Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Article
  • Published:

Learning-induced autonomy of sensorimotor systems

Nature Neuroscience volume 18, pages 744–751 (2015)Cite this article

Subjects

Abstract

Distributed networks of brain areas interact with one another in a time-varying fashion to enable complex cognitive and sensorimotor functions. Here we used new network-analysis algorithms to test the recruitment and integration of large-scale functional neural circuitry during learning. Using functional magnetic resonance imaging data acquired from healthy human participants, we investigated changes in the architecture of functional connectivity patterns that promote learning from initial training through mastery of a simple motor skill. Our results show that learning induces an autonomy of sensorimotor systems and that the release of cognitive control hubs in frontal and cingulate cortices predicts individual differences in the rate of learning on other days of practice. Our general statistical approach is applicable across other cognitive domains and provides a key to understanding time-resolved interactions between distributed neural circuits that enable task performance.

This is a preview of subscription content, access via your institution

Access options

Buy this article

  • Purchase on Springer Link
  • Instant access to full article PDF
Buy now

Prices may be subject to local taxes which are calculated during checkout

Figure 1: Schematic of methods.
Figure 2: Summary architecture of learning.
Figure 3: Dynamic brain architecture associated with task practice.
Figure 4: Recruitment and integration were modulated by training.
Figure 5: Individual differences in brain network architecture map to task learning.
Figure 6: The release of a fronto-cingulate control network predicts individual differences in learning.

Similar content being viewed by others

References

  1. Ajemian, R., D'Ausilio, A., Moorman, H. & Bizzi, E. A theory for how sensorimotor skills are learned and retained in noisy and nonstationary neural circuits. Proc. Natl. Acad. Sci. USA 110, E5078–E5087 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  2. Grafton, S.T. & Hamilton, A.F. Evidence for a distributed hierarchy of action representation in the brain. Hum. Mov. Sci. 26, 590–616 (2007).

    Article  PubMed  PubMed Central  Google Scholar 

  3. Rowe, J.B. & Siebner, H.R. The motor system and its disorders. Neuroimage 61, 464–477 (2012).

    Article  PubMed  Google Scholar 

  4. Sun, F.T., Miller, L.M., Rao, A.A. & D'Esposito, M. Functional connectivity of cortical networks involved in bimanual motor sequence learning. Cereb. Cortex 17, 1227–1234 (2007).

    Article  PubMed  Google Scholar 

  5. Xiong, J. et al. Long-term motor training induced changes in regional cerebral blood flow in both task and resting states. Neuroimage 45, 75–82 (2009).

    Article  PubMed  Google Scholar 

  6. Büchel, C., Coull, J.T. & Friston, K.J. The predictive value of changes in effective connectivity for human learning. Science 283, 1538–1541 (1999).

    Article  PubMed  Google Scholar 

  7. Fedorenko, E. & Thompson-Schill, S.L. Reworking the language network. Trends Cogn. Sci. 18, 120–126 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

  8. Bassett, D.S. & Bullmore, E.T. Small-world brain networks. Neuroscientist 12, 512–523 (2006).

    Article  PubMed  Google Scholar 

  9. Bassett, D.S. & Bullmore, E.T. Human brain networks in health and disease. Curr. Opin. Neurol. 22, 340–347 (2009).

    Article  PubMed  PubMed Central  Google Scholar 

  10. Bassett, D.S. & Bullmore, E. Brain Anatomy and Small-world Networks (Betham, 2010).

  11. Bullmore, E.T. & Bassett, D.S. Brain graphs: graphical models of the human brain connectome. Annu. Rev. Clin. Psychol. 7, 113–140 (2011).

    Article  PubMed  Google Scholar 

  12. Sporns, O. Networks of the Brain (MIT Press, 2010).

  13. Bassett, D.S. et al. Dynamic reconfiguration of human brain networks during learning. Proc. Natl. Acad. Sci. USA 108, 7641–7646 (2011).

    Article  PubMed  PubMed Central  Google Scholar 

  14. Bassett, D.S. et al. Task-based core-periphery structure of human brain dynamics. PLoS Comput. Biol. 9, e1003171 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  15. Holme, P. & Saramäki, J. Temporal networks. Phys. Rep. 519, 97–125 (2012).

    Article  Google Scholar 

  16. Kivelä, M. et al. Multilayer networks. J. Complex Netw. 2, 203–271 (2014).

    Article  Google Scholar 

  17. Bassett, D.S. et al. Robust detection of dynamic community structure in networks. Chaos 23, 013142 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  18. Bassett, D.S. & Lynall, M.-E. Network methods to characterize brain structure and function. In The Cognitive Neurosciences 5th edn (eds Gazzaniga, M.S. & Mangun, G.R.) Ch. 79 (MIT Press, 2013).

  19. Doron, K.W., Bassett, D.S. & Gazzaniga, M.S. Dynamic network structure of interhemispheric coordination. Proc. Natl. Acad. Sci. USA 109, 18661–18668 (2012).

    Article  PubMed  PubMed Central  Google Scholar 

  20. Mantzaris, A.V. et al. Dynamic network centrality summarizes learning in the human brain. J. Complex Netw. 1, 83–92 (2013).

    Article  Google Scholar 

  21. Bassett, D.S., Wymbs, N.F., Porter, M.A., Mucha, P.J. & Grafton, S.T. Cross-linked structure of network evolution. Chaos 24, 013112 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

  22. Logan, G.D. Toward an instance theory of automatization. Psychol. Rev. 95, 492–527 (1988).

    Article  Google Scholar 

  23. Hikosaka, O., Nakamura, K., Sakai, K. & Nakahara, H. Central mechanisms of motor skill learning. Curr. Opin. Neurobiol. 12, 217–222 (2002).

