Mesoscale cortex-wide neural dynamics predict self-initiated actions in mice several seconds prior to movement

  1. Catalin Mitelut
  2. Yongxu Zhang
  3. Yuki Sekino
  4. Jamie D Boyd
  5. Federico Bollanos
  6. Nicholas V Swindale
  7. Greg Silasi
  8. Shreya Saxena
  9. Timothy H Murphy

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    The neural correlates of voluntary action is one of the most intriguing questions in neuroscience, but studying it at laboratory settings is incredibly difficult. Here, the authors have used an impressive range of methods and analyses approaches in mice to investigate the neural activity preceding voluntary action in mice. Using widefield calcium imaging in mice to study volition is novel and welcome but the great strength of this paper is its wide range of analyses approaches. There remains a question to what extent the findings reveal specific properties of 'voluntary action,'.

    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.

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Volition – the sense of control or agency over one’s voluntary actions – is widely recognized as the basis of both human subjective experience and natural behavior in nonhuman animals. Several human studies have found peaks in neural activity preceding voluntary actions, for example the readiness potential (RP), and some have shown upcoming actions could be decoded even before awareness. Others propose that random processes underlie and explain pre-movement neural activity. Here, we seek to address these issues by evaluating whether pre-movement neural activity in mice contains structure beyond that present in random neural activity. Implementing a self-initiated water-rewarded lever-pull paradigm in mice while recording widefield [Ca++] neural activity we find that cortical activity changes in variance seconds prior to movement and that upcoming lever pulls could be predicted between 3 and 5 s (or more in some cases) prior to movement. We found inhibition of motor cortex starting at approximately 5 s prior to lever pulls and activation of motor cortex starting at approximately 2 s prior to a random unrewarded left limb movement. We show that mice, like humans, are biased toward commencing self-initiated actions during specific phases of neural activity but that the pre-movement neural code changes over time in some mice and is widely distributed as behavior prediction improved when using all vs. single cortical areas. These findings support the presence of structured multi-second neural dynamics preceding self-initiated action beyond that expected from random processes. Our results also suggest that neural mechanisms underlying self-initiated action could be preserved between mice and humans.

Article activity feed

  1. Version published to 10.7554/elife.76506 on eLife
  2. Version published to 10.1101/2021.12.17.473193v3 on bioRxiv
  3. eLife

    Author Response

    Reviewer 2

    Weaknesses: While I applaud the use of a "simplified" task in rodents to disambiguate controversial questions traditionally addressed in human studies, I found that the behavioral data were underanalyzed and thus not strongly supporting the central claim of the manuscript. Below are my main comments:

    1. One of the goals of the authors was to study the neural mechanisms underlying "voluntary" movements. While they acknowledge (in the discussion) that they do not have evidence that actions are "intentional", they make the assumption that mice do "form the intent to act near the lever pull time". To back up this assumption, the authors should at least present some evidence that the action of interest (i.e., the rewarded lever-pull) is not just a random jerky movement that happens to be rewarded once in a while. In fact, mice seemed to pull the lever very frequently and impulsively (the majority of inter-pull intervals were way below 3 s in Supplementary Fig. 1.2) even for the last sessions of the training. Therefore, it is not readily apparent that mice apply any control to their lever-pull actions. Providing evidence that the action is goal-directed is important if the goal of the paper is to study neural signatures of the intention to act. A somewhat compelling analysis could be to compare rewarded lever-pulls with "spontaneous" movements, provided that these two types of movement can be convincingly characterized as goal-directed vs. incidental. In contrast, throughout the manuscript, the neural activity aligned to rewarded lever-pull events (which are assumed to be "voluntary" actions) is compared to the neural activity aligned to random times during the task (whether or not it involved movements), which may not be the most convincing control.

    a. We agree with the reviewer and have provided additional explanation and evidence for the learning component of our study.

    1. The learning trajectory of mice is also not well characterized (e.g. changes in inter-pull intervals are not quantified, nor the relative increase in rewarded actions across training sessions, etc.). Yet, several claims in the paper are directly based on the fact that mice have learned to pull the lever after 3 s interval to receive water rewards (which relates to point 1). In particular, one important assumption in the paper is that as mice learn, the lever-pull movements become more stereotyped, but this has not been shown explicitly. It would be helpful, for example, to see how analog traces of lever-pulling change throughout the learning stages and how the variance of the movement across trials decreases in late sessions.

    a. We agree and have provided additional analysis and figures.

