LSD test, DAGLα^−/−β^−/−versus DAGLα^−/− mice, P = 0.089). These results suggest that DAGLα-driven 2-AG synthesis accounts for most of the basal content of 2-AG in multiple brain regions.

Previous studies showed that in addition to reducing 2-AG, DAGLα deficiency also significantly decreased anandamide, when measured in total brain tissue or in hippocampal tissue (Gao et al. 2010; Tanimura et al. 2010). Such a decrease of anandamide by DAGLα deficiency suggests, first, that some form of cross-talk between anandamide and 2-AG biosynthesis may exist (Di Marzo, 2011) and, second, that anandamide might still participate in DSI (Di Marzo, 2011). Because anandamide levels were not previously examined in neocortex of DAGLα^−/− mice (Gao et al. 2010; Tanimura et al. 2010) we next determined anandamide levels in neocortex and other brain regions in wild-type, DAGLα^−/− and DAGLβ^−/− mice. We found that anandamide tissue concentration was nearly an order of magnitude lower than that of 2-AG, as shown previously (Kano et al. 2009). Moreover, anandamide content was higher in hippocampus than in neocortex, cerebellum or hypothalamus (Fig. 5D) showing a significant effect of brain region (F_(3,62) = 19.018, P < 0.0001). Genotype also had a significant effect on anandamide levels (F_(3,74) = 3.491, P < 0.02) which was driven mostly by a decrease of anandamide in the hippocampus of DAGLα^−/− mice (Fig. 5D). Indeed, as reported previously (Tanimura et al. 2010), hippocampal anandamide content was strongly (∼65%) and significantly reduced by DAGLα deficiency (Fisher's LSD test, DAGLα^−/−versus wild-type, P < 0.00001), whereas DAGLβ deficiency produced a weaker (∼25%) reduction in anandamide (Fisher's LSD test, DAGLβ^−/−versus wild-type, P < 0.05). In contrast to hippocampus, anandamide levels in neocortex were not affected by genotype (Fig. 5D, Fisher's LSD test versus wild-type, DAGLα^−/−, P = 0.459, DAGLβ^−/−, P = 0.312; DAGLα^−/−β^−/−, P = 0.473). In summary, our analysis of endocannabinoid content showed that in multiple brain regions 2-AG is significantly reduced by DAGLα deficiency but not by DAGLβ deficiency. Moreover, we found that the decrease in anandamide content in the brain of DAGLα^−/− mice is region-specific, since it was observed in hippocampus but not in the neocortex.

Supplemental Fig. 2 (available online only) shows examples of three-dimensional reconstructions of PFC pyramidal neurons from DAGLα^−/− and DAGLβ^−/− mice, as well as from wild-type littermates. Such reconstructions qualitatively suggest that genotype had no major effects on the dendritic and axonal architecture of PFC neurons. However, the absence of endocannabinoid signalling in CB1R-deficient mice significantly decreased the length of the basal dendrites of PFC pyramidal neurons (Hill et al. 2011). Moreover, DAGLα or DAGLβ knock-down decreased, and their overexpression increased, neurite outgrowth in neuroblastoma cell cultures (Jung et al. 2011). Therefore, the possibility that DAGL deficiency changes the morphology of PFC neurons remains to be investigated quantitatively.

DAGLα deficiency abolished DSI without generally affecting inhibitory synaptic transmission

