Distinct Roles of Deep and Superficial Cortical Layers in Tone Prediction, Comparison, and Adaptation in Human Auditory Cortices

Auditory perception relies on prior sensations to form expectations and adapt to repeated inputs. Yet, how these processes interact with frequency-specific preference across cortical depth remains unclear. Using ultra-high-field functional magnetic resonance imaging (at 7T), participants listened to stochastic tone sequences varying in their repetition structure and predictability. We modelled voxel-wise responses for stimulus preference, repetition suppression, and expectations, partitioning their unique and shared variance across three depth bins (infragranular, granular, supragranular). Expectation-related signals explained most unique variance in deep layers, aligning with internal model representations. Variance jointly explained by stimulus preference and expectation was instead strongest in superficial layers, consistent with the content-specific alignment of top-down predictions (prediction error signalling). Finally, repetition suppression accounted for variance uniformly across depth. Together, the data suggest that deep layers contain content-specific predictive models, superficial layers register prediction-input alignment, and repetition suppression provides a depth-invariant, local gain control that complements predictive processing.