Decoding locomotion from population neural activity in moving C. elegans
Curation statements for this article:-
Curated by eLife
Evaluation Summary:
This paper will be of interest to a wide range of systems neuroscientists seeking to understanding the relationship between neuronal activity and behavior. Building on previous technical advances in brain-wide imaging of neuronal activity (Ca signals) in freely moving animals (Caenorhabditis elegans), it demonstrates that a linear regression model is sufficient reconstruct key parameters of locomotion - velocity and body curvature - from the imaging data and documents differences in activity between freely moving and immobilized worms.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- @aksharma_118's saved articles (aksharma_118)
Abstract
We investigated the neural representation of locomotion in the nematode C. elegans by recording population calcium activity during movement. We report that population activity more accurately decodes locomotion than any single neuron. Relevant signals are distributed across neurons with diverse tunings to locomotion. Two largely distinct subpopulations are informative for decoding velocity and curvature, and different neurons’ activities contribute features relevant for different aspects of a behavior or different instances of a behavioral motif. To validate our measurements, we labeled neurons AVAL and AVAR and found that their activity exhibited expected transients during backward locomotion. Finally, we compared population activity during movement and immobilization. Immobilization alters the correlation structure of neural activity and its dynamics. Some neurons positively correlated with AVA during movement become negatively correlated during immobilization and vice versa. This work provides needed experimental measurements that inform and constrain ongoing efforts to understand population dynamics underlying locomotion in C. elegans .
Article activity feed
-
-
Author Response:
Reviewer #3:
A. Summary of what the authors were trying to achieve
The authors seek to understand how whole-animal behavior is represented in the nervous system. They approach this problem utilizing high-speed volumetric calcium imaging in freely moving nematodes (C. elegans). In recording from a majority of neurons in the head, this approach is state-of-the art in C. elegans and, arguably, far beyond what is likely to be achieved in most other organisms in the foreseeable future. Imaging data are analyzed by training a linear decoder to predict the instantaneous locomotion velocity and body curvature from instantaneous neuronal activity at single neuron resolution.
B. Major strengths and weaknesses of the methods and results
The paper has numerous strengths:
- State-of-the art simultaneous imaging of …
Author Response:
Reviewer #3:
A. Summary of what the authors were trying to achieve
The authors seek to understand how whole-animal behavior is represented in the nervous system. They approach this problem utilizing high-speed volumetric calcium imaging in freely moving nematodes (C. elegans). In recording from a majority of neurons in the head, this approach is state-of-the art in C. elegans and, arguably, far beyond what is likely to be achieved in most other organisms in the foreseeable future. Imaging data are analyzed by training a linear decoder to predict the instantaneous locomotion velocity and body curvature from instantaneous neuronal activity at single neuron resolution.
B. Major strengths and weaknesses of the methods and results
The paper has numerous strengths:
- State-of-the art simultaneous imaging of brain-wide neuronal activity and unrestrained behavior.
- The overall approach has been published in two papers by this group and one from another group, but this is the first paper that actually takes the next logical step: connecting the recordings back to behavior. This is a major strength.
- Comparison of neuronal dynamics during locomotion and immobilization in the same worm.
- Rigorous data collection and modeling.
The paper in its current form has a number of weaknesses:
- Several of the main findings of the paper seem rather obvious. (i) "We report that a neural population more accurately decodes locomotion than any single neuron (Abstract)". Similarly, "We conclude that neural population codes are important for understanding neural dynamics of behavior in moving animals." (ii) "Our measurements suggest that neural dynamics from immobilized animals may not entirely reflect the neural dynamics of locomotion." Consider rephrasing, as this sentence is almost a tautology: "…neural dynamics in the absence of locomotion may not entirely reflect the dynamics in the presence of locomotion (line 379)." Can these conclusions be rephrased, or put in a more significant context?
Thank you for this feedback. We have completely rewritten the relevant portion of the discussion to better place our findings in context and better convey the implications.
"That C. elegans neural dynamics exhibit different correlation structure during movement than during immobilization has implications for neural representations of locomotion. For example, it is now common to use dimensionality reduction techniques like PCA to search for low-dimensional trajectories or manifolds that relate to behavior or decision making in animals undergoing move- ment (Churchland et al., 2012; Harvey et al., 2012; Shenoy et al., 2013) or in immobilized animals undergoing fictive locomotion (Briggman et al., 2005; Kato et al., 2015). PCA critically depends on the correlation structure to define its principal components. In C. elegans, the low-dimensional neural trajectories observed in immobilized animals undergoing fictive locomotion, and the un- derlying correlation structure that defines those trajectories, are being used to draw conclusions about neural dynamics of actual locomotion. Our measurements suggest that to obtain a more complete picture of C. elegans neural dynamics related to locomotion, it will be helpful to probe neural state space trajectories recorded during actual locomotion: both because the neural dy- namics themselves may differ during immobilization, but also because the correlation structure observed in the network, and consequently the relevant principal components, change upon im- mobilization. These changes may be due to proprioception (Wen et al., 2012), or due to different internal states associated with fictive versus actual locomotion."