The reduction of basal 2-AG in cortical tissue of DAGLα^−/− but not DAGLβ^−/− mice suggests that DAGLβ is not significantly involved in 2-AG synthesis in neocortex. Therefore, if 2-AG mediates DSI in PFC, then DSI should be reduced or abolished in neurons from DAGLα^−/− but not DAGLβ^−/− mice. However, it is not clear if baseline 2-AG content represents a 2-AG pool associated with production of DSI. For instance, DSI may be associated with a 2-AG pool synthesized ‘on-demand’, whereas baseline 2-AG concentration may reflect an independent ‘pre-stored’ 2-AG pool unrelated to DSI mechanisms (Alger & Kim, 2011; Di Marzo, 2011). Therefore, a reduction of basal 2-AG content by DAGLα, but not DAGLβ deficiency (Fig. 5C and D), does not rule out a contribution of DAGLβ to DSI. Consequently, we next studied DSI of carbachol-induced sIPSCs in PFC slices from DAGLα^−/−, DAGLβ^−/− and wild-type mice (Fig. 6A). We found that depolarizing steps completely failed to induce DSI in pyramidal neurons from DAGLα^−/− mice (Fig. 6A–C; sIPSC suppression, wild-type littermates: 27.35 ± 3.98%, n = 28; DAGLα^−/−: 3.69 ± 3.20%, n = 16, P < 0.05 versus wild-type). In contrast, DSI magnitude was normal in PFC neurons from DAGLβ^−/− mice (Fig. 6A–C; sIPSC suppression: 38.59 ± 5.96%n = 17, P > 0.05 versus wild-type). The distribution of sIPSC suppression values (Fig. 6D) showed that DSI was significantly reduced in neurons from DAGLα^−/− mice (P < 0.01, versus wild-type, Kolmogorov–Smirnov test) but not from DAGLβ^−/− mice (P = 0.372 versus wild-type, Kolmogorov–Smirnov test). Moreover, compared with wild-type mice, the proportion of DSI-negative cells (Fig. 6E) was significantly reduced by DAGLα deficiency but not by DAGLβ deficiency (wild-type, 28.6%; DAGLα^−/−, 87.5%, P < 0.0005 versus wild-type, χ² test; DAGLβ^−/−, 17.7%, P = 0.408 versus wild-type, χ² test). The analysis of DSI in DAGL-deficient mice therefore confirms the findings of our pharmacological experiments; that is, DAGL-mediated 2-AG synthesis is required for DSI and DAGLα is the critical DAGL isoform involved. Moreover, our data suggest that the basal 2-AG tissue content reflects a DAGLα-dependent and DAGLβ-independent 2-AG pool that is required for DSI.

Figure 6. DSI is abolished in PFC neurons from DAGLα^−/− but not from DAGLβ^−/− mice.

A, examples of the effect of depolarizing steps on carbachol-induced sIPSCs recorded from layer 2/3 PFC pyramidal neurons from wild-type, DAGLα^−/− and DAGLβ^−/− mice. Note that in neurons of wild-type and DAGLβ^−/− genotype we commonly observed a significant long-lasting outward current (large arrows) shortly after the end of the depolarizing steps. In DAGLα^−/− neurons, such outward current was significantly reduced (see Fig. 5) and frequently an inward current was observed instead (small arrow). Independently of the long-lasting inward or outward currents, depolarizing steps produced significant sIPSC suppression in wild-type and DAGLβ^−/− neurons, whereas in DAGLα^−/− neurons sIPSC suppression was absent. B, time course plot showing the effects of depolarizing steps on IPSCs recorded from pyramidal cells from mice of the three genotypes shown in A. C, bar graph summarizing the effects of DAGL genotype on DSI magnitude. Single-factor ANOVA revealed a statistically significant effect of genotype (F_(2,58) = 12.660, P < 0.00005). P < 0.005 compared with the wild-type or DAGLβ^−/− groups, post hoc comparison using Fisher's LSD test. D, cumulative distribution histogram for the initial IPSC suppression values observed for experiments in neurons from wild-type, DAGLα^−/− and DAGLβ^−/− neurons. Analysis using Kolmogorov–Smirnov test revealed that IPSC suppression was significantly decreased in neurons from DAGLα^−/− (P < 0.05 versus wild-type), whereas IPSC suppression was not significantly affected in neurons from DAGLβ^−/− mice (P > 0.05 versus wild-type). E, percentage of DSI-positive and DSI-negative neurons observed in mice from each genotype. Note that the percentage of DSI-positive cells was significantly decreased in DAGLα^−/− mice, whereas in DAGLβ^−/− mice the percentage of DSI-positive cells was not significantly affected. The significance of differences in the percentage of DSI-positive and DSI-negative neurons was assessed using χ² test.

