Flexible integration of natural stimuli by auditory cortical neurons

Grace Wan Yu Ang, Claudia Clopath, Andreï S. KozlovbioRxiv 2024 · DOI

TL;DR

This paper investigates how auditory cortical neurons integrate complex, naturalistic sounds (mouse vocalizations) to perform computations like pattern recognition. They found that individual neurons are surprisingly flexible, capable of implementing different logical operations (like AND and OR) depending on the input, a flexibility that could be key for robust sensory processing.

What Problem Does This Paper Address?

Understanding how neurons process information is fundamental to neuroscience. While we know neurons combine inputs to generate outputs, much of this research has been done in vitro using simplified, artificial stimuli. This leaves a gap in our understanding of how neurons integrate complex, real-world stimuli in their natural environment. Specifically, how do neurons combine the rich, spatiotemporal patterns of natural sounds to achieve functions like pattern recognition, which often relies on logical operations (AND, OR)? This paper aims to bridge this gap by studying neuronal integration using ethologically relevant stimuli and exploring the underlying mechanisms.

Methods

The researchers employed a multi-pronged approach:

  • In Vivo Electrophysiology: They recorded single-unit activity from the auditory cortex of anesthetized mice while presenting pairs of ultrasonic mouse vocalization (USV) syllables.
  • Stimulus Design: They used natural USV syllables and manipulated their temporal alignment to probe integration. They also used pitch-shifted versions of syllables to investigate how spectral dissimilarity affects integration.
  • Computational Modeling: They built a biophysical neuron model with dendritic subunits to simulate how spatiotemporal patterns of synaptic inputs influence neuronal integration. This model incorporated aspects of auditory periphery processing and dendritic integration.
  • Analysis of Integration Functions: They quantified neuronal responses using the Summation Index (SmI), a metric that characterizes how inputs are combined, relating it to Boolean logic operations (MAX-like for OR, AND-like for AND). They also analyzed factors like stimulus dissimilarity, temporal correlation, and cross-adaptation.

Datasets

  • Mouse Ultrasonic Vocalizations (USVs): The study used natural USV syllables recorded from mice as the primary stimuli. These were presented individually and in pairs, with varying temporal overlaps.
  • Pitch-shifted USVs: To explore spectral dissimilarity, syllables were pitch-shifted by different amounts.

Key Insights and Ideas

  • Neuronal Flexibility: Individual neurons are not fixed in their computational mode. They can flexibly switch between different integration strategies (e.g., MAX-like, AND-like) depending on the characteristics of the incoming stimuli. This is a significant departure from artificial neural networks, where units typically have fixed activation functions.
  • Stimulus-Dependent Integration: The type of integration observed is strongly influenced by the input. More similar or strongly activating stimuli tend to elicit MAX-like (OR-like) responses, promoting invariance. Conversely, spectrally distinct or weakly activating stimuli tend to elicit AND-like responses, promoting selectivity.
  • Dendritic Role: The study highlights the crucial role of dendritic properties in shaping integration. The spatiotemporal pattern of synaptic inputs arriving at different dendritic branches can determine whether a neuron performs a MAX or AND operation.
  • Bridging Levels: The research successfully bridges the gap between low-level biophysical mechanisms (dendritic integration) and high-level computational functions (pattern recognition), showing how the former enables the latter.

Key Results

  • Predominance of MAX-like responses: The majority of observed integration responses in the auditory cortex peaked around a Summation Index (SmI) of 0, indicating a prevalence of MAX-like (OR-like) operations.
  • AND-like responses with dissimilarity: Integration tended to be more AND-like when stimulus pairs were spectrally distinct.
  • Input independence: A significant portion of neurons exhibited input independence, meaning the response to one syllable was not significantly affected by prior exposure to another, suggesting distinct processing pathways.
  • Model recapitulates findings: The computational model successfully reproduced the observed range of integration functions and demonstrated how dendritic properties and input spatiotemporal patterns influence the resulting computation. For instance, spectrally separated inputs targeting different dendrites led to AND-like responses in the model.

Limitations and Future Work

  • In Vivo vs. In Vitro: While the study used natural stimuli, the experiments were conducted in anesthetized mice. Future work could explore these phenomena in awake, behaving animals to understand how behavioral context influences integration.
  • Model Simplifications: The biophysical model, while sophisticated, still involves simplifications. For example, it uniformly represents stimuli across time and frequencies, neglecting more nuanced tuning-specific effects. Future models could incorporate more detailed receptive field properties and explore how different regions of a neuron's spectrotemporal receptive field contribute to computation.
  • Network Dynamics: The study focused on single-neuron integration. Future research could investigate how this single-neuron flexibility contributes to network-level computations and emergent behaviors.
  • Active Dendrites: While the model explored passive dendritic integration, the authors suggest exploring how active dendritic properties (like NMDA conductances) further influence integration, potentially shifting responses towards MAX-like computations.

Summary updated

Abstract

ABSTRACT Neurons have rich input-output functions for processing and combining their inputs. Although many experiments characterize these functions by directly activating synaptic inputs on dendrites in vitro , the integration of spatiotemporal inputs representing real-world stimuli is less well studied. Using ethologically relevant stimuli, we study neuronal integration in relation to Boolean AND and OR operations thought to be important for pattern recognition. We recorded single-unit responses in the mouse auditory cortex to pairs of ultrasonic mouse vocalization (USV) syllables. We observed a range of integration responses, spanning the sublinear to supralinear regimes, with many responses resembling the MAX-like function, an instantiation of the OR operation. Integration was more MAX-like for strongly activating features, and more AND-like for spectrally distinct inputs. Importantly, single neurons could implement more than one integration function, in contrast to artificial networks which typically fix activation functions across all units and inputs. To understand the mechanism underlying the flexibility and heterogeneity in neuronal integration, we modelled how dendritic properties could influence the integration of inputs with complex spectrotemporal structure. Our results link nonlinear integration in dendrites to single-neuron computations for pattern recognition. Significance statement Sensory neurons compute over their inputs, combining them in ways to achieve selectivity and invariance for pattern recognition. Using real-world stimuli, we show that single cortical neurons are flexible, being capable of implementing more than one computation, unlike artificial neural-network units with fixed activation functions. We investigate this flexibility by modeling how synaptic activation patterns of real-world stimuli affect dendritic integration and the resultant neuronal computation. Our work bridges the gap between biophysical mechanisms and computation, linking neuronal input integration to pattern recognition.
BibTeX
@misc{ang2024flexible,
  author = {Grace Wan Yu Ang and Claudia Clopath and Andreï S. Kozlov},
  title = {{Flexible integration of natural stimuli by auditory cortical neurons}},
  year = {2024},
  doi = {10.1101/2024.04.18.590153},
  howpublished = {bioRxiv},
}
Figure