    1. The central claim of the paper is that rewarded lever-pulls can be predicted from pre-movement neural activity several seconds (even up to 10 s) prior to the action. However, obvious motor confounds and other alternative explanations have not been convincingly ruled out. In fact, the action of lever pulling may require a series of complex movements (like changing posture, extending the forelimb, reaching the lever, grabbing the lever, etc.). The authors themselves mentioned that they found strong correlations between lever pulls and body movements in all mice, but the data is not used nor shown in the paper. The motor commands preceding but related to lever-pull could unfold at least a few hundreds of milliseconds prior to the detection of lever-pull in the task, and thus be reflected in the neural activity that is predictive of the lever pull. Moreover, if this series of movements is highly stereotyped, and in turn leads to stereotyped neural activity (like the slow oscillations observed before the lever-pulls), it could explain why the detection of lever pulling actions always occurs at a given phase of the neural oscillation. Such observations that stereotyped movements occur way before the lever-pull detection could partially rule out the fully "cognitive" explanation proposed in the paper, but would concur with recent findings that showed that ramping neural activity can be, for the most part, explained by movement-related activity (Musall et al., 2019).

    a. We agree with the reviewer and have added analysis panels showing cross correlations for behavior as well as additional panels showing there are no behavior initiation sequences in the data.

    1. Toward the end of the result section (Fig. 6), the authors briefly begin to address the issue about whether pre-movement activity can really be considered movement free. Here, "lockouts", i.e. periods where other movements (like licking, or previous lever-pulls) did not occur, were introduced in the analysis. The lockouts altered the earliest-decoding-time (EDT) of the lever-pull (in some mice EDT was even divided by half: from -4 s to -2 s). However, the effects of "micro-movements" like facial movements or changes in body posture may not be taken into account with the lockout approach. Such micro-movements have been shown to explain a large variance of the neural activity (see Stringer et al. 2019 and Musall et al. 2019). Therefore, to fully control for movement confounds, the effect of high dimensional/micro-movements extracted from video recordings should be removed from the neural activity. These analyses could yield a much shorter EDT (e.g., -0.15 s), more consistent with previous reports.

    a. We agree and have added additional discussion about sequences of behaviors or micromovements.

  4. eLife

    Evaluation Summary:

    The neural correlates of voluntary action is one of the most intriguing questions in neuroscience, but studying it at laboratory settings is incredibly difficult. Here, the authors have used an impressive range of methods and analyses approaches in mice to investigate the neural activity preceding voluntary action in mice. Using widefield calcium imaging in mice to study volition is novel and welcome but the great strength of this paper is its wide range of analyses approaches. There remains a question to what extent the findings reveal specific properties of 'voluntary action,'.

    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.

  5. eLife

    Reviewer #1 (Public Review):

    Here the authors examined volitional neural signals during an un-cued lever pulling task. The authors impressively monitored cortical activity using widefield Ca++ imaging over many sessions (days to months).

    Mice received water-rewards to motivate them to pull the lever. The major aim in this study was to understand when neural signals corresponding to the upcoming level pull appeared within the cortex. This is an important question in sensorimotor control, namely, how and where do volitional signals associated with our actions arise in the brain? The authors compare their results at various points to human motor control, where readiness potentials appear prior to the execution of movement. Thus, the authors' study could make a meaningful contribution to understanding how neural activity changes prior to the initiation of a voluntary movement (i.e. there is no cue in their task).

    Prior to each lever pull, neural activity exhibited oscillatory patterns that sharpened in amplitude with proximity to the pulling event. These oscillations could be observed throughout the cortex: in retrosplenial, barrel, somatosensory, visual, and motor regions. As previously reported, neural activity exhibited a reduction in variance prior to movement initiation. The collapsing in variance was observed prior to the movement in all areas, excepting visual cortex. These changes in variance were echoed by convex hull analyses, which aimed to summarize the space spanned by pre-pull neural activity. As the movement approached, the convex hull gradually narrowed. The intersection between the lever pull's convex hull and the convex hull associated with all paw movements appeared to decrease with training in the task. This suggested a restructuring in neural activity whereby lever pull movements became more distinct in their neural activity patterns.

    To understand whether pre-pull neural activity was associated with the upcoming movement, the authors used an SVM to predict whether neural activity over some pre-pull window could predict the upcoming lever pull. They observed that SVM classifiers could indeed predict the upcoming action well in advance of the behavior, in some cases 10-15 sec prior to the lever pull. This result is quite notable, given that previous evidence in humans suggests that readiness potentials arise 0.5-1.5 seconds prior to movement. Thus, the authors' study suggests a much longer time horizon for volitional signals in the brain.

