The experimental protocols were approved by Institutional Animal Care and Use Committee (IACUC) from Stockholm (Sweden) or Lisbon (Portugal), and conducted in accordance with Portuguese and Swedish legislation on animal care and the European Community guidelines (Directive 2010/63/EU).

Most of the experiments were performed using male C57Bl6/J mice, aged between 8 and 18 weeks (most frequently 9–13 weeks) (Charles River Laboratories, Paris). In some cases, male and female mice were used to maximize the use of A1R knockout mice; because of this, control experiments using male and female mice have been performed. No appreciable differences between data obtained in males or females were detected (Supplementary Figure S1). The adenosine A₁ receptor knockout (A₁R^-/-) and wild-type (A₁R^+/+) mice were generated by inactivating the second protein coding exon of the mouse A₁R gene, from heterozygous breeding pairs with C57Bl6/J background strain (Johansson et al., 2001), obtained from a breeding colony derived from this original line that is housed at Karolinska Institutet, Sweden, and genotyped as described previously (Yang et al., 2015). All animals were social housed under standardized conditions of light (12-h light/12-h dark cycle), temperature (22–24°C), humidity (55–65%), and environmental enrichment (cardboard tubes plus nest material) and had free access to food and tap water.

Hippocampal slices were prepared as previously (e.g., Diógenes et al., 2004). The animals were sacrificed by decapitation under deep isoflurane anesthesia. The hippocampus was dissected free within ice-cold artificial cerebrospinal fluid (aCSF) solution composed of (millimeter): NaCl 124, KCl 3, NaHCO₃ 26, Na₂HPO₄ 1.25, MgSO₄ 1, CaCl₂ 2; and glucose 10, previously gassed with 95% O₂ and 5% CO₂, pH 7.4. Slices (400-μm thick) were cut perpendicularly to the long axis of hippocampus with a McIlwain tissue chopper and allowed to recover functionally and energetically for 1 h in a resting chamber filled with the same solution, at room temperature and continuously gassed.

For electrophysiological recordings of field excitatory post-synaptic potentials (fEPSP), individual slices were transferred into a submerged recording chamber (dual submerged chamber) over the nylon mesh and continually superfused with gassed aCSF solution at a constant flow (3 ml/min) and temperature (32°C). This allows oxygenation in both slice surfaces while permitting a relatively fast flow rate to facilitate drug replacement. Stimulation (rectangular 0.1 ms pulses, once every 20 s) was delivered through a concentric electrode placed on Schaffer collateral-commissural fibers, in the stratum radiatum near the CA3–CA1 border. The intensity of the stimulus was set to the one eliciting near 50% of the maximal response, and was maintained throughout the experiment except in those experiments designed to perform input–output curves. In such experiments, after a stabilization period under the standard stimulation conditions, the stimulus intensity was increased by 20 μA every 6 min, within a range of 80–300 μA. fEPSP recording was through a microelectrode filled with NaCl 4 M (2–6 MΩ resistance), placed in CA1 stratum radiatum, coupled to an Axoclamp 2B Amplifier (Axon Instruments) and digitized BNC-2110 (National Instruments). Individual responses were monitored, and averages of six consecutive responses were continuously stored on a personal computer with the WinLTP Software (Anderson and Collingridge, 2001). fEPSPs were continuously recorded under basal stimulation frequencies and LTP was induced only after obtaining stable fEPSP slope values for at least a 15 min. Test drugs were added to the perfusing aCSF at least 30 min before LTP induction, or initiation of the input–output curves. Changes in stimulus frequency (LTP induction) or intensity (input–output curves) were only initiated after at least a 15 min stable baseline in the presence of the drugs.

Long-term potentiation was induced by θ-burst stimulation. Two different stimulation paradigms were used in different experiments, weak-θ-burst and strong-θ-burst protocols, which differed only in the number of trains delivered. The weak-θ-burst consisted of five trains whereas the strong-θ-burst was composed of 10 trains, in both cases the stimulation trains were separated by 200 ms. In both paradigms each train was composed of four stimuli delivered at 100 Hz. LTP magnitude was quantified as the % change in the average fEPSP slopes recorded from 50 to 60 min after LTP induction, taking as 0% the averaged fEPSP slope recorded for 10 min immediately before LTP induction. Throughout the text, while referring to weak LTP or to strong LTP we mean LTP induced by a weak-θ-burst or by a strong-θ-burst, respectively.

