Neuronal encoding of ranked items in primate prefrontal cortex during different phases of transitive inference task

Problem solving based on deduction using logical reasoning is one of the aspects of intelligence in humans and non-human primates. Provided with the knowledge that A > B and B > C, the subject is able to infer that A > C. This ability to link the partially overlapping information and extend it to deduce a novel relationship is the basis of a Transitive Inference (TI) capability. Previous neurophysiological studies have described the pattern of neuronal activity from the Prefrontal Cortex being modulated while an abstract mental schema of ranked items is accessed during the inferential reasoning test phase. However, the question of how the neuronal encoding of the individual items subtending this representation is shaped by the learning of higher reciprocal relationships is relatively unexplored. In this study, we aimed to answer this question by investigating the single-neuron activity recorded from the dorsolateral PFC of 2 monkeys that learned the rank ordered series of items as (A>B>C>D>E>F) for solving TI problems. Each session was organized around two consequential Learning phases, where the relationship between the adjacent items (e.g. A>B, B>C) of the series were first learned separately and then intermingled, and a Test phase where inferential pairs were included. In each of the experimental sessions, we first searched for a signature of the acquisition of the mental schema in the monkey's performance during Inference. This was behaviorally assessed by observing a symbolic distance effect in each session, which characterizes the comparisons between items with greater rank difference as easier, than the ones with smaller rank differences, subtending a comparison between their positions on the mental schema. Then, we studied the neuronal activity of 80 neurons recorded in these sessions while the monkeys fixated each item of the series to define a tuning curve of activity of these neurons during the Test Phase. The tuning activity from the same neurons was then observed during the different Learning phases of the task to study if and how the changed with the acquisition of the items rank order. We observed that the progression of the task from Learning to Inference led to significantly different activity patterns in 50% of the studied neurons. A greater majority (43%) exhibited different tuning preferences for the series items, while only about 7% cells mirrored the previous tuning preferences but with different magnitudes of activity. These results emphasize the involvement of PFC neurons in the learning phases of the TI task by reorganization of rank ordered activity.