Ranking and serial thinking: evidence of a geometrical solution in premotor cortex

Serial thinking is a cognitive function underpinning almost any of our daily actions. Our brain continuously encodes and learns ordered sequences that we can remember, process and replay as in motor planning and language production. Despite the pervasiveness of such computation, a full understanding of the neuronal machinery underpinning this function is still lacking. Psychophysical evidence suggests that the ordering of learned information can be solved resorting to a linear mental workspace, namely the mental line, in which the relevant items are mapped according to their arbitrary assigned rank. Here, we prove that a “geometric” mental line (GML) exists and it can be encoded as a suited linear combination of the item representations in the neuronal state space of cortical networks, as long as their independence can be assumed. In associative cortices, such independence is known to be guaranteed by quasi-randomly projecting sensory information in the high-dimensional space of neuronal activity. Accordingly, we found that abstract symbols involved in a transitive inference task, were represented in the dorsal premotor (PMd) cortex of two rhesus monkeys as independent and almost orthogonal engrams. Indeed, the multi-unit activity (MUA) recorded in the left PMd from a 96-channel microelectrode array displayed mixed selectivity to the symbols and their locations. The monkeys were required to learn the ordinal position of a set of seven visual abstract symbols (A, B, C, …), arbitrarily rank ordered. In trials and error sessions they first learned that A > B and B > C, subsequently accomplishing the task request to transitively infer the never experienced relationship A > C. Composing the GML based on the decoded representations, we show that PMd activity projected on this linear subspace is predictive of the motor decision, eventually mapping the symbols in adjacent positions along this mental line.