The role of adaptation in generating monotonic rate codes in auditory cortex

Jong Hoon Lee, Xiaoqin Wang, Daniel BendorPLOS Computational Biology 2020 · DOI

TL;DR

Auditory cortex neurons represent sound timing with precise spike patterns, but also use firing rate to encode information. This study uses a computational model to show that short-term synaptic depression, a common biological mechanism, can explain how some neurons develop monotonic firing rate responses (always increasing or decreasing with stimulus rate) that complement their precise timing codes. This offers a simple biophysical explanation for how the brain transforms temporal sound information into a more robust rate code.

What Problem Does This Paper Address?

The auditory system needs to process complex sounds, which involves understanding both the timing of sound events and their overall rate. In the auditory cortex, neurons exhibit two main ways of representing this information:

  1. Temporal Coding: Precisely timed spikes that lock to individual sound events, effective for tracking rapid changes up to about 40-50 Hz (the "flutter" range).
  2. Rate Coding: The average firing rate of a neuron changes with a stimulus feature, providing a more "processed" and instantaneous readout.

While temporal coding is precise, it requires time to decode. Rate coding is faster but can be ambiguous, as a single rate change might represent frequency, loudness, or location. To be useful, rate codes need to be demultiplexed. A particularly interesting form of rate coding observed in the auditory cortex is monotonic tuning, where a neuron's firing rate consistently increases (Sync+) or decreases (Sync-) as a stimulus parameter (like repetition rate) changes. This "opponent coding" offers advantages like robustness to adaptation and multiplexing information.

The key question this paper addresses is: What biophysical mechanisms in single neurons can generate these complementary monotonic rate codes (Sync+ and Sync-) that coexist with stimulus-locked temporal coding? Previous models had limitations, and the authors sought to find a simple, biologically plausible mechanism.

Methods

The researchers employed a computational single-neuron model based on the leaky integrate-and-fire (LIF) framework, incorporating both excitatory and inhibitory inputs (an E-I model).

  1. Base E-I Model: Initially, they used a model with adjustable parameters for the excitatory-inhibitory (E/I) ratio, the delay between excitation and inhibition (I-E delay), and overall excitation strength. This model could generate stimulus-synchronized responses.
  2. Introducing Short-Term Depression (STD): To account for adaptation observed in real neurons (where responses decrease over time during repeated stimulation), they incorporated short-term depression. This mechanism models the decrease in neurotransmitter release probability with repeated synaptic activity. They introduced two key parameters:
  3. Amplitude of Depression (AD): Controls the strength of adaptation after each stimulus pulse.
  4. Time Constant of Recovery (Tp): Governs how quickly the synapse recovers from depression. These parameters were independently varied for both excitatory and inhibitory inputs.

  5. Exploring Other Mechanisms: They also briefly modeled short-term facilitation (STF) and spike-frequency adaptation (SFA) to compare their effectiveness in generating Sync+ and Sync- responses.

  6. Model Robustness Testing: The model's performance was tested against noise (Gaussian noise added to conductances) and temporal jitter (variations in stimulus timing).

Datasets

The study used electrophysiological data from a previous published dataset [11] collected from awake, semi-restrained common marmosets (Callithrix jacchus). This dataset contained single-unit recordings from the auditory cortex. The authors did not introduce new datasets but rather used existing experimental data to validate their computational model.

Key Insights and Ideas

  • Synaptic Depression as a Unifying Mechanism: The most significant insight is that short-term synaptic depression, particularly in excitatory inputs, is sufficient to generate both Sync+ and Sync- monotonic rate codes in neurons that also exhibit stimulus-synchronized activity. This provides a parsimonious explanation for a complex neural phenomenon.
  • Opposite Roles of Depression for Sync+ vs. Sync-: The direction of monotonicity (positive or negative) is determined by the relative strength of depression on excitatory versus inhibitory inputs. Weak or absent depression of excitation leads to Sync+ responses, while strong depression of excitation leads to Sync- responses.
  • Adaptation is Key for Rate Coding: The study highlights that adaptation mechanisms, specifically synaptic depression, are crucial for transforming precise temporal codes into more robust rate codes, especially at higher repetition rates.
  • Temporal Dynamics Predict Neuron Type: The temporal profile of responses to pure tones (sustained for Sync+, onset-suppressed for Sync-) could predict a neuron's classification, a finding that the model successfully emulated.

Key Results

  • The computational model, with the addition of short-term depression, successfully reproduced both Sync+ and Sync- responses observed in real auditory cortex neurons.
  • Sync- neurons showed stronger adaptation (a greater decrease in firing rate between the start and end of stimulus presentation) than Sync+ neurons, especially at higher repetition rates.
  • The model demonstrated that depression of excitatory inputs was the primary factor determining the direction of monotonicity, with inhibition playing a modulating role.
  • The model's simulated responses to pure tones matched the temporal dynamics of real Sync+ and Sync- neurons, with Sync+ showing sustained responses and Sync- showing transient, suppressed responses.
  • The model was robust to noise and temporal jitter, maintaining monotonic coding properties under these conditions, though Sync- neurons were slightly more affected.

Limitations and Future Work

  • Model Simplifications: The model uses a single-compartment LIF neuron and does not account for different sources of inhibition (e.g., top-down feedback) or the specific roles of different receptor types (AMPA, GABA-A, NMDA). The use of data from anesthetized rats for model parameters, rather than marmosets, is also a limitation.
  • Stimulus Specificity: The model primarily focuses on acoustic pulse trains and pure tones. It does not explicitly model responses to more complex stimuli like amplitude-modulated (AM) tones.
  • Future Directions: The authors suggest that future work could incorporate NMDA receptors, explore the role of different inhibitory interneurons, and investigate how these mechanisms contribute to processing more complex auditory stimuli. Further studies using optogenetics to manipulate inhibitory neurons could provide deeper insights.

Summary updated

BibTeX
@article{lee2020role,
  author = {Jong Hoon Lee and Xiaoqin Wang and Daniel Bendor},
  title = {{The role of adaptation in generating monotonic rate codes in auditory cortex}},
  journal = {PLOS Computational Biology},
  year = {2020},
  doi = {10.1371/journal.pcbi.1007627},
}
Figure