    Article  CAS  PubMed  Google Scholar 

  24. Petersen, S.E., van Mier, H., Fiez, J.A. & Raichle, M.E. The effects of practice on the functional anatomy of task performance. Proc. Natl. Acad. Sci. USA 95, 853–860 (1998).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  25. Kelly, A.M. & Garavan, H. Human functional neuroimaging of brain changes associated with practice. Cereb. Cortex 15, 1089–1102 (2005).

    Article  PubMed  Google Scholar 

  26. Otto, A.R., Skatova, A., Madlon-Kay, S. & Daw, N.D. Cognitive control predicts use of model-based reinforcement learning. J. Cogn. Neurosci. 27, 319–333 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

  27. Mucha, P.J., Richardson, T., Macon, K., Porter, M.A. & Onnela, J.-P. Community structure in time-dependent, multiscale, and multiplex networks. Science 328, 876–878 (2010).

    Article  CAS  PubMed  Google Scholar 

  28. Power, J.D. et al. Functional network organization of the human brain. Neuron 72, 665–678 (2011).

    CAS  PubMed  PubMed Central  Google Scholar 

  29. Dayan, E. & Cohen, L.G. Neuroplasticity subserving motor skill learning. Neuron 72, 443–454 (2011).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  30. Bassett, D.S., Nelson, B.G., Mueller, B.A., Camchong, J. & Lim, K.O. Altered resting state complexity in schizophrenia. Neuroimage 59, 2196–2207 (2012).

    Article  PubMed  Google Scholar 

  31. Floyer-Lea, A. & Matthews, P.M. Distinguishable brain activation networks for short- and long-term motor skill learning. J. Neurophysiol. 94, 512–518 (2005).

    Article  CAS  PubMed  Google Scholar 

  32. Sakai, K. et al. Presupplementary motor area activation during sequence learning reflects visuo-motor association. J. Neurosci. 19, RC1 (1999).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  33. Grafton, S.T., Hazeltine, E. & Ivry, R.B. Motor sequence learning with the nondominant left hand. A PET functional imaging study. Exp. Brain Res. 146, 369–378 (2002).

    Article  PubMed  Google Scholar 

  34. Honda, M. et al. Dynamic cortical involvement in implicit and explicit motor sequence learning. A PET study. Brain 121, 2159–2173 (1998).

    Article  PubMed  Google Scholar 

  35. Lehéricy, S. et al. Distinct basal ganglia territories are engaged in early and advanced motor sequence learning. Proc. Natl. Acad. Sci. USA 102, 12566–12571 (2005).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  36. Siebenhühner, F., Weiss, S.A., Coppola, R., Weinberger, D.R. & Bassett, D.S. Intra- and inter-frequency brain network structure in health and schizophrenia. PLoS ONE 8, e72351 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  37. Patel, R., Spreng, R.N. & Turner, G.R. Functional brain changes following cognitive and motor skills training: a quantitative meta-analysis. Neurorehabil. Neural Repair 27, 187–199 (2013).

    Article  PubMed  Google Scholar 

  38. Esslinger, C. et al. Neural mechanisms of a genome-wide supported psychosis variant. Science 324, 605 (2009).

    Article  CAS  PubMed  Google Scholar 

  39. Orban, P. et al. The multifaceted nature of the relationship between performance and brain activity in motor sequence learning. Neuroimage 49, 694–702 (2010).

    Article  PubMed  Google Scholar 

  40. Elton, A. & Gao, W. Divergent task-dependent functional connectivity of executive control and salience networks. Cortex 51, 56–66 (2014).

    Article  PubMed  Google Scholar 

  41. Stoet, G. & Snyder, L.H. Neural correlates of executive control functions in the monkey. Trends Cogn. Sci. 13, 228–234 (2009).

    Article  PubMed  PubMed Central  Google Scholar 

  42. Shenhav, A., Botvinick, M.M. & Cohen, J.D. The expected value of control: an integrative theory of anterior cingulate cortex function. Neuron 79, 217–240 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  43. Chrysikou, E.G., Weber, M.J. & Thompson-Schill, S.L. A matched filter hypothesis for cognitive control. Neuropsychologia 62, 341–355 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  44. Thompson-Schill, S.L., Ramscar, M. & Chrysikou, E.G. Cognition without control: when a little frontal lobe goes a long way. Curr. Dir. Psychol. Sci. 18, 259–263 (2009).

    Article  PubMed  PubMed Central  Google Scholar 

  45. Collins, A.G. & Frank, M.J. Cognitive control over learning: creating, clustering, and generalizing task-set structure. Psychol. Rev. 120, 190–229 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  46. Galea, J.M., Albert, N.B., Ditye, T. & Miall, R.C. Disruption of the dorsolateral prefrontal cortex facilitates the consolidation of procedural skills. J. Cogn. Neurosci. 22, 1158–1164 (2010).

    Article  PubMed  PubMed Central  Google Scholar 

  47. Frey, S.H. et al. Neurological principles and rehabilitation of action disorders: computation, anatomy, and physiology (CAP) model. Neurorehabil. Neural Repair 25, 6S–20S (2011).

