Here we present results supporting a new integrative role for dopamine and eCBs in modulating BG output. Using electrophysiology, optogenetics, and modeling, we presented evidence suggesting interdependence between the direct, indirect, and hyperdirect afferents to the EP. We demonstrated that eCB modulates synaptic strength of the indirect pathway (Figure 1), the direct pathway (Figure 2), and glutamatergic (Figure 3) inputs to the EP, and that dopamine modulates eCB induced plasticity of all inputs to the EP (Figure 5). We also showed that eCBs diffuse, modulating synaptic strength of neighboring EP neurons (Figure 4). Thus, despite the lack of axon collaterals, information is transferred between neurons in the EP via endocannabinoid diffusion. Our numerical simulations demonstrate that the combination of STP and eCB mediated long-term plasticity enhances the transmission of direct pathway information while reducing indirect and hyper-direct pathway information. Thus, our results propose that the passage of BG information through the bottleneck of the EP allows modulation of the information using small neuronal populations. Our data suggest that this cellular mechanism controls the balance between direct and indirect pathway information in the EP output.

eCB mediated synaptic plasticity is critically involved in neural functions ranging from homeostasis to cognition (Katona and Freund, 2012). Behavioral adaptations rely on changes in synaptic strength and the prevalence of eCB-mediated long-term depression (eCB-LTD) at synapses (Castillo et al., 2012). eCBs have been identified as triggers for short and long-term plasticity at synapses throughout the brain. There is a notable expression of CB1Rs in the basal ganglia, mainly in the Striatum, GP, EP, and SNr (Herkenham et al., 1990). Activation of the CB1R modulates LTD and also short-term plasticity (Llano et al., 1991; Pitler and Alger, 1992; Morishita and Alger, 2000; Kreitzer and Regehr, 2001; Ohno-Shosaku et al., 2001; Wilson and Nicoll, 2001). This modulation of short-term plasticity appears in several basal ganglia regions including the striatum (Kreitzer and Malenka, 2005; Freiman et al., 2006; Narushima et al., 2006a,b, 2007; Yin and Lovinger, 2006; Centonze et al., 2007; Uchigashima et al., 2007; Maccarrone et al., 2008), GP (Engler, 2005), SNr (Yanovsky et al., 2003; Engler, 2005; Szabo et al., 2006), SNc (Szabo et al., 2006), and NAc (Rancz and Ha, 2006). Furthermore, eCB induced LTD has been reported in several brain regions such as the dorsal striatum (Kreitzer and Malenka, 2005, 2007; Ronesi and Lovinger, 2005; Wang et al., 2006), NAc (Robbe et al., 2002; Mato et al., 2008), the cortex (Panikashvili et al., 2005; Nevian and Sakmann, 2006; Lafourcade et al., 2007), the cerebellum (Tzounopoulos et al., 2007), amygdala (Marsicano et al., 2002; Azad et al., 2004), and the hippocampus (Chevaleyre and Castillo, 2003, 2004; Chevaleyre et al., 2007).

Here we demonstrated that glutamatergic input to the EP exhibit eCB-LTD that is induced by post-synaptic firing. The eCB induced LTD reported here shares similarities with striatal LTD. HFS of corticostriatal glutamatergic inputs to SPNs is known to induce LTD that requires post-synaptic Ca²⁺ elevation and leads to a decrease in the probability of glutamate release (Calabresi et al., 1992, 1994; Choi and Lovinger, 1997; Kreitzer and Malenka, 2005). Striatal LTD is dependent on group I mGluRs and L-type Ca²⁺ channels. Ca²⁺ elevation by an LTD induction protocol induces eCB release. Results presented here suggest a calcium dependent LTD. The depression level of the STN-EP synapses depends on the frequency of the APs in the EP neurons. This matches the AP frequency dependence of Ca²⁺ concentration in the soma and the dendrite of the EP neurons (Gorodetski et al., 2018), suggesting that the observed eCB-LTD is indeed calcium-dependent. Post-synaptic release of eCBs is sufficient to induce CB1R-mediated depression at GABAergic synapses (Adermark et al., 2009). Similarly, the synaptic depression reported here is blocked by AM-251, implicating CB1R in this LTD. On the other hand, in the direct pathway, Str-EP synapses exhibit eCB-LTP that also depends on the frequency of the APs and is blocked by CB1Rs antagonist. An eCB-LTP also has been observed in the dorsal striatum (Cui et al., 2015). These similarities suggest that HFS induced by high frequency firing of STN neurons could induce synaptic plasticity.

After identifying the eCBs effect on the EP pathways, we demonstrated that dopamine modulates the eCB induced plasticity (Figure 5). Striatal evoked IPSCs exhibit short-term facilitation that is modulated by dopamine via D1LR, which are co-localized with striatal axon terminals (Lavian et al., 2018). Furthermore, GP evoked IPSCs that show short-term depression are modulated by dopamine via D2LR, which are co-localized with GP axon terminals (Lavian and Korngreen, 2016; Lavian et al., 2017). In the current study, we showed, both pharmacologically and optogenetically, that eCB effects require D1R (at Str-EP synapses) or D2R (at STN-EP synapses). These results indicate that DA modulates the generation of eCB-LTD by synaptic activity and comply with previous work implicating dopamine modulation of state-dependent eCB release in the striatum (Kreitzer and Malenka, 2005, 2007). However, in the GP-EP synapses, the block of D2R resulted in LTP instead of LTD, which differs from corticostriatal synapses (Xu et al., 2018), where eCB-LTP requires D2Rs. Our observations, suggests that these G protein coupled dopamine receptors are interacting with eCB production pathways, as has been described in the striatum, though the molecule mechanisms in the EP have not yet been characterized.