In addition to impairing DSI, the constitutive reduction of cortical 2-AG tissue levels in DAGLα^−/− mice (Fig. 5) altered other aspects of neuronal and synaptic function. For instance, in neurons from wild-type mice, we commonly observed a long-lasting outward current shortly after the depolarizing steps applied to induce DSI (see Figs 2 and 6). At the time of peak for such current in wild-type neurons, its magnitude was significantly smaller in DAGLα^−/− neurons (Fig. 7A and B), and was unaffected in DAGLβ^−/− neurons (peak current, wild-type: 24.0 ± 3.1 pA, n = 17; DAGLα^−/−: –0.52 ± 6.5 pA, n = 14; DAGLβ^−/−: 31.1 ± 6.3 pA, n = 13; single-factor ANOVA, F_(2,41) = 9.575, P < 0.0005; Dunnett's test versus wild-type, DAGLα^−/−: P < 0.005, DAGLβ^−/−: P = 0.533). Such outward current may be similar to the hyperpolarizing current underlying endocannabinoid-mediated slow self-inhibition in pyramidal cells and interneurons (Marinelli et al. 2009; Min et al. 2010b). If so, the outward current should be abolished or strongly depressed by CB1R antagonists or DAGL inhibitors, which block slow self-inhibition (Bacci et al. 2004; Marinelli et al. 2009; Min et al. 2010b). However, we found that the DAGLα-dependent hyperpolarizing current was insensitive to acute application of THL, SR141617A or NAM (Fig. 7C and D; outward current amplitude, DMSO control: 22.1 ± 4.2 pA, n = 12; SR141617A: 17.7 ± 7.3 pA, n = 9; extracellular THL: 14.8 ± 2.7 pA, n = 17; intracellular THL: 26.0 ± 5.0 pA, n = 9; NAM: 20.7 ± 2.3, n = 15; single-factor ANOVA, F_(4,57) = 1.343, P = 0.265; Dunnett's test versus DMSO, SR141617A: P = 0.851, extracellular THL: P = 0.378, intracellular THL: P = 0.896; NAM: P = 0.995), even though in the same experiments DSI was blocked by THL and SR141617A and enhanced by NAM (Fig. 3). Therefore, unlike DSI or slow self-inhibition, the long-lasting outward current affected by chronic DAGL deficiency is not directly mediated by endocannabinoid activation of CB1Rs, showing that the constitutive decrease of 2-AG levels in DAGLα^−/− mice may produce alterations that are not mimicked by short-term blockade of 2-AG-mediated signalling. The presence of such outward current may decrease the membrane resistance and potentially decrease sIPSC amplitude if the voltage clamp conditions are imperfect. However, most of the cannabinoid-sensitive IPSCs originate in perisomatic synapses located in a membrane compartment with relatively good voltage control. In addition, the fact that DSI was blocked by pharmacological manipulations that do not affect the outward current suggests that there is no mechanistic relation between the outward current and sIPSC suppression.

The absence of DSI in PFC neurons from DAGLα^−/− mice does not indicate whether cannabinoid-sensitive PFC synapses maintain a normal response to CB1R receptor activation. For instance, chronic elevation of 2-AG levels in vivo decreases the sensitivity to CB1R activation of IPSCs evoked in hippocampal neurons by cannabinoid-sensitive synapses (Schlosburg et al. 2010). Conversely, chronic reduction of 2-AG levels in DAGLα-deficient mice could upregulate the sensitivity to CB1R activation. To test the effect of DAGLα deficiency on the response to CB1R stimulation of cannabinoid-sensitive synapses in PFC, we evoked IPSCs (eIPSCs) with extracellular stimulation in the presence of ω-agatoxin-IVA (250 nm), which blocks P/Q-type Ca²⁺ channel-dependent GABA release from cannabinoid-insensitive synapses (Freund & Katona, 2007), including those of parvalbumin-positive interneurons in PFC (Zaitsev et al. 2007). In wild-type PFC neurons, agatoxin-resistant eIPSCs showed significant CB1R-mediated DSI produced by depolarizing steps in experiments performed in the absence of carbachol (Fig. 8A). Moreover, DSI of agatoxin-resistant eIPSCs was strongly reduced by DAGLα deficiency or by application of the CB1R antagonist AM251 (10 μm, Fig. 8A and B). Therefore, in PFC pyramidal neurons agatoxin-resistant eIPSCs are elicited by cannabinoid-sensitive GABA synapses that show DAGLα-dependent DSI in the absence of carbachol. The agatoxin-resistant eIPSCs were significantly suppressed by the CB1R agonist WIN55212-2 (1 μm), via an effect that in most neurons was reversed by the CB1R antagonist SR141716A (10 μm, Fig. 8C and D), although the time course of reversal of the WIN55212-2 effect by SR141716A varied significantly between experiments. Here, we only quantified the effect of WIN55212-2 if SR141716A reversed its effect by at least 80%. Interestingly, the SR141716A-reversible suppression of agatoxin-resistant eIPSCs by WIN55212-2 (Fig. 8E) was similar in wild-type and DAGLα^−/− PFC neurons (eIPSC suppression, wild-type: 52.23 ± 5.36%, n = 8 cells; DAGLα^−/− 58.90 ± 8.29%, n = 8 cells, P = 0.51). These results show that chronic reduction of cortical 2-AG levels in DAGLα^−/− mice does not change the sensitivity of cannabinoid-sensitive synapses to CB1R activation by WIN55212-2.

