Population coding strategies in human tactile afferents

Giulia Corniani, Miguel A. Casal, Stefano Panzeri, Hannes P. Saal

TL;DR

This study uses a sophisticated simulation to explore how populations of tactile nerve fibers (afferents) in the human hand encode information about touch. They found that optimal information encoding depends on both the density and the specific mix of different types of afferents, and that combining information from multiple afferent types is often more efficient than simply increasing the number of one type. This work highlights the importance of a population-level view for understanding tactile sensation.

What Problem Does This Paper Address?

Understanding how our sense of touch works is crucial for fields ranging from robotics to prosthetics. While we know that touch information is transmitted by specialized nerve fibers called afferents, much of our understanding comes from studying individual neurons. However, this approach misses the bigger picture: how do populations of these afferents work together to encode complex tactile information? This paper tackles this gap by investigating how the density and composition of these afferent populations influence the encoding of various tactile features, moving beyond single-neuron studies to a more holistic, population-level perspective.

Methods

The researchers employed a computational model called Touchsim to simulate the spiking activity of tactile afferents. They then used Non-Negative Matrix Factorization (NMF) to reduce the dimensionality of the complex, spatiotemporal spike train data, extracting key patterns of neural activity. A Generalized Linear Model (GLM) was used to model the relationship between these extracted patterns and stimulus features. Finally, information-theoretic analysis was employed to quantify how much information about different stimulus features (size, frequency, ramp length, indentation) was encoded in the simulated neural activity. To assess the importance of spatial and temporal coding, they systematically corrupted either the spatial arrangement of afferents or the precise timing of their spikes and measured the resulting information loss.

Datasets

Not discussed in the paper. The study relies on simulated data generated by the Touchsim model, rather than real-world experimental datasets.

Key Insights and Ideas

  • No Single Best Strategy: The optimal density and mix of tactile afferents for encoding information isn't one-size-fits-all. It heavily depends on the specific tactile feature being sensed.
  • Teamwork Makes the Dream Work: Combining information from different types of afferents (SA1, RA, PC) is often more effective for transmitting tactile information than simply increasing the density of a single, highly informative afferent type. This suggests a sophisticated division of labor.
  • Timing is Crucial: Precise spike timing is a surprisingly important component of tactile coding, even at the population level. Losing this temporal information is more detrimental than losing spatial information about afferent arrangement.
  • Complementarity is Key: Different afferent classes often encode complementary information, meaning they capture aspects of the stimulus that others miss. This necessitates integration of signals from multiple classes for a complete picture.
  • Redundancy for Robustness: While complementarity is important, there's also redundancy in coding. This might serve as a backup mechanism, ensuring information is reliably transmitted even if some afferents fail.

Key Results

  • Optimal Density Varies: The density of afferents required for maximal information encoding differs significantly across tactile features. For spatial features like size, high densities are crucial, while for temporal features like frequency, sparser but specialized afferents (like PC) can be highly effective.
  • Combining Classes Outperforms Density Increase: In most cases, combining information from two different afferent classes yielded more information than doubling the density of the most informative single class.
  • Spatial vs. Temporal Coding: Destroying the spatial arrangement of afferents reduced information by about 26%, while disrupting precise spike timing led to a much larger information loss of approximately 54%. This emphasizes the critical role of temporal coding.
  • Feature-Specific Coding: SA1 afferents were most informative for stimulus size and ramp indentation, while PC afferents excelled at encoding frequency. RA afferents showed a more mixed performance, often being more informative at low densities.

Limitations and Future Work

The authors acknowledge that their study is based on simulations and that experimental validation is necessary. They also note that their model focused on three main classes of tactile afferents (SA1, RA, PC) and did not include others like SA2 or unmyelinated afferents, which might contribute to tactile coding. Furthermore, the study primarily focused on the glabrous skin of the hand, and tactile coding might differ in hairy skin or other body regions.

Future work could explore tactile inputs arising from natural behaviors like grasping and manipulating objects, which involve more complex interactions. Investigating how the shape of the hand and varying afferent distributions across different body areas influence population coding strategies is also a promising direction. The authors also suggest exploring how these findings can inform the design of downstream decoding mechanisms for artificial systems.

Summary updated

Abstract

Sensory information is conveyed by populations of neurons, and coding strategies cannot always be deduced when considering individual neurons. Moreover, information coding depends on the number of neurons available and on the composition of the population when multiple classes with different response properties are available. Here, we study population coding in human tactile afferents by employing a recently developed simulator of mechanoreceptor firing activity. First, we demonstrate that the optimal afferent density for conveying maximal information depends on the tactile feature under consideration and the afferent class coding this feature. Second, we find that information is spread across different classes for all tactile features, such that combining information from multiple afferent classes improves information transmission, and is often more efficient than increasing the density of afferents from the same class. Finally, we test the importance of timing precision and afferent identity in the population code to probe whether temporal and spatial information can be traded against each other. Destroying temporal information turns out to be more destructive than removing spatial information, and the contribution of either cannot be completely recovered from the other. Overall, our results suggest that both optimal afferent innervation densities and the composition of the population depend in complex ways on the tactile features in question, potentially accounting for the variety in which tactile peripheral populations are assembled in different regions across the body.
BibTeX
@article{corniani2022population,
  author = {Giulia Corniani and Miguel A. Casal and Stefano Panzeri and Hannes P. Saal},
  title = {{Population coding strategies in human tactile afferents}},
  journal = {PLOS Computational Biology},
  year = {2022},
  doi = {10.1371/journal.pcbi.1010763},
}
Figure