Investigating Motor Skill Learning Processes with a Robotic Manipulandum

Skilled reaching tasks are commonly used to study motor skill acquisition as well as impairment of motor function under pathological conditions⁶. Reliable and unambiguous analysis of reaching behavior is essential for the study of cellular mechanisms underlying motor skill acquisition, as well as neurophysiological processes involved in loss and subsequent recovery of function in animal models of neurological disease. The results presented here show how spatial and temporal aspects of the pulling movement show distinct profiles during motor skill training. These may reflect different subprocesses within the motor skill learning process^7,12.

In the results presented here, we show that even a small change in training protocol, such as a different starting position of the handle (SP1 vs. SP2), results in altered movement parameter profiles. By adding two FP sessions prior to straight pull training to our previously published training protocol¹⁰, we were able to dissociate the effects of learning to pull from the skill to pull straight. In addition, the off-center placement of the robot's handle in this improved training protocol (SP2) results in a task with a shallower learning curve that is more advantageous for studying mechanisms of motor learning as it allows time for interventions before plateau performance levels are reached. Moreover, it is possible to distinguish factors related to task execution from factors related to task difficulty that are not immediately reflected in success rates, but may indicate further refinement of motor skill⁴. Smooth execution of the pulling movements, reflected in the number of submovements may be considered a measure of skilled, straight pulling. By contrast, spatial variability of the movement drops rapidly as the number of valid pulls increases in all three versions of the reach-and-pull task, but is not directly related to pulling success in straight pull tasks and may reflect general motor function or convergence on the correct movement required for a successful trial, rather than skilled task execution once the rules of the task have been learned.

Execution of our robotic reach-and-pull tasks is remarkably reliable and reproducible between animals. All trained animals acquire the free pull task, and a majority (90-95% of animals) is able to learn the straight pull task. Even animals that show consistently low success rates during straight pulling continue to execute a high number of valid pulling attempts. These failed, but valid attempts are fully recorded. Failed reaching attempts in a classical skilled reaching task typically results in an incomplete reaching movement. It is therefore not only possible to analyze aspects of the movement related to successful task acquisition, but also to assess which movement parameters result in failure.

While the measurements presented here give detailed information about pulling movements, not all aspects of the reach-and-pull forelimb movement are captured. Any changes in reaching behavior that occur prior to the first touch or after release of the handle at the end of a valid pulling attempt are not recorded and can therefore not be analyzed with equal precision. For example, the number of grabbing attempts prior to a pulling movement are not measured, even though they may be relevant in relation to functional recovery models. Analysis of high-speed video recordings provides this additional information. Previously described methods for movement tracking^13,14 can be adapted for this purpose.

Moreover, our recordings do not provide information about the quality of the rat's grip of the handle. Rotation of the handle, indicating pronation or supination of the paw, could provide some insight in combination with high-speed video. Since rotation and grip are particularly affected in rat stroke models^9,15,16, future experiments are needed to determine how effective the reach-and-pull tasks presented here are in capturing post-stroke motor deficits.

The tasks presented here were designed to mimic conventional single pellet reaching tasks: the required pulling distance is based on the typical distance between window and pellet in these tasks and the free movement of the handle within the horizontal plane enables measurement of the rat's natural paw trajectories over a 10 mm distance in valid trials.

Likewise, invalid trials where the handle is moved outside of the robot's workspace (e.g., the handle is swiped aside during pulling) or where the pulled distance is insufficient could be interpreted as being similar to dropped pellets in single pellet reaching tasks, even though the handle does not drop to the floor when released.

This design captures more aspects of the pulling movement than automated tasks aimed at measuring a single simple movement. However, it also enables interaction between pulling and rotational movements and gives animals the opportunity for compensation. Insight into compensatory movements during motor function recovery may be valuable, but also complicates interpretation of the results.

