Abstract
The allocation of attention can be decoded from the activity of lateral prefrontal cortex neuronal ensembles (Tremblay et al., 2015). One issue that remains unclear is the impact of a neural population's size and composition on decoding of attention. To investigate this, we recorded the responses of neurons in lateral prefrontal cortex of two macaques using microelectrode arrays while they performed a visuospatial attention task. During the task, the animals had to direct attention to a cued target stimulus positioned in one of the four visual quadrants while ignoring 3 identical distractors positioned in the remaining quadrants. We systematically changed the size and composition of the neuronal ensembles, as well as the pattern of noise correlations, and evaluated their information content using a linear decoder. First, we found that the location of visuospatial attention was reliably decoded from ensembles of approximately 50 units (mean accuracy = 76%, p< 0.05, Permutation test). We progressively increased the number of neurons in an ensemble from 1 to 50 units and assessed decoding performance using two methods; first, we built subnetworks of most informative neurons, and second, we built subnetworks that maximized information of the ensemble. We found that the decoding performance of the most informative subnetworks was higher than those composed of the most informative units. Interestingly, the most informative subnetworks were not necessarily comprised of most informative units, including in many cases non-selective units (Kruskal-Wallis test P>0.05). Finally, removing noise correlations increased the decoding performance of ensembles of most informative units (6%, Signed rank P< 0.01), whereas removing correlations in most informative subnetworks of equivalent size had no effect on performance (Signed rank P>0.05). These results indicate a complex effect of ensemble size and composition on the coding of attention in lateral prefrontal cortex neuronal ensembles.
Meeting abstract presented at VSS 2016