Finally, we demonstrate that eCBs diffuse and modulate synaptic plasticity in neighboring neurons (Figure 4). The diffusion of eCBs implies that all three types of synapses likely undergo plasticity at the same time. Moreover, nearby neurons that do not fire at high frequency may still experience modulation of synaptic inputs. These results suggest that the EP is not only a feedforward nucleus (Parent et al., 2000), despite containing no axon collaterals, but also has lateral interactions. In other words, high frequency STN inputs could trigger EP firing and heterosynaptic plasticity of GP or Str synapses, modulating the entire recombination process of direct and indirect pathways to the EP.

We created a data-driven model of EP neurons, and measured the response to in vivo like inputs to Str, STN, and GPe synapses simultaneously, to evaluate the functional effect of the simultaneous plasticity of all three synapses, as might occur with high frequency STN input and eCB diffusion. Our modeling results show that eCB mediated long term plasticity enhances information transmission from the direct pathway, compared to indirect and hyper-direct pathways, confirming that eCB-mediated plasticity controls the balance of information transmission through the basal ganglia, and thus may directly influence decision making behavior. We further show that the eCB-mediated enhanced direct pathway information is eliminated when dopamine is blocked, confirming that dopamine is critical for controlling basal ganglia output directly. Many models of the basal ganglia have evaluated the role of dopamine in action selection. In most of these models (Humphries et al., 2006; Leblois et al., 2006; Lindahl and Kotaleski, 2016), dopamine influences direct vs. indirect pathway information by its effect on striatal activity—either excitability or corticostriatal synaptic plasticity. One model implemented short term depression of inputs to both GPe and SNr (another BG output region) (Lindahl and Kotaleski, 2016); however, the contribution SNr input modulation to action selection was not evaluated. Several network models evaluated the effect of dopamine on GPi (primate analog of EP) oscillatory firing, to investigate mechanisms underlying Parkinson's beta oscillations and normalization by DBS (Hahn and McIntyre, 2010; Humphries and Gurney, 2012; Kumaravelu et al., 2016). In these models, dopamine modulates striatal or GPe neurons or synapses, and beta oscillations in the GPi are a readout of basal ganglia network state. In contrast, our model revealed a direct effect of dopamine on beta oscillations. Specifically, the model exhibits a small increase in power at beta frequency, produced solely by dopamine mediated changes in synaptic plasticity of EP neurons. Thus, this suggests that lack of dopamine in the GPi can contribute to the production of beta oscillations in Parkinson's.

Overall, our results suggest the EP has two modes of operation, based on the anatomical and functional polarity of EP neurons. In one mode, GABAergic inputs from the GP act, due to their short-term depression kinetics and high baseline firing rates, as somatic shunting inhibition delivering an almost constant inhibition to the soma (Bugaysen et al., 2013; Lavian and Korngreen, 2019). The second mode occurs periodically during brief periods of striatal activation by cortical and thalamic inputs. The consequent amplification of dendritic inputs produces phase locking of EP firing to that of the Str (Lavian and Korngreen, 2016; Lavian et al., 2017), overcomes the somatic shunt, and allows short bursts of Str firing to affect the firing of neurons in the EP. It is possible to visualize GABAergic neurons in the EP as having a somatic obstacle surmounted by a dendritic amplifier. It is worth noting that neurons in the EP fire spontaneously over a wide range of frequencies. Figures 1D, 2D, 3D show a functional relationship between the firing frequency and the change to synaptic strength. Thus, the spontaneous firing of EP neurons probably modulates synaptic strength continuously. We induced plasticity using a 10-s train of action potentials ranging between 10 and 100 Hz covering the entire range of firing for rodent EP neurons, with 100 Hz being an upper boundary for EP firing rate in rodent. Furthrmore, it is likely that STN inputs would be able to produce synaptic plasticity at lower firing frequencies because the concomitant activation of group 1 mGluRs would facilitate eCB production (Covey et al., 2017).

The observation that dopamine interacts with eCBs to modulate all synaptic inputs to the EP has highly significant implications. In the presence of dopamine, an increase in EP firing (probably due to glutamatergic input from the STN) lowers the somatic barrier (eCB-LTD in GP-EP synapse) while boosting the dendritic amplifier (eCB-LTP in Str-EP synapse). In other words, a burst of dopamine plus STN inputs will increase the weight of input from the direct pathway while lowering that from the indirect pathway (Figure 10); thereby allowing Str to control EP firing for a longer duration than predicted by STP alone. Moreover, increased EP firing lowers the weight of glutamatergic input to the EP, further increasing the control of EP output by input from the striatum. The concomitant release and diffusion of eCBs may modulate neighboring synapses on the same neuron or on nearby neurons. This spatial effect may lower the somatic barrier even more while boosting the dendritic amplifier for a population of EP neurons. The net result of this cellular mechanism for “selection” is to determine whether information from the direct or indirect pathway will dominate EP output. These results transform the prevailing view of the entopeduncular nucleus as a feedforward “relay” nucleus to an intricate control unit, which may play a vital role in the process of action selection.