Body size and the neural control of movement

Abstract

Nervous systems, muscles, and motor organs (biomechanics) generate movement cooperatively and should therefore change coordinately as body size changes. Proving this hypothesis requires a description of the nervous system, muscle, and biomechanics of individual species and comparison of these properties across a collection of species spanning a large size range. Obtaining these data in even one species is daunting, and we are far from reaching a general theory of how nervous system, muscle, and biomechanics depend on body size. Sufficient cross-species data nonetheless exist to identify some consequences of body size on the neural control of movement, which form the focus of this Primer: for example, locomotor cycle period depends on leg length; large animals should devote more neural resources to hazard avoidance and stride correction; the importance of gravity and momentum increases, and of passive muscle properties decreases, with limb size; and the medium in which movements occur affects these size dependencies. These considerations suggest that, in applying biological data to robotic design, proper ‘size’ should be maintained across all levels (i.e., if small-animal neural mechanisms are used, so should small-animal muscle and biomechanical properties). Nervous systems, muscles, and motor organs (biomechanics) generate movement cooperatively and should therefore change coordinately as body size changes. Proving this hypothesis requires a description of the nervous system, muscle, and biomechanics of individual species and comparison of these properties across a collection of species spanning a large size range. Obtaining these data in even one species is daunting, and we are far from reaching a general theory of how nervous system, muscle, and biomechanics depend on body size. Sufficient cross-species data nonetheless exist to identify some consequences of body size on the neural control of movement, which form the focus of this Primer: for example, locomotor cycle period depends on leg length; large animals should devote more neural resources to hazard avoidance and stride correction; the importance of gravity and momentum increases, and of passive muscle properties decreases, with limb size; and the medium in which movements occur affects these size dependencies. These considerations suggest that, in applying biological data to robotic design, proper ‘size’ should be maintained across all levels (i.e., if small-animal neural mechanisms are used, so should small-animal muscle and biomechanical properties). An illustrative example of the interaction between neural control and biomechanics is terrestrial locomotion in birds and mammals. The legs of all species examined here are located under the body (as opposed to the sprawling posture of, for instance, alligators). The legs could thus move, at least in part, as pendulums during locomotion, and passive two-legged mechanical models can walk down inclined planes. Neurally induced muscle contractions also drive animal locomotion, and the importance of biomechanics vs. neural driving in locomotion is not fully resolved. Within the angles used in locomotion, pendulum period in air is approximately 2πL/g, where L is limb length and g gravitational acceleration. Leg cycle period depends on length less than pendulum period (exponents of ∼0.35 vs. 0.5; Figure 1A ). However, fits to the data with the exponent fixed at 0.5 only moderately decrease the goodness of the fits (decreasing the R values to 0.87 for trot and 0.93 for gallop), and whether this difference is relevant is thus unclear. Leg cycle periods are unambiguously much (∼75%) shorter than those of matched-length pendulums. This shortening arises in part from leg mass not being concentrated at the toe (the periods of ‘pendulums’ whose mass is distributed along the lever arm are shorter than those of ideal pendulums, in which all the mass is concentrated at the pendulum end). However, this effect should only decrease limb pendulum period by about 25%. These data suggest that neural driving decreases the cycle period that would arise from pendular mechanics alone by ∼50% across the entire size range shown (from mice to horses). Moreover, passive damping of muscle should prevent small legs from acting as pendulums at all (see below). Pendular mechanics should thus play a role only in long legs, with the dependence of cycle period on length in small legs arising from other sources. Regardless, a prediction from these data is that the inherent period of locomotor neural network should also scale with body size, but insufficient data from isolated nervous systems are presently available to test this prediction. Despite locomotion period being longer in large animals, their ground speed nonetheless increases because of their longer legs. Their kinetic energy (m·v2/2, where m is mass and v velocity) thus also increases. Surface area scales as mass2/3. Assuming as a worst case that the animal runs into an immovable, incompressible object (a brick wall) and does not bounce (thus absorbing all the kinetic energy), the energy per unit surface area that galloping animals (in the case considered here) absorb in the collision can therefore be calculated. The units of this calculation (J/m2) have no intuitive meaning. I therefore plot the data in terms of the energy per unit surface area that a prone human falling various distances in vacuum would absorb (Figure 1B). Small animals absorb inconsequential energies, but sheep absorb the equivalent of a human falling 2.4 meters, and horses the equivalent of a human falling 24 meters. Not running into brick walls thus matters more for horses than mice. Quantifying how the damage from stepping into holes or stumbling depends on size is more difficult because of the importance in these cases of mishap specifics. Nonetheless, the above considerations suggest that such mishaps would be much more dangerous for large animals. Large animals would therefore be predicted to engage in high-speed locomotion less often and, when doing so, to plan their routes in greater detail and further into the future. These animals would also be predicted to devote more neural machinery to recovery from locomotor mishaps. A concern with this hypothesis is that large animals would have insufficient time to perform these calculations because of increased action potential transit time. Full recovery from a locomotor mishap often takes several strides. However, to avoid a fall an animal must in general make initial corrections (either so that the disturbed leg can nonetheless provide sufficient support, or by altering other leg movements to compensate for the lack of the disturbed one) before the disturbed leg's next stance phase. This time is at most one cycle period. Using this duration as a guide allows comparing the times that small and large animals have to make such corrections. Myelinated nerve action potential velocity is essentially independent of animal size and ranges from 10 to 55 m/s. Using a value of 30 m/s, toe to spinal cord transit time would be 1 ms in mice vs. 50 ms in horses, perhaps twice that to the brain; round trip 2 ms vs. 100 ms (spinal cord) and 4 ms vs. 200 ms (brain). Turning motor nerve action potentials into muscle contractions takes another 40 to 50 ms. Gallop period for mice is 130 ms and for horses is 600 ms. Mice thus have about 78 ms and horses 450 ms for spinal-based calculations, and 76 and 350 ms for brain-based calculations, to alter leg movements in single locomotor cycles. Assuming, for the sake of argument, 10 ms from neuron input to output and linearly arranged neural circuitry (changing any of these assumptions alters only the absolute numbers, not the inter-species comparison being made here), horses have enough time to use a network with 35 to 45 neurons, and mice a network with 8 neurons, for intra-stride movement corrections. Long-legged animals can thus devote greater neural resources to intra-stride alterations. Larger animals also have straighter legs, which helps to maintain acceptable bone stress by more closely aligning the limbs with the ground reaction force. This closer alignment might also simplify joint control calculations, providing another selective pressure for leg straightness in large animals. Large animals nonetheless sometimes fall. Given the much graver consequences of falls in large animals, this argument thus does not suggest that animals became large to allow for increased stride recovery times. The shorter times for corrective neural computations in small animals may also largely explain why in very small animals, such as insects, stride ‘correction’ during high-speed locomotion is largely mediated by purely passive, biomechanical mechanisms. Locomotor, heartbeat and respiration periods all increase with mass. These dependencies likely arise from scaling issues (limb length as mass1/3; surface area as mass2/3). In cases without these issues, however, there is no a priori reason that small things should move more rapidly. This is well demonstrated by the lobster pyloric neuromuscular system, whose motor neural network maintains a cycle period of ∼1 s as body size changes 20-fold. Unstimulated muscles generate restoring force when stretched beyond rest length. This force increases with muscle cross-section and thus, with perfect scaling, varies as (limb mass)2/3. The effect of these properties on neural control can be demonstrated by setting one's elbow on a table with the forearm held vertically, palm facing the shoulder, with the fingers relaxed. This posture requires biceps contraction, and relaxing the biceps results in the forearm falling to the table. The ‘C’-shaped posture of the fingers will not have changed, even though gravity worked to flex them when the forearm was vertical and worked to extend them when the forearm was horizontal. The hypothesis is that the force of gravity on the fingers is always much smaller than the passive forces of the finger muscles. Regardless of the direction of gravity, fingers therefore always assume the equilibrium posture at which the antagonistic muscles exert equal but opposite force. This hypothesis has been verified in the stick insect by inserting electrodes to record leg muscle activity, lifting the animals off the substrate and holding them head up, and rotating the animals 180° (head down) and then 360° (back to head up). Throughout this procedure ‘shoulder’ joint (arrows) angle remained at ∼90° to the body long axis without any leg motor nerve activity (Figure 2A ). If subjected to this treatment, relaxed human shoulders would rotate from fully flexed (0°, arms next to side) when the head is up, to fully extended (180°, arms hanging toward the floor) when the head is down. Amputated stick insect ‘knees’ also assume gravity-independent rest angles and return to angles near this angle if moved from it (Figure 2B). These angles are in the range at which the joint's antagonistic muscles exert approximately equal passive forces (Figure 2C). Because of the much greater increase in limb mass with size than muscle cross-section, muscle passive force is inconsequential in large limbs compared with the force of gravity (hence the elbow extension above with biceps relaxation). This difference has two consequences for motor control. First, for large limbs, nervous systems must monitor gravity direction to calculate what muscle contractions will produce a given movement. For instance, when a human arm hangs alongside the body, no muscle contraction results in a straight arm and biceps contraction flexes the elbow. Alternatively, when the arm is held above the head, a straight arm requires triceps contraction and triceps relaxation flexes the elbow. No gravity monitoring is necessary for small limbs — relaxed fingers assume the rest ‘C’ posture regardless of gravity's direction, and finger flexion/extension always requires flexor/extensor contraction. The second consequence is that nervous systems can use momentum-based strategies for large, but not small limbs (Figure 2D). In horse and human locomotion the swing muscles accelerate the leg at swing onset with a brief action potential burst and then go silent, relying on momentum to maintain leg movement. Stick insects cannot use this strategy because, when the leg is not moving toward its equilibrium angle (e.g., late in swing), if the agonist stops generating force, antagonistic muscle passive force rapidly (in 0.09 ms) stops the leg and then moves it back to the equilibrium angle. Mouse and cat swing muscles similarly fire throughout the entirety of the swing. An immediate prediction of the sliding filament theory of muscle contraction was that sarcomere number determines contraction and relaxation velocities (Figure 3A ). This prediction was first tested using muscle fibers of equal length but different sarcomere sizes. In such fibers sarcomere length is a proxy for sarcomere number, and over time this resulted in frequent statements in the literature that sarcomere size determined contraction velocity. Recent work in muscles with different sarcomere lengths, muscle lengths, and sarcomere numbers has reconfirmed that sarcomere number is the determining parameter (Figure 3B). Because limb speed depends on muscle activation and lever arm considerations as well as muscle intrinsic contraction velocity, this observation does not imply that limbs powered with muscles containing few sarcomeres must move more slowly than limbs powered by muscles containing many sarcomeres. Nonetheless, how quickly muscles can contract is presumably important to nervous systems in calculating how many and how rapidly motor neurons should fire. This issue is a particular concern in mammals and birds, in which sarcomere length is uniform across muscles, and thus muscle length must change by altering sarcomere number. Mouse leg muscles should therefore contract and relax more slowly than horse leg muscles, and toddler leg muscles more slowly than adult leg muscles. To my knowledge whether motor network activity varies in ways consistent with this predicted size dependence of muscle contraction velocity has not been examined, but these considerations suggest such investigation might be useful. Several of the properties noted above are strongly affected by whether movements occur in air or water. For instance, since the density of most biological materials is near that of water, limb weight in water is near zero. The limb sizes at which gravity, not muscle antagonism, determines the rest angle of a joint will therefore be much greater in water. With respect to momentum-based movements, momentum depends on mass and is thus independent of medium. However, fluid drag increases with medium density. Air's 1 kg/m3 density is small enough that, for biological limb velocities and durations, drag forces inconsequentially decelerate momentum-based swings. Water's 1,000-fold greater density results in drag forces large enough that momentum-based strategies are very unlikely in aquatic limbed organisms. Aquatic animals must have evolved neuromuscular and biomechanical systems taking into account that, for them, drag matters. And, since drag depends on area, the importance of this issue should vary with size. Biological motor control principles are increasingly being applied to robotic design. The size dependencies noted above suggest that building ‘biologically-inspired’ robots requires matching the neural, muscle, and biomechanical components according to size. For instance, consider the momentum-based vs. non-momentum-based motor control strategies in Figure 2D, with, at the extremes of body size, their very different motor neuron firing patterns. Engineers can design robots capable or incapable of the free movement upon which momentum-based strategies depend. But having made that choice, if biological networks are inspiring the robot's movement control, it is critical that a matching (momentum- or non-momentum-based) neural network be used. Analogous concerns would apply to pairing gravity-independent (small limb) robotic actuator and limb assemblies with gravity-dependent (large limb) neural networks, or components from systems in which drag is important with ones in which it is inconsequential. It is also important to stress that nervous system, muscle, and biomechanics evolved in a concerted, interdependent, and incremental manner. Computer programs, actuators, and limb materials fundamentally differ from neurons, muscle, and bone. Even if biological movement generation were completely understood, these differences make it unclear that this knowledge alone would allow for the design of robots with equal functionality. Although biology will undoubtedly make useful contributions, ultimate success in robotic design will likely require an analogous concerted evolution of robotic control mechanisms, actuators, and structural materials. The data above suggest that neural control mechanisms may vary significantly with body size, and that the details of this dependence will differ in terrestrial vs. aquatic animals. In terms of the size range considered here, most neurobiological research, particularly at the cellular level, is performed on small to tiny animals, with aquatic and terrestrial species being approximately equally represented. Determining which principles gleaned from these data are truly general is at present difficult. Similar motor control differences also likely apply to differently sized limbs in single animals, e.g., fingers vs. legs. A general understanding of biological motor control would thus benefit from more studies of larger animals and of differently sized limbs in a single species, and from a greater consideration of possible media-specific effects.