Injection of a dopamine type 2 receptor antagonist into the dorsal striatum disrupts choices driven by previous outcomes, but not perceptual inference

Abstract

Decisions are often driven by a combination of immediate perception and previous experience. In this study, we investigated how these two sources of information are integrated and the neural systems that mediate this process. Specifically, we injected a dopamine type 1 antagonist (D1A; SCH23390) or a dopamine type 2 antagonist (D2A; eticlopride) into the dorsal striatum while macaques performed a task in which their choices were driven by perceptual inference and/or reinforcement of past choices. We found that the D2A affected choices based on previous outcomes. However, there were no effects of the D2A on choices driven by perceptual inference. We found that the D1A did not affect perceptual inference or reinforcement learning. Finally, a Bayesian model applied to the results suggested that the D2A may be increasing noise in the striatal representation of value, perhaps by disrupting the striatal population that normally represents value.

Keywords: Parkinson's disease; action value; dorsal striatum; neuromodulation; reinforcement learning; sequential decision making.

Figure 1.

Task. A, Events in a single trial. The beginning of a new trial was indicated by a green dot at the center of the initial fixation frame. A stimulus with blue and red pixels indicated the correct decision to the peripheral target matching the dominant color of the stimulus. The inset shows an example of a single frame from the stimulus. B, A monkey was trained to execute eight possible sequences of saccadic eye movements. Each movement occurred in at least two sequences. S1, S2, etc., indicate sequence 1, sequence 2, etc. C, In the fixed condition (top), the sequence of eye movements was fixed for eight correct trials and then switched to a new sequence and remained fixed again. In the random condition (bottom), the sequence changed every trial. D, MRI of the anterior portion of the macaque brain with approximate bilateral injection areas. The red squares indicate the location of the chambers. Blue, green, and red lines indicate injection cannulae, electrodes in PFC, and electrodes in dStr, respectively. E, Locations of the injection sites for saline (green dots and circles), the D2R antagonist (red dots and circles), and the D1R antagonist (black dots and circles). The dots show the center points of the four cannulae for each session. The shaded circles show approximate injected areas. Injection locations were symmetric in the two hemispheres, so we only show them for one hemisphere. The coronal section is taken from +29 mm anteroposterior (AP) (center of the chamber was +28 AP, 17 medial-lateral). Four cannulae were used for each hemisphere. For one of the sessions, the locations of the four individual cannulae are shown in blue to show the orientation. The cannulae were spaced by ∼2.5 mm dorsoventral and 2.7 mm mediolateral.

Figure 2.

Effect of bilateral injection of saline, D2R antagonist eticlopride, and D1R antagonist SCH23390 in dStr. Heavy lines are means. Each shaded region is 1 SEM with n given by the number of sessions from all animals for each condition (see Materials and Methods). A, Fraction of correct decisions as a function color bias for fixed and random conditions for the presaline injection (blue, red) and postsaline injection (green, pink) conditions. B, Reaction times as a function of color bias for the random condition for the presaline injection (red) and postsaline injection (pink) conditions. C, Fraction of correct decisions as a function of trials after switch for fixed conditions. D, Reaction time as a function of trials after switch for fixed conditions. Data shown here are averaged across color bias conditions. E–H, Same as A–D for the D₂ antagonist. I–L, Same as A–D for performance in the D₁ antagonist condition.

Figure 3.

Model fits for perceptual inference and reinforcement learning. A, Fraction correct in the random condition, for each color bias (CB), as a function of time after stimulus presentation (time 0). Bin size, 50 ms. Dotted lines are data, and solid lines are model fits. B, Before (Etic Pre) and after (Etic Post) action values in the fixed condition as a function of trials after switch. C, Inverse temperature (β parameter) before and after injection for the D₂ antagonist. One point was excluded for plotting (before, 2.21; after, 175.33). D, Gamma parameter representing weight of color bias in the fixed condition. E, Learning rate parameter for positive feedback. F, Learning rate parameter for negative feedback.

Figure 4.

Behavioral performance and model prediction. Data plotted are observed and predicted fraction corrected across trials after switch. Color indicates the color bias level, as in C. All data are from D2A sessions. Pre–post injections are indicated on the corresponding plots. A, Scatter plot of choice accuracy before injection for the model and data. The inset shows the estimated variance of the noise on value integration. B, Scatter plot of choice accuracy after injection for the model and data. C, D, Same data as plotted in A and B, plotted as a function of trials after switch, broken out by the color bias level.