Abstract
Although the generation of movements is a fundamental function of the nervous system, the underlying neural principles remain unclear. As flexor and extensor muscle activities alternate during rhythmic movements such as walking, it is often assumed that the responsible neural circuitry is similarly exhibiting alternating activity1. Here we present ensemble recordings of neurons in the lumbar spinal cord that indicate that, rather than alternating, the population is performing a low-dimensional ‘rotation’ in neural space, in which the neural activity is cycling through all phases continuously during the rhythmic behaviour. The radius of rotation correlates with the intended muscle force, and a perturbation of the low-dimensional trajectory can modify the motor behaviour. As existing models of spinal motor control do not offer an adequate explanation of rotation1,2, we propose a theory of neural generation of movements from which this and other unresolved issues, such as speed regulation, force control and multifunctionalism, are readily explained.
This is a preview of subscription content, access via your institution
Access options
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$29.99 / 30 days
cancel any time
Subscribe to this journal
Receive 51 print issues and online access
$199.00 per year
only $3.90 per issue
Buy this article
- Purchase on Springer Link
- Instant access to full article PDF
Prices may be subject to local taxes which are calculated during checkout
Similar content being viewed by others
Data availability
Data are available through a link to an online repository, which can be found on the laboratory web page (https://berg-lab.net/) or on reasonable request from the corresponding authors.
Code availability
The Python code that was used for simulating the BSG network is available in an online repository (https://github.com/BergLab/BSG). The Matlab code for analysing the experimental data is available on the laboratory web page (https://berg-lab.net/) or on reasonable request from the corresponding authors.
References
Grillner, S. & El Manira, A. Current principles of motor control, with special reference to vertebrate locomotion. Physiol. Rev. 100, 271–320 (2020).
McCrea, D. A. & Rybak, I. A. Organization of mammalian locomotor rhythm and pattern generation. Brain Res. Rev. 57, 134–146 (2008).
Arber, S. & Costa, R. M. Connecting neuronal circuits for movement. Science 360, 1403–1404 (2018).
Gosgnach, S. et al. Delineating the diversity of spinal interneurons in locomotor circuits. J. Neurosci. 37, 10835–10841 (2017).
Goulding, M. Circuits controlling vertebrate locomotion: moving in a new direction. Nat. Rev. Neurosci. 10, 507–518 (2009).
Stein, P. S. G. & Daniels-McQueen, S. Modular organization of turtle spinal interneurons during normal and deletion fictive rostral scratching. J. Neurosci. 22, 6800–6809 (2002).
Petersen, P. C. & Berg, R. W. Lognormal firing rate distribution reveals prominent fluctuation-driven regime in spinal motor networks. eLife 5, e18805 (2016).
Radosevic, M. et al. Decoupling of timescales reveals sparse convergent CPG network in the adult spinal cord. Nat. Commun. 10, 2937 (2019).
Russo, A. A. et al. Motor cortex embeds muscle-like commands in an untangled population response. Neuron 97, 953–966 (2018).
Churchland, M. M. et al. Neural population dynamics during reaching. Nature 487, 51–56 (2012).
Sussillo, D. et al. A neural network that finds a naturalistic solution for the production of muscle activity. Nat. Neurosci. 18, 1025–1033 (2015).
Bruno, A. M. et al. A spiral attractor network drives locomotion in Aplysia. eLife 6, e27342 (2017).
Berkinblit, M. B. et al. Generation of scratching. I. Activity of spinal interneurons during scratching. J. Neurophysiol. 41, 1040–1057 (1978).
Auyong, N. et al. Preferred locomotor phase of activity of lumbar interneurons during air-stepping in subchronic spinal cats. J. Neurophysiol. 106, 1943–1953 (2011).
Kwan, A. C. et al. Spatiotemporal dynamics of rhythmic spinal interneurons measured with two-photon calcium imaging and coherence analysis. J. Neurophysiol. 104, 3323–3333 (2010).
