Item

Abstract

The influential theoretical framework of predictive coding states that low-level sensory areas subtract top-down predictions from bottom-up signals to calculate prediction errors, which are used for perceptual inference. So far, only few single-unit studies have directly tested predictions of this model. Here, we measured responses of single neurons in mouse primary visual cortex (V1) and asked how temporal predictability affects the processing of a visual stimulus. Head-fixed mice were free to run or sit on a spherical treadmill in front of a gray screen, on which we flashed a visual stimulus (black square, 100 ms duration) in the absence of any task. To manipulate stimulus predictability, we presented an auditory cue (5 KHz pure tone) shortly before the stimulus appeared. In a standard condition (80 of the trials), the interval between cue and stimulus (ISI) was fixed at 500 ms. In 20 of the trials, ISIs deviated from this value (± 100, ±200 ms) to test the precision of prediction. In an unpredictable condition, the auditory cue was omitted. We recorded from V1 neurons simultaneously across cortical layers and found that stimulus predictability robustly decreased V1 responses. This decrease was seen in both the initial transient and later components of neural activity. The standard ISI condition led to the strongest decrease, resulting in a “U-shaped” curve of evoked activity versus ISIs. In control experiments conducted under anesthesia the difference in evoked activity between predictable and unpredictable conditions disappeared. Furthermore, we replaced the auditory cue by a distant visual cue, which, when presented by itself, did not affect the firing rates of the recorded neurons. Here, predictable stimuli still decreased evoked responses, ruling out low-level, multi-sensory interactions. To investigate the local neural circuitry, we expressed Archaerhodopsin (Arch) in V1 parvalbumin-positive inhibitory interneurons. We suppressed interneuron activity by shining yellow light (590nm, 500ms duration) at the time of stimulus onset. Our preliminary data show a reversed effect, with evoked activity being larger for predictable stimuli and suggest that local inhibitory neurons might convert the effect of the external modulatory signal from facilitation to suppression. One interesting observation is that we found no modulation of responses in ‘cue only’ conditions. This observation is inconsistent with the predictive coding model, which predicts response suppression at the time when the omitted stimulus would have appeared. Overall, these results point to an external modulatory signal that is broadly aligned with the timing of the visual stimulus.