Neuronal Representation of a Working Memory-Based Decision Strategy in the Motor and Prefrontal Cortico-Basal Ganglia Loops

Behavioral results. A, Diagram of the block alternation schedule. The order of left-right reward probabilities was randomized across sessions. CC-L and CC-R indicate that the left and right choices were more rewarded in the CC block, respectively. IC-R and IC-L indicate that the left and right choices were more rewarded in the IC block, respectively. P(L) was the probability of the rat choosing the left side. B, A representative example of rat performance. Blue vertical lines indicate individual choices in choice trials. Red vertical lines indicate no-choice trials. Long lines and short lines represent rewarded and nonrewarded trials, respectively. The green trace in the middle indicates the probability of a left choice in choice trials (average of the last 10 choice trials). C, Averaged learning curves in one sequence of CC (upper) and IC (lower) on all sessions. A sequence consisted of 20 trials, and after 10 trials, reward probabilities were reversed. The vertical axis indicates the frequency of the action associated with the higher reward probability in the first 10 blocks. Filled circles and open circles show that the action frequency was significantly different from 0.5 (p < 0.05; Mann–Whitney U test). D, Distributions of the choice probability of 75% reward side in one session of CC (upper) and IC (lower). The optimal action probability is the frequency of selecting the action associated with the larger reward probability in one session. Medians of both distributions are significantly different from 0.5 (p < 0.01 for CC and IC; Mann–Whitney U test). E, Effects of interaction between past actions and outcomes on the subsequent action. The subsequent action was regressed by action in the previous trial and interaction of actions and outcomes in the past nine trials. **p < 0.01, *p < 0.05, Wilcoxon signed-rank test. F, Win-stay lose-switch (WSLS) indexes. The horizontal axis represents a win-stay index, the frequency that rats selected the same action after the rewarded trial. The vertical axis represents a lose-switch index, the frequency that rats switch the action after the no-reward trial. WSLS indices of CC sessions are plotted with green dots, while indices of IC sessions are shown with pink dots. Vertical lines in histograms indicate the medians of win-stay or lose-switch probabilities in CC and IC. **p < 0.01, Mann–Whitney U test.

Representative activity patterns of neurons during choice trials. A, Nissl-stained coronal sections showing recording locations for the DMS (blue) and DLS (red). B, Tracks of accepted electrode bundles for all rats are indicated by rectangles. Neurons recorded from blue, red, cyan, and magenta rectangles were classified as DMS, DLS, mPFC, and M1 neurons, respectively. Each diagram represents a coronal section referenced to the bregma (Paxinos and Watson, 1998). C–F, Perievent time histograms (PETHs) of a representative DMS neuron (C), DLS neuron (D), mPFC neuron (E), and M1 neuron (F). PETHs were calculated based on timings of five task events (onset of center-hole poking, onset of the tone, offset of center-hole poking, onset of left or right hole-poking, offset of left-hole or right-hole poking), and the following four task phases were defined: phase 1, the period from the start of the center-hole poking to the onset of the cue tone; phase 2, the choice tone presentation period; phase 3, the action execution period between exiting the center hole and entry into the left or right hole; phase 4, the feedback period when a reward or no-reward tone was presented after left-hole or right-hole poking. PETHs of all choice trials (black), and of trials in CC (green) and trials in IC (cyan; upper panel). PETHs of CC and IC choice trials in which left was selected and rewarded (L1), left was selected and not rewarded (L0), right was selected and rewarded (R1), or right was selected and not rewarded (R0; lower panel). All PETHs (50-ms bins) were smoothed with a Gaussian kernel with a 150-ms SD. G–J, Normalized activity patterns of all recorded neurons from the DMS (G), DLS (H), mPFC (I), and M1 (J). An activity pattern for each neuron was normalized so that the maximum PETH was 1 and represented by pseudo-color (values from 0 to 1 are represented from blue to red). Indexes of neurons were sorted based on the time that the normalized PETH first surpassed 0.8.

Proportions of neurons coding actions, rewards, and conditions. A, Meanings of regressors in each task phase. B, Means of time duration in each phase do not differ significantly between CC and IC. Results are means ± SEM; n.s.: p > 0.05 (paired t test). C, D, Proportions of neurons significantly correlated with a regressor for action (C) and reward (D) in each phase. Upward and downward bars indicate proportions of neurons with positive and negative regression coefficients, respectively. **p < 0.01, *p < 0.05, χ² test between the DMS and DLS or mPFC and M1. ##p < 0.01, #p < 0.05, Binominal test. E, Mean regression coefficients of lateralized reward. Mean ± SEM. **p < 0.01, *p < 0.05, paired t test compared with zero.

Neuronal representation of WSLS strategy in the prefrontal and motor cortico-basal ganglia loops. A–D, Proportions of neurons coding actions, rewards, and A×R s of the previous choice trial and coding WSLS and current action in phases 1 (A), 2 (B), 3 (C), and 4 (D). Filled and hatched bars indicate proportions in CC and IC, respectively. When rats employed the WSLS strategy, DLS neurons more strongly conveyed information about the previous choice trial during action preparation than DMS neurons (yellow arrows). Neuronal activities of all areas excluding the mPFC represented WSLS action during action execution (green arrow). **p < 0.01, *p < 0.05, χ² test between DMS and DLS or mPFC and M1 in each task condition. ##p < 0.01, #p < 0.05, χ² test between CC and IC in each recording area. E, The time course mutual information between previous A×R and neuronal firing in 100 ms bins before and after the offset of center-hole poking in the CC. Mean ± SEM. F, The average mutual information between previous A×R and neuronal firing during 500 ms before and after the offset of center-hole poking in the CC. Mean ± SEM; **p < 0.01, *p < 0.05, unpaired t test.