    Article  PubMed  PubMed Central  Google Scholar 

  48. Beeler, J.A., Petzinger, G. & Jakowec, M.W. The enemy within: propagation of aberrant corticostriatal learning to cortical function in Parkinson's disease. Front. Neurol. 4, 134 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  49. Cumberland Consensus Working Group et al. The future of restorative neurosciences in stroke: driving the translational research pipeline from basic science to rehabilitation of people after stroke. Neurorehabil. Neural Repair 23, 97–107 (2009).

  50. Sandrini, M. & Cohen, L.G. Noninvasive brain stimulation in neurorehabilitation. Handb. Clin. Neurol. 116, 499–524 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  51. Desmond, J.E. & Glover, G.H. Estimating sample size in functional MRI (fMRI) neuroimaging studies: statistical power analyses. J. Neurosci. Methods 118, 115–128 (2002).

    Article  PubMed  Google Scholar 

  52. Schmidt, R.A. & Lee, T.D. Motor Control and Learning: A Behavioral Emphasis 4th edn (Human Kinetics, 2005).

  53. Rosenbaum, D.A. Human Motor Control (Elsevier, 2010).

  54. Yarrow, K., Brown, P. & Krakauer, J.W. Inside the brain of an elite athlete: the neural processes that support high achievement in sports. Nat. Rev. Neurosci. 10, 585–596 (2009).

    Article  CAS  PubMed  Google Scholar 

  55. Snoddy, G.S. Learning and stability: a psychophysical analysis of a case of motor learning with clinical applications. J. Appl. Psychol. 10, 1–36 (1926).

    Article  Google Scholar 

  56. Crossman, E.R.F.W. A theory of the acquisition of speed-skill. Ergonomics 2, 153–166 (1959).

    Article  Google Scholar 

  57. Newell, K.M. & Rosenbloom, P.S. Mechanisms of skill acquisition and the law of practice. In Cognitive Skills and Their Acquisition (ed Anderson, J.R.) 1–55 (Lawrence Erlbaum Associates, 1981).

  58. Heathcote, A., Brown, S. & Mewhort, D.J. The power law repealed: the case for an exponential law of practice. Psychon. Bull. Rev. 7, 185–207 (2000).

    Article  CAS  PubMed  Google Scholar 

  59. Murphy, K., Birn, R.M., Handwerker, D.A., Jones, T.B. & Bandettini, P.A. The impact of global signal regression on resting state correlations: are anti-correlated networks introduced? Neuroimage 44, 893–905 (2009).

    PubMed  Google Scholar 

  60. Saad, Z.S. et al. Trouble at rest: how correlation patterns and group differences become distorted after global signal regression. Brain Connect. 2, 25–32 (2012).

    Article  PubMed  PubMed Central  Google Scholar 

  61. Chai, X.J., Castañón, A.N., Ongür, D. & Whitfield-Gabrieli, S. Anticorrelations in resting state networks without global signal regression. Neuroimage 59, 1420–1428 (2012).

    Article  PubMed  Google Scholar 

  62. Grinband, J., Wager, T.D., Lindquist, M., Ferrera, V.P. & Hirsch, J. Detection of time-varying signals in event-related fMRI designs. Neuroimage 43, 509–520 (2008).

    Article  PubMed  Google Scholar 

  63. Steffener, J., Tabert, M., Reuben, A. & Stern, Y. Investigating hemodynamic response variability at the group level using basis functions. Neuroimage 49, 2113–2122 (2010).

    Article  PubMed  Google Scholar 

  64. Calhoun, V.D., Stevens, M.C., Pearlson, G.D. & Kiehl, K.A. fMRI analysis with the general linear model: removal of latency-induced amplitude bias by incorporation of hemodynamic derivative terms. Neuroimage 22, 252–257 (2004).

    Article  CAS  PubMed  Google Scholar 

  65. Bullmore, E. & Sporns, O. Complex brain networks: graph theoretical analysis of structural and functional systems. Nat. Rev. Neurosci. 10, 186–198 (2009).

    Article  CAS  PubMed  Google Scholar 

  66. Bassett, D.S., Brown, J.A., Deshpande, V., Carlson, J.M. & Grafton, S.T. Conserved and variable architecture of human white matter connectivity. Neuroimage 54, 1262–1279 (2011).

    Article  PubMed  Google Scholar 

  67. Zalesky, A. et al. Whole-brain anatomical networks: does the choice of nodes matter? Neuroimage 50, 970–983 (2010).

    Article  PubMed  Google Scholar 

  68. Wang, J. et al. Parcellation-dependent small-world brain functional networks: a resting-state fMRI study. Hum. Brain Mapp. 30, 1511–1523 (2009).

    Article  PubMed  Google Scholar 

  69. Wig, G.S., Schlaggar, B.L. & Petersen, S.E. Concepts and principles in the analysis of brain networks. Ann. NY Acad. Sci. 1224, 126–146 (2011).

    Article  PubMed  Google Scholar 

  70. Smith, S.M. et al. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage 23, S208–S219 (2004).

    Article  PubMed  Google Scholar 

  71. Woolrich, M.W. et al. Bayesian analysis of neuroimaging data in FSL. Neuroimage 45, S173–S186 (2009).

    Article  PubMed  Google Scholar 

  72. Hermundstad, A.M. et al. Structural foundations of resting-state and task-based neural activity in the human brain. Proc. Natl. Acad. Sci. USA 110, 6169–6174 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  73. Hermundstad, A.M. et al. Structurally-constrained relationships between cognitive states in the human brain. PLoS Comput. Biol. 10, e1003591 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  74. Diedrichsen, J., Balster, J.H., Flavell, J., Cussans, E. & Ramnani, N. A probabilistic MR atlas of the human cerebellum. Neuroimage 46, 39–46 (2009).