The following drugs were used: WIN55,212-2 (WIN) mesylate, AM251, 1,3-dipropyl-8-cyclopentyl-xanthine (DPCPX), picro toxin (PTX), JZL 184, and JZL 195 from Tocris. SR141716A (Rimonabant), tetrahydrolipstatin (Orlistat) from Biogen, URB597 from Cayman Chemicals, barium salt of DL-fluorocitric acid from Sigma-Aldrich.

WIN55,212-2 was used as CB₁R agonist at a concentration (500 nM) 250 times higher than its K_i value for these receptors (Kuster et al., 1993). AM251 was used as a CB₁ receptor inverse agonist at a concentration (1 μM) 100 times higher than its K_i value for these receptors (Lan et al., 1999). Rimonabant was used as a CB₁ receptor antagonist at a concentration (1 μM) 500 times higher than its K_i for this receptor (Rinaldi-Carmona et al., 1994). DPCPX was used as an adenosine A₁ receptor antagonist at a concentration (50 nM) 100 times higher than its K_i value for this receptor (Bruns et al., 1987). PTX was used as a GABA_A receptor antagonist at a concentration (50 μM) 100 times higher than its K_i value this receptor (Mehta and Ticku, 2013). JZL 184 was used as potent and selective monoacylglycerol lipase (MAGL) inhibitor at concentration (1 μM) 125 times higher than its IC₅₀ for this enzyme (Long et al., 2009a). JZL 195 was used as potent inhibitor of both fatty acid amide hydrolase (FAAH) and of MAGL at a concentration (1 μM), respectively, 500 and 250 times higher than its IC₅₀ for these enzymes (Long et al., 2009b). Orlistat was used as a diacylglycerol (DAG) lipase inhibitor at a concentration (10 μM) 100 times higher than the IC₅₀ to inhibit DAG lipases α (Bisogno et al., 2006). URB597 was used as a selective FAAH inhibitor at a concentration (1 μM) 200 times the IC₅₀ to inhibit this enzyme (Kathuria et al., 2003). Care was taken to use drug concentrations within selectivity ranges and according to previously published work using the same drugs for similar purposes. Solutions of all these drugs were prepared as stock solutions in 100% dimethylsulfoxide (DMSO). Aliquots of these stock solutions were diluted in aCSF in the day of the experiment. The concentration of the stock solution was chosen so that the final concentration of DMSO in the perfusion solutions was ≤0.1% (v/v).

Sodium fluorocitrate, an astrocyte metabolism inhibitor (Bonansco et al., 2011), was prepared as described by Paulsen et al. (1987): 8 mg of the barium salt of DL-fluorocitric acid was dissolved in 0.1 M HCl, precipitated by the addition of 0.1 M Na₂SO₄, buffered with 0.1 mM Na₂HPO₄ and centrifuged at 1000 × g for 5 min; the supernatant containing fluorocitrate was added to aCSF at a final concentration of 200 μM (pH 7.4).

Data are expressed as the mean ± SEM; n corresponds to the number of experiments; in each experiment, only one slice was used per drug condition. At least one drug condition and the corresponding control was tested in each experimental day. Statistical significance was assessed by two-tailed Student’s t-test when comparing two groups, or by one-way ANOVA with treatment as the between-subject factor, followed by Sidak’s post hoc test when comparing multiple experimental groups. For the input–output curves, statistical significance was assessed by two-way ANOVA with treatment as the between-subject factor, followed by Sidak’s post hoc test when comparing multiple experimental groups. A p-value of <0.05 was considered to account for significant difference. Analyses were performed with the GraphPad Prism 6 Software.

The first series of experiments was designed to evaluate the influence of eCBs upon weakly induced LTP. The influence of eCBs was assessed by testing the consequences of drugs that prevent CB₁R activation by eCBs or the synthesis of eCBs. We focused upon the synthesis of a predominant eCB at the hippocampus, 2-AG (Piyanova et al., 2015).

