The neural basis of goal-directed action in decision-making is theorized to incorporate the cognitive representations of external stimuli; the representations of future possibilities and the individual value assigned to each prospective outcome. In complex multimodal scenarios the same set of external stimuli may be represented in multiple ways, some of which more efficient for achieving the goal than others. For example, in spatial navigation, there are numerous ways by which space may be represented. It is believed that both the cognitive representation of space and of future possibilities are implemented in the hippocampus as a cognitive map. And yet, factors that determine which of the myriad possible representations will be learned and implemented by the hippocampus in a cognitive map to guide behavior are still unknown. Therefore, we devised two tasks in which the reward rule was manipulated to require the representation of different sets of stimuli: In the first manipulation, we recorded the activity of CA1 neurons of rats performing a spatial navigation task where olfactory cues and spatial dimensions were selectively altered to predict rewarding outcomes. We found that when an olfactory rule governed reward, neuronal activity gradually shifted to represent olfactory coordinates. In contrast, when the reward rule was a spatial one, a gradual shift in neuronal representation occured towards better encoding of spatial coordinates. This reorganization of representation in CA1 preceded marked behavioral improvement, suggesting that the critical step in learning is downstream of CA1 rearrangement. These findings reveal that as learning progresses, CA1 place cell activity rearranges to encode those aspects of experience that are most relevant for goal-directed behavior. In the second manipulation, we trained mice in a complex odor discrimination task in a four- armed maze with an attentinal set-shifting paradigm. In this task, the reward rule was associated with one of two odors, which were presented in random positions in the maze. The correct arm choice in the maze was irrespective of the spatial dimensions of the task. After initial learning was completed, we introduced an interdimensional shift (ID) by using novel pairs of reward-relevant odors. We compared the learning performance of two mouse lines expressing different levels of the kainate receptor subunit, GluK2. GluK2 is a kainate type glutamate receptor with metabotrophic characterictics, whose disfunction has been implicated in humans with cognitive disabilities (Motazacker et al., 2007). The GluK2 wildtype mice successfully learned to associate a particular odor with the rewarding outcome and reached asymptomatic levels after 10 days of training. Moreover, they applied the reward rule to a different set of odors (ID shift) and reached asymptomatic levels after 5 days of training. In contrast, GluK2 knockout mice could only perform at chance levels. These data show that, mice are capable of associating olfactory information with rewarding outcomes. Furthermore, they can learn a rule and apply it to similar settings by forming an attentional set shift. These findings show that intact cortico-hippocampal processing as well as unimpaired Kainate receptor function in those structures are crucial in storing and recalling reward relevant information for goal-directed behavior.