    Article  PubMed  Google Scholar 

  75. Diedrichsen, J. et al. Imaging the deep cerebellar nuclei: a probabilistic atlas and normalization procedure. Neuroimage 54, 1786–1794 (2011).

    Article  CAS  PubMed  Google Scholar 

  76. Bullmore, E. et al. Wavelets and statistical analysis of functional magnetic resonance images of the human brain. Stat. Methods Med. Res. 12, 375–399 (2003).

    Article  PubMed  Google Scholar 

  77. Bullmore, E. et al. Wavelets and functional magnetic resonance imaging of the human brain. Neuroimage 23, S234–S249 (2004).

    Article  PubMed  Google Scholar 

  78. Brammer, M.J. Multidimensional wavelet analysis of functional magnetic resonance images. Hum. Brain Mapp. 6, 378–382 (1998).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  79. Achard, S., Salvador, R., Whitcher, B., Suckling, J. & Bullmore, E. A resilient, low-frequency, small-world human brain functional network with highly connected association cortical hubs. J. Neurosci. 26, 63–72 (2006).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  80. Bassett, D.S., Meyer-Lindenberg, A., Achard, S., Duke, T. & Bullmore, E. Adaptive reconfiguration of fractal small-world human brain functional networks. Proc. Natl. Acad. Sci. USA 103, 19518–19523 (2006).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  81. Achard, S. & Bullmore, E. Efficiency and cost of economical brain functional networks. PLoS Comput. Biol. 3, e17 (2007).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  82. Achard, S., Bassett, D.S., Meyer-Lindenberg, A. & Bullmore, E. Fractal connectivity of long-memory networks. Phys. Rev. E 77, 036104 (2008).

    Article  CAS  Google Scholar 

  83. Bassett, D.S., Meyer-Lindenberg, A., Weinberger, D.R., Coppola, R. & Bullmore, E. Cognitive fitness of cost-efficient brain functional networks. Proc. Natl. Acad. Sci. USA 106, 11747–11752 (2009).

    Article  PubMed  PubMed Central  Google Scholar 

  84. Lynall, M.E. et al. Functional connectivity and brain networks in schizophrenia. J. Neurosci. 30, 9477–9487 (2010).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  85. Percival, D.B. & Walden, A.T. Wavelet Methods for Time Series Analysis (Cambridge University Press, 2000).

  86. Sun, F.T., Miller, L.M. & D'Esposito, M. Measuring interregional functional connectivity using coherence and partial coherence analyses of fMRI data. Neuroimage 21, 647–658 (2004).

    Article  PubMed  Google Scholar 

Download references

Acknowledgements

D.S.B. acknowledges support from the John D. and Catherine T. MacArthur Foundation; the Alfred P. Sloan Foundation; the Army Research Laboratory (contract W911NF-10-2-0022); the Institute for Translational Medicine and Therapeutics; the US Army Research Office (contract W911NF-14-1-0679); the National Institute of Mental Health through award 2-R01-DC-009209-11; and National Science Foundation awards CRCNS BCS-1441502 and BCS-1430087 through the ENG, CISE and SBE directorates. M.Y. acknowledges support from the Applied Mathematics and Computational Science Graduate Program at the University of Pennsylvania. N.F.W. and S.T.G. were supported by PHS grant NS44393 and the Institute for Collaborative Biotechnologies through grant W911NF-09-0001 from the US Army Research Office. The content is solely the responsibility of the authors and does not necessarily represent the official views of any of the funding agencies. We thank D. Baker, S. Feldt Muldoon and Q. Telesford for comments on an earlier version of the manuscript, and we thank M.A. Porter and P.J. Mucha for helpful discussions.

Author information

Authors and Affiliations

  1. Department of Bioengineering, University of Pennsylvania, Philadelphia, Pennsylvania, USA

    Danielle S Bassett & Muzhi Yang

  2. Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia, Pennsylvania, USA

    Danielle S Bassett

  3. Applied Mathematics and Computational Science Graduate Group, University of Pennsylvania, Philadelphia, Pennsylvania, USA

    Muzhi Yang

  4. Department of Psychological and Brain Sciences and UCSB Brain Imaging Center, University of California, Santa Barbara, Santa Barbara, California, USA

    Nicholas F Wymbs & Scott T Grafton

  5. Department of Physical Medicine and Rehabilitation, Human Brain Physiology and Stimulation Laboratory, Johns Hopkins Medical Institution, Baltimore, Maryland, USA

    Nicholas F Wymbs

Authors
  1. Danielle S Bassett
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Muzhi Yang
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Nicholas F Wymbs
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Scott T Grafton
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

D.S.B. formulated the project; N.F.W. and S.T.G. performed the experiments; D.S.B., N.F.W. and M.Y. did the computations; and D.S.B., M.Y., N.F.W. and S.T.G. wrote the manuscript.

Corresponding author

Correspondence to Danielle S Bassett.

Ethics declarations

Competing interests

The authors declare no competing financial interests.

Integrated supplementary information

Supplementary Figure 1 Trial structure and stimulus-response (S-R) mapping.

