Neuro-Shazam: Unlocking Complex Auditory Signal Recognition through Combination-Sensitive Neurons

Abstract

Biologically relevant stimuli are often highly complex in nature, containing multiple spectral and temporal components that carry information crucial for survival. Several studies suggest a strong relationship between accurate perception of communication sounds and the fidelity of the spectrotemporal auditory processing mechanisms. Combination sensitivity in central auditory neurons reflects a highly selective form of spectrotemporal integration. In this process, specialized neurons, generally known as combination-sensitive neurons (CSNs), integrate auditory inputs in a highly nonlinear fashion across both temporal and spectral domains. This non-linear integration of sound elements is a hallmark of complex auditory processing, in which the neuronal response to a combination of stimuli is much larger than the sum of the individual responses. The precise temporal coincidence of inputs, occurring within few tens of milliseconds, serves as the fundamental mechanism driving this process. Bats, among other species that exhibit this neural behavior, rely on such rapid coincidence to enhance their integration capabilities for effective echolocation. Yet one of the most complex examples of combination sensitive behavior exist in species of songbirds. Neurons in higher-order brain areas are found to be highly tuned to the selective combination of song-syllables, exhibiting a more complex processing behavior, with extended integration periods of hundreds of milliseconds, paralleling the complex processing demands of human speech. In this complex system, CSNs play a key role in allowing songbirds to recognize their own song, differentiate between conspecifics, and decode the subtle social information encoded in song. Despite their crucial role in songbird auditory processing, the mechanisms underlying the combination-sensitive neurons are mostly unexplored. In particular, it is not known how such extended temporal integration can be achieved by these neurons and yet remain sensitive to precisely timed inputs. With these complexities in mind, we developed three biophysically realistic conductance-based neural network models, representing different underlying mechanisms. Our network models are largely based on well-established principles of spectrotemporal processing and coincidence detection, which are particularly important in complex auditory processing. Our first network extends the mechanism of postsynaptic facilitation/priming by adding network-level mechanisms that allow the precisely timed convergence of both inhibitory and excitatory inputs onto the combination-sensitive neuron, which in turn drive spiking activity reflective of their integrated influence. Our second network model examines a scheme in which network-level mechanisms, specifically precisely timed offset responses, gate the integration of a dual-inhibitory inputs at the level of CSN, giving rise to its own offset response, signifying successful combination. Our third network model extends temporal coding principles to show the dynamic transition of the CSN between subthreshold excitatory temporal integration and coincidence detection mode, gated by the shift from asynchronous to precisely timed synchronous inputs, where a subsequent, spectrally distinct input elicits a rapid fluctuation that enables coincidence detection. For each network model, we conducted in-depth simulations to explore how this precise behavior is orchestrated by the interplay of synaptic and intrinsic mechanisms in the combination-sensitive network. We have identified key intrinsic parameters, like the T-type calcium channel conductance and the hyperpolarization-activated current (Ih), that strongly impact the generation of successful network behavior. We further explored how dynamic modulation of these parameters, in response to synaptic influences, further contributing to the selective spectrotemporal integration exhibited by these networks. Overall, our findings suggest that songbirds employ a two-stage process of auditory temporal coding, unlike the single-stage process described previously in several species, including bats. In songbirds, the classical coincidence detection window is preceded by a pre-coincidence integration stage. The first stage is gated by a network of interconnected neurons that can hold on to the carried information from the first stimulus, allowing for an extended temporal integration window. Under a precisely timed release of the held input, the system then transitions into a high-precision coincidence detection mode, during which combination-sensitive neurons temporally integrate incoming sensory information within a narrow window. This work provides a generalized framework for understanding the principles of temporal coding and combination sensitivity across sensory modalities but also provides valuable insights to guide future experimental investigations.