Abstract
The lateral prefrontal cortex (LPFC) plays a key role in higher-order cognition. Electrophysiology studies in behaving animals show good decoding of task-relevant information from patterns of activity across neural populations in LPFC. However, functional magnetic resonance imaging (fMRI) studies in humans report only weak decoding of task-relevant information from LPFC activity. The limited access to prefrontal population codes in humans has hampered progress in understanding the neural computations that support human cognition. We hypothesize that the spatial topography of prefrontal population codes is too fine-grained to be effectively resolved by standard fMRI (2×2×2 mm3). We analyzed microelectrode array data (Utah array, 4×4 mm2, 10×10 channels spaced 0.4 mm apart) recorded from LPFC areas 8A and 9/46 of two macaques performing three visuospatial tasks: an oculomotor delayed response task, a visuospatial working memory task, and an associative learning task. For each monkey, task, and measurement session, we estimated each channel’s tuning profile by computing its mean firing rate for each condition. To assess the spatial scale of the population codes, we estimated tuning profile similarity between channel pairs as a function of distance, using global Moran’s I. To inspect the spatial topography of the population codes, we visualized channel tuning similarity on the arrays. We find that tuning profiles are spatially autocorrelated up to a distance of 3.5 mm across all tasks in most sessions for both monkeys. The array map visualizations suggest moderate spatial clustering of channels with similar tuning and a somewhat irregular topography. The observed topography and fine-grained spatial scale of the prefrontal population codes, which falls above the Nyquist frequency of standard fMRI sampling, may limit the sensitivity of standard fMRI to prefrontal population codes. High-field fMRI, with increased spatial resolution, may be able to bridge the gap between monkey electrophysiology and human fMRI.