In the flexible value procedure (Figure 1A), the saccade to one object was followed by a reward and the other object was associated with no reward, and this contingency

was reversed frequently. To examine the short-term behavioral learning, we measured the target acquisition time after a go cue (the disappearance of the fixation point). As the value of each object changed blockwise, the target acquisition time changed accordingly: the monkeys made saccades more quickly to the high-valued object Dabrafenib clinical trial than the low-valued object (Figure 1B) (difference of target acquisition time: 57.7 ms, p < 0.001, two-tailed t test). On choice trials (see Experimental Procedures), the monkeys mostly chose the high-valued object (average: 83.9% ± 0.8%). These saccades can be called “controlled saccades,” because they were controlled by reinforcing feedbacks delivered just after the saccades. During learning of stable values (Figure 1C), the saccades to a set of objects were always followed by a reward (high valued) and the saccades to a different set of objects were always followed by no reward (low valued), and this was repeated across days (see Figure S2 for detail). To examine the long-term behavioral memory, we used a free-looking task (Figure 1D) and a free-viewing procedure (Figure S2D). These tests were done at least 1 day after the learning session, and the saccades were followed by no reward. Yet, see more the monkey made saccades to the objects

automatically and did so more likely to high-valued objects than low-valued objects. The preference to the high-valued objects emerged slowly across several daily learning sessions and then remained stable after four daily sessions of learning (Figure S2D), as reported previously (Yasuda et al., 2012). Therefore, to analyze the neuronal and behavioral coding of stable object values, we used fractal objects that the monkey had learned for more than four daily sessions. Cytidine deaminase When such well-learned objects were used in the free-looking task, the likelihood of saccades to high-valued

objects was significantly higher than to low-valued ones (Figure 1D, right) (difference of automatic looking: 18.9%, p < 0.01, two-tailed t test). These saccades can be called “automatic saccades,” because they were not followed by any reinforcing feedbacks delivered just after the saccades. To test whether the caudate nucleus controls the saccade behavior to choose high-valued objects, we recorded spike activity of single neurons in the caudate nucleus using the flexible and stable value procedures. We first found that many neurons in the caudate nucleus responded to visual objects, confirming previous studies (Brown et al., 1995, Caan et al., 1984, Rolls et al., 1983 and Yamamoto et al., 2012). The ratios of neurons that responded to fractal objects relative to the encountered neurons in the three caudate regions (Figure 3A) were: head 163/845 (19.3%), body 109/381 (28.6%), and tail 107/205 (52.2%).