In rodent neocortex, we report here the existence of a Hebbian coincidence-activity dependent LTP induced by a low number of pairings (~10), which involved the eCB system. eCB-tLTP induction relies on activation of CB₁R and TRPV1 triggered by coupled rises of calcium mediated by mGluR5 (via PLCß, diacylglycerol lipase-α activation and calcium released from internal stores) and VSCCs. Most of the steps of eCB synthesis and release (mainly 2-AG and anandamide) tightly depend on postsynaptic calcium levels (and time course; Piomelli et al., 2007; Di Marzo, 2008; Kano et al., 2009; Alger and Kim, 2011). Due to their on-demand intercellular signaling (Piomelli et al., 2007; Alger and Kim, 2011), eCB action is expected to be controlled by precisely timed stimuli. Here we show that STDP, a Hebbian synaptic learning rule (Dan and Poo, 2006; Sjöström et al., 2008; Feldman, 2012; Froemke, 2015), efficiently triggers eCB signaling, even for a low number of pairings, and can promote the expression of eCB-tLTP. Therefore, in addition, to the widespread eCB-LTD, eCBs can aslo mediate potentiation. Bidirectionality of eCB-plasticity is a crucial property of eCB system because it allows eCB-LTP and -LTD to reverse each other at a single synapse. We have previously proposed a mechanism accounting for corticostriatal eCB-tLTP using a combination of patch-clamp recordings and a mathematical model (Cui et al., 2016). In our model, low to moderate peak levels of eCB would lead to tLTD whereas high eCB levels would yield tLTP. It is thus expected that the first 10 pairings produce large peak levels of eCB synthesis, thus inducing tLTP. If the amplitude of the 2-AG peaks decreases for subsequent pairings, this initial tLTP would be de-potentiated; its expression would be thus restricted for few coincident activity and eCB would have no significant impact on tLTP induced with 100 pairings. Supporting this, it has been reported that in the somatosensory cortex NMDAR-tLTP expression (induced with 100 pre-post pairings) was not modified after the inhibition of CB₁Rs (Bender V. A. et al., 2006; Nevian and Sakmann, 2006).

We observed eCB-tLTP in both somatosensory cortex and in dorsolateral striatum, demonstrating that this form of plasticity is not restricted to a single type of synapses, but could serve as a general system in various brain structures to allow the engram of salient events from few spikes. Neocortical eCB-tLTP is similar to those we recently described in the dorsal striatum (Cui et al., 2015, 2016) with the noticeable difference that this new form of plasticity displays, a distinct polarity depending on its synaptic site of expression: Hebbian in neocortex and anti-Hebbian in striatum. Interestingly, eCB-tLTP only requires a single postsynaptic spike per paired stimulation in the dorsal striatum but 2–3 in the neocortex, possibly consistent with the general observation that learning dynamics are usually faster in sub-cortical structures than in the neocortex (Pasupathy and Miller, 2005). It remains to be investigated whether neuromodulators control eCB-tLTP expression and/or polarity, as reported for dopamine for the control of the NMDAR-tLTP in prefrontal or visual cortex (Seol et al., 2007; Xu and Yao, 2010). Furthermore, it has been reported that in a STDP paradigm noradrenaline and serotonin transform eligibility traces into plasticity in the visual cortex (He et al., 2015). In the same line, it is important to explore whether neuromodulators could promote the emergence of eCB-tLTP for lower number of pairings or with a single backpropagating action potential (instead of 2–3 as it is the case in the present study).

