Measuring Stimulus-Related Redundant and Synergistic Functional Connectivity with Single Cell Resolution in Auditory Cortex

Loren Koçillari, Marco Celotto, Nikolas A. Francis, Shoutik Mukherjee, Behtash Babadi, Patrick O. Kanold, Stefano Panzeri

TL;DR

This study investigates how different types of functional connectivity (FC) – redundant and synergistic – relate to information processing in the mouse auditory cortex during a sound discrimination task. They found that connections identified by Granger Causality (GC) are primarily redundant, and this redundancy is more pronounced during correct behavioral choices, suggesting its importance for accurate perception.

What Problem Does This Paper Address?

Understanding how information is communicated within the brain relies heavily on measuring functional connectivity (FC). Traditionally, FC studies focused on "redundant" connectivity, where the activity of different brain regions or neurons is similar. However, recent work suggests that "synergistic" connectivity, where the combined activity of neurons provides information not present in individual neurons, might also be crucial. This paper tackles two key questions: 1) How do redundant and synergistic connectivity relate to each other and to established measures of directed FC like Granger Causality (GC)? and 2) What is the spatial scale at which these interactions occur, specifically at the single-neuron level? Existing FC studies often use coarse measures like fMRI or EEG, leaving the fine-grained, single-neuron dynamics less understood.

Methods

The researchers employed a combination of advanced techniques:

  • Granger Causality (GC): This statistical method was used to identify directed functional connections between pairs of neurons. GC assesses whether the past activity of one neuron can predict the future activity of another, going beyond simple correlations by considering directionality.
  • Partial Information Decomposition (PID): This information-theoretic framework was crucial for dissecting the joint information carried by pairs of neurons into distinct, non-negative components: redundancy and synergy. Unlike simpler measures like co-information, PID allows for a precise quantification of each component separately.
  • Two-photon calcium imaging: This technique was used to record the activity of individual neurons in the primary auditory cortex (A1) of mice.
  • Stimulus Information Calculation: Shannon mutual information was used to quantify how much information each neuron carried about the auditory stimuli. This was then extended to pairs of neurons to measure joint stimulus information.

Datasets

The study reanalyzed a previously published dataset [20] involving two-photon calcium imaging of A1 L2/3 neurons in transgenic mice performing a pure-tone discrimination task. The dataset comprised recordings from 2792 neurons across 34 sessions.

Key Insights and Ideas

  • GC connections are predominantly redundant: The study found that pairs of neurons exhibiting Granger Causality (GC) connections showed a higher proportion of redundant information and a lower proportion of synergistic information compared to unconnected pairs. This suggests that GC primarily captures redundant communication pathways.
  • Redundancy is key for correct behavior: The proportion of redundancy was significantly higher in trials where the mice made correct behavioral choices compared to incorrect ones. This highlights the functional importance of redundant connectivity for accurate perceptual discrimination.
  • Synergy's role is less clear for discrimination: While synergy was present, its contribution to correct behavioral choices appeared less pronounced than redundancy. This suggests that synergistic information might be harder to decode or less directly relevant for simple discrimination tasks.
  • Single-neuron resolution reveals fine-grained dynamics: By analyzing connectivity at the single-neuron level, the study provides a more detailed understanding of how information is processed and shared within neural circuits.

Key Results

  • GC-connected pairs of neurons carried proportionally more redundancy (29.2%) and less synergy (23.9%) than GC-unconnected pairs (redundancy: 26.2%, synergy: 26.1%) in correct trials.
  • In correct trials, the normalized redundancy for GC-connected pairs was significantly higher (-0.053 ± 0.013) than for GC-unconnected pairs (-0.001 ± 0.002).
  • In incorrect trials, the joint stimulus information was significantly reduced (~30% of correct trials), and fractional redundancy dropped considerably, while fractional synergy increased proportionally.
  • The study found that only redundant FC associated with GC links appeared to be useful for correct sensory discriminations.

Limitations and Future Work

  • Focus on a specific FC measure: The study primarily used Granger Causality. While powerful, other directed FC measures exist, and future work could explore how redundancy and synergy relate to them.
  • Task simplicity: The perceptual task was relatively simple. Future research could investigate whether synergy plays a more significant role in more complex cognitive behaviors.
  • Decoder capabilities: The authors suggest that synergistic information might be difficult for current decoders to read out, implying that future work could explore how different decoder architectures might better utilize synergistic signals.
  • Broader implications: The findings suggest that FC generally creates prevalent redundancy, which is beneficial for correct sensory judgments. Further research is needed to fully understand the contributions of synergy and redundancy in different neural contexts and for various cognitive functions.

Summary updated

Abstract

Abstract Measures of functional connectivity have played a central role in advancing our understanding of how information is communicated within the brain. Traditionally, these studies have focused on identifying redundant functional connectivity, which involves determining when activity is similar across different sites. However, recent research has highlighted the potential importance of also identifying synergistic connectivity—that is, connectivity that gives rise to information not contained in either site alone. Here, we measured redundant and synergistic functional connectivity with individual-neuron resolution in the primary auditory cortex of the mouse during a perceptual task. Specifically, we identified pairs of neurons that exhibited directed functional connectivity between them, as measured using Granger Causality. We then used Partial Information Decomposition to quantify the amount of redundant and synergystic information carried by these neurons about auditory stimuli. Our findings revealed that functionally connected pairs carry proportionally more redundancy and less synergy than unconnected pairs, suggesting that their functional connectivity is primarily redundant in nature. Furthermore, we observe that the proportion of redundancy is higher for correct than for incorrect behavioral choices, supporting the notion that redundant connectivity is beneficial for behavior.
BibTeX
@inproceedings{kocillari2023measuring,
  author = {Loren Koçillari and Marco Celotto and Nikolas A. Francis and Shoutik Mukherjee and Behtash Babadi and Patrick O. Kanold and Stefano Panzeri},
  title = {{Measuring Stimulus-Related Redundant and Synergistic Functional Connectivity with Single Cell Resolution in Auditory Cortex}},
  booktitle = {BI},
  year = {2023},
  doi = {10.1007/978-3-031-43075-6\_5},
}

Loading related content...

Figure