And we have rewritten portions of the introduction, for example:
"There has not yet been a systematic exploration of the types and distribution of locomotor related signals present in the neural population during movement and their tunings. So for example, it is not known whether all forward related neurons exhibit duplicate neural signals or whether a variety of distinct signals are combined. Interestingly, results from recordings in immobile animals suggest that population neural state space trajectories in a low dimensional space may encode global motor commands (Kato et al., 2015) , but this has yet to be explored in moving animals. Despite growing interest in the role of population dynamics in the worm, their dimensionality, and their relation to behavior (Costa et al., 2019; Linderman et al., 2019; Brennan and Proekt, 2019; Fieseler et al., 2020) it is not known how locomotory related information contained at the population level compares to that contained at the level of single neurons. And importantly, current findings of population dynamics related to locomotion in C. elegans are from immobilized animals. While there are clear benefits in studying fictive locomotion (Ahrens et al., 2012; Briggman et al., 2005; Kato et al., 2015), it is not known for C. elegans how neural population dynamics during immobile fictive locomotion compare to population dynamics during actual movement."
- The rationale for the decoding exercises seems underdeveloped. Figs. 3-6 are motivated by the question of whether "activity of the neural population might be more informative of the worm's locomotion than an individual neuron." It just seems obvious this will be the case. There might be a missed opportunity, here. Perhaps a stronger motivation would be to ask whether locomotion related signals can be found in the subset of neurons found in the head. The alternative hypothesis would be that head neurons alone are not sufficient, the implication being that the ventral cord and/or tail ganglia must be included.
We have added rationale for decoding in the results section:
“...because an effective strategy adopted by the decoder may also be available to the brain, understanding how the decoder works also illustrates plausible strategies that the brain could employ to represent locomotion.”
And added motivation in the introduction:
“...Despite growing interest in the role of population dynamics in the worm, their dimensionality, and their relation to behavior (Costa et al., 2019; Linderman et al., 2019; Brennan and Proekt, 2019; Fieseler et al., 2020) it is not known how locomotory related information contained at the population level compares to that contained at the level of single neurons. ”
The ideas about head vs ventral cord and tail are interesting, but since we are limited in what we can say about signals beyond the head we hesitated to pursue that path.
- The logic of how decoding exercises are interpreted also seems underdeveloped: (i) Why isn't the finding of locomotion-related signals in the head a forgone conclusion? After all, the worm's head is literally "carving the furrow" that the rest of the body follows, leading to body curvatures that ought to be correlated with with neuronal activity in the head. Furthermore, a substantial fraction of head neurons are nose and neck muscle motor neurons. These contribute to overall thrust, which in the worm's fluidic regime is proportional to velocity. Thus, as stronger head motor neuron activation would generate more thrust, there a correlation with velocity is expected. (ii) What does it mean to say, "The distribution of weights assigned by the decoder provides information about how behavior is represented in the brain (p. 8)"? Who or what is reading this representation? Is the representation detected by the decoder necessarily in the same or similar language used by the worm's brain? If not, how are the decoder findings significant for understanding locomotion in the worm? (iii) It seems likely that the decoder picks up signals of neurons that causally regulate locomotion, but also signals that follow from it (e.g., efference copy, proprioception, re-entrant signals, etc.). Assuming this is true, again: how are the decoder findings significant for understanding locomotion in the worm? (iv) In what ways, if at all, is the decoder a model for worm locomotion? If it's not a model, how does it improve our understanding of locomotion, or our future ability to construct and informative model?
Response to items 3 and 4 are combined below.
- The Discussion seems to miss key points: (i) What are the main limitations of the approach (paucity of identified neurons, inability of Ca imaging to report inhibition, etc)? (ii) Why are the limitations non-fatal? (iii) What are the broader impacts of the main conclusions? For example, what is this significance of the finding of locomotion representations in the C. elegans nervous system or, indeed, in any nervous system? How do the results illuminate neural mechanisms of behavior?
We thank the reviewer for posing these thoughtful questions. We have rewritten the discussion to better explore some of the implications of our finding that a linear model works to decode locomotion and we explicitly highlight limitations including those related to:
Neural identities: “ Future studies using newly developed methods for identifying neurons (Yemeni et al., 2020) are needed to reveal the identities of those neurons weighted by the decoder for decoding velocity, curvature, or both.”
Linear vs nonlinear models: “...This does not preclude the brain from using other methods for representing behavior.”