A critical step in acquisition of the reach-and-pull tasks is successful conditioning of the reward touch step of training. Without reliable interaction between rat and robot, any further training steps are difficult to carry out and quantify reliably. It is equally important to not overtrain the rats during transitional training steps, however. While animals may continue to show improvements in performance during reward-touch training sessions for more than 3-4 days, over-consolidation prevents effective shaping of behavior in subsequent reach-and-pull training.

Constant placement of the handle is essential for correct execution and reliable analysis of any data obtained using the robotic reach-and-pull task presented here. While the position of the handle relative to rest of the robot and to midline is software-defined, the handle's position relative to the cage window is easily varied by moving either sideways. Here, we have shown how alignment of the handle with either the edge or center of the window alters task acquisition and can be used to study processes underlying motor skill learning. Inconstant alignment of the handle within a training period, however, will introduce confounding effects and yields unreliable behavioral readouts.

In the protocol described here, rats are trained during daily sessions comprised of 100 trials, similar to our previous single pellet reaching experiments^4,17. With automated training setups the number of trials per session can easily be increased, without requiring much more effort on the part of the researcher. While higher numbers of trials per session can result in lower intra-individual variability, the effects of increasing the number of trials per training session on the speed of learning and recovery has to be taken into account. Moreover, factors like satiety and fatigue will become more relevant and could interfere with performance in very long training sessions.

Although strain and sex differences in motor skill learning ability have been described in rats^4,18,19, we have obtained reliable performance in both male Sprague-Dawley and Long-Evans rats. Older animals (4-5 months old) are generally slower than younger ones (8-10 weeks) and show longer intervals between trials. Moreover, we have observed poor performance in older animals when food pellets are replaced with a sucrose-water reward. Older animals that receive liquid rewards experience difficulties when transitioning from reward-touch to pulling. This may be caused by age-related reward-preferences, in which case using a more palatable liquid reward such as yoghurt or peanut oil may yield better results. Alternatively, the behavior of older animals might indicate difficulties in understanding a more abstract reaching task (compared to conventional pellet reaching), although we do observe reliable learning behavior in these animals when solid food rewards are used. Thus, the effect of the chosen reward should particularly be considered when designing experiments where the use of older animals is preferable (i.e. in models of stroke or neurodegenerative disease).

We have not yet studied sex differences in motor skill learning using ETH Pattus. It is unlikely that female rats will not be able to acquire the SP or FP tasks, although their learning strategies may differ from males. However, although the estrous cycle affects spine density in female rats, effects on learning curves and learning-related plasticity in primary motor cortex are limited²⁰.

The robotic tasks can be varied in several ways: the required movement may be more or less precise (for example, a limited amount of deviation from a prescribed trajectory in a reach-and-pull task), or the robot may interfere with or assist the executed movement in any one or all of three dimensions (lengthwise or sideways movements in the horizontal plane and rotation angle of the handle). Other than variations on the reach-and-pull tasks presented in this paper, it is possible to design motor skill learning tasks where for example rotation angle of the handle, maximum movement velocity, or acceleration profile define the success of a trial.

Other than allowing easy variation of task parameters, the experimental setup presented here spatially separates the motor action from the given reward, which is dispensed on the opposite side of the cage. Adapting the reward size is not possible in classical reaching tasks without affecting task difficulty²¹ (a smaller food pellet is more difficult to grasp than a larger one), nor is it possible to vary the chance of a reward independently of the animal's skill level. Using a robotic task, the reward obtained for a motor action may be varied based on skill, current performance, or may be varied to assess factors such as motivation.

In conclusion, automated training combined with kinematic movement analysis provides an automated, objective method for studying motor skill learning that closely mimics conventional skilled reaching tasks, but yields additional data of both successful and failed pulling attempts. This approach opens up new avenues of investigation in combination with electrophysiological, pharmacological or optogenetic interventions aimed at enhancement of or interference with reach-and-pull movements or resulting food rewards.