    The authors' question is both intriguing and important to the field of motor control, but certain details about their task complicate interpretations of their data. Most importantly, in the pre-pull period, behavior was not generally quiescent. Because the task did not use a cue, animals engaged in many behaviors in the windows preceding rewarded lever pulls. Thus, it is hard to know whether pre-pull neural activity relates to the upcoming rewarded lever pull, or earlier lever pull events (and other behaviors) that were likely to occur within the SVM window itself. While the authors used a 3-second lockout in their analysis (only considered rewarded pulls that were not preceded by lever pulls in the past 3 seconds), it remains challenging to interpret neural activity prior to threshold value (and it is currently unclear whether this lockout period excludes all lever pulls, or only some that met certain criteria). Along these lines, when the lockout window was extended in a control analysis, the SVM's time horizon for volitional signals shortened, suggesting that pre-pull behaviors indeed influenced the primary results. Thus, it remains unclear exactly when volitional signals arise in this task. The authors could greatly strengthen their paper with additional neural and behavioral analyses on these matters.

  6. eLife

    Reviewer #2 (Public Review):

    Gradual increases in neural activity have been observed prior to execution of movements in several species, including in humans. Yet, the nature of this ramping activity and whether it signals the intent to act is debated. To bring novel insights into this debate, the authors examined the cortical activity that preceded self-initiated actions in mice and evaluated the predictive relationship between pre-movement activity in different regions of the dorsal cortex and the lever-pull behavior of mice.

    Strengths: The manuscript addresses a timely and controversial topic in the field of neuroscience. The authors provide a novel and relatively rich dataset in mice with the goal of tackling a question that remains challenging to answer with current techniques used in human subjects. The data consists of longitudinal widefield imaging across the entire dorsal cortex of head-fixed mice performing a behavioral task paired with video monitoring of body movements. This dataset could be quite versatile and potentially useful to the community if shared publicly as the authors intend to. The manuscript is well written and easy to read, and the figures are quite self-explanatory.

    Weaknesses: While I applaud the use of a "simplified" task in rodents to disambiguate controversial questions traditionally addressed in human studies, I found that the behavioral data were under-analyzed and thus not strongly supporting the central claim of the manuscript. Below are my main comments:

    1. One of the goals of the authors was to study the neural mechanisms underlying "voluntary" movements. While they acknowledge (in the discussion) that they do not have evidence that actions are "intentional", they make the assumption that mice do "form the intent to act near the lever pull time". To back up this assumption, the authors should at least present some evidence that the action of interest (i.e., the rewarded lever-pull) is not just a random jerky movement that happens to be rewarded once in a while. In fact, mice seemed to pull the lever very frequently and impulsively (the majority of inter-pull intervals were way below 3 s in Supplementary Fig. 1.2) even for the last sessions of the training. Therefore, it is not readily apparent that mice apply any control to their lever-pull actions. Providing evidence that the action is goal-directed is important if the goal of the paper is to study neural signatures of the intention to act. A somewhat compelling analysis could be to compare rewarded lever-pulls with "spontaneous" movements, provided that these two types of movement can be convincingly characterized as goal-directed vs. incidental. In contrast, throughout the manuscript, the neural activity aligned to rewarded lever-pull events (which are assumed to be "voluntary" actions) is compared to the neural activity aligned to random times during the task (whether or not it involved movements), which may not be the most convincing control.

    2. The learning trajectory of mice is also not well characterized (e.g. changes in inter-pull intervals are not quantified, nor the relative increase in rewarded actions across training sessions, etc.). Yet, several claims in the paper are directly based on the fact that mice have learned to pull the lever after 3 s interval to receive water rewards (which relates to point 1). In particular, one important assumption in the paper is that as mice learn, the lever-pull movements become more stereotyped, but this has not been shown explicitly. It would be helpful, for example, to see how analog traces of lever-pulling change throughout the learning stages and how the variance of the movement across trials decreases in late sessions.

