Consistent coordination patterns provide near perfect behavior decoding in a comprehensive motor program for insect flight

Joy Putney, Marko Angjelichinoski, Robert Ravier, Silvio Ferrari, Vahid Tarokh, Simon SponbergbioRxiv 2021 · DOI

TL;DR

This study demonstrates that precise spike timing in a comprehensive set of hawk moth flight muscles can nearly perfectly decode six distinct flight behaviors. Remarkably, this decoding accuracy is maintained even when using only half the muscles, suggesting significant redundancy and conserved coordination patterns across different flight maneuvers.

What Problem Does This Paper Address?

Decoding behavior from motor activity is a fundamental challenge in neuroscience. While precise spike timing is known to encode information in motor systems, its utility for decoding complex behaviors and understanding muscle coordination across different actions has been underexplored. Traditional methods often rely on broader measures like EMG or incomplete neural recordings, limiting our ability to examine how fine-grained temporal patterns in motor commands contribute to diverse behaviors and how muscle coordination might be shared across them. This paper aims to bridge this gap by using a complete, spike-resolved motor program in hawk moths to decode flight behaviors and investigate conserved coordination patterns.

Methods

The researchers developed a comprehensive, spike-resolved motor program for hawk moth flight, recording from all 10 primary muscles that control wing movements. They employed a Gaussian kernel filtering approach to convert individual spike trains into continuous time-domain representations. These representations were then fed into a Principal Component Analysis (PCA) pipeline to reduce dimensionality and capture key patterns of muscle covariation. Finally, a Linear Discriminant Analysis (LDA) classifier was used to decode the behavioral states elicited by visual stimuli. This pipeline allowed for the investigation of how different levels of detail in spike timing and the number of muscles used affect decoding accuracy. They also explored decoding using simpler representations like spike rate and first-spike timing.

Datasets

The study utilized data collected from 9 tethered hawk moths (Manduca sexta). These moths were presented with wide-field drifting visual stimuli designed to elicit six distinct flight behaviors: pitch up, pitch down, roll left, roll right, yaw left, and yaw right. The data collected included simultaneous recordings of muscle action potentials (EMG) from 10 muscles, as well as force and torque measurements.

Key Insights and Ideas

  • Spike timing is king for decoding: The study strongly suggests that precise spike timing, down to the millisecond scale, is crucial for accurately decoding complex motor behaviors, outperforming simpler spike rate representations.
  • Conserved coordination across behaviors: Muscle coordination patterns, identified through PCA, are not unique to specific behaviors. Patterns learned from one pair of opposing movements (e.g., pitch up/down) can be effectively used to decode movements in a different pair (e.g., roll left/right), indicating a fundamental underlying coordination strategy.
  • Redundancy is the name of the game: Near-perfect decoding accuracy can be achieved using as few as half of the muscles in the motor program. This highlights significant redundancy in the motor system, where information is distributed across multiple muscles.
  • Linear models are sufficient: Despite the potential for non-linear control in biological systems, a simple linear decoding pipeline was sufficient to achieve near-perfect behavioral decoding, making the results more interpretable.

Key Results

  • Near-perfect decoding: The researchers achieved an average decoding accuracy of 99.7 ± 0.2% for six distinct flight behaviors using all 10 muscles.
  • Robustness to muscle reduction: Decoding accuracy remained high (above 98%) even when using only 5 muscles (half of the motor program), with a standard deviation encompassing 99.5% decoding accuracy.
  • Transferability of coordination patterns: Muscle coordination patterns derived from one pair of visual stimulus conditions could be used to decode behaviors in other pairs with high accuracy (above 95% with ~70 principal components).
  • Optimal timing precision: Peak decoding accuracy was achieved with a Gaussian kernel width (σ) on the millisecond scale, emphasizing the importance of precise spike timing.

Limitations and Future Work

  • Animal-to-animal variation: While overall decoding accuracy was high, there was inter-individual variation in how muscle contributions defined the LDA axes of best separation, suggesting unique muscle activation patterns for individuals. Future work could explore how to better account for this variation.
  • Force/torque correlation: The study primarily focused on decoding from neural activity. Investigating how these coordination patterns directly relate to force and torque output and how animal-to-animal variation in these outputs influences decoding would be valuable.
  • Tethered vs. Free flight: The experiments were conducted on tethered moths. Future research could explore whether these conserved coordination patterns hold true in free flight, where a broader range of motor control is required.
  • Functional implications of redundancy: While redundancy was demonstrated, the specific functional advantages of this redundancy (e.g., robustness to injury, fine-tuning) warrant further investigation.

Summary updated

Abstract

Abstract Patterns of motor activity can be used to decode behavior state. Precise spike timing encoding is present in many motor systems, but is not frequently utilized to decode behavior or to examine how coordination is achieved across many motor units. Testing whether the same coordinated sets of muscles control different movements is difficult without a complete motor representation at the level of the currency of control – action potentials. Here, we demonstrate nearly perfect decoding of six hawk moth flight behaviors elicited in response to wide-field drifting visual stimuli about the flight axes – pitch, roll, and yaw – using a comprehensive, spike-resolved motor program and a simple linear decoding pipeline. A complex decoding scheme is not necessary, even if the functional patterns of control are nonlinear. We show that muscle covariation present in one pair of visual stimulus conditions can be used to decode behavior in a different pair of visual stimulus conditions, indicating the presence of conserved muscle coordination patterns at the level of motor neuronal timings in functionally distinct behaviors. We also demonstrate that as few as half the muscles can be used to retain decoding performance, linking coordination to redundancy in encoding, if not function, across the entire moth flight motor program.
BibTeX
@misc{putney2021consistent,
  author = {Joy Putney and Marko Angjelichinoski and Robert Ravier and Silvio Ferrari and Vahid Tarokh and Simon Sponberg},
  title = {{Consistent coordination patterns provide near perfect behavior decoding in a comprehensive motor program for insect flight}},
  year = {2021},
  doi = {10.1101/2021.07.13.452211},
  howpublished = {bioRxiv},
}
Figure