Kiehn, O. et al. Excitatory components of the mammalian locomotor CPG. Brain Res. Rev. 57, 56–63 (2008).
Rancic, V. & Gosgnach, S. Recent insights into the rhythmogenic core of the locomotor CPG. Int. J. Mol. Sci. 22, 1394 (2021).
Dayan, P. & Abbott, L. F. Theoretical Neuroscience: Computational and Mathematical Modeling of Neural Systems (MIT Press, 2001).
Hennequin, G., Agnes, E. J. & Vogels, T. P. Inhibitory plasticity: balance, control, and codependence. Annu. Rev. Neurosci. 40, 557–579 (2017).
Berg, R. W., Willumsen, A. & Lindén, H. When networks walk a fine line: balance of excitation and inhibition in spinal motor circuits. Curr. Opin. Physiol. 8, 76–83 (2019).
Berg, R. W., Alaburda, A. & Hounsgaard, J. Balanced inhibition and excitation drive spike activity in spinal half-centers. Science 315, 390–393 (2007).
Machado, T. A. Probing Circuits for Spinal Motor Control. PhD thesis, Columbia Univ. (2015).
Ramirez, J.-M. & Baertsch, N. A. The dynamic basis of respiratory rhythm generation: one breath at a time. Annu. Rev. Neurosci. 41, 475–499 (2018).
Rajan, K., Abbott, L. F. & Sompolinsky, H. Stimulus-dependent suppression of chaos in recurrent neural networks. Phys. Rev. E 82, 011903 (2010).
Hennequin, G., Vogels, T. P. & Gerstner, W. Optimal control of transient dynamics in balanced networks supports generation of complex movements. Neuron 82, 1394–1406 (2014).
van Vreeswijk, C. & Sompolinsky, H. Chaos in neuronal networks with balanced excitatory and inhibitory activity. Science 274, 1724–1726 (1996).
Beiran, M. & Ostojic, S. Contrasting the effects of adaptation and synaptic filtering on the timescales of dynamics in recurrent networks. PLoS Comput. Biol. 15, e1006893 (2019).
Stroud, J. P., Porter, M. A., Hennequin, G. & Vogels, T. P. Motor primitives in space and time via targeted gain modulation in cortical networks. Nat. Neurosci. 21, 1774–1783 (2018).
Gosgnach, S. et al. V1 spinal neurons regulate the speed of vertebrate locomotor outputs. Nature 440, 215–219 (2006).
Callahan, R. A. et al. Spinal V2b neurons reveal a role for ipsilateral inhibition in speed control. eLife 8, e47837 (2019).
Berkowitz, A. et al. Roles for multifunctional and specialized spinal interneurons during motor pattern generation in tadpoles, zebrafish larvae, and turtles. Front. Behav. Neurosci. 4, 36 (2010).
Briggman, K. L. & Kristan, W. B. Multifunctional pattern-generating circuits. Annu. Rev. Neurosci. 31, 271–294 (2008).
Sussillo, D. & Abbott, L. F. Generating coherent patterns of activity from chaotic neural networks. Neuron 63, 544–557 (2009).
Logiaco, L., Abbott, L. & Escola, S. Thalamic control of cortical dynamics in a model of flexible motor sequencing. Cell Rep. 35, 109090 (2021).
Mortin, L. I., Keifer, J. & Stein, P. S. Three forms of the scratch reflex in the spinal turtle: movement analyses. J. Neurophysiol. 53, 1501–1516 (1985).
Elsayed, G. F., Lara, A. H., Kaufman, M. T., Churchland, M. M. & Cunningham, J. P. Reorganization between preparatory and movement population responses in motor cortex. Nat. Commun. 7, 13239 (2016).
Gallego, J. A., Perich, M. G., Miller, L. E. & Solla, S. A. Neural manifolds for the control of movement. Neuron 94, 978–984 (2017).
Aljadeff, J., Stern, M. & Sharpee, T. Transition to chaos in random networks with cell-type-specific connectivity. Phys. Rev. Lett. 114, 088101 (2015).