Distinguishing motor commands from signals that monitor: “... the measurements here do not distinguish between neural signals that drive locomotion, such as motor commands; and neural signals that monitor locomotion generated elsewhere, such as proprioceptive feedback”
-
Evaluation Summary:
This paper will be of interest to a wide range of systems neuroscientists seeking to understanding the relationship between neuronal activity and behavior. Building on previous technical advances in brain-wide imaging of neuronal activity (Ca signals) in freely moving animals (Caenorhabditis elegans), it demonstrates that a linear regression model is sufficient reconstruct key parameters of locomotion - velocity and body curvature - from the imaging data and documents differences in activity between freely moving and immobilized worms.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)
-
Reviewer #1 (Public Review):
This paper is generally well written, and it represents a lot of hard work.
Key strengths:
- The design and methods for extraction of time-series of neuronal activity, corrections for motion artifacts, and correlations with locomotion;
- The direct evidence for the dramatic impact on population dynamics performed under different experimental conditions (moving versus paralyses);
- The conclusion that the simplest linear regression model predicts locomotion the best.
Key weakness:
The current paper seems to emphasize the following conclusions: two largely distinct and small neuron populations predict the two features of locomotion, and population-based prediction outperforms single neuron prediction. To me, scientifically, neither offers surprises nor reveals truly exciting new insights. It was perhaps too …
Reviewer #1 (Public Review):
This paper is generally well written, and it represents a lot of hard work.
Key strengths:
- The design and methods for extraction of time-series of neuronal activity, corrections for motion artifacts, and correlations with locomotion;
- The direct evidence for the dramatic impact on population dynamics performed under different experimental conditions (moving versus paralyses);
- The conclusion that the simplest linear regression model predicts locomotion the best.
Key weakness:
The current paper seems to emphasize the following conclusions: two largely distinct and small neuron populations predict the two features of locomotion, and population-based prediction outperforms single neuron prediction. To me, scientifically, neither offers surprises nor reveals truly exciting new insights. It was perhaps too predictable from the perspective of the systems neuroscience, but too vague (lacking the necessary biological details for the groups of neurons) to be more informative for experimental neuroscience. The most informative new takeaway for me was that the simple linear model works the best in behavioral prediction, however I did not see insightful discussions on its potential implication on the property of the C. elegans neural network or the brain's locomotory presentation.
-
Reviewer #2 (Public Review):
In the submitted manuscript, Hallinen et al. dissect how neural activity across a large population of C. elegans neurons gives rise to the animal's locomotion. First, they analyze single neurons in their population-level recordings and find that different cells have different "tunings" with respect to the animal's velocity and curvature. They also show that individual identified neurons (AVAL/R) display their well-known activity patterns in these brain-wide datasets. They then use ridge regression to predict velocity and curvature from single neurons, as well as the full set of neurons, and show that the prediction is better when the full population is used. They present exemplary data suggesting that different neurons predict different aspects of the animal's behavior (for example, forward/backward …
Reviewer #2 (Public Review):
In the submitted manuscript, Hallinen et al. dissect how neural activity across a large population of C. elegans neurons gives rise to the animal's locomotion. First, they analyze single neurons in their population-level recordings and find that different cells have different "tunings" with respect to the animal's velocity and curvature. They also show that individual identified neurons (AVAL/R) display their well-known activity patterns in these brain-wide datasets. They then use ridge regression to predict velocity and curvature from single neurons, as well as the full set of neurons, and show that the prediction is better when the full population is used. They present exemplary data suggesting that different neurons predict different aspects of the animal's behavior (for example, forward/backward transition vs. high-frequency changes in forward velocity). They also estimate the number of neurons that are necessary to fully predict these behavioral variables by training a range of models with different #s of neurons. Finally, the authors perform recordings where animals are immobilized partway through the recording and show that the correlational structure of neural activity changes after immobilization.
This paper uses state-of-the-art methods to address an interesting problem -- how the brain gives rise to behavior. As of now, there have been very few large-scale C. elegans brain recordings performed in freely-moving animals in order to address a biological problem (just a handful of papers primarily focused on methodology), so this work represents an important advance for the field. The analysis of single neuron "tunings" could be improved (or at least further unpacked) and concerns about noise levels in the data also impact some of the interpretations of these otherwise very interesting datasets.
-
Reviewer #3 (Public Review):
A. Summary of what the authors were trying to achieve
The authors seek to understand how whole-animal behavior is represented in the nervous system. They approach this problem utilizing high-speed volumetric calcium imaging in freely moving nematodes (C. elegans). In recording from a majority of neurons in the head, this approach is state-of-the art in C. elegans and, arguably, far beyond what is likely to be achieved in most other organisms in the foreseeable future. Imaging data are analyzed by training a linear decoder to predict the instantaneous locomotion velocity and body curvature from instantaneous neuronal activity at single neuron resolution.