    3. The central claim of the paper is that rewarded lever-pulls can be predicted from pre-movement neural activity several seconds (even up to 10 s) prior to the action. However, obvious motor confounds and other alternative explanations have not been convincingly ruled out. In fact, the action of lever pulling may require a series of complex movements (like changing posture, extending the forelimb, reaching the lever, grabbing the lever, etc.). The authors themselves mentioned that they found strong correlations between lever pulls and body movements in all mice, but the data is not used nor shown in the paper. The motor commands preceding but related to lever-pull could unfold at least a few hundreds of milliseconds prior to the detection of lever-pull in the task, and thus be reflected in the neural activity that is predictive of the lever pull. Moreover, if this series of movements is highly stereotyped, and in turn leads to stereotyped neural activity (like the slow oscillations observed before the lever-pulls), it could explain why the detection of lever-pulling actions always occurs at a given phase of the neural oscillation. Such observations that stereotyped movements occur way before the lever-pull detection could partially rule out the fully "cognitive" explanation proposed in the paper, but would concur with recent findings that showed that ramping neural activity can be, for the most part, explained by movement-related activity (Musall et al., 2019).

    4. Toward the end of the result section (Fig. 6), the authors briefly begin to address the issue about whether pre-movement activity can really be considered movement free. Here, "lockouts", i.e. periods where other movements (like licking, or previous lever-pulls) did not occur, were introduced in the analysis. The lockouts altered the earliest-decoding-time (EDT) of the lever-pull (in some mice EDT was even divided by half: from -4 s to -2 s). However, the effects of "micro-movements" like facial movements or changes in body posture may not be taken into account with the lockout approach. Such micro-movements have been shown to explain a large variance of the neural activity (see Stringer et al. 2019 and Musall et al. 2019). Therefore, to fully control for movement confounds, the effect of high dimensional/micro-movements extracted from video recordings should be removed from the neural activity. These analyses could yield a much shorter EDT (e.g., -0.15 s), more consistent with previous reports.

  7. eLife

    Reviewer #3 (Public Review):

    The neural correlates of voluntary action is one of the most intriguing questions in neuroscience. But, at the same time, studying it at laboratory settings is incredibly difficult. Here, Mitelut et al., have used an impressive range of methods and analyses approaches in mice to investigate the neural activity preceding voluntary action in mice. They showed that: 1) Self-initiated behaviour could be decoded up to 10s in advance. 2) Decoding works best when using information from across the cortex rather than any specific region of interest. 3) The neural dynamics become increasingly stereotyped prior to movement. 4) This latter effect becomes stronger with weeks of increased task performance. 5) Single trial variance decreases prior to movement, another sign of increase stereotypy in neural dynamics. 6) Random body movements do not influence the findings. 7) The pre-movement neural dynamics are in slow frequency range. The authors then went on to conclude that neural mechanisms underlying self-initiated voluntary action is preserved between mice and humans and suggested that mice could be an adequate model for studying the neural correlates of self-initiated action.

    Using widefield calcium imaging in mice to study volition is novel and welcome but the great strength of this paper is its wide range of analyses approaches. Importantly, all these approaches, more or less, point to same direction: neural dynamics become increasingly stereotyped prior to movement. However, I am not convinced the findings reveal any specific property of 'voluntary action', nor it is clear what the authors refer to by 'voluntary action'. The authors first define volition as "the sense of control or agency over one's voluntary actions", but then they do not clarify how one can measure this 'sense of control' in rodents. They acknowledge, however, that it is not possible to measure the subjective experience of intention in mice but then insist on referring to the behaviour under study as "voluntary behaviour", even when it does not correspond with their own definition of volition. The question then is what makes the behaviour that is under study 'voluntary'. It is perhaps 'voluntary' because it is goal-directed. If this is the case, then the neural signal should be compared with another condition where action is not goal-directed but is habitual. This is, however, unlikely to be the case, because even though the authors did not directly examine the goal-directedness of the behaviour (by violating the value of the outcome or action-outcome contingency), they showed that the signal gets stronger with weeks of increased task performance, which actually makes the behaviour more stereotypical and habitual. Alternatively, the behaviour is perhaps 'voluntary' because it is self-initiated as opposed to externally triggered. If this is the case, then neural signal should be compared with another condition where action is performed in response to an external cue. The authors then should show that increase stereotypy in neural dynamics is only observed prior to self-initiated but not externally triggered actions. Only then one can conclude the presence of a specific signal preceding voluntary action.

    Importantly, it is not enough to compare data with a random section of the neural activity. Therefore, to make any conclusion regarding the specificity of the neural signal to 'voluntary' behaviour, the activity needs to be compared with a similar condition where behaviour is not 'voluntary', whatever its definition is.

  8. Version published to 10.1101/2021.12.17.473193v2 on bioRxiv
  9. Version published to 10.1101/2021.12.17.473193v1 on bioRxiv