In control slices, fEPSP slopes recorded 50–60 min after inducing LTP with a weak-θ-burst, were 26.7 ± 5.5% higher than before LTP induction (n = 18; Figure 1). In slices where the CB₁R inverse agonist, AM251 (1 μM) was added to the perfusion at least 30 min before LTP induction, the magnitude of LTP was 46.5 ± 5.4% (n = 17, t = 2.6, p < 0.05 vs. control, Figure 1), which corresponds to near 80% increase in LTP magnitude. A similar result was obtained in the presence of another CB₁R blocker, the selective CB₁R antagonist, rimonabant (1 μM) (LTP magnitude: 53.5 ± 12.8%, n = 6, t = 2.5, p < 0.05 vs. control, Figure 1). In the presence of Orlistat (10 μM), an inhibitor of DAG lipase, the enzyme responsible for the conversion of DAG into 2-AG, the magnitude of LTP was also enhanced toward 50.7 ± 7.2% (n = 8, t = 2.5, p < 0.05; Figure 1). Importantly, when both CB₁R activation and 2-AG synthesis were prevented together, by the simultaneous presence of AM251 (1 μM) and Orlistat (10 μM) the magnitude of LTP was enhanced at the same degree as obtained with each of the drugs alone (t = 0.3, p > 0.05, Figure 1C). This lack of additivity indicates that both drugs facilitate LTP due to their common ability to prevent eCB signaling.

Summarizing, the above results show that drugs known to prevent the activation of CB₁R by eCBs or drugs known to inhibit the synthesis of 2-AG, the predominant eCB in the hippocampus (Piyanova et al., 2015), cause a marked facilitation of LTP induced by a weak-θ-burst, thus suggesting that eCBs inhibit such form of LTP.

We then assessed the influence of eCBs on LTP induced by a strong-θ-burst protocol, all other experimental conditions being similar to those used before. Fifty–sixty minutes after the strong-θ-burst stimulation, LTP magnitude in control conditions was 68.1 ± 3.7% (n = 22) of pre-θ-burst stimulation. LTP dropped off by around 40% in the presence of AM251 (29.6 ± 6.8%, n = 9, t = 5.1, p < 0.001; Figure 2) or of rimonabant (28.5 ± 7.4%, n = 5, t = 4.2, p < 0.01; Figure 2). In the presence of Orlistat, the magnitude of LTP also decreased toward similar values (30.3 ± 8.4%, n = 5, t = 4.0, p < 0.01; Figure 2). It is worthwhile to note that in what concerns to the inhibition of LTP induced by a strong-θ-burst, the effect of AM251 was also not additive with that of Orlistat. Indeed, when both drugs were present, the LTP magnitude was 38.5 ± 6.4% (n = 6), a value significantly different (t = 0.8, p < 0.05) from that obtained in control conditions, but of similar magnitude as that obtained in the presence of each of the drugs separately (Figure 2). Again, this suggests that the ability of these drugs to inhibit strongly induced LTP results from their common ability to prevent eCB signaling.

Summarizing, the results reported in this section show that drugs known to prevent CB₁R activation by eCBs or to inhibit 2-AG synthesis lead to an inhibition of LTP induced by strong-θ-burst. These data are in clear contrast with what was observed when inducing LTP with a weak-θ-burst, and suggest that LTP induced by a strong-θ-burst is facilitated by eCBs.

The approach described in the previous sections was always directed toward the consequences of preventing CB₁R activation by eCBs. On the light of what is known about the inhibitory action of cannabinoids on neuronal activity, our finding that LTP induced by a strong-θ-burst is reduced by preventing CB₁R activation was unexpected. We thus decided to assess how this form of LTP is affected by continuous activation of CB₁Rs. To do so we used two approaches: (1) test the influence of inhibitors of eCB hydrolysis and in such way create conditions for sustained enhanced levels of eCBs, or to (2) use a CB₁ receptor agonist to exogenously activate CB₁Rs in a sustained way.

Data shown in Figure 3 summarize the findings while using inhibitors of enzymes that prevent hydrolysis of eCBs. When using JZL 184 (1 μM), a selective inhibitor of MAGL, the enzyme that hydrolyses 2-AG, the magnitude of LTP was enhanced toward 92.3 ± 11.2% (n = 9, t = 2.6, p < 0.05 as compared with absence of drugs, Figure 3), corresponding to a value about 40% higher than that obtained in the absence of any drug. This finding suggests that enhancement of the levels of the predominant eCB in the hippocampus, 2-AG (Piyanova et al., 2015), facilitates strong LTP, thus in line with previous results showing that blockade of CB₁R or inhibition of synthesis of 2-AG inhibit strong LTP.

Remarkably, in the presence of URB 597, which at the concentration used (1 μM) inhibits FAAH, but not MAGL, the magnitude of LTP decreased toward 33.3 ± 8.6% (n = 7, t = 3.4, p < 0.05 as compared with absence of drugs, Figure 3), thus toward near half of the value obtained in control conditions. Since FAAH hydrolyses anandamide, this data suggest that accumulation of the non-predominant eCB in the hippocampus, anandamide (Piyanova et al., 2015), inhibits LTP in clear contrast with what occurs with the influence of the most abundant eCB in the hippocampus, 2-AG. This conclusion is further supported by the experiments where a non-selective inhibitor of both enzymes, FAAH and MAGL, was used. Thus, in the presence of JZL 195 (1 μM), the LTP magnitude was decreased toward a value (44.2 ± 8.8%, n = 8, Figure 3) between that obtained with URB 597 and that obtained in the absence of any drug, being not significant different (t = 2.4, p > 0.05) from any of these conditions. Altogether, the data with JZL 184, URB 597, and JZL 195 allow to suggest that enhanced production of 2-AG and enhanced production of anandamide affect strong LTP in an opposed way. However, the possibility that the inhibitory action of URB 597 results from non-CB1-related mechanisms (Kathuria et al., 2003; Ratano et al., 2017) cannot be fully excluded.

Secondly, we tested the effect of WIN (500 nM), a compound known to activate CB₁R. Since the effect of WIN upon synaptic transmission is known to be rather slow (Serpa et al., 2009), the slices were pre-incubated with WIN for at least 60 min before transfer to the acquisition chamber. Then, the slices were stabilized for at least 20 min, LTP being only induced when fEPSP slope values remained stable for at least 15 min. In such experiments LTP was virtually abolished (LTP magnitude: 5.5 ± 10.5%, n = 7, t = 6.5, p < 0.05 vs. pre-LTP induction; p < 0.0001 vs. LTP magnitude in control conditions; Figure 4). This inhibitory effect of WIN was prevented when the slices had been pre-incubated with the CB₁R inverse agonist, AM251, before addition of WIN. Indeed, under such conditions the inhibitory effects of both the agonist and the antagonists seem to be reciprocally canceled since LTP magnitude obtained in slices in the presence of AM251 and WIN (55.2 ± 11.2%, n = 7, Figure 4) was similar (t = 1.3, p > 0.05) to that obtained in the absence of any drug.

Summarizing, the data reported in this section suggest that sustained activation of CB₁Rs induced by adding an exogenous agonist as well as prevention of degradation of the non-prevalent eCB in the hippocampus leads to inhibition of LTP induced by the strong-θ-burst. This is in clear contrast with the conclusions that could be drawn while assessing the action of a drug known to prevent the hydrolysis or prevent the formation of the predominant eCB in the hippocampus as well as when accessing the action of transiently released eCBs by using CB₁R blockers. Altogether, the data indicate that while physiologically released eCBs are required to facilitate LTP induced by a strong θ-burst, non-physiological activation of CB₁R leads to inhibition of this form of LTP.

The above results indicating that physiologically released eCBs can facilitate LTP lead us to hypothesize that the strong-θ-burst could lead to the recruitment of other neuromodulators that would affect the neuromodulatory influence of eCBs. Purines are released during high-frequency neuronal firing (Cunha et al., 1996) and adenosine is known to inhibit LTP through activation of A₁R (De Mendonça and Ribeiro, 2000; Wang et al., 2016), which are abundantly expressed in the hippocampus. Furthermore, A₁R can affect CB₁R signaling (Hoffman et al., 2010; Sousa et al., 2011). To test if the apparent facilitatory action of eCBs upon strong-θ-burst-induced LTP could be due to any interference with endogenous adenosine acting on A₁R, we used two different approaches: genetic (A₁R^-/- mice) or pharmacological (selective A₁R antagonist, DPCPX) prevention of A₁R activity.

The magnitude of LTP induced by the strong-θ-burst in slices from A₁R^-/- mice (65.5 ± 6.7%, n = 13) was not significantly different (t = 0.35, p > 0.05) from that obtained in A₁R^+/+ (68.1 ± 3.7%, n = 22, Figure 5). Remarkably, the CB₁R inverse agonist AM251 inhibited LTP toward a similar value in both genotypes [A₁R^+/+: 29.6 ± 6.8%, n = 9; A₁R^-/-: 25.1 ± 9.3%, n = 8, t = 0.4, p > 0.05 when comparing genotypes; t = 4.1, p < 0.05 when assessing the effect of AM251 in A₁R^-/- (control A₁R^-/- vs. A₁R^-/-: AM251), Figure 5]. These data suggest that the inhibition of strong LTP caused by the inverse agonist of CB₁R does not result from an enhanced A₁R activation by released adenosine. To further confirm this, and to preclude any adaptation-like process due to genetic removal of A₁R, we tested action of AM251 in A₁R^+/+ mice following inhibition of the A₁R with DPCPX. Again, and in spite the presence of DPCPX at a concentration (50 nM) near 100 times its K_i for A₁R (Bruns et al., 1987), thus expected to fully block A₁R signaling, AM251 caused a marked inhibition of LTP induced by the strong-θ-burst (DPCPX: 68.0 ± 9.3%, n = 9; DPCPX+AM251: 32.5 ± 5.2%, n = 7, t = 3.3, p < 0.05, Figure 5).

Altogether the above data seem to indicate that the inhibition of LTP caused by preventing CB₁R activation by eCBs is not due to enhanced activation of A₁R by endogenous adenosine.

The absence of A₁Rs also did not affect the inhibitory action of the CB₁R agonist (500 nM WIN) upon LTP magnitude. Thus, in A₁R^(-/-) mice the magnitude of LTP in the presence of 500 nM WIN (-0.57 ± 9.4, n = 5) was significantly lower (t = 4.1, p < 0.0001) than in the absence of WIN and not different (t = 0.5, p > 0.05) from LTP magnitude in slices from WT mice in the presence of 500 nM WIN.

Astrocytes are known to contribute to the facilitatory action of eCBs upon glutamatergic transmission (Navarrete and Araque, 2010). In addition, it is known that astrocytes, by releasing gliotransmitters, which then act in pre- and post-synaptic receptors, affect neuronal signaling and plasticity (Pascual et al., 2005; Henneberger et al., 2010). We thus hypothesized that the apparent facilitatory action of eCBs upon LTP induced by the strong-θ-burst would involve the astrocytes. To address that possibility we incubated the slices with the metabolic gliotoxin fluorocitrate (200 μM) for at least 20 min and allowed the fEPSP slopes to stabilize for at least 15 min before inducing LTP either in the presence or absence of AM251. As expected from previous reports (Bonansco et al., 2011) LTP magnitude was reduced in slices incubated with fluorocitrate (cf. data in Figure 6 with Figure 2). Remarkably, however, under such conditions the CB₁R inverse agonist, AM251, was still able to markedly inhibit LTP (fluorocitrate: 43.0 ± 13.5%, n = 7; fluorocitrate +AM251: 7.9 ± 15.8%, n = 9, t = 2.3, p < 0.05 vs. fluorocitrate alone; Figure 7).

These data suggest that the apparent facilitatory action of eCBs upon strong-θ-burst-induced LTP does not involve the astrocytes. This data also show that, at least under some experimental conditions, the strength of LTP induction may be even more determinant of the direction of the influence of eCBs upon LTP than the magnitude of LTP itself.

Next we hypothesized that the apparent facilitatory action of eCBs upon LTP caused by the strong-θ-burst stimulation could be due to a preponderant inhibition of GABA release over glutamate release. To test this possibility, experiments were performed in the presence of the GABA_AR antagonist, PTX (50 μM). LTP magnitude was smaller in the presence of PTX, as compared with its absence (cf. Figures 2, 7), which may result from overactivity of glutamatergic transmission even before LTP induction. Remarkably, in slices in the presence of PTX, the inhibitory action of AM251 upon LTP was lost (PTX: 34.3 ± 9.7%, n = 6; PTX+AM251: 31.8 ± 7.7%, n = 5; 0.1, p > 0.05, Figure 7). The ability of PTX to prevent the inhibitory action of AM251 upon LTP should not be attributed to its ability to diminish LTP, since fluorocitrate also inhibited LTP and did not prevent the action of AM251 (cf. Figures 6, 7).

The above data suggest that the apparent facilitatory action of eCBs upon LTP induced by the strong-θ-burst involves GABA_AR-mediated GABAergic transmission, most probably resulting from eCB-induced inhibition of GABA release with consequent disinhibition of glutamatergic neurons.

Long-term potentiation can be influenced by changes in basal synaptic transmission. To evaluate if manipulation of 2-AG signaling could have a global influence upon excitability, input/output (I/O) curves were compared in the absence and presence of AM251, Rimonabant, or Orlistat. As illustrated in Figure 8, none of these drugs appreciably modified I/O curves compared with the control. Exogenous activation of CB₁R using the CB₁R agonist WIN, did however clearly altered the I/O curve, an action that can be attributed to its well-known ability to inhibit synaptic transmission at the CA1 area of the hippocampus (Serpa et al., 2009). The absence of influence of AM251, Rimonabant, or Orlistat in I/O curves but their influence upon LTP, indicates that transiently released 2-AG have a predominant influence over LTP rather than over basal synaptic transmission.