Bikoff, J. B. et al. Spinal inhibitory interneuron diversity delineates variant motor microcircuits. Cell 165, 207–219 (2016).
Levine, A. J. et al. Identification of a cellular node for motor control pathways. Nat. Neurosci. 17, 586–593 (2014).
Berg, R. W. in The Neural Control of Movement (eds Whelan, P. J. & Sharples, S. A.) 205–219 (Elsevier, 2020).
Walloe, S. et al. Stereological estimate of the total number of neurons in spinal segment D9 of the red-eared turtle. J. Neurosci. 31, 2431–2435 (2011).
Bannatyne, B. A. et al. Neurotransmitters and motoneuron contacts of multifunctional and behaviorally specialized turtle spinal cord interneurons. J. Neurosci. 40, 2680–2694 (2020).
Stein, P. S. G. Central pattern generators in the turtle spinal cord: selection among the forms of motor behaviors. J. Neurophysiol. 119, 422–440 (2018).
Kadir, S. N. et al. High-dimensional cluster analysis with the masked EM algorithm. Neural Comput. 26, 2379–2394 (2014).
Berg, R. W. et al. Exploratory whisking by rat is not phase locked to the hippocampal theta rhythm. J. Neurosci. 26, 6518–6522 (2006).
Percival, D. B. & Walden, A. T. Spectral Analysis for Physical Applications - Multitaper and Conventional Univariate Techniques (Cambridge Univ. Press, 1998)
Bonizzato, M. et al. Multi-pronged neuromodulation intervention engages the residual motor circuitry to facilitate walking in a rat model of spinal cord injury. Nat. Commun. 12, 1925 (2021).
Hochberg, L. R. et al. Neuronal ensemble control of prosthetic devices by a human with tetraplegia. Nature 442, 164–171 (2006).
Serruya, M. D. et al. Instant neural control of a movement signal. Nature 416, 141–142 (2002).
Dougherty, K. J. & Ha, N. T. The rhythm section: an update on spinal interneurons setting the beat for mammalian locomotion. Curr. Opin. Physiol. 8, 84–93 (2019).
Lafreniere-Roula, M. & McCrea, D. A. Deletions of rhythmic motoneuron activity during fictive locomotion and scratch provide clues to the organization of the mammalian central pattern generator. J. Neurophysiol. 94, 1120–1132 (2005).
Zhong, G. et al. Neuronal activity in the isolated mouse spinal cord during spontaneous deletions in fictive locomotion: insights into locomotor central pattern generator organization. J. Physiol. 590, 4735–4759 (2012).
Suresh, A. K. et al. Neural population dynamics in motor cortex are different for reach and grasp. eLife 9, e58848 (2020).
Vogels, T. P. & Abbott, L. F. Gating multiple signals through detailed balance of excitation and inhibition in spiking networks. Nat. Neurosci. 12, 483–491 (2009).
Kanaka, R. & Abbott, L. F. Eigenvalue spectra of random matrices for neural networks. Phys. Rev. Lett. 97, 188104 (2006).
Mardia, K. V. & Jupp, P. E. Directional Statistics (Wiley, 2000).
Acknowledgements
This work was supported by The Independent Research Fund Denmark and Mobilex (number 8020-00436B, DFF–1333-00226A), the Carlsberg Foundation (number CF18-0845) and the Lundbeck Foundation (number R366-2021-233).