eCB system is deeply involved in learning and memory (Mechoulam and Parker, 2013) because its major role in synaptic plasticity expression (Augustin and Lovinger, 2018). Indeed, eCBs (2-AG and anandamide) have been extensively reported to mediate short- or long-term depression, via the activation of CB₁R (Kano et al., 2009; Castillo et al., 2012; Katona and Freund, 2012; Melis et al., 2014) or TRPV1 (Gibson et al., 2008; Maione et al., 2009; Chávez et al., 2010; Grueter et al., 2010; Puente et al., 2011). Nevertheless, it exists now as a growing body of evidence showing an indirect role of eCBs in promoting short- or LTP (reviewed in: Araque et al., 2017; Augustin and Lovinger, 2018): at mixed synapses of the goldfish Mauthner cell via intermediary dopaminergic neurons (Cachope et al., 2007) or at CA1 synapses in hippocampus via a GABA_A receptor-mediated mechanism (Lin et al., 2011; Xu et al., 2012). It has also been reported in the hippocampus that eCB-induced presynaptic depression of GABAergic transmission facilitates LTP (Carlson et al., 2002; Chevaleyre and Castillo, 2004; Zhu and Lovinger, 2007), or that eCBs mediate heterosynaptic short-term potentiation via intermediary astrocytes (Navarrete and Araque, 2010). More recently, it has been reported a direct role of eCBs in promoting LTP at cortical inputs to the granule cells of the dentate gyrus (Wang et al., 2016, 2018), to the medium-sized spiny neurons of the dorsolateral striatum (Cui et al., 2015, 2016) or to the basal dendrites of layer 5 pyramidal cells (in this later case the eCB-LTP is also BDNF- and NMDAR-mediated; Maglio et al., 2018). At hippocampal CA1 synapses, eCB-mediated LTP induced with high-frequency (Lin et al., 2011), low-frequency (Zhu and Lovinger, 2007) or paired (Xu et al., 2012) stimulations were prevented by inhibition of CB₁R and GABA_A receptors. Here, we show at neocortical synapses that GABA, which does not affect EPSC amplitude during baseline, controls the plasticity magnitude by exerting a brake on eCB-tLTP. In line with previous studies showing the existence of eCB-LTP at striatal (Cui et al., 2015, 2016), hippocampal (Wang et al., 2016, 2018) and cortical (Maglio et al., 2018) synapses, the present neocortical eCB-tLTP constitutes an example of a paired-activity eCB-LTP with a direct implication of eCBs (see i-THL experiments) in the induction of tLTP of the stimulated synapse itself. It remains to be determined whether eCB-tLTP expression could be extended to other brain structures and synapses.

Evidence for TRPV1 activation by physiological neuronal activity remains unclear. In rodent neocortex, it is fair to say that the expression of TRPV1 in pyramidal cells was debated. Indeed, although an immunoelectronmicroscopy study reported TRPV1 protein in neocortex at the postsynaptic dendritic spines of pyramidal cells (Tóth et al., 2005), a multi-approach investigation (in situ hybridization, calcium imaging) detects TRPV1 only at the level of the arteriolar smooth muscle cells (Cavanaugh et al., 2011). Recent studies using real-time PCR and western blot (Huang et al., 2014) and patch-clamp recordings (Pezzoli et al., 2014) showed expression of TRPV1 by pyramidal cells of the neocortex. Here, EPSC recordings, before and after TRPV1 agonist (capsaicin) application, show a marked decrease of EPSC amplitude in presence of capsaicin, in line with previous observations (Pezzoli et al., 2014). In addition, we found that two competitive TRPV1 antagonists prevent eCB-tLTP expression. As previously described in striatum (Cui et al., 2015), our study confirms that STDP efficiently triggers eCB signaling and is able to recruit the TRPV1 signaling pathway. TRPV1 is a cationic channel highly permeable to calcium (Starowicz et al., 2007; Di Marzo, 2008) and may contribute to eCB-tLTP induction by boosting the calcium transients. Here, TRPV1 is most likely activated by anandamide arising from the stimulated postsynaptic cell. In the dorsolateral striatum, anandamide appears necessary but not sufficient for tLTP induction (Cui et al., 2016). In the neocortex, it remains to be determined whether anandamide alone could trigger eCB-tLTP induction following limited occurrences of coincident activity. As described for eCB-mediated LTD (Puente et al., 2011), our results illustrate the bidirectionality of eCBs as a system exhibiting polymodal activation through CB₁R and TRPV1, to induce LTD and LTP.

