Attentional selection drives information convergence within macaque lateral prefrontal cortex

Abstract

Attention is a coherent state, in which multiple brain regions converge to represent selected features of a focal object or event. Lateral prefrontal cortex (LPFC), with its flexible coding of multiple task features and their conjunctions, is widely believed to play a key role in this process. While much research has investigated biased competition between features within a brain region, and cofluctuations of activity between regions, less is known about the representational dynamics through which a coherent neural state emerges. Here, we examine directed mutual information concerning multiple feature-specific population codes, between dorsal and ventral LPFC, across three phases of an attentional selection task. We find bidirectional convergence of information regarding multiple task features, but specifically following the period of selection from the visual display. The results show that neural processes driving inter-region coherence are especially salient during a period of cued object selection, despite comparable local information representation during other task phases. Significance statementIn the primate brain, lateral prefrontal cortex (LPFC) is thought to play a key role in selective attention, which is fundamental to goal-directed behaviour, and implies inter-region convergence towards a coherent representational state. Combining neural recordings, multivariate decoding, and analysis of information dynamics, we find local representations of multiple task features during multiple task phases, in separate regions of LPFC, with strong inter-region representational coherence emerging when attention could be directed to a chosen target. Results show that information convergence between these regions is bidirectional, sustained, and reflects multiple features of a chosen object, but, unlike local information representation, is highly specific to the choice phase of the task.