Because endocannabinoids regulate inhibitory transmission at synapses from only a few interneuron subtypes (Glickfeld & Scanziani, 2006; Galarreta et al. 2008; Lee et al. 2010), most GABA synapses are not directly affected by decreased 2-AG levels in DAGLα-deficient mice. However, endocannabinoids also regulate cellular excitability and glutamate-mediated transmission (Kano et al. 2009), therefore DAGLα deficiency may change the patterns of network activity and indirectly produce a general alteration of synaptic inhibition. For instance, prolonged CB1R agonist exposure generally decreases the strength of GABA_A receptor-mediated inhibition (Deshpande et al. 2011). We therefore recorded sIPSCs and carbachol-induced sIPSCs in PFC pyramidal neurons of DAGLα^−/− and wild-type mice (Fig. 8F). We found that neither the frequency, amplitude nor kinetics (rise and decay times) of sIPSCs and carbachol-induced sIPSCs differed significantly between DAGLα^−/−, DAGLβ^−/− and wild-type neurons (Fig. 8F). Moreover, carbachol similarly increased the IPSC frequency and amplitude (without changing IPSC kinetics) in either wild-type or DAGL-deficient neurons (Fig. 8G). These results suggest that the constitutive decrease of 2-AG in DAGLα-deficient mice has a very selective effect, leaving baseline transmission unaltered at the majority of GABA synapses.

Retrograde suppression of inhibition by action potential-mediated stimulation of endocannabinoid synthesis requires DAGLα activity

DSI is typically studied by voltage-clamping neurons for a prolonged time (∼1–5 s) at largely depolarized membrane potentials (∼0 mV), a stimulation protocol that effectively reveals the basic properties of DSI, but is probably non-physiological. In hippocampus and somatosensory cortex, retrograde suppression of inhibitory synaptic transmission can be produced by action potential-mediated depolarization (Pitler & Alger, 1992b; Fortin et al. 2004). Importantly, DSI induced by repetitive firing of PFC neurons is hypothesized to be crucial to produce persistent neural activity in PFC during working memory tasks (Carter & Wang, 2007). However, whether persistent firing of PFC neurons produces DSI has not been determined. Therefore we examined if action potential firing can suppress inhibition in PFC neurons and if so, whether firing-induced suppression depends on DAGLα, the DAGL isoform that is necessary for DSI in PFC neurons and in other brain regions (Gao et al. 2010; Tanimura et al. 2010). Interestingly, pyramidal cells from DAGLα^−/− mice displayed reduced excitability (Fig. 9A), firing action potentials at lower frequency than wild-type neurons throughout a wide range of excitatory input currents (Fig. 9B, two-factor ANOVA, input current: F_(40,533) = 31.3, P < 0.00001, genotype: F_(1,533) = 207.5, P < 0.000001). Multiple changes seem to contribute to the lower excitability of DAGLα^−/− neurons, which had a reduced slope of the voltage–current relation in the hyperpolarizing range, suggestive of a decrease in their membrane resistance (Fig. 9C, two-factor ANOVA, input current: F_(4,65) = 43.5, P < 0.0001; genotype: F_(1,65) = 4.95, P < 0.05). In addition, DAGLα^−/− cells had higher voltage threshold for action potential firing for input currents up to 400 pA (Fig. 9D; input current: F_(30,403) = 0.83, P = 1.0; genotype F_(1,403) = 170.1, P < 0.0001) and stronger spike frequency adaptation (Fig. 9E; input current F_(48,629) = 14.1, P < 0.00001, genotype: F_(1,629) = 48.6, P < 0.00001) for input currents up to ∼100 pA above threshold. The mechanisms by which DAGLα deficiency decreases the intrinsic excitability of PFC pyramidal cells remain to be determined.