(a) Each trial began with the presentation of a sequence-identity cue that remained on screen for 2 s. Each of the six trained sequences was paired with a unique identity cue. A discrete sequence-production (DSP) event structure was used to guide sequence production. The onset of the initial DSP stimulus (thick square, colored red in the task) served as the imperative to produce the sequence. A correct key press led to the immediate presentation of the next DSP stimulus (and so on) until the ten-element sequence was correctly executed. Participants received a ‘+’ as feedback to signal that a sequence was completed and to wait (approximately 0–6 s) for the start of the next trial. This waiting period was called the ‘intertrial interval’ (ITI). At any point, if an incorrect key was hit, a participant would receive an error signal (not shown in the figure), and the DSP sequence would pause until the correct response was received. (b) There was direct S-R mapping between a conventional keyboard or an MRI-compatible button box (lower left) and a participant’s right hand, so that the leftmost DSP stimulus cued the thumb and the rightmost stimulus cued the pinky finger. Note that the button location for the thumb was positioned to the lower left for maximum comfort and ease of motion.

Supplementary Figure 2 Experiment timeline.

Training sessions in the MRI scanner during the collection of blood oxygen level–dependent (BOLD) signals were interleaved with training sessions at home. Participants first practiced the sequences in the MRI scanner during a baseline training session (top). After every approximately ten training sessions, participants returned for another scanning session. During each scanning session, a participant practiced each sequence for 50 trials. Participants trained at home between the scanning sessions (bottom). During each home training session, participants practiced the sequences in a random order. (We determined a random order using the Mersenne Twister algorithm of Nishimura and Matsumoto as implemented in the random-number generator rand.m of MATLAB version 7.1.) Each EXT sequence was practiced for 64 trials, each MOD sequence was practiced for 10 trials, and each MIN sequence was practiced for 1 trial.

Supplementary Figure 3 Recruitment and integration are modulated by training, as shown by thresholded module-allegiance matrices.

In the article, we report the modulation of recruitment and integration by training, where ‘training’ is specifically measured by the number of trials practiced. In our experiment, we were able to obtain 12 different levels of training by using a combination of four different scanning sessions (each separated by approximately 2 weeks) and three different sequence types (extensive, EXT; moderate, MOD; and minimal, MIN). In the article, we show motor-visual integration and non-motor, non-visual recruitment as a function of the 12 training levels. For completeness, here we replot the same data with consideration of both scanning session and sequence type. Results shown in this figure were obtained using thresholded module-allegiance matrices, as opposed to those shown in Supplementary Figure 4, which were obtained using unthresholded module-allegiance matrices; the two figures show comparable results. (a) Motor and (b) visual recruitment was unaffected by training intensity (EXT (maroon), MOD (red), and MIN (orange) sequences) and duration (naive, early, middle, and late). (c) Integration between motor and visual modules and (d) recruitment of non-motor and non-visual cortices decreased with training intensity and duration. Error bars indicate s.d. of the mean across participants.

Supplementary Figure 4 Recruitment and integration are modulated by training, as shown by unthresholded module-allegiance matrices.

In the article, we use thresholded module-allegiance matrices to compute the recruitment and integration in the motor and visual modules and the non-motor, non-visual set, and to determine how these diagnostics are modulated by training. Here we demonstrate that these results are robust to the choice to statistically threshold the module-allegiance matrices. Results shown in this figure were obtained using unthresholded module-allegiance matrices and are comparable to those illustrated in Figure 4 and Supplementary Figure 3 (obtained using statistically thresholded module-allegiance matrices). (a) Motor and (b) visual recruitment was unaffected by training intensity (EXT (maroon), MOD (red), and MIN (orange) sequences) and duration (naive, early, middle, and late). (c) Integration between motor and visual modules and (d) recruitment of non-motor and non-visual cortices decreased with training intensity and duration. (e,f) The observations in c and d (i.e., that recruitment and integration depended on training intensity and duration) can be parsimoniously described by a single latent variable: the number of trials practiced (i.e., ‘depth’). Solid diagonal lines indicate the best linear fit, and r values indicate Pearson correlation coefficients. The Pearson correlation coefficient for the data in e is r = –0.92, P = 0.000022, and for the data in f it is r = –0.94, P = 0.0000055. Error bars indicate s.d. of the mean across participants.

Supplementary Figure 5 Individual differences in brain network architecture map to task performance and task learning.

In the article, we use thresholded module-allegiance matrices to compute the recruitment and integration in the motor and visual modules and the non-motor, non-visual set, and to determine how these diagnostics are modulated by training. Here we show that these results are robust to the choice to statistically threshold the module-allegiance matrices. Results shown in this figure were obtained using unthresholded module-allegiance matrices and are comparable to those illustrated in Figure 5 (obtained using statistically thresholded module-allegiance matrices). (a) Box plots of task-dependent modulation (correlation between network diagnostic and number of trials practiced) for both motor-visual integration (left) and non-motor, non-visual recruitment (right). (b) Scatter plot of learning rate and training-dependent modulation of non-motor, non-visual recruitment (Pearson's r = 0.51, P = 0.021).

Supplementary Figure 6 Implications of module dissociation for time courses of BOLD activation.

In the article, we claim that (i) the dissociation of motor and visual modules indicates that brain areas in these two systems display significantly different time courses of BOLD activation, (ii) the non-motor, non-visual regions display significantly different time courses of BOLD activation than regions in the motor and visual systems, and (iii) the time courses of BOLD activation for regions of the non-motor, non-visual set differ significantly from one another. These claims are mathematically consistent with the formulation of the module-allegiance matrix and the utilization of the underlying community-detection approaches. In addition to these theoretical considerations, here we provide direct quantification of exactly how different the time courses of BOLD activation were in motor, visual, and non-motor, non-visual regions. In this figure, we show the average coherence between pairs of brain regions in which (i) both regions of the pair are within the motor module (green), (ii) both regions of the pair are within the visual module (yellow), (iii) both regions of the pair are within the non-motor, non-visual set (purple), (iv) one region of the pair lies in the motor module and the other region of the pair lies in the visual module (gray), (v) one region of the pair lies in the motor module and the other region of the pair lies in the non-motor, non-visual set (orange), and (vi) one region of the pair lies in the visual module and the other region of the pair lies in the non-motor, non-visual set (blue). These average values were computed per subject. P values for two-sided two-sample t-tests across subjects between these scenarios are given by the gray bars; note that all P values passed Bonferroni correction for multiple comparisons. Error bars indicate s.d. of the mean across participants.

