Neural representation of rules at different hierarchical levels

In everyday life, humans use rules to organize their thoughts and actions in order to achieve specific goals (Bunge & Wallis, 2007). For simple situations, single rules can be used to link a sensory stimulus (the traffic light is green) to its appropriate response (cross the road). More complex situations, however, require the application of multiple rules organized in hierarchies, where high level rules influence the selection or application of lower level rules. Previous studies have demonstrated that Prefrontal Cortex (PFC) is one of the key areas underlying rule processing and control of action. However, it is still unclear whether distinct brain regions within PFC systematically encode qualitatively different task features. In the present study we investigated whether different features defining a complex rule set are represented in different brain areas depending on the level of control they enforce. To this purpose, we devised an experiment in which participants (N = 20) learnt complex rule sets composed by rules at two different levels of control: low (e.g., “if you see a banana, then press left”) and high (e.g., “If you see a star, then only consider red targets”). The task required participants to retrieve, maintain, and apply two rule sets (one low and one high level) to target stimuli. At the beginning of each trial two cues associated with low (or high) level rules were displayed, followed by a delay (delay 1). Then a second pair of cues standing for high (or low) level rules were presented followed by a second delay (delay 2), after which the target was shown. Participants had to apply all the rules to the target stimuli and respond accordingly. The paradigm allowed us to: (i) independently assess the encoding of high and low level rules, (ii) evaluate the difference between the encoding of the two types of rules (comparing high vs. low level rule representations during delay 1, when only one type of rule was maintained), and (iii) decode rule integration (by comparing rule representations during delay 1 vs. delay 2, in which the two levels of rules had to be integrated in order to respond). We applied multivariate decoding analysis (e.g., Haynes et al., 2007) to functional magnetic resonance imaging data to perform the above-described comparisons. Behavioral as well as preliminary decoding results suggest that rules at different levels of abstraction are indeed processed differently in distinct brain regions within a large-scale brain network comprising parietal and prefrontal areas.