Muscle synergies during locomotion define a flexible substrate for neural control Nedialko Krouchev, John Kalaska, Trevor Drew GRSNC, Faculté de Médecine, Université de Montréal, Canada Despite the now classic idea that the central nervous system (CNS) may use muscle synergies to simplify the control of movement1, a widely accepted definition of “synergy” has remained 3-5 elusive. In one wide-ranging series of experiments to study this issue, Bizzi and colleagues have used mathematical techniques for linear decomposition to produce temporal waveforms and corresponding muscle loadings. These studies defined synergies and their representation by basis sets of primitives (temporal waveforms) according to different sets of constraints. Either each muscle was reconstructed using its own unique set of primitives (time-varying synergies), or the same waveforms were reused with a full weight matrix, fitting all EMG’s simultaneously (synchronous synergies.). In both cases, the constraint that all muscles are activated by each 3-9 synergy means that the predicted descending commands are likely to be very complex, and in the case of locomotion, they might be expected to vary continuously throughout the step cycle. However, complex signals varying continuously throughout the step cycle, were rarely seen in 2 the discharge patterns of individual neurones . During locomotion instead, the activity of most motor cortical neurones is characterised by phasic bursts activated sequentially - some at the onset of swing, others discharging progressively later in the same phase. We suggest that the base synergies during locomotion are likewise discrete and active during confined sub-intervals of the step cycle and that descending signals act to modulate these synergies to produce different gait patterns. In the present study we perform an initial examination of this premise by making a detailed study of the temporal patterns of cat fore- and 10 hindlimb muscle activity during treadmill locomotion by using a novel cluster analysis . This analysis differs from previous approaches using blind decomposition methods in its more restricted and classical definition of a synergy as a group of muscles that are temporally coactivated and whose period of activity begins and ends synchronously. Analysis of the patterns of EMG onset and offset in both single trials or in averages showed that most muscles could be grouped into a number of clusters (Fig.1) that are activated during different phases of the step cycle. Nine of the defined synergies were activated sequentially during the swing phase of locomotion, in a similar manner to that reported for neurones in the spinal cord, motor cortex and red nucleus. Synchronous activation and suppression of multiple muscles allows descending controls that can be described by a single state variable. Therefore, for a subset of the clusters identified in the (onset, offset) phase-space, we used the mean onset and offset for all data points in a cluster to construct basis primitives: we call these Direct Components (Fig.2). We suggest that these direct components (black lines in Fig. 2) reflect the descending signal that is required to modulate activity in each group of synergists. To determine how the synergies and putative control signals change in different conditions we repeated our cluster analysis on the EMG patterns obtained when cats stepped over an obstacle attached to a moving belt, both when the limb was the first to step over the obstacle (lead limb) and when it was the second (trail limb). The results (green and red lines in Fig, 2) show that most synergies maintain their structure and relative timing within the step cycle. However, in some cases (e.g. DC1 and DC6) there are changes either in duration and/or the phase of activation. These changes are similar to those observed in some populations of motor cortical neurones during the same behaviours, supporting our suggestion that descending controls act to modify the phase and level (not illustrated) of synergistic groups of muscles. Our approach simplifies neural control as all temporal controls could stem from a cascade of simple modules. Their activation is interdependent, follows task state and is affected by motor adaptation and learning.

1.2

A

C 1.0

Figure 1. 1

Phase of EMG offset

1

0.8

0.6

2

Stance

2 3 3

0.4

4

Prepare for Landing Dorsiflex Paw

0.2

5 6 6 7

Flex Elbow

0.0

ECR(2) Bic(2) TrM(2) PrT(2) EDC(2) LtD(2) SpD(2) ECU TriL

8 9

Lift Leg

9

-0.2 0.6

0.8

9

1.0

9 9

B

Lift Leg

{

Transport Limb Flex Elbow

Dorsiflex Paw

10

Stance

11 11

Tri SSp PaL AcD(2)

11 11 11 11

Figure 2. DCA - a truly sparse representation of multiple behaviours A. For a subset of the synergies in Figure 1 (data coming from a single cat, see inset B for details), we performed the cluster analysis across three different behaviours: baseline locomotion (control walk) and stepping over obstacles with the recorded forelimb leading (lead, L), or trailing (trail,T). We used the mean onset and offset for all data points of the resulting clusters to construct basis primitives. We call these Direct Components (DC). Their waveforms may be a plausible match to descending signals driving the corresponding muscles in the synergy. Change of extracted components is expressed by showing the closest-matching pairs of control versus lead or trail DC’s.

A. Results of cluster analysis Each point, displayed using a unique combination of colour and symbol (e.g. *), presents the processed activity of a given muscle during one trial. The phase plane is defined by the onset and offset of EMG bursts, relative to the start and duration of the step cycle. B. Muscle synergies active at various consecutive stages of the step cycle, as derived from A. Identical colours represent the same synergies in A and B.

SpD(1) AcD(1) BrR PrT(1) Br Bic(1) ClB ClT LvS

2

Transport Limb

Explicit definition of synergies

EDC(1) ECR(1) LtD(1) TrM(1)

1

Comparison of DCA for 3 behaviours

A 1

L1 T2

DC1 0.5

B

Control Walk (DC) Lead (L)

0 1

Trail (T) L2 T3

DC2 0.5 0 1

1.TrM(1)

L3 T3

DC3 0.5 0 1

L4 T4

DC4 0.5

2.PrT(1)

1.TrM(1)

3.BrR

2.PrT(1)

3.Br

3.BrR

4.Bic(1)

3.Br

5.LvS

4.Bic(1)

6.Bic(2)

4.LvS

6.TrM(2)

5.Bic(2) 5.TrM(2)

6.PrT(2)

5.PrT(2)

6.EDC(2)

0 1

5.EDC(2)

7.SpD(2)(2)

L5

DC5 0.5

T4

L6

T5

0 1

6.PaL

8.Tri

0 1 DC6 0.5

6.SpD(2)(2)

8.PaL

6.Tri

8.TriL

6.TriL

8.SSp

6.SSp

8.ECU

6.ECU

L7

DC7 0.5

T6

0 1

L8 T6

DC8 0.5

0 -0.4

References 1. Bernstein N. The Coordination and Regulation of Movements ( New York:Pergamon Press, 1967). 2. Lavoie S & Drew T. Journal of Neurophysiology 88, 1791814 (2002). 3. d'Avella, A., Saltiel, P. & Bizzi, E. Nature Neuroscience 6, 300-308 (2003). 4. Cheung, V.C.K., d'Avella, A., Tresch, M.C. & Bizzi, E. Journal of Neuroscience 25, 6419-6434 (2005). 5. d'Avella A, Portone A, Fernandez L, Lacquaniti F. Journal of Neuroscience 26, 7791-810 (2006).

-0.2

0

0.2

0.4 step-cycle phase

0.6

0.8

1

6. Hart, C.B. & Giszter, S.F. Journal of Neuroscience 24, 5269-5282 (2004). 7. Ivanenko, Y.P., Poppele, R.E. & Lacquaniti, E. Journal of Physiology-London 556, 267-282 (2004). 8. Lee, D.D. & Seung, H.S. Nature 401, 788-791 (1999). 9. Bell, A.J. & Sejnowski, T.J. Neural Computation 7, 11291159 (1995). 10. Krouchev NI, Kalaska JF & Drew T. Journal of Neurophysiology, Epub July 5 (2006).