Supplementary Figure 7 Quantifying the change in sensorimotor autonomy induced by learning.

In the article, we describe data that support our conclusion that learning induces sensorimotor autonomy. In light of this conclusion, it is interesting to note that the motor and visual systems were quite pronounced as network communities even during early learning. Yet they were far from autonomous in the sense that regions of the two systems often tended to be present in the same network community or putative functional module. As learning progressed, the two systems became increasingly autonomous in the sense that the probability that regions of the two systems would be found in the same modules decreased significantly. To illustrate these points, we examined the motor-visual integration in the naive state (extensively practiced sequences in scan 1) compared to the motor-visual integration in the late-learning state (extensively practiced sequences in scan 4). The integration in the naive state (scan 1) was 0.58 (s.d. over subjects = 0.29), meaning that the probability that motor and visual modules would be in the same community in any given time window was greater than 50% (left; error bars indicate s.d.). In contrast, the integration in the late-learning state (scan 4) was 0.31 (s.d. over subjects = 0.30), meaning that the probability that motor and visual modules would be in the same community in any given time window was roughly 30% (right; error bars indicate s.d.). We observed that the two distributions of integration (in naive and late-learning states) were significantly different (two-sided two-sample t-test, t = 2.92, P = 0.0058), supporting our conclusion that learning induces relative autonomy of sensorimotor systems.

Supplementary Figure 8 Training-dependent modulation of intramodule integration for motor and visual systems.

Here we ask whether regions in the motor or visual systems integrate further or disengage from their respective modules with increasing task practice. Following the approach we used for the modules, we quantified, for each subject, the intramodule integration of a region with its own module as the sum of all functional connections between that region and other regions in its module. For example, the intramodule integration of the left SMA with the rest of the motor module is given by the sum over all j (not equal to i) in the motor module of Pij for i = left SMA and j indexing the other 11 regions in the motor module. To determine whether this intramodule integration of a brain region increased or decreased with task practice, we calculated the training-dependent modulation of that region as –1 times the Pearson correlation coefficient between the group-averaged intramodule integration and the number of trials practiced. This figure shows the results for the motor module; 7 of the 12 regions showed significant disengagement from the motor module with training, as evidenced by associated P values that passed Bonferroni correction for 12 multiple comparisons: left SMA (r = 0.79, P = 0.0023), left precentral (r = 0.79, P = 0.0024), left postcentral (r = 0.80, P = 0.0019), left superior parietal (r = 0.77, P = 0.0036), right precentral (r = 0.79, P = 0.0023), right SMA (r = 0.80, P = 0.0020), and left parietal operculum (r = 0.78, P = 0.0030). No regions showed significantly enhanced integration with training. In the visual module, no regions showed significantly enhanced integration or disengagement with training.

Supplementary Figure 9 Relationship between functional connectivity and activation.

The relationships between functional connectivity and functional activation during training are of critical interest. Here we ask whether the visual, motor, and non-motor, non-visual sets show different patterns of BOLD activation over the course of training. We performed a standard GLM analysis (see “Online Methods”) to quantify the degree to which brain regions showed a linear decrease in activity over the course of training, and then we asked whether average beta weights of this GLM were different between functional modules. In this figure, we observe that the motor module showed predominantly positive beta weights (one-sided one-sample t-test showed that the mean was significantly greater than 0: t = 3.01, P = 0.0071), indicating a decrease in activation during learning. The visual module showed predominantly negative beta weights (t = 5.14, P = 5.8 × 10−5), indicating an increase in activation during learning. The non-motor, non-visual set of regions on average showed beta weights close to zero (t = 1.27, P = 0.22). Box plots indicate variation of the mean across participants. See “Online Methods” for a description of the GLM analysis used to estimate beta weights. It is interesting to compare these results to those obtained from functional connectivity measurements. Motor module: from the GLM analysis, we learned that the motor module showed a decrease in BOLD activation over the course of training; from the functional connectivity analysis, we learned that the motor module showed no change in network recruitment over the course of training (Supplementary Fig. 3a) but did show a decrease in integration with the visual module (Supplementary Fig. 3c). Visual module: from the GLM analysis, we learned that the visual module showed an increase in BOLD activation over the course of training; from the functional connectivity analysis, we learned that the visual module showed no change in network recruitment over the course of training (Supplementary Fig. 3b) but did show a decrease in integration with the motor module (Supplementary Fig. 3c). Non-motor, non-visual set: from the GLM analysis, we learned that the non-motor, non-visual set showed no change in BOLD activation over the course of training; from the functional connectivity analysis, we learned that the non-motor, non-visual set showed an exponential decrease in network recruitment over the course of training (Supplementary Fig. 3d). This pattern of results clearly shows that functional connectivity–based analyses provide complementary (and not redundant) information relative to activation-based analyses. In particular, the two features of learning-related changes in connectivity reported in the article—decreases in motor-visual integration and decreases in non-motor, non-visual recruitment—are not observable in activation-based analyses. Finally, the prediction of individual differences in learning on the basis of decreases in non-motor, non-visual recruitment could not be done using activation-based analyses in which non-motor, non-visual activity did not change as a function of training. These results highlight the utility of network-based approaches in uncovering previously hidden features of cognitive processes and individual variation in the performance of those processes.

