The inferior colliculus (IC) is an obligatory relay for ascending auditory inputs from the brainstem and receives descending input from the auditory cortex. The IC comprises a central nucleus (CNIC), surrounded by several shell regions, but the internal organization of this midbrain nucleus remains incompletely understood. We used two-photon calcium imaging to study the functional microarchitecture of both neurons in the mouse dorsal IC and corticocollicular axons that terminate there. In contrast to previous electrophysiological studies, our approach revealed a clear functional distinction between the CNIC and the dorsal cortex of the IC (DCIC), suggesting that the mouse midbrain is more similar to that of other mammals than previously thought. We found that the DCIC comprises a thin sheet of neurons, sometimes extending barely 100 m below the pial surface. The sound frequency representation in the DCIC approximated the mouse's full hearing range, whereas dorsal CNIC neurons almost exclusively preferred low frequencies. The response properties of neurons in these two regions were otherwise surprisingly similar, and the frequency tuning of DCIC neurons was only slightly broader than that of CNIC neurons. In several animals, frequency gradients were observed in the DCIC, and a comparable tonotopic arrangement was observed across the boutons of the corticocollicular axons, which form a dense mesh beneath the dorsal surface of the IC. Nevertheless, acoustically responsive corticocollicular boutons were sparse, produced unreliable responses, and were more broadly tuned than DCIC neurons, suggesting that they have a largely modulatory rather than driving influence on auditory midbrain neurons.

Topographic representation of the receptor surface is a fundamental feature of sensory cortical organization. This is imparted by the thalamus, which relays information from the periphery to the cortex. To better understand the rules governing thalamocortical connectivity and the origin of cortical maps, we used in vivo two-photon calcium imaging to characterize the properties of thalamic axons innervating different layers of mouse auditory cortex. Although tonotopically organized at a global level, we found that the frequency selectivity of individual thalamocortical axons is surprisingly heterogeneous, even in layers 3b/4 of the primary cortical areas, where the thalamic input is dominated by the lemniscal projection. We also show that thalamocortical input to layer 1 includes collaterals from axons innervating layers 3b/4 and is largely in register with the main input targeting those layers. Such locally varied thalamocortical projections may be useful in enabling rapid contextual modulation of cortical frequency representations.

Hearing involves analyzing the physical attributes of sounds and integrating the results of this analysis with other sensory, cognitive and motor variables in order to guide adaptive behavior. The auditory cortex is considered crucial for the integration of acoustic and contextual information and is thought to share the resulting representations with subcortical auditory structures via its vast descending projections. By imaging cellular activity in the corticorecipient shell of the inferior colliculus of mice engaged in a sound detection task, we show that the majority of neurons encode information beyond the physical attributes of the stimulus and that the animals' behavior can be decoded from the activity of those neurons with a high degree of accuracy. Surprisingly, this was also the case in mice in which auditory cortical input to the midbrain had been removed by bilateral cortical lesions. This illustrates that subcortical auditory structures have access to a wealth of non-acoustic information and can, independently of the auditory cortex, carry much richer neural representations than previously thought.

Models of behavior typically focus on sparse measurements of motor output over long timescales, limiting their ability to explain momentary decisions or neural activity. We developed data-driven models relating experimental variables to videos of behavior. Applied to mouse operant behavior, they revealed behavioral encoding of cognitive variables. Model-based decoding of videos yielded an accurate account of single-trial behavior in terms of the relationship between cognition, motor output and cortical activity.

No abstract