Abstract
During perception, the complex and noisy visual information relayed by our eyes is interpreted within the context of prior knowledge and expectations. The neural circuitry underpinning this process of inference is currently unknown. Recent neural theories of perceptual inference propose that neurons in the deep layers of cortex represent expectations, which in turn modulate sensory processing in the middle and superficial layers. However, direct support for this idea is limited due to the significant challenge of resolving cortical layers using human neuroimaging. Here, we capitalise on ultra-high resolution 7T functional magnetic resonance imaging (fMRI) to probe stimulus-specific activity induced by prior expectations in deep, middle and superficial layers of the human primary visual cortex (V1). In order to induce expectations, we presented participants (N=18) with coloured dot cues that predicted the likely orientation of an upcoming grating stimulus. On 75% of trials, gratings with the predicted orientation were presented. On the remaining 25% of trials, the predicted gratings were omitted. Crucially, as no stimulus was presented to the eyes in these ‘omission’ trials, any activity in V1 could be considered to arise solely from top-down inputs. Results showed that such predicted-but-omitted gratings led to an orientation-specific BOLD response in the deep, but not the middle and superficial layers of V1. In contrast, actually presented stimuli activated all layers of V1. These results provide novel insights into the neural circuitry by which the brain integrates prior expectations with sensory inputs in the service of perceptual inference.