B. Major strengths and weaknesses of the methods and results
The paper has numerous strengths:
State-of-the art simultaneous imaging of brain-wide neuronal …
Reviewer #3 (Public Review):
A. Summary of what the authors were trying to achieve
The authors seek to understand how whole-animal behavior is represented in the nervous system. They approach this problem utilizing high-speed volumetric calcium imaging in freely moving nematodes (C. elegans). In recording from a majority of neurons in the head, this approach is state-of-the art in C. elegans and, arguably, far beyond what is likely to be achieved in most other organisms in the foreseeable future. Imaging data are analyzed by training a linear decoder to predict the instantaneous locomotion velocity and body curvature from instantaneous neuronal activity at single neuron resolution.
B. Major strengths and weaknesses of the methods and results
The paper has numerous strengths:
State-of-the art simultaneous imaging of brain-wide neuronal activity and unrestrained behavior.
The overall approach has been published in two papers by this group and one from another group, but this is the first paper that actually takes the next logical step: connecting the recordings back to behavior. This is a major strength.
Comparison of neuronal dynamics during locomotion and immobilization in the same worm.
Rigorous data collection and modeling.
The paper in its current form has a number of weaknesses:
Several of the main findings of the paper seem rather obvious. (i) "We report that a neural population more accurately decodes locomotion than any single neuron (Abstract)". Similarly, "We conclude that neural population codes are important for understanding neural dynamics of behavior in moving animals." (ii) "Our measurements suggest that neural dynamics from immobilized animals may not entirely reflect the neural dynamics of locomotion." Consider rephrasing, as this sentence is almost a tautology: "...neural dynamics in the absence of locomotion may not entirely reflect the dynamics in the presence of locomotion (line 379)." Can these conclusions be rephrased, or put in a more significant context?
The rationale for the decoding exercises seems underdeveloped. Figs. 3-6 are motivated by the question of whether "activity of the neural population might be more informative of the worm's locomotion than an individual neuron." It just seems obvious this will be the case. There might be a missed opportunity, here. Perhaps a stronger motivation would be to ask whether locomotion related signals can be found in the subset of neurons found in the head. The alternative hypothesis would be that head neurons alone are not sufficient, the implication being that the ventral cord and/or tail ganglia must be included.
The logic of how decoding exercises are interpreted also seems underdeveloped: (i) Why isn't the finding of locomotion-related signals in the head a forgone conclusion? After all, the worm's head is literally "carving the furrow" that the rest of the body follows, leading to body curvatures that ought to be correlated with with neuronal activity in the head. Furthermore, a substantial fraction of head neurons are nose and neck muscle motor neurons. These contribute to overall thrust, which in the worm's fluidic regime is proportional to velocity. Thus, as stronger head motor neuron activation would generate more thrust, there a correlation with velocity is expected. (ii) What does it mean to say, "The distribution of weights assigned by the decoder provides information about how behavior is represented in the brain (p. 8)"? Who or what is reading this representation? Is the representation detected by the decoder necessarily in the same or similar language used by the worm's brain? If not, how are the decoder findings significant for understanding locomotion in the worm? (iii) It seems likely that the decoder picks up signals of neurons that causally regulate locomotion, but also signals that follow from it (e.g., efference copy, proprioception, re-entrant signals, etc.). Assuming this is true, again: how are the decoder findings significant for understanding locomotion in the worm? (iv) In what ways, if at all, is the decoder a model for worm locomotion? If it's not a model, how does it improve our understanding of locomotion, or our future ability to construct and informative model?
The Discussion seems to miss key points: (i) What are the main limitations of the approach (paucity of identified neurons, inability of Ca imaging to report inhibition, etc)? (ii) Why are the limitations non-fatal? (iii) What are the broader impacts of the main conclusions? For example, what is this significance of the finding of locomotion representations in the C. elegans nervous system or, indeed, in any nervous system? How do the results illuminate neural mechanisms of behavior?
C. Appraisal of whether the authors achieved their aims, and whether the results support their conclusions.
The authors convincingly demonstrate that locomotion-related signals are present in their recordings. The effects are fairly robust. But if an implied aim was also to elucidate mechanisms of locomotion in C. elegans, this was not achieved.
D. A discussion of the likely impact of the work on the field, and the utility of the methods and data to the community.
The authors have not made the case that their main findings are broadly significant. We learn what the linear decoder finds in the neuronal data - sustained and transient locomotion signals and distinct populations of velocity and curvature tuned neurons - but we do not learn what these properties of the decoder have to say about biological mechanisms. This problem is especially acute given: (i) the likelihood of neural correlations with behavior that are not functional representations of behavior and (ii) the absence of evidence that the decodable information is in fact used by the worm. We also learn that, as one would expect, immobilization alters the correlation structure of neural activity, but this finding has not been placed in a mechanistic context.
-
-