Author information
Authors and Affiliations
Contributions
R.W.B. conceived the original experiments. P.C.P. set up and performed the experiments, collected the data and analysed some of the data. M.V., H.L. and R.W.B. conceived the original theory. H.L. performed the model simulations. H.L. and R.W.B. designed and developed the theory, analysed the experimental data and wrote the manuscript.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data figures and tables
Extended Data Fig. 1 Firing rate is rhythmic with rotational population dynamics across trials in the lumbar spinal motor network during rhythmic movement.
a, The firing rates of 3 sample units (black) with their spike times indicated as blue dots. A sinus function was fitted to the firing rate (red) and the mean square error is indicated (bottom right). b, More sample units, with the mean square error indicated to the left. c, Distribution of mean square errors for the population (n = 214). The mean square error is unitless, since the firing rates were high pass filtered and normalized (maximal firing is 1). d, The firing rates (normalized, color coded) of 214 spinal neurons in laminae VII-VIII as a function of time and sorted according to phase with respect to the nerve activity (hip flexor). Eight consecutive trials from same experiment with a 5 min pause in between each. e, The phase distribution across the neuronal population. f, The population activity has rotational dynamics, as demonstrated by the circular motion of the first two PCs. The PCs were calculated by the data of one trial (trial 3, "*") and the applied to the rest of the trials. The sorting of neurons was according to their phase relation with representative nerve for one trial (also trial 3, "*") and this order was maintained for the rest of the trials. Bottom scale bars represent 1000.
Extended Data Fig. 2 Rotational population dynamics in a spinal motor network across animals.
a, The rhythmic firing rates in populations of spinal neurons in laminae VII-VIII shown in colors as a function of time and sorted according to phase with respect to a nerve (hip flexor). A representative trial from 5 experiments of approximately 10 s demonstrate similar sequential/rotational population activity. Animal used in Extended Data Fig. 1 is marked "*". b, The corresponding distribution of neurons having preferred phases among the population of rhythmic neurons. c, Population activity represented by first two PCs exhibit rotational dynamics. Scale bars: 250. d, Cumulative explained variance by principal components, indicating the population dynamics is low-dimensional, i.e. most of the variance is captured by few components.
Extended Data Fig. 3 Neuronal population trajectories in PC-space have lower tangling than the corresponding motor nerve trajectories.
a, Illustration that during rotational dynamics the points in the trajectory that move in opposite direction are also far apart, i.e. they have low tangling (left), whereas during alternation the points of the trajectory that move in opposite direction are also close, i.e. have high tangling (right). b, The ratio of tangling metric of the PC trajectory of the nerves (Qnerve) to that of the network (Qnetwork). This ratio is close to 100%, which indicates most trials and animals had a larger tangling of the motor nerves than the network. (N = 11, data set 1; N = 10, data set 2; N = 10, data set 3; N = 4, data set 4; N = 3, data set 5). c–g, Sample trials from 5 different data sets. Left is shown the phase sorted firing rate activity (top) and the associated nerves (bottom). The nerves were rectified and low-pass filtered (red) on temporal scale matching the firing rates. The PCs of network (middle left) and nerves (middle right, green). Scales of PCs are variance normalized. The tangling metric (Q) for the nerve PCs (in 3 dimensions) is calculated as a function of time (t) through the trial and plotted versus that for the network. The ratio of points below the x = y–line (pale blue) is indicated in percent and form one point in panel (b). Note that the nerve trajectories more resemble "alternation" whereas the network more resembles "rotation"-scheme of (a).
Extended Data Fig. 4 Rotational ensemble activity within the excitatory and inhibitory sub-populations in the BSG-model.
a, Activation of the motor circuit by descending drive. b, The firing rates of 10 sample excitatory neurons oscillate because of the descending input. (c) Sorting the excitatory neurons according to phase of firing rates reveals a sequential activity like the previously observed for all neurons. d-e, Activity and similar sorting of the inhibitory sub-populations reveals similar sequential and rotational dynamics within that sub-population. f, The network eigenmode for the whole network: Each dot represent both the phase (the polar angle) and the peak firing rate (the radius) for a given neuron (n = 200). g-h, Similar plot for the excitatory and inhibitory populations. i-k, the distribution of phases in linear histograms for all neurons (i), excitatory (j) and inhibitory neurons (k). To be compared with experimental distributions (Extended Data Figs. 1 and 2).
