Decision timing in the face of changing sensory input: behavior and neural correlates

Most real-life situations require organisms to extract information from incoming stimuli to predict future events and, from the prediction, to precisely time the appropriate motor act. In the present study we designed a new behavioral task that requires subjects (humans and rodents) to extract temporal information characterizing a continuous stream of sensory input and execute a precisely timed motor act. Furthermore, we recorded neurons in the premotor cortex of rats performing this task to investigate the involvement of this brain area in timing actions as a response to incoming stimulation. In our experiment rats received vibrations on their whiskers and responded by withdrawing from the nose-poke hole, while humans received the stimuli on their fingertips and responded by pressing a button. The stimuli were formed by multiplying pink-noise velocity values by an envelope sine wave. Responses made around the peak of the envelope (40% of each cycle) were rewarded. The parameters of the envelope (frequency, amplitude and phase at stimulus onset) changed from trial to trial to ensure that subjects could not set an absolute amplitude threshold or use timing alon e (e.g. “w ait 1 secon d after stimu lu s on set”) to solve th e task. Rats and humans learned to time their responses to the envelope peak at above-chance levels across different envelope parameters. Both rats and humans responded in later cycles in high frequ ency and low amplitude stimuli, suggesting that these stimuli were more difficult and thus required integration of more evidence to support the response. Furthermore, rats benefited from collecting more information about the stimulus, as shown by better -timed responses made in the second than in the first cycle of stimulation. As expected, the activity of premotor cortex neurons was predictive of th e im m in en ce of th e an im al’s action , in th e tim e p eriod p reced in g th e withdrawal. Moreover, neurons carried information regarding the stimulus, with a large proportion coding for the overall stimulus amplitude. A small percentage of the recorded premotor cortex neurons also showed a correlation between firing rate and the stimulus amplitude at any given point in th e trial. The strategy rats were likely to use for solving the task emerging from these results was to understand the global amplitude of the trial and set an amplitude threshold against which to compare the perceived stimulus. Interestingly, the activity of premotor cortex neurons at different moments in the trial was correlated with the time at which the rat withdrew, carrying information both regarding how much time the rat is willing to wait and how much time has passed since the stimulation started. We used an artificial neural network (ANN) implemented in MATLAB to predict withdrawal time from the firing rates at different time bins of all the neurons recorded simultaneously within a behavioral session, and found a good network performance in the time bin s p reced in g th e an im al’s action . Performance was better in incorrect trials, indicating that in some trials rats only engaged in timing, while in others they paid attention to the stimulus and did not keep track of time. In summary, we designed a new behavioral paradigm to investigate how the brain times decisions in response to changing incoming sensory stimulation. Both rats and humans learned to align their responses to the peak of the envelope and chose to gather more stimulus information in trials characterized by low amplitude and high frequency. Finally, neurons in the premotor cortex of rats performing the task carried signals related to key aspects of the task: the time of withdrawal and stimulus properties.