Harvard University COVID-19 updates

Department News

Who Moved My Arm? Learning from Performance Error [Uchida Lab]

Who Moved My Arm? Learning from Performance Error [Uchida Lab]

You head to your favorite coffee shop, order a cappuccino, and when the barista calls your name, you grab your coffee. Imagine, however, you grab the wrong cup on the bar and it was empty. You lift the cup and it “flies” towards you, as you applied too much force. When you grab the correct cup to drink, you carefully adjust the strength of your grasp and force you apply to lift the cup to your mouth, and can seamlessly account for changes in the weight of the cup as you consume the coffee.

Our ability to monitor the outcomes of our actions and compare it with predictions is critical in executing movements. For instance, just grasping a cup requires that the brain predicts how your arm will move taking into account the dynamics of the arm, including how heavy your arm is (and whether you are wearing a watch that could add weight) or perhaps whether you wear a stiff leather jacket (that could add resistance), which could otherwise perturb your arm movement. We can execute simple motor movements like reaching because the brain effortlessly learns to predict these recurring perturbations and “adapts” to them. This process of the brain predicting the consequences of a particular movement, monitoring whether the prediction was correct, and adjusting future actions is called motor adaptation.

How the brain adapts to perturbations has been studied in the laboratory by introducing systematic disturbances either to a movement or to sensory feedback. About 20 years ago, two researchers, Shadmehr and Mussa-Ivaldi, asked subjects to move a joystick that was connected to motors. These motors can, for instance, add predetermined forces that may “kick” the arm laterally during a movement. They investigated how subjects learned to adapt their movements to reach a target in a similar manner as before the perturbations. The simplicity and elegance of their paradigm received much attention, which led to many experimental findings and influential computational theories about how the motor system adapts to systematic perturbations. Yet, understanding the neural mechanisms that support motor adaptation has been lagging, largely because previous animal models, i.e. primates, which are known to manipulate joysticks, are not particularly amenable to modern circuit analysis using population recording and manipulation techniques employing molecular and genetic methods.

In the present study (Mathis et al., Neuron, 2017, PDF), we established the first mouse model of motor adaptation that naturally lends itself to powerful neural circuit studies. Head-fixed mice were trained to manipulate a joystick: They had to reach towards a joystick, grab it and pull it from a starting location to a target location to receive a reward. The joystick can be moved in two independent directions. To get rewards mice had to pull it from the center towards themselves. To study motor adaptation, we used a magnetic force to apply a brief “kick”, or force-field pulse, halfway through a pull that deflects the joystick perpendicularly to the pull direction. Over the course of 100 trials the mice learned to predictively steer away from the kick before the onset of the pulse and reduced motor errors during the force-pulse period.  Furthermore, when the force field was (unpredictably) turned off, they moved the joystick in the opposite direction of the force pulse. Such a movement, called “aftereffect”, reveals that the mice were executing a counteracting force against the fore field, rather than stiffening their arm to counteract the perturbation.  These results demonstrated that mice could adapt to recurring perturbation to a forelimb movement, that bears a striking similarity to the adaptation observed in humans and non-human primates.

What in the brain enabled this adaptation? In our task the mice cannot see the joystick, therefore mice only receive proprioceptive feedback (from the “sense of posture”). However, there are two main sensory feedback pathways from touch and posture sensors: direct projections from the spinal cord to the cerebellum and thalamo-cortical projections targeting all the way to the neocortex. Previous studies in humans have implicated the cerebellum as a critical brain center in regulating motor adaptation, although its role remains debated. These studies have limitations: because most patients with impairments to the cerebellum also have movement disorders, dissociating deficits in adaptation from those in fine motor control has been difficult. Moreover, no studies had directly tested the role of the other feedback pathway to the neocortex, so its role in adaptation remained unclear.

To test whether the subcortical pathway was sufficient for motor adaptation in our paradigm, we first tested whether the thalamo-cortical pathway plays any role. To do this, we took advantage of optogenetics: By expressing a light-gated ion channel, channelrhodopsin-2, in GABAergic inhibitory neurons, one can inactivate a target area with high temporal and spatial specificity. Brief inactivation of S1, applied concurrently with the force field, abolished predictive steering, the reduction in errors and the aftereffect. Our results therefore demonstrate an essential role for the primary somatosensory cortex (S1) in motor adaptation. Remarkably, the lack of motor adaptation came with striking specificity – the execution of forelimb movements and reward-based learning were not impaired by inactivating S1. These results showed that subcortical processing by the cerebellum alone is not sufficient to support motor adaptation in our task. Interestingly, inactivating S1 after mice partially adapted did not interfere with the execution of already-adapted motor commands, consistent with the idea that S1 plays a critical role in learning to predict the force field, but not in storing the memory about the force field.

Motor adaptation is an extremely exciting field of study, with elaborate behavioral experiments, as well as attractive theories and computational models. Nonetheless, the mechanisms at the biological level are far less understood. The mouse model for adaption, which we developed, opens up new avenues to study detailed neural circuit mechanisms, and in turn to test, refine and further develop theories of motor adaptation.



by Mackenzie Weygandt Mathis and Alexander Mathis and Nao Uchida

Uchida lab website

Nao Uchida’s faculty profile

(l to r) Nao Uchida, Mackenzie Weygandt Mathis, and Alexander Mathis

(l to r) Nao Uchida, Mackenzie Weygandt Mathis, and Alexander Mathis