Mechanisms Underlying Complex Sound Processing in the Inferior Colliculus

Abstract

One of the primary functions of the brain is to translate information about the environment into behaviors that ensure an animal’s survival and reproduction. To do this, representations of complex sensory objects are constructed from a relatively limited set of features encoded by the sensory periphery. However, the mechanisms governing how this process occurs in the auditory system remain poorly understood. In the lower auditory brainstem, a host of neuron types respond to different features of sound, including sound frequency, intensity, and location, and these neurons primarily project to the inferior colliculus (IC), a nearly obligatory integration hub for ascending auditory information. In the IC, this ascending auditory information is shaped by extensive local circuitry to produce novel representations of complex sound features that are not rendered in upstream auditory neurons. However, the cellular and circuit mechanisms that govern how IC neurons integrate multiple streams of information and represent complex sound features remains poorly understood. Here, we first investigated the cellular mechanisms underlying how amplitude-modulated sounds (AMs) are encoded in the IC. AMs are common features of complex sounds and are critical for interpreting both human speech and mouse vocalizations. The IC is the first site where rate coding is the primary strategy for encoding AM information, as earlier brain regions primarily encode AMs using temporal codes (Joris et al., 2004). However, how rate codes are computed within the IC is not well understood. Using whole-cell patch-clamp electrophysiology in brain slices, we found that many IC neurons express NMDA receptors (NMDARs) containing GluN2D subunits, which endow NMDARs with slow kinetics and the ability to activate at resting membrane potential. We next showed that GluN2D-containing NMDARs enhance temporal integration of optogenetically-evoked synaptic inputs, providing a cellular mechanism supporting the transition from temporal to rate coding in the IC. In line with this, we found that GluN2D-containing receptors increase the firing rate of IC neurons in response to AM sounds both in vivo and in a computational neuron model, suggesting that GluN2D-containing NMDARs promote rate coding of AM stimuli in the IC by providing additive gain to input-output functions. Natural sounds also contain rapid changes in frequency known as frequency-modulated (FM) sweeps, and previous research has shown that many neurons in the IC respond selectively to the direction of frequency sweeps (Xie et al., 2007; Kuo and Wu, 2012). However, how this selectivity is encoded within individual cells and whether IC neurons simultaneously encode other FM sweep features such as intensity, speed, and frequency range are poorly understood. To examine this, we performed in vivo recordings in awake mice and trained a machine learning model to predict sound feature identity from neuron spike times. We found that individual IC neurons exhibit robust selectivity for FM sweep frequency ranges and rates, forming a representation of potential stimulus identities across the IC neuron population that relies on both spike counts and spike timing. In line with this, we show that decoding from small groups of neurons provides higher accuracy than decoding from individual cells, suggesting that populations of IC neurons can convey stable stimulus-specific information to downstream targets. Overall, these results establish new cellular mechanisms by which the IC encodes complex sound features and provide insights into how individual cells and populations of IC neurons form representations of complex auditory objects.