Extended Data Fig. 5 BSG-model: Correlation between descending drive and radius of rotation as well as amplitude of nerve output without affecting the period.
a, For low neuronal gain (top), the eigenvalue spectrum does not have any eigenvalues that cross the stability line (broken vertical line). As the gain increases (downward direction) the spectrum expands, and eigenvalues cross the stability line. For larger gain the eigenvalues cross the stability line farther. b, The associated population dynamics (sorted firing rates) exhibit oscillation of increasing magnitude as the drive increase. c, The rotational dynamics also has a radius that increases with increasing drive. d, The resulting motor nerve output is also increasing in amplitude. e, Descending drive (gain) versus the population firing rate (RMS), radius of rotation in PC space, f, and amplitude of nerve output (flexor RMS), g. h, the radius of rotation (PC1 RMS) vs. the nerve amplitude (flexor RMS).
Extended Data Fig. 6 Radius of rotation correlates with nerve output in experiment.
a, Sample trial where the population activity was divided up in pieces with the corresponding nerve output b. c, The PC manifolds had rotation with varying radius. d-f, other pieces with same organization. g, The RMS of the nerve activity versus the RMS of the first two PCs for various pieces of activity had a significant correlation (F-statistic of rejection of no trend at p«0.01). h, The R2 values for all animal tested (n=5). *: F-statistic of rejection of null hypothesis of zero correlation at p«0.01. f, Scale bar: 1000.
Extended Data Fig. 7 Representation of one behavior in the subspace of another behavior.
a, The variance captured by the projection of the network dynamics onto the first three PCs of another trial (green) normalized by the variance captured by the PCs of its own dynamics. Orange: the subspace-overlap of a different behavior. Independent samples: N= Same behavior/ different behavior, N = 5/6, data set 1; N = 3/3, data set 2; N=8/10, data set 3; N = 3/4, data set 4; N = 2/3, data set 5). Whisker plots represent min and max values. Box plots represent median −25% and +75% quartiles. b, the subspace representation of the nerve activity of same behavior (green) and a different behavior (orange). N=Same behavior / different behavior: N = 5/6, data set 1; N = 3/3, data set 2; N = 8/10, data set 3; N = 3/4, data set 4; N = 2/3, data set 5). Whisker plots represent min and max values. Box plots represent median −25% and +75% quartiles. c, Nerve overlap plotted against the network overlap. A large overlap in nerve output is associated with a large overlap in network overlap. Gray line represents a linear fit, red region represents 95% confidence. d-aa, The flexor/extensor nerve output from the BSG-network. e, The sorted neuronal population firing rate (n = 400 neurons) with rotational dynamics. (cc) Color map of the population firing rate. dd, Mean (red) and variance of the population activity. e, same organization as in (d), but for experimental data. Animal no. 3 trial 8.
Extended Data Fig. 8 Multifunctionalism in the BSG-model.
a, Five examples of specific activation/modulation of selected neurons in the network ("activation profiles"). The top profile has an even distribution, whereas all the below profiles have selective modulation of specific neurons. b, The ensemble activity as a result of the activation profile show a sequential activity, with similar but not identical sequence of activity. c, The first two PCs, based on the top activation profile, all exhibit rotational dynamics, albeit with different radius and trajectories. (d) the output motor patterns associated with the different activation profiles and ensemble activities.
Extended Data Fig. 9 "Deletions" both in the experimental data and in the BSG model.