Supplementary Figure 10 Summary architecture of learning in finer-grained parcellation.

It was important to examine the robustness of our results to changes in the granularity of the whole-brain atlas used. In the next few supplementary figures, we report a reanalysis of our data with a new parcellation scheme composed of smaller regions of interest. We combined two separate atlases: (i) the AAL-derived 600-region atlas that we developed for use in refs. 68 and 69, which subdivides the 90 AAL anatomical regions into regions of roughly similar size via a spatial bisection method, and (ii) a high-resolution probabilistic 26-region atlas of the cerebellum in the anatomical space defined by the MNI152 template, obtained from T1-weighted MRI scans (1-mm isotropic resolution) of 20 healthy young participants70,71 (this latter atlas is provided by SPM8). The combination of these two atlases provided a high-resolution, 626-region atlas of cortical, subcortical, and cerebellar regions. We called this new atlas the AAL-626 atlas. Following the same procedures outlined for the Harvard-Oxford atlas, we constructed a module-allegiance matrix and observed the presence of two strong modules, which corresponded to the motor and visual systems. (a) The module-allegiance matrix that indicates, for a pair of nodes, the probability that those two nodes will be located in the same functional community across subjects, scanning sessions, sequence types, and trial blocks. Similar to that reported in the article for the coarser-grained parcellation, this module-allegiance matrix displays two putative functional modules composed of brain regions that were consistently grouped into the same network community: (b) one composed of primary and secondary sensorimotor areas, and (c) one composed of primary visual cortex. Brain regions in a have been ordered to maximize strong associations along the diagonal.

Supplementary Figure 11 Recruitment and integration modulated by training in finer-grained parcellation.

Consistent with results reported in the article, we observed that (i) motor-visual integration and (ii) non-motor, non-visual recruitment were modulated by training in the AAL-626 atlas. (a) Motor-visual integration and (b) non-motor, non-visual recruitment depended on the number of trials practiced. Solid diagonal lines indicate the best linear fit, and r values indicate Pearson correlation coefficients. Error bars indicate s.d. of the mean across participants. To quantify these observations, we again defined the training-dependent modulation of brain network diagnostics as –r, where r is the Pearson correlation coefficient between the logarithm of the number of trials practiced and the diagnostic value. Motor-visual integration showed significant training-dependent modulation (r = –0.90, P = 5.54 × 10−5), as did non-motor, non-visual recruitment (r = –0.84, P = 4.05 × 10−4).

Supplementary Figure 12 Individual differences in brain network architecture map to task performance and task learning in finer-grained parcellation.

In the AAL-626 atlas, we observed a similar relationship between brain network architecture and individual differences in learning (compare to Fig. 5). (a) Box plots of task-dependent modulation (correlation between network diagnostic and number of trials practiced) for both motor-visual integration (left) and non-motor, non-visual recruitment (right). (b) Scatter plot of learning rate and training-dependent modulation of non-motor, non-visual recruitment (r = 0.57, P = 0.0062).

Supplementary Figure 13 The release of a fronto-cingulate control network in finer-grained parcellation predicts individual differences in learning.

In the AAL-626 atlas, we observed that the release of a fronto-cingulate control network predicted individual differences in learning, corroborating our results obtained with the Harvard-Oxford atlas. The 8,286 elements in the driver network are given by the statistically significant (P < 0.05, uncorrected) Pearson correlation coefficients between individual differences in training-induced modulation and individual differences in learning rate. The distribution of strength values over brain areas was highly skewed (s = 1.30)—significantly more so than would be expected in a random-network null model (non-parametric test, P < 0.00000). Here we show the strength of brain areas in this driver network, mapped onto the cortical surface using Caret software (http://brainvis.wustl.edu). The strength of area i is given by the sum of column i in the driver network. Warm colors indicate high strength in the driver network, and cool colors indicate low strength in the driver network. In addition to those regions observed in the 112-region template, here we also notice the high strength of the temporo-parietal junction, and the entorhinal cortex.

Supplementary Figure 14 Module-allegiance versus functional connectivity matrices.

When introducing any new method, it is important to ask whether similar results could have been uncovered using a simpler approach. Here we show that module-allegiance matrices provide a level of sensitivity to learning-related changes in brain network architecture that is not observed in functional connectivity matrices. We subdivide this exposition into discussions of (i) summary and dynamic architecture of learning (in this figure caption) and (ii) recruitment and integration modulated by training (in the Supplementary Math Note). In Figure 2 we show the module-allegiance matrix P, whose elements Pij give the probability that area i and j are in the same community across all subjects, scanning sessions, sequence types, and trial blocks. At the top of this figure, we show the average functional connectivity matrix W for comparison. The elements Wij give the wavelet coherence values averaged over all subjects, scanning sessions, sequence types, and trial blocks. Motor and visual areas are less delineated in the average functional connectivity matrix than in the module-allegiance matrix. In Figure 3 we show the module-allegiance matrices for the naive, early, middle, and late learning sessions separately. For the naive module-allegiance matrix Pnaive, the elements Pij give the probability that area i and j are in the same community over all subjects, sequence types, and trial blocks in the naive scanning session. Module-allegiance matrices for the other three scanning sessions were constructed similarly. At the bottom of this figure, we show the average functional connectivity matrices for comparison. For the naive functional connectivity matrix Wnaive, the elements Wij give the wavelet coherence averaged over all subjects, sequence types, and trial blocks in the naive scanning session. Average functional connectivity matrices for the other three scanning sessions were constructed similarly. We observed that (i) motor and visual areas were less delineated, (ii) changes in motor-visual integration were less clear, and (iii) the coherent involvement of the non-motor, non-visual areas was less pronounced in the average functional connectivity matrices than in the module-allegiance matrices. We normalized the functional connectivity matrices in this figure by dividing by the mean for comparability of visualizations. In the Supplementary Math Note, we show that alternative estimates of recruitment and integration based on the functional connectivity matrix rather than the module-allegiance matrix did not provide sensitivity to learning-related changes in brain network architecture.