The neocortex receives a broad range of cortical activity patterns, from isolated trains of few spikes to sustained bursting events. At neocortical layer 4-2/3 synapses, STDP exhibits symmetric Hebbian tLTP during the first two postnatal weeks (Itami and Kimura, 2012), then asymmetric Hebbian STDP (Feldman, 2000; Sjöström et al., 2001, 2003; Bender K. J. et al., 2006; Bender V. A. et al., 2006; Froemke et al., 2006; Nevian and Sakmann, 2006; Banerjee et al., 2014); a lack of tLTD has been reported for older animals (Banerjee et al., 2009 but see Min and Nevian, 2012). It should be noted that depending on the site of stimulation (layer 4 or at within layer 2/3), the temporal window and the locus of NMDAR involved of tLTD are different: for post-pre pairings, stimulation in layer 4 induced a tLTD dependent on presynaptic NMDAR which is expressed in a broad Δt_STDP, whereas stimulation within layer 2/3 triggered a tLTD dependent on postsynaptic NMDAR which is expressed in a more restricted Δt_STDP (Banerjee et al., 2014). Although cortical plasticity under prolonged activation (low- and high-frequency stimulations, theta bursts or 100 pairings STDP) is well elucidated, its expression in response to few spikes remained elusive. Nevertheless, it has been observed that dendritic spike(s) induced by single-shock and single-burst were responsible for, respectively, LTD in visual cortex (Holthoff et al., 2004) and LTP in hippocampus (Remy and Spruston, 2007). These both plasticity was NMDAR-mediated and could thus account for single-trial learning. eCB-LTP is promoted by about 10 of pairings, allowing for the synapses to react to the first occurrences of incoming activity. In the same line, ~20–25 STDP pairings induced LTP in hippocampal neurons (Zhang et al., 2009) or at layer 2/3 cortical pyramidal cells (Froemke et al., 2006); interestingly, in hippocampal neurons bath-application of dopamine allows the induction of tLTP with a lower number of pairings (10 instead of 20 pairings; Zhang et al., 2009). Associative memories and behavioral rules can be learned with few trials (5–10) or even with one trial (Schultz et al., 2003; Pasupathy and Miller, 2005; Armstrong et al., 2006; Rutishauser et al., 2006; Whitlock et al., 2006; Tse et al., 2007; Quilodran et al., 2008; Cook and Fagot, 2009; Ito and Doya, 2009; Izquierdo et al., 2016). Upon behaviorally pertinent events, neurons with behavior-related activities fire a few spikes during each trial (one to a dozen) upon each trial (i.e., a discharge at frequency 5–10 Hz during 0.1–0.5 s; Schultz et al., 2003; Pasupathy and Miller, 2005; Quilodran et al., 2008). This suggests that a low number (2–50) of spikes should be sufficient for the expression of synaptic plasticity. The present results suggest that eCB-tLTP could be involved for learning salient events from a low number of action potentials and may constitute a neuronal substrate for single-trial or online learning, such as cortical episodic memory. It remains to be investigated if eCB-tLTP occurs in vivo and to evaluate the involvement of eCB-tLTP in the initial phases of online learning, which could be thereafter reinforced by NMDAR-LTP when stimuli are subsequently repeated.

Frequent cannabis use leads to impairment of working memory in the left superior parietal cortex (Jager et al., 2006) as well as long-term memory via activation of CB₁R (Mechoulam and Parker, 2013; Augustin and Lovinger, 2018). This impairment was mainly interpreted as the effect of cannabinoids on the induction of short- and long-term synaptic depression. Our results, in line with other studies (Lin et al., 2011; Xu et al., 2012; Cui et al., 2015; Wang et al., 2016; Maglio et al., 2018), indicate that potentiation of synaptic transmission may also be involved in the effects of marijuana intoxication.