Since DAGLα^−/− pyramidal neurons have reduced excitability, we used trains of brief and largely suprathreshold current steps to evoke a controlled number of action potentials in each neuron independent of its genotype, and determine if action potential firing produces endocannabinoid-mediated DSI. Action potentials were elicited at 20 Hz, a frequency within the range of firing of medial PFC neurons in vivo including during performance of cognitive tasks that require working memory (Baeg et al. 2003; van Aerde et al. 2008; Lapish et al. 2008; however, see van Aerde et al. 2009; Hyman et al. 2010). In PFC pyramidal neurons from wild-type mice, suprathreshold depolarizing steps produced significant DSI of inhibitory postsynaptic potentials (IPSPs) induced by carbachol (Fig. 10A and B). Interestingly, depolarizing steps driving the cells’ membrane potential to values just below spike threshold did not suppress the IPSPs, even though action potential-eliciting steps delivered in an alternating manner produced significant IPSP suppression (Fig. 10A and B). These data indicate that depolarization by action potentials is necessary to suppress IPSPs. In wild-type PFC pyramidal neurons, suppression of carbachol-induced IPSPs by action potential firing at 20 Hz was significant for stimulus trains of durations ≥0.5 s (Fig. 10C and D). Interestingly, the DSI duration increased as the duration of 20 Hz stimulus trains increased (Fig. 10D). In contrast, in pyramidal cells from DAGLα^−/− mice, 20 Hz stimulation failed to suppress the IPSPs independently of stimulus duration (Fig. 10E and F). Analysis of IPSP suppression by DSI as a function of stimulus duration (stimulus frequency 20 Hz) showed that DSI magnitude in wild-type neurons increased between 0.1 s and 1.0 s stimulus duration, remaining essentially independent of stimulus duration for longer stimulus trains (Fig. 10G). Therefore, in PFC pyramidal neurons action potential firing produced, in a DAGLα-dependent manner, an IPSP suppression that was somewhat graded as a function of stimulus duration.

Discussion

2-AG synthesized by postsynaptic DAGLα is the endocannabinoid mediating DSI in PFC pyramidal neurons

Identifying the endocannabinoid mediating DSI has been complicated by some discrepancies in findings across studies. For instance, DAGL inhibitors abolished DSI in cultured hippocampal neurons (Hashimotodani et al. 2007, 2008) or Purkinje cells (Szabo et al. 2006), but did not consistently block DSI in hippocampal or cerebellar slices, even though other forms of endocannabinoid-mediated synaptic plasticity were abolished (Chevaleyre & Castillo, 2003; Safo & Regehr, 2005; Edwards et al.2006, 2008; Min et al. 2010b; Chavez et al. 2010).

DAGL inhibitors, including THL, also produced inconsistent results in part depending on the manner of application. For example, bath-applied THL inhibited DSI in hippocampal neurons (Edwards et al. 2008; Min et al. 2010b), but intracellularly applied THL did not (Edwards et al. 2008; Min et al. 2010b), whereas in other studies both THL and the DAGL inhibitor OMDM-169 applied intracellularly blocked hippocampal DSI (Zhang et al. 2011). Conversely, the DAGL inhibitor RHC-80267 did not block DSI when bath or intracellularly applied (Chevaleyre & Castillo, 2003; Safo & Regehr, 2005; Edwards et al. 2006, 2008). The fact that THL and RHC-80267 may inhibit other lipases in addition to DAGL (Hoover et al. 2008) may confound data interpretation.

Two recent studies reported the absence of DSI in the hippocampus and cerebellum of DAGLα^−/− mice (Gao et al. 2010; Tanimura et al. 2010). In such mice, 2-AG levels were markedly reduced, a finding consistent with the idea that 2-AG is the endocannabinoid mediating DSI. However, in those studies DAGLα^−/− mice also had reduced anandamide levels in total brain tissue (Gao et al. 2010), and specifically in cerebellum, hippocampus or striatum (Tanimura et al. 2010). If both 2-AG and anandamide levels are reduced, identifying the endocannabinoid mediating DSI is complicated. Interestingly, Tanimura and colleagues showed that combined application of a metabotropic glutamate receptor agonist and high extracellular potassium increases 2-AG, but not anandamide, levels in cerebellar slices from wild-type mice (Tanimura et al. 2010). Whereas such data suggest that anandamide synthesis in cerebellum is not activated by stimuli that produce endocannabinoid mobilization, previous studies showed that anandamide synthesis is increased by electrical stimulation, high potassium, carbachol or metabotropic glutamate receptor agonists in striatal, cortical or hypothalamic neurons (Kano et al. 2009).

Here, using pharmacology and genetically modified mice, we found convergent evidence indicating that in PFC circuits 2-AG synthesized by DAGLα in postsynaptic PFC pyramidal cells mediates DSI independently of anandamide. First, extra- or intracellularly applied THL strongly inhibited DSI. Second, DSI was absent in PFC neurons from DAGLα-deficient mice in which cortical 2-AG was significantly reduced but cortical anandamide was unaffected. Third, acute pharmacological inhibition of 2-AG degradation enhanced DSI, whereas inhibition of anandamide degradation did not affect DSI in PFC. Similarly, pharmacological inhibition of anandamide degradation, or fatty acid amide hydrolase deficiency did not affect DSI in the hippocampus (Hashimotodani et al. 2007; Pan et al. 2009; Kim & Alger, 2010; Schlosburg et al. 2010) and cerebellum (Szabo et al. 2006; Pan et al. 2009). Our data thus favour the idea that in PFC circuits, DSI is primarily mediated by 2-AG with little or no anandamide contribution.

The decrease in anandamide content previously found in the hippocampus of DAGLα^−/− mice (Tanimura et al. 2010) suggests the presence of a cross-talk between 2-AG and anandamide synthesis (Di Marzo, 2011). Since quantification of brain endocannabinoids is methodologically challenging, we used liquid chromatography coupled with mass spectroscopy, a technique that offers high sensitivity and selectivity (Buczynski & Parsons, 2010). Using this technique and a different strain of DAGLα^−/− mice, we confirmed that hippocampal anandamide levels are decreased by DAGLα deficiency, as previously shown (Tanimura et al. 2010). However, we did not find lower anandamide levels in the neocortex of DAGLα^−/− mice, suggesting that such cross-talk between 2-AG and anandamide synthesis is region-specific. Interestingly, monoacylglycerol lipase-deficient mice have elevated 2-AG but normal anandamide levels (Schlosburg et al. 2010), suggesting that the mechanisms underlying the interactions between 2-AG and anandamide synthesis are complex. DAGLα deficiency may reduce arachidonic acid levels (Gao et al. 2010), therefore decreasing the phospholipid precursors for anandamide synthesis (Di Marzo, 2011).

We found that in addition to blocking DSI, congenital reduction of 2-AG levels in DAGLα^−/− mice decreased the excitability of PFC neurons. In neocortex, endocannabinoids down-regulate synaptic excitation (Fortin & Levine, 2007; Lafourcade et al. 2007) and pyramidal cell excitability (Marinelli et al. 2009). Therefore, in DAGLα^−/− mice cortical networks may be hyperexcitable, as in CB1R^−/− mice (Monory et al. 2006). If so, then the reduced excitability of PFC neurons in DAGLα^−/− mice could be part of a compensatory mechanism triggered by network hyperexcitability. Importantly, whether the decrease of pyramidal cell excitability in the PFC of DAGLα^−/− mice requires long-term deficits in 2-AG levels or CB1R signalling versus short-term inhibition of 2-AG synthesis remains to be determined.

In neurons from DAGLα^−/− mice, some membrane currents observed shortly after the depolarizing steps used to induce DSI were altered. Whether such currents are related to the DSI-inducing mechanisms is at present unclear but, interestingly, the changes found in DAGLα^−/− mice were not mimicked by acute inhibition of DAGL. These data suggest that long- versus short-term decreases of DAGL activity have different effects. For example, congenital DAGL deficiency may affect both a ‘pre-stored’ and an ‘on-demand’ 2-AG pool (Alger & Kim, 2011; Di Marzo, 2011), whereas acute DAGL inhibition may only affect the latter.

Functional implications

Here, and in previous studies (Yoshida et al. 2006; Volk et al. 2010), DAGLβ mRNA was found in the frontal cortex; however, the function of DAGLβ remains enigmatic. We found that genetic manipulation of DAGLα or DAGLβ expression did not alter the other transcript, suggesting they are regulated independently. Previous studies showed that DAGLα mediates depolarization-induced suppression of excitation without DAGLβ contribution (Tanimura et al. 2010). Whether DAGLβ mediates endocannabinoid-dependent long-term depression of excitatory (Sjostrom et al. 2003) or inhibitory (Heifets & Castillo, 2009) synaptic transmission remains to be addressed.

The requirement of DAGLα for DSI is somewhat puzzling, given its predominant localization in dendritic spines (Katona et al. 2006; Yoshida et al. 2006; Lafourcade et al. 2007; Ludanyi et al. 2011). The lipid-soluble nature of 2-AG would severely limit its diffusion from spines onto cannabinoid-sensitive GABA synapses, most of which are perisomatic (Lee et al. 2010). Since DAGLα deficiency abolished DSI, one possibility is that near CB1R-containing GABA synapses DAGLα levels are low and undetectable by electron microscopy (Yoshida et al. 2006; Lafourcade et al. 2007), but sufficient to support DSI. Similarly, inhibition of monoacylglycerol lipase strongly affects DSI, although by electron microscopy the enzyme is predominant in glutamatergic axon terminals (Gulyas et al. 2004; Ludanyi et al. 2011).

In PFC neurons, monoacylglycerol lipase inhibition with NAM prolonged DSI and increased the magnitude of the suppression of carbachol-induced sIPSCs from ∼30% to ∼55%. In hippocampus and somatosensory cortex, selective stimulation of cannabinoid-sensitive inputs in paired recordings revealed a ∼100% IPSC suppression by DSI in the absence of NAM (Glickfeld & Scanziani, 2006; Galarreta et al. 2008; Lee et al. 2010; Min et al. 2010b). Whether DSI suppresses IPSCs by ∼100% also at isolated cannabinoid-sensitive inputs in PFC is not yet known. If so, then the less than 100% initial suppression in our data suggests that a fraction of the carbachol-induced IPSCs originate in cannabinoid-insensitive synapses.

Carbachol may enhance DSI by stimulating the firing of cholecystokinin-positive GABA neurons (Kawaguchi, 1997; Cea-del Rio et al. 2010) that are the main source of cannabinoid-sensitive synapses (Glickfeld & Scanziani, 2006; Neu et al. 2007; Galarreta et al. 2008; Lee et al. 2010; Ali, 2011). However, synapses from some cholecystokinin-positive cells onto pyramidal neurons are cannabinoid-insensitive (Lee et al. 2010; Ali, 2011) possibly because they lack CB1Rs (Eggan et al. 2010). Moreover, carbachol also stimulates somatostatin- and vasoactive intestinal peptide-positive GABA neurons (Kawaguchi, 1997; Fanselow et al. 2008), which are not thought to provide cannabinoid-sensitive synapses, although somatostatin neurons contain CB1R mRNA (Hill et al. 2007). Therefore, carbachol-induced IPSCs indeed seem to originate in cannabinoid-sensitive and -insensitive synapses.

We found that the DSI magnitude varied largely between cells and was undetectable in some neurons, but that blockade of 2-AG degradation with NAM made DSI detectable in every case, suggesting that all layer 2/3 PFC neurons can produce DSI. In somatosensory cortex DSI is found in most layer 2/3 neurons, but is weak or absent in layer 5 cells (Bodor et al. 2005; Fortin & Levine, 2007). Moreover, layer 5 has a low density of CB1R-positive axons in PFC (Lafourcade et al. 2007) and somatosensory cortex (Bodor et al. 2005). Therefore, DSI may regulate pyramidal cell output more strongly in superficial than in deep cortical layers which target cortical versus subcortical sites, respectively.

In a computational model, disinhibition by DSI counteracted spike frequency adaptation in pyramidal cells, stabilizing working memory-related attractor network states of persistent firing and increasing the accuracy of working memory-guided responses (Carter & Wang, 2007). Consistent with that role, we found that persistent firing of mouse PFC neurons at 20 Hz, a firing frequency typical of PFC neurons during working memory tasks (Goldman-Rakic, 1995), induced significant DSI. However, the typical working memory-related neuronal firing was originally described in the monkey PFC (Goldman-Rakic, 1995), and whether it is observed in mouse medial PFC is debatable (Preuss, 1995; Seamans et al. 2008). Furthermore, whereas DSI-mediated disinhibition requires that cannabinoid-sensitive GABA synapses have a strong inhibitory effect, those GABA synapses are instead thought to affect pyramidal cell activity in subtle manners (Freund & Katona, 2007).

If DSI in fact mediates disinhibition, then its absence in the PFC of DAGLα^−/− mice may trigger a compensatory decrease of inhibitory synaptic strength. However, the fact that neither the amplitude, frequency nor kinetics of GABA_A receptor-mediated IPSCs were significantly altered in the PFC of DAGLα^−/− mice argues against this possibility. Because endocannabinoids suppress both synaptic inhibition (this study) and excitation (Lafourcade et al. 2007, 2011) in the PFC, the net changes in PFC network activity produced by DAGLα deficiency may be too small to trigger compensatory responses.