Supplementary Figure 15 Alternative quantifications of behavior.

Movement time is the canonical behavioral measure of motor-sequence learning. However, error rates and reaction times can in some experimental designs provide additional insights. Here we show that error rates and reaction-time rates were not correlated with (i) learning rate, (ii) training-dependent modulation of motor-visual integration, or (iii) training-dependent modulation of non-motor, non-visual recruitment. These results suggest that our findings relating learning rates to training-dependent changes in network reconfiguration are specific and do not relate more broadly to other behavioral characteristics such as error rates and reaction-time rates.

In this particular experiment, subjects were instructed to remain as accurate as possible, and therefore error rates are quite low in this data set. Nevertheless, errors did occur, and their number varied from subject to subject. To address the effects of errors, we performed an additional analysis to (i) estimate errors as a function of training, (ii) determine the relationship between error rate and movement time, and (iii) quantify the relationship between the training-dependent modulation of error rate and training-dependent modulation of motor-visual integration and non-motor, non-visual recruitment. To estimate errors as a function of training, we ordered all trials (including all six sequences) according to the amount of training. For each subject, we divided the number of trials into 45 equally sized bins (similar results were obtained for 15–120 bins) and calculated the error rate (e.g., percent correct) in each bin. Because subjects had an almost identical number of trials, the number of trials per bin was comparable across subjects. We fit the data of error rate versus trial bin using a first-degree polynomial with a single slope parameter ς (used to measure the training-dependent modulation of error rate) and constant. We next determined the relationship between the training-dependent modulation of error rate ς and the learning rate κ. We observed no significant relationship between these two variables: the Pearson correlation coefficient between ς and κ was r = 0.34, P = 0.13 (a; each data point indicates a single subject). This result indicates that individuals who are good learners do not show an increase or decrease in error rate over training in comparison to poor learners. This finding shows that our subjects were indeed learning the sequences: they moved much more quickly with training while remaining similarly accurate. Our data therefore do not support a simple speed-accuracy tradeoff. Finally, we determined the relationship between the training-dependent modulation of error rate ς and the training-dependent modulation of motor-visual integration and non-motor, non-visual recruitment. We observed no significant relationship between ς and training-dependent modulation of motor-visual integration (r = 0.04, P = 0.87) or non-motor, non-visual recruitment (r = 0.15, P = 0.52). This indicates that individuals who show greater training-dependent modulation of motor-visual integration and non-motor, non-visual recruitment do not tend to show an increase or decrease in error rates over training in comparison to individuals with less training-dependent modulation of motor-visual integration and non-motor, non-visual recruitment. Again, this finding is evidence for learning.

Reaction-time estimates have in some experimental scenarios provided additional behavioral insights into the mechanisms of learning. In our study, subjects were first presented with a symbolic cue indicating which of the six sequences to expect. After a variable delay, one of five spatial cues appeared, indicating the first element of the sequence. RT is the time between the appearance of the spatial cue and the first motor response. In the analysis of error rate, we binned error data of trial blocks to obtain continuous measurements from what was inherently binary data (e.g., error or no error). In contrast, reaction time is itself a continuous measure, and therefore we can treat it the same as movement time by performing a double exponential fit with robust outlier correction to home training session behavioral data. We next determined the relationship between the reaction-time rate κRT and the learning rate κ. We observed no significant relationship between the two variables: r = –0.08 and P = 0.73 (b; each data point indicates a single subject). Because κRT values appeared to be heavily skewed, we confirmed this lack of relationship using a Spearman rank correlation (ρ = 0.09, P = 0.70). This result indicates that individuals who are good learners do not show a different reaction-time rate than poor learners. Finally, we determined the relationship between the reaction-time rate κRT and the training-dependent modulation of motor-visual integration and non-motor, non-visual recruitment. We observed no significant relationship between κRT and training-dependent modulation of motor-visual integration (r = –0.17, P = 0.46) or non-motor, non-visual recruitment (r = –0.35, P = 0.13). This result indicates that individuals who show greater training-dependent modulation of motor-visual integration and non-motor, non-visual recruitment do not tend to show unusually high or low reaction-time rates.

These results suggest that the findings reported in the article relating learning rates to training-dependent changes in network reconfiguration are specific and do not relate more broadly to other behavioral characteristics such as error rates and reaction-time rates.

Supplementary information

Supplementary Text and Figures

Supplementary Figures 1–15 and Supplementary Note (PDF 7606 kb)

Supplementary Methods Checklist (PDF 487 kb)

Source data

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Bassett, D., Yang, M., Wymbs, N. et al. Learning-induced autonomy of sensorimotor systems. Nat Neurosci 18, 744–751 (2015). https://doi.org/10.1038/nn.3993

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/nn.3993

This article is cited by

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing