Article activity feed

  1. Evaluation Summary:

    This manuscript bridges neurophysiology and biomechanics and is of broad interest in improving our understanding of insect flight control. Here, Putney et al. record the activity of the flight muscles of tethered hawkmoths and demonstrate that the direction of the visual stimulus to which the insect responds can be classified using precisely timed information on muscle activity from a subset of the flight muscles. This is an important step in identifying the mapping from visual input to motor output, albeit that the mapping identified here is qualitative (i.e. classification of visual stimulus direction) rather than quantitative (i.e. prediction of output torque or apparent angular velocity of self-motion).

    (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. Reviewer #1 agreed to share their name with the authors.)

    Read the original source
    Was this evaluation helpful?
  2. Reviewer #1 (Public Review):

    This is a clearly written paper, the results of which are also clear and well supported. Putney et al. recorded the electrical activity of a subset of the flight muscles in the tobacco hawkmoth Manduca sexta during tethered flight. The moths were presented with six visual stimuli, comprising up or down pitch motions, right- or left-handed roll motions, and right- or left-handed yaw motions. The muscle recordings were analysed by: (i) filtering the signals using a gaussian kernel of variable width; (ii) linearly combining the filtered signals by projecting them into a new basis using principal components analysis (PCA); (iii) applying linear discriminant analysis (LDA) on a subset of these principal components, to classify the recordings according to stimulus condition.

    The authors demonstrate that their method allows robust classification of the visual stimulus associated with each electrophysiological recording (better than 99.5% classification accuracy for n=4 moths with complete recordings from all 10 muscles that were studied). This well-supported result is not in itself surprising given the causality of the input-output relationship and the high dimension of the output data relative to the low dimension of the input classification. It follows that the key strengths of the manuscript lie in the extent to which it explores the details of this result. This is done in several ways:

    First, the authors explore the effect of varying the width of their gaussian smoothing kernel. They demonstrate convincingly that the accuracy of classification is lost if the gaussian kernel is too broad, which amounts to showing that lowpass filtering the data eliminates important information. The authors interpret this result as showing that the analysis of their muscle recordings is "sufficient to predict behavior, but only if precise timing information in included". This summary statement, expressed in the time domain, may invite the reader to assume that the key information lies in the arrival time of each spike, so it is worth noting that it could be reframed equivalently in terms of the spectral content of the signal - as indeed it is in Fig. 4C.

    It is also worth adding that two of the N=4 moths with complete data appear to behave differently with respect to the drop-off in decoding accuracy with increasing gaussian kernel width (Fig. 3A), although there seems to be no discussion of this point.

    Second, the authors explore the effect of data reduction in terms of: (i) the number of principal components required to capture the variation in the signals, and (ii) the number of muscles included in the analysis. This is a strength of the paper insofar as it provides a high level of model reduction, and also allows the authors to include results from N=5 moths with incomplete data (i.e. moths with missing muscle recordings). Of course, this is also a weakness insofar as there is missing data and a small absolute number of individuals sampled, but this needs to be viewed in the context of the extreme challenge of making multiple muscle recordings simultaneously.

    Third, the authors use the results of their PCA-LDA analysis to identify muscle coordination patterns, from which they conclude that "while the realization of muscle activity in each of the functionally distinct pairs of behavioral states about each flight axis changes, the underlying muscle coordination patterns are conserved." This conclusion is reached by calculating the inner product of the vectors providing the best separation (i.e classification) of different directions of motion within and between flight axes. As the inner product of these unit vectors is constrained to take values between 0 and 1, it would be helpful to consider whether there is some form of randomisation analysis that could be used to identify a null distribution against which the apparently quite weak co-directionality of the vectors could be assessed. This is important because the weakness of the co-directionality forms the basis of the conclusion referred to above.

    In summary, the authors' methodological pipeline will be useful in future studies and lays important groundwork for future work combining analysis of muscle activity, wing kinematics, and force-torque output in the visuomotor responses of insects.

    Finally, whilst the context of this work is explained well in relation to previous electrophysiological studies, other relevant work exploring the mapping from visual input to force-torque in hawkmoths using similar stimulus arrangements (Windsor et al., 2014; doi:10.1098/rsif.2013.0921) is ignored, and may provide some insight into the neuromechanical couplings relevant to the discussion. This earlier work showed that whereas visual stimuli in pitch and roll produce almost pure pitch and roll torques, yaw stimuli produce coupled roll and yaw moments. This coupling ought in principle to be present within the details of muscle coordination identified in the PCA-LDA analysis.

    Read the original source
    Was this evaluation helpful?
  3. Reviewer #2 (Public Review):

    Putney et al. has recorded EMG from 10 muscles within tethered moths, manduca sexta, and then played them rolling, yawing, and pitching visual stimuli while recording the reaction torques of the tethered animal and the EMG. They then applied statistical models to correlate the neural activity of these muscles with the directions of roll, yaw, and pitches that were measured from the animal. They found that steering of the animal was mostly correlated with activity in the steering muscles, and found that the models needed time-related correlations in order to correlate an individual animal with its individual behaviour. There were, however, large variations between animals, preventing one animal's parameters from being able to be used for another animal.

    The figures are professionally done, and it should be commented upon that the text is of high quality.

    Read the original source
    Was this evaluation helpful?