a, Six trials (1–6) shown with the neuronal firing rates sorted according to phase (color map, top) and the 6 nerves (bottom) during a motor behavior (pocket scratching). The absence of a burst, i.e. a deletion, was observed in the hip extensor nerve recording (red dots) whereas the hip flexor (bottom trace) seems to continue and combine two cycles although with a small decrease. Regular bursts are also indicated (blue dots). The corresponding trajectories in PC-space are shown with the corresponding dots matching the time in the nerve activity. A selected period around the occurrence of one delete is indicated in the nerve traces (red vertical lines). The corresponding time in the trajectory is also indicated in red. Note that deletions tend to occur at smaller radius of the rotation and the population firing rates (color map) are dimmer at those instances. b, A proper motor behavior devoid of "deletions" can be produced by the balanced sequence generator despite receiving a varying input (top). The firing rates for the sorted neuronal population (middle), and the resulting motor nerve output pattern (bottom). c, The appropriate motor program shown in (b) is achieved by a selective gain-modulation, i.e. gain-profile (y-axis), across the neuronal population (x-axis). d, Population activity from (b) represented by in PC-space by the first two components. e, When the varying input transiently becomes too low at a certain phase the nerve cycle is absent, i.e. a "deletion" has occurred (red dots). The firing rates of the neuronal population will be lower at these instances and hence appear dimmer in the color map (middle). A consequent absence of a burst in the nerve is seen (nerve 3, compare red and blue dots). f, the PC-trajectories corresponding to (e) indicated as 1,2 and 3. The temporary distortion of the trajectory at a particular phase is associated with a deletion (red dots). Red parts of the trajectory represent the period between vertical red bars indicated in the nerve activities (d–f). Compare with (a).
Extended Data Fig. 10 Reconstruction of nerve output based on linear decoding of neuronal population activity.
a, A linear decoder function was estimated using a training set consisting of 9 trials of same behavior. Top: color coded firing rates for the neuronal population (sorted according to phase) with 9 trials concatenated. Bottom: the rectified and low-pass filtered motor nerve output of 6 nerves. b, A trial, that was not included in the training set, is used for validation of the linear decoder. Top: the firing rates of the population, like (a). Bottom: The nerve output of 3 selected nerves (rectified and LP-filtered in green, nerves 3, 4 and 6). The reconstructed standard deviations of the nerves (orange) are multiplied by white Gaussian noise to imitate nerve output (gray). c, the correlation between predicted and actual nerve output for the six nerves (individual dots) are shown for two different motor behaviors (right and left pocket scratching) across the 5 experiments. The median value across all nerves and experiments is R = 0.6. All 5 data sets had correlations, which were found significantly different than zero using a t-test of Pearson linear correlation, p << 0.01, N = 6. d, Training set consisting of 9 bouts, ie. trials of different motor behaviors, which is used to train a linear decoder function. Top: color coded firing rates for the neuronal population (sorted according to phase) with 9 concatenated bouts. Bottom: the rectified and low-pass filtered motor nerve output of 6 nerves. e, two trials, that was not included in the training set, contained instances of "deletions". Top: the firing rates of the population, like (a). Bottom: The nerve output of 3 selected nerves (rectified and LP-filtered in green, nerves 3, 4 and 6). The reconstructed standard deviations of the nerves (orange) are multiplied by white Gaussian noise to imitate nerve output (gray). f, the correlation between predicted and actual nerve output for the six nerves (individual dots) are shown for two different motor behaviors (right and left pocket scratching) across the 5 experiments. The median value across all nerves and experiments is R = 0.6. All 6 correlations were found significantly different than zero using a t-test of Pearson linear correlation, p << 0.01, N = 1300 temporal-measurements.
Supplementary information
Supplementary Information
This file contains Supplementary Table 1, Discussion, Equations and legends for Videos 1–3.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Lindén, H., Petersen, P.C., Vestergaard, M. et al. Movement is governed by rotational neural dynamics in spinal motor networks. Nature 610, 526–531 (2022). https://doi.org/10.1038/s41586-022-05293-w
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41586-022-05293-w
This article is cited by
-
Preparatory activity and the expansive null-space
Nature Reviews Neuroscience (2024)
-
Latent neural population dynamics underlying breathing, opioid-induced respiratory depression and gasping
Nature Neuroscience (2024)
-
Neuronal travelling waves explain rotational dynamics in experimental datasets and modelling
Scientific Reports (2024)
-
Emergence of task-related spatiotemporal population dynamics in transplanted neurons
Nature Communications (2023)
-
Reconstructing computational system dynamics from neural data with recurrent neural networks
Nature Reviews Neuroscience (2023)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.