a) Dorsal view of the mouse brain showing the basal ganglia subnuclei. The dorsal striatum (dSTR), globus pallidus external and internal segment (GPe and GPi, respectively), subthalamic nucleus (STN), substantia nigra pars reticulata and pars compacta (SNr and SNc, respectively) are shown in relative sizes. b) Diagram of the connectivity within the striatum between direct and indirect SPN cells, FS cells and cortex (Ctx) projections to the striatum. Inset: relative distribution of each cell-type within the striatum. SPN cells correspond to about 95% of the total, while the rest corresponds to striatal interneurons. c) Example of the response of an dSPN cell to the injection of current of different amplitude for experiments (left) and for the AdEx neuronal model (right). d) Example of a fit of the AdEx model on experimental data for dSPN cells. Panels (a) and (c) adapted from Ref.[4].

Numerical transfer function (a,b,c) and the corresponding semi-analytical approximation fitted from Eq.5 (d,e,f) for each cell-type. Solid lines in panels d,e,f correspond to the firing rates obtained from Eq.5 while filled-circles correspond to the numerical results. The colormaps in the panels a,b,c indicate the output firing rates obtained from each cell-type for the indicated input. Different circle-colors in panels a,b correspond to different input values of νSPN, while in panel c correspond to different input values of νFS. For illustration purposes the panels a,b,d,e correspond to the particular case νFS = 0, nevertheless the fit was performed for the full range in the three variables.

Response of the systems to δ (a), θ (b), β (c), and γ (d) rhythms. Results from the meanfield (solid blue lines) are superimposed to the firing rates obtained from the spiking-network model of the striatum.

Dopaminergic effects on striatal cells. a) Response of the system to dopaminergic driven current affecting dSPN and iSPN cells. The currents considered excitatory (positive) for dSPN cells exhibiting D1 dopamine (DA) receptors and inhibitory (negative) for iSPN cells exhibiting D2 receptors. We show the raster plot obtained from network simulations and the corresponding mean firing rates obtained from the network and from the mean-field for dSPN and iSPN cells. The dopaminergic current is simulated as a square function. b) Response of the system to a transient change of synaptic conductance. As in panel (a), the change is considered positive for dSPN cells and negative for iSPN cells. In this case the variation in conductance is modeled as a gaussian function on time. c) Response of the system to a combined DA current and conductance variation accounting for the multiple effects of dopamine in SPN cells.

Basic implementation of reinforcement learning (RL) with the mean-field model of the striatal microcircuit. a) Diagram of the RL model. Two different sensory stimulus (A and B) are represented by two independent cortical projections towards the striatum. The output of dSPN and iSPN cells are used to decide between two possible actions (Action 1 and Action 2) according to Eq.10. The resulting action will generate a reward or punishment dopaminergic signal (DA) which can modify corticostriatal synaptic weights according to Eq.11. b) Simulation of the RL process. We show the cortical activity (Stim. A and Stim.B, bottom panel), the dSPN, iSPN (red and blue lines respectively, second plot from the top) and the FS response. On the right column we show the detailed for timeperiod indicated with the squared area (between 8-to-10 seconds approximately). We can see that by the end the simulations the dSPN and iSPN cells respond selectively to each stimulus type. c) Action selection for each stimulus presentation. During the first trials the system is incapable of distinguishing the correct action to take under each stimulus type. By the end of the simulation, the system can correctly select the Action 1 for Stim.A (magenta circles) and Action 2 for Stim.B (cyan circles).