Permanent Present Tense

by Suzanne Corkin

  A striking example of coordinating sensory and motor circuits is prism adaptation. To perform this task, participants wear glasses with prisms that bend the light a few degrees to the left or right, making objects appear displaced to the left or right of their true locations. Before donning the prisms, participants practice pointing to a target with normal vision. Once they are proficient, the experimenter asks them to wear the prism glasses, thereby changing the visual environment in which the target is presented. If the prism glasses displace the target slightly to the left, participants at first point to the right of the target. But after practicing for a few minutes, they update their movements and eventually reach the target. When they remove the prisms and point again, they show an aftereffect of adaptation—pointing in the opposite direction, indicating that they had adapted to the altered visual information.

  Research in the late 1990s helped uncover the brain circuits required for the adaptive process. To pinpoint the specific area critical for accommodating changes in the visual environment, neuroscientists administered a prism-adaptation task to patients with disorders of the cerebellum. In a 1996 experiment, participants threw balls at a target under three conditions: before donning the prisms, while they wore them, and immediately after they removed them. The researchers assessed learning during the third condition. Because the prisms made the target appear to the left of where it really was, participants initially threw the balls to the left side of the target. With practice, they gradually threw more and more to the right, and the impact points moved back toward the center of the target. After removing the prisms, the control participants continued throwing to the right side of the target, as if they were still wearing the prisms, indicating that they had adapted to the visual shift. This negative aftereffect is the measure of learning. The patients with cerebellar disorders did not show a negative aftereffect, convincing proof that their brains had not retained the altered map created by the prisms. This experiment shows that the cerebellum integrates two kinds of information, perceptual and motor, to accommodate vicissitudes in the visual world.21

  Remarkably, when we tested him in the mid 1990s, Henry showed normal adaptation to prisms, despite having marked cerebellar atrophy. The prism-adaptation task was ideal for testing the effects of his cerebellar damage on a kind of nondeclarative learning that grows from interactions between brain circuits specialized for visual perception and for movement. Our experiment tested whether Henry’s motor system could adapt to a situation in which prisms displaced everything in his workspace eleven degrees to the left. To accomplish this visual shift, we asked him to wear glass prisms set in a pair of laboratory goggles. His task was to point quickly with his right index finger to a vertical line at arm’s length in three situations: a baseline condition without prisms, an exposure condition with prisms, and a post-exposure condition without prisms. In each condition, Henry pointed to nine different targets, one straight ahead and four to each side. We presented each target four times in random order. For each trial, we recorded the position of Henry’s finger and then determined how far it was from the target. As in other prism-adaptation experiments, the measure of learning was the amount of negative aftereffect in the post-exposure condition—how far the point he touched deviated from the target.

  Henry performed just like the ten control participants. In the exposure condition, he could clearly see that he was pointing far to the left of the target and gradually shifted his pointing to the right to hit the target dead-on. When the prisms were removed, he continued to point to the right of each target as if the prisms were still in place—clear evidence of a normal aftereffect. During the experiment, sensory and motor circuits in Henry’s brain interacted successfully to accomplish this nondeclarative learning.

  Although we do not yet know what residual cerebellar function supported Henry’s good performance, we hope to understand these results more fully as we examine his postmortem brain and identify the specific cerebellar circuits that were left intact. Of particular interest are structures that transmit information to the cerebellum—the deep cerebellar nuclei—which if spared may have provided the necessary machinery for prism adaptation. Figuring out the anatomical substrate for prism adaptation will be a noteworthy accomplishment.

  Our investigations of motor-skill learning contrasted Henry’s performance with that of other patients who had damage to areas outside the medial temporal lobes. We learned that motor-skill learning and declarative memory are assigned to different compartments in the brain. The hippocampal region contains the circuits critical for recalling and recognizing facts and events, but not for learning new motor skills. In contrast, circuits in the caudate nucleus, putamen, and cerebellum are necessary for motor-skill learning, but not for retrieving facts and events.

  Although Henry could acquire new skills in the laboratory, this ability did not bring much benefit to his everyday life, except for mastering the walker. The symptoms caused by the damage to his cerebellum, on top of his epilepsy, were not conducive to dancing or learning new sports. He did play croquet, but we do not know whether his game improved with practice.

  In addition to studying brain-damaged patients to deconstruct the neural architecture of motor skills, scientists have proposed theoretical models to explain how the brain learns and then performs these tasks. In 1994, neuroscientists Reza Shadmehr and Ferdinando Mussa-Ivaldi at MIT provided a major breakthrough in understanding motor memory, introducing the idea that when the body makes reaching movements, the motor-control system adapts to unanticipated changes in the environment. The brain accomplishes this feat by constructing an internal model that, with experience, estimates the forces in the environment—the pushing and pulling. The concept of an internal model has become a popular explanation for how the brain represents and modifies learned skills.22

  To understand internal models, imagine that you are thirsty; you pour a glass of water, grasp it, bring it to your lips, and drink. This simple action, which you have carried out many times in many different places, is not as straightforward as it seems. Before you move your arm, your brain receives and processes basic information about the glass: its shape, how heavy it is likely to be, where it is located, and where your hand is. The problem for your brain is to translate the location of the glass on the table and your goal, grasping the glass, into the pattern of muscle activity necessary to bring the glass to your lips. We execute this kind of motor command constantly as we move through our day—brushing our teeth, using a knife and fork, driving a car, browsing the web. Over our lifetimes, we interact with innumerable different objects in a vast set of environments, and each time our brains must transform information from our senses into movement, and luckily can readily adapt to changes from one situation to another.

  Internal models represent circuits in the brain that process the relation between the motion of the hand and the motor commands. For example, an inverse model embodies the relation between the desired motion of the hand and the motor outputs required to achieve this motion. This sort of internal model is a major component of a system that can guide your hand to grasp the glass. Another type of internal model, a forward model, allows the brain to predict the likely outcomes of a motor command and choose those necessary to perform specific motor tasks successfully—in our example, having a drink of water. In 1998, a computational neuroscientist in Japan, in collaboration with a colleague in London, adopted the idea of internal models and proposed that acquiring a new motor skill depends on establishing such internal models for motor-task performance. Motor learning is a process of translating the spatial characteristics of the target or the goal of the movement into an appropriate pattern of muscle activations.23

  The computational neuroscientists proposed that the two kinds of internal models work cooperatively to track what we are actually doing and create a mental picture of the movement we want to achieve. One model registers the link between motor outputs—reaching for and grasping the glass—and the ensuing sensory inputs—th
e glass and the position and velocity of your arm. This model makes step-by-step predictions about the next position and velocity of your arm, given the present state of your arm and the reaching command (go to the glass). The other model provides the actual motor command needed to grasp the glass.24

  When these two internal models interact, the brain compares the actual state of the arm with the desired state of the arm; the discrepancy provides critical information about errors in performance. Error messages facilitate learning by indicating how to adapt the movement to reduce errors and to achieve the desired goal. The brain can switch from one internal model to another based on contextual information—new location of the glass—or on error information—sensorimotor feedback about accuracy. This switching mechanism guarantees flexible adaptation in constantly and rapidly changing environments.25

  When Henry was learning to trace around a star while seeing the pattern, the stylus, and his hand only in mirror-reversed view, he was building up new internal models in his brain depicting the relation between what he saw and how he moved his pencil. These novel internal models had dedicated circuits in his brain, so they did not interfere with all the other motor behaviors he had learned previously. In everyday life, we accumulate many such internal models to build an enormous repertoire of complex motor behaviors.

  Based on evidence from computational modeling, cognitive science, and neurophysiology, researchers in Kyoto, Japan, predicted that internal models are predominately created and stored in the cerebellum. This large, complicated structure is qualified for this assignment because it has the physiological capability to compare the desired movement with the actual movement, and then to use this difference—an error signal—to guide the next movement.

  In 2007, when the Japanese researchers tested this hypothesis with functional MRI, they obtained the first physiological evidence that internal models are formed in the cerebellum. These scientists conducted a series of experiments in which participants executed a tracking task, moving a computer mouse to keep the cursor on a target that moved randomly on a computer screen. In the baseline condition, the mouse was in the normal orientation, but in the test condition, it was rotated 120 degrees, changing the relation between the mouse and cursor and forcing participants to learn how to control the mouse in a new way. Training occurred over eleven sessions, with functional MRI scanning on the odd-numbered sessions to capture the neural activity associated with the learning process from beginning to end.

  During testing, the researchers found two separate regions of activity in the cerebellum. One was an error-related region where neural activity decreased as learning progressed and tracking became more accurate. The second was unrelated to tracking errors; instead, it was an internal-model-related region where activity continued to appear late in training, and seemed to be the site where an enduring internal model of the new tracking skill was stored. The neural activation for this motor-learning task occurred in many areas on both sides of the cerebellum, some of which receive helpful information from the frontal and parietal cortices about planning, strategy, and reaching movements.26

  Considering the evidence about the critical role of the cerebellum in motor-skill learning, it was surprising to me that Henry, whose cerebellum was severely damaged by his medicine, did as well as he did on mirror tracing, rotary pursuit, and bimanual tracking. My initial studies had been limited because they provided only crude measures of Henry’s performance—how many errors he made and how long it took him to complete a task. I sought a deeper understanding of how Henry’s brain controlled his movements throughout the skill-learning process. In 1998, an exciting and fruitful collaboration with Shadmehr, a researcher at Johns Hopkins University, enabled us to examine Henry’s motor memory processes in greater detail during the course of learning. Shadmehr had been a postdoctoral fellow in my department, and I was impressed by his research and expertise in the field of motor control. I, therefore, invited him and two of his students to MIT to conduct a skill-learning experiment.