We show here that chronic 2-AG deficiency does not affect the sensitivity of cannabinoid-sensitive synapses to CB1R stimulation, suggesting the absence of compensatory changes in CB1R function. In contrast, chronic 2-AG elevation produced CB1R down-regulation (Schlosburg et al. 2010). Endocannabinoids tonically modulate GABA release (Losonczy et al. 2004; Neu et al. 2007; Kim & Alger, 2010) via an effect possibly mediated by anandamide (Kim & Alger, 2010). Since anandamide levels are not significantly altered in the neocortex of DAGLα^−/− mice, tonic activation by anandamide could maintain cortical CB1R density and efficacy in the absence of 2-AG.

In most experiments, we studied DSI in the presence of carbachol, which can induce depolarization-independent endocannabinoid mobilization (Kim et al. 2002). Depolarization in the presence of carbachol could mobilize a 2-AG pool different from the 2-AG pool activated by depolarization alone (Zhang et al. 2011). We found that DSI of agatoxin-resistant eIPSCs evoked in the absence of carbachol was abolished in DAGLα^−/− mice, suggesting that DAGLα also mediates carbachol-independent DSI. Endocannabinoid mobilization by carbachol produces IPSC depression (Kim et al. 2002). Here, we found that carbachol increased sIPSC amplitude instead of producing sIPSC depression (see Figs 2 and 8); however, it is possible that compared with IPSCs evoked by electrical stimulation, recordings of sIPSCs are less likely to detect IPSC depression by carbachol-induced endocannabinoid mobilization. Thus, whether carbachol produces endocannabinoid mobilization in PFC remains to be addressed, for instance, by studying its effect on agatoxin-resistant eIPSCs. Importantly, we found that DAGLα deficiency abolished DSI of both sIPSCs induced by carbachol and eIPSCs recorded in the absence of carbachol, showing that in PFC DAGLα activity is necessary for DSI produced by more than one form of endocannabinoid mobilization (depolarization alone or combined with carbachol).

We found that in PFC pyramidal neurons, carbachol enabled DSI, an effect consistent with the idea that DSI suppresses IPSCs originating in GABA neurons whose firing is activated or facilitated by acetylcholine. In hippocampal neurons, acetylcholine receptor activation by carbachol strongly depressed or completely silenced, via tonic endocannabinoid effects, IPSCs from some cholecystokynin-positive basket cells (Neu et al. 2007; Szabóet al. 2010; Gulyas et al. 2010). Identifying the GABA neuron subtypes whose output is regulated via tonic endocannabinoid effects versus DSI is thus crucial for understanding the role of endocannabinoids in PFC-dependent cognitive function and dysfunction.

Glossary

Abbreviations

2-AG

2-arachidonoylglycerol

CB1R

cannabinoid receptor 1

DAGL

diacylglycerol lipase

DSI

depolarization-induced suppression of inhibition

eIPSCs

evoked IPSCs

ES

embryonic stem cells

i-

intracellularly applied

NAM

N-arachidonoyl maleimide

OST

OmniBank Sequence Tag

PFC

prefrontal cortex

sIPSCs/sIPSPs

spontaneous IPSCs/IPSPs

THL

tetrahydrolipstatin

Author contributions

H.Y., T.M. and G.G.-B. designed and performed all the electrophysiology experiments in the Translational Neuroscience Program, University of Pittsburgh. H.Y., T.M. and G.G.-B. analysed the electrophysiological data. G.H. and B.Z. conceived, designed, analysed and interpreted the data relating to the production and quality control of the DAGL knockout mice at Lexicon Pharmaceuticals. D.P., M.F. and Y.B. collected and analysed data in non-electrophysiology experiments performed at Bristol-Myers Squibb. R.S.W. and R.Z. conceived, designed and interpreted the data of non-electrophysiology studies performed at Bristol-Myers Squibb. H.Y., T.M., D.A.L and G.G.-B. wrote the paper. All authors approved the final version of the manuscript.

Author's present address

H. Yoshino: Department of Psychiatry, Nara Medical University, Nara, Japan.

Supplementary material

Supplemental Figure 1

Supplemental Figure 2

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer-reviewed and may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors