What is common to memory retrieval and relational reasoning? Testing neural overlap in anterior lateral prefrontal cortex

Abstract

Anterior lateral prefrontal cortex (aLPFC) has been consistently implicated in retrieval monitoring and relational reasoning. However, it is unknown whether aLPFC supports a cognitive process that is common to memory retrieval and reasoning tasks. To inform this question, a first step is to ascertain whether there is, within individuals, spatial overlap in aLPFC between tasks. A prior study found overlap in activation of left aLPFC between verbal analogy and source memory retrieval tasks, using tightly matched word stimuli (Westphal et al., 2016). To assess whether this overlap in activation extends to other stimulus types and task designs, and another sensory modality, we collected fMRI data from 35 healthy young adults as they completed an associative memory retrieval task along with four variants of a relational reasoning paradigm that reliably engages left aLPFC. The reasoning tasks followed a 2 (visual/auditory modality) x 2 (spatial/featural relations) x 2 (low/high relational complexity) design. In the associative memory task, participants studied word pairs and decided whether the two words were related. At retrieval, they viewed pairs of previously seen words and judged whether the pair was Intact (words paired together at encoding) or Recombined (previously seen words recombined into new pairs). Response times were slower for Recombined vs. Intact pairs, consistent with heightened retrieval monitoring. As predicted, aLPFC was more active for Recombined than Intact pairs, as well as for the manipulation of relational complexity across the four reasoning tasks. A group-level conjunction analysis demonstrated substantial overlap between memory and reasoning manipulations in left aLPFC (Glasser parcels a47r, p47r, a10p, and p10p). Within individuals, clusters defined from the reasoning task were also sensitive to retrieval monitoring demands. Further, there was within-participant overlap in the pattern of active vertices, as measured via Dice coefficients. Bayes factor testing within each parcel of interest demonstrated greater vertex-level overlap than expected by chance, particularly in p47r. Importantly, degree of overlap varied widely among individuals, and exploratory results showed that both cluster-level activation and vertex overlap in p47r correlated with memory discriminability (d-prime). Thus, we hypothesize that neural overlap in this area reflects engagement in relational thinking during memory retrieval monitoring. Together, these results move beyond general characterizations of aLPFC function by testing whether it meets a precondition -- individual-level neural overlap across varied tasks -- of a unified neurocognitive function in support of reasoning and memory retrieval.