  The motivation for our experiment was a 1996 study demonstrating that consolidation of the motor-learning experience continues after learning, an insight that came from examining motor-memory consolidation in healthy young adults. When the participants performed a motor skill they had practiced in a previous session, they immediately performed better than they had at the end of the last training trial, indicating that the memory had improved over the intervening time. This gain, however, was disrupted when participants were instructed to learn a second motor task right after the first. Their consolidation of the first task was disturbed because of interference from the second task. In contrast, no disruption occurred if four hours elapsed between learning the two motor skills. The study suggests that consolidation in motor memory happens rapidly—over a period of just four hours after practice, the memory of a new skill was transformed from an initial fragile state to a more solid state. This rapid time-course contrasts with the consolidation of declarative memories, which may require years.27

  This discovery made in the laboratory often carries over into personal experiences. One of my editors tells her ski instructors that she can learn only one new skill per lesson. If they try to teach her two or more, she does not learn anything at all because the consolidation of one new skill is interrupted by switching to another.

  Shadmehr’s 1996 experiment in healthy young adults raised important questions about skill learning: does the consolidation of motor memories require that the participants remember declarative information about the task? Does the medial temporal-lobe need to function normally for the interference effect to occur? Because declarative memory was operational in the young adults we studied, the answers had to come from a memory-impaired participant. Studying Henry, whose declarative memory was decimated, could tell us definitively whether this source of knowledge mattered. Our study was the first to examine the process of interference associated with motor memories in amnesic patients. If declarative memory played no role in the interference of motor memories after practice, then the consequences of learning multiple motor skills should have been the same for Henry and the control participants.28

  During a two-day experiment, we studied Henry’s ability to learn a novel motor skill. The task was not a video game, but it resembled the Wii game Link’s Crossbow Training, in which players shoot at bull’s-eyes as they pop up on the screen. Initially, the targets are stationary, and when the player hits one, it explodes; as the game advances, the bull’s-eyes move, making the task more difficult. In our experiment with Henry, the targets were always stationary. After he became proficient at firing straight at the targets, we introduced an unexpected change by mechanically jolting his arm as he moved, thus throwing him off-course. We wanted to know whether with training, his moves to the target would become straight again (see Fig. 14).

  The apparatus for the task was a mechanical arm with a video monitor located just above it. When the researchers first sat Henry in front of the arm, he, like all inexperienced volunteers, sat quietly without touching the machine. They asked him to grasp the handle of the mechanical arm and move it around a bit to get used to it. At first, he kept his gaze on his hand as he moved the handle, but then they told him to look at the monitor, where a cursor was present. After Henry moved the cursor around for a minute or so, the researchers illuminated a target at the center of the screen and asked him to move the cursor to that location. They then showed him other individual targets and asked him to move the cursor to those locations as quickly as possible. His goal was to reach each target within one second. Each time he succeeded, the target exploded.

  1. Four lobes of the cerebral cortex

  MRI scan of a healthy forty-one year old man, showing his whole brain viewed from his left side, with the four lobes of the cerebral cortex delineated. The frontal lobe (Fr) modulates basic motor functions and cognitive control processes (setting goals, making decisions, solving problems); the temporal lobe (Te) complex visual and auditory process
es, memory, language, and emotion; the parietal lobe (Pa) touch, pain, and other body sensations as well as spatial ability and language; and the occipital lobe (Oc) basic visual processes. The peaks in the cortex are the gyri and the valleys between them are the sulci. The cerebellum (Cb) is specialized for balance and movement coordination. In Henry’s brain, this structure was badly shrunken as a side effect of Dilantin. The brainstem (BrSt) connects the spinal cord to the rest of the brain. It is the entryway for information from the senses, and circuits in the brain stem control vital bodily functions, such as heart rate, blood pressure, respiration, and level of consciousness.

  2a. Medial temporal-lobe structures

  The same healthy brain showing the medial temporal-lobe structures that were removed from Henry’s brain. The amygdala (Am, dark gray outlined in white) and the head and body of the hippocampus (Hp, light gray outlined in white). The tail of the hippocampus has been removed for simplicity but would normally continue upward forming the fornix (Fx), which arches forward to the mammillary bodies of the hypothalamus (Hy). The back part of the parahippocampal cortex (Ph) was spared in Henry, but the front part was removed. Other structures shown are the cerebellum (Cb), the striatum (St), an area engaged in motor control and motor learning, and the thalamus (Th), a structure where information from the eyes, ears, and skin is relayed to pathways that carry the input to the cortex.