Cues for Sound Localization Are Encoded in Multiple Aspects of Spike Trains in the Inferior Colliculus

Steven M. Chase, Eric D. YoungJournal of Neurophysiology 2008 · DOI

TL;DR

This study investigates how neurons in the inferior colliculus (IC) encode sound localization cues (ITD, ILD, SN) using different aspects of their spike trains. They found that ITD is primarily encoded by spike rate, while spectral notch (SN) cues are more effectively represented by spike timing, especially at lower frequencies. This suggests that different neural codes are used for different localization cues, allowing for multiplexing of information within single neurons.

What Problem Does This Paper Address?

Sound localization, our ability to determine the source of a sound in space, relies on integrating multiple auditory cues. The primary cues are interaural timing differences (ITDs), interaural level differences (ILDs), and spectral notch cues (SNs). The inferior colliculus (IC) is a key brain region for processing these cues, receiving input from various auditory nuclei. However, it's not fully understood how the IC integrates these diverse cues. Specifically, while previous research has focused on how spike rate encodes these cues, this paper explores whether other aspects of neural activity, such as the timing of spikes (first-spike latency and ongoing temporal patterns), play a significant role and how this coding might differ across the various localization cues. The central question is how information about these distinct localization cues is represented and potentially combined (multiplexed) within the spike trains of IC neurons.

Methods

The researchers used information-theoretic techniques to quantify the amount of information that different aspects of neuronal responses carried about specific sound localization cues. They recorded from single neurons in the inferior colliculus of decerebrate cats.

Key methods include:

  • Virtual Space Stimuli: They used broadband noise stimuli manipulated to simulate different ITD, ILD, and SN values, mimicking how these cues vary with sound source location. These stimuli were presented in sets where two parameters were varied simultaneously.
  • Information Theory: Mutual Information (MI) was calculated to measure the statistical dependency between the stimulus parameters and the neural response. This allowed them to quantify how much information about a specific cue (e.g., ITD) was present in the neural response.
  • Multiple Encoding Schemes: They analyzed information conveyed by:
  • Spike Rate: Standard measure of average firing rate over a time window.
  • First-Spike Latency (FSL): The time of the very first spike after stimulus onset. This was further decomposed into information about whether a spike occurred (MI<sub>sp</sub>) and information about the timing of that spike (MI<sub>tim</sub>).
  • Spike Distance Metric (SDM): A more sophisticated code sensitive to spike timing across the entire response duration, capturing temporal patterns beyond just the first spike.
  • Decomposition of Information: They used mathematical decompositions (e.g., Eq. 2, Eq. 4) to isolate the independent contributions of rate, FSL, and ongoing spike timing to the overall information about a cue. This allowed them to determine if FSL or SDM provided information beyond what was already present in the spike rate.

Datasets

Not discussed in the paper. The study used custom-generated virtual space stimuli.

Key Insights and Ideas

  • Differential Coding: The most striking finding is that different localization cues are not encoded uniformly across all aspects of a neuron's spike train. This suggests a specialized coding strategy for each cue.
  • ITD is Rate-Dominant: Interaural timing differences (ITDs) are primarily encoded by the neuron's firing rate. While there's a small amount of information in first-spike latency and ongoing timing, it's largely redundant with the rate information.
  • SN Cues Leverage Timing: Spectral notch (SN) cues are more effectively represented by spike timing, particularly ongoing temporal patterns captured by the SDM code. This effect is more pronounced at lower best frequencies (BFs) of the neurons.
  • ILD is a Mix: Interaural level differences (ILDs) show a more balanced encoding across rate and timing, with both contributing significantly.
  • Multiplexing is Key: The differential coding allows neurons to carry information about multiple localization cues simultaneously within a single spike train. For example, a neuron might use its firing rate to signal ITD and its spike timing to signal SN.
  • Latency's Role: First-spike latency provides independent information for ILD and SN cues, suggesting it's not just a byproduct of rate changes but a distinct coding mechanism.

Key Results

  • ITD Information: Primarily encoded by spike rate (average MI<sub>rate</sub> ≈ 0.68 bits for the full stimulus set). FSL and SDM provided minimal independent information.
  • ILD Information: Encoded by both rate and FSL. FSL provided a significant amount of information beyond rate (average FSL gain ≈ 0.17 bits). SDM also captured additional information.
  • SN Information: Showed the most diverse coding. Rate was important at higher BFs, while FSL and especially SDM provided significant information, particularly at lower BFs. SDM gain was highest at low BFs (average SDM gain ≈ 0.31 bits).
  • FSL vs. Rate: For ILD and SN, FSL provided substantial information independent of rate. For ITD, FSL's independent contribution was minimal.
  • SDM vs. Rate: For SN, spike timing beyond the first spike (captured by SDM) provided significant information not present in rate or FSL alone.

Limitations and Future Work

  • Artificial Stimuli: The use of virtual space stimuli, while controlled, may not perfectly replicate natural auditory environments. The paper acknowledges that the utility of temporal coding for SNs might be limited in natural conditions where stimulus envelopes are unknown and not repeatable.
  • Decerebrate Cats: Experiments were conducted on decerebrate cats, which lack higher brain centers. This limits direct conclusions about how these cues are integrated in conscious perception.
  • Future Work: The authors suggest that future research should investigate how these differentially coded cues are integrated by downstream neurons. They also highlight the need for further exploration into the precise mechanisms by which temporal codes like latency are decoded by neural circuits. The potential for multisensory integration, where auditory spatial processing might interact with other sensory modalities, is also mentioned as a promising avenue.

Summary updated

Abstract

To localize sound, information from three cues--interaural timing differences (ITDs), interaural level differences (ILDs), and spectral notch cues (SNs)--must be properly integrated. The inferior colliculus (IC) receives convergent input from neurons encoding all three cues. Using virtual space stimuli and information theoretic techniques, we investigated the coding of the various localization cues in single neurons of the IC under different encoding schemes. Here we focus on the analysis of information encoded by first-spike latency, in comparison to previous results on discharge rate and ongoing spike timing. The results show that the localization cues converge to different degrees in particular neurons. ITD information is conveyed most strongly by spike rate, with small amounts of independent information in latency and ongoing spike timing. ILD information shows a similar pattern, with larger mutual information values for all three cues. For these cues, ongoing spike timing does not typically contribute independent information over that captured by a joint rate/first-spike latency code. SNs are coded by both rate and first-spike latency, but ongoing spike timing significantly enhances their representation in a best frequency-dependent manner, as long as the temporal envelope of the stimulus can be used in the decoder. The differential coding of the localization cues suggests that information about multiple cues could be multiplexed onto the responses of single neurons.
BibTeX
@article{chase2008cues,
  author = {Steven M. Chase and Eric D. Young},
  title = {{Cues for Sound Localization Are Encoded in Multiple Aspects of Spike Trains in the Inferior Colliculus}},
  journal = {Journal of Neurophysiology},
  year = {2008},
  doi = {10.1152/jn.00644.2007},
}
Figure