Research Shows How Brain Can Tell Magnitude of Errors

University of Pennsylvania researchers have made another advance in understanding how the brain detects errors caused by unexpected sensory events. This type of error detection is what allows the brain to learn from its mistakes, which is critical for improving fine motor control.  

Their previous work explained how the brain can distinguish true error signals from noise; their new findings show how it can tell the difference between errors of different magnitudes. Fine-tuning a tennis serve, for example, requires that the brain distinguish whether it needs to make a minor correction if the ball barely misses the target or a much bigger correction if it is way off.

The study was led by Javier Medina, an assistant professor in the Department of Psychology in Penn’s School of Arts & Sciences, and Farzaneh Najafi, then a graduate student in the Department of Biology. They collaborated with postdoctoral fellow Andrea Giovannucci and associate professor Samuel S. H. Wang of Princeton University.

It was published in the journal eLife.

Our movements are controlled by neurons known as Purkinje cells. Each muscle receives instructions from a dedicated set of hundreds of these brain cells. The instructions sent by each set of Purkinje cells are constantly fine tuned by climbing fibers, a specialized group of neurons that alert Purkinje cells whenever an unexpected stimulus occurs.

“An unexpected stimulus is often a sign that something has gone wrong,” Medina said, “When this happens, climbing fibers send signals to their related Purkinje cells that an error has occurred. These Purkinje cells can then make changes to avoid the error in the future.”

These error signals are mixed in with random firings of the climbing fibers, however, and researchers were long mystified about how the brain tells the difference between this noise and the useful, error-related information it needs to improve motor control.

Medina and his team showed the mechanism behind this differentiation in a study published earlier this year. By using a non-invasive microscopy technique that could monitor the Purkinje cells of awake and active mice, the researchers could measure the level of calcium within these cells when they received signals from climbing fibers.

The unexpected stimuli in this experiment were random puffs of air to the face, which caused the mice to blink. The researchers located Purkinje cells that control the mice’s eyelids and saw that calcium levels necessary for neuroplasticity, i.e., the brain’s ability to learn, were greater when the mice got an error signal triggered by a puff of air than they were after a random signal.

While being able to make such a distinction is critical to the brain’s ability to improve motor control, more information is needed to fine-tune it.  

“We wanted to see if the Purkinje cells could tell the difference not just between random firings and true errors signals but between smaller and bigger errors,” Medina said.

In their new study, the researchers used the same experimental set-up, with one key difference. They used air puffs of different durations: 15 milliseconds and 30 milliseconds.

What they found was that the eyelid-associated Purkinje cells filled with different amounts of calcium depending on the length of the puff; the longer ones produced larger spikes in calcium levels.        

In addition, the researchers saw that different percentages of eyelid-related Purkinje cells respond depending on the length of the puff.  

“Though there is a large population of climbing fibers that can give error-related information to the relevant Purkinje cells when they encounter something unexpected, not all of them fire each time,” Medina said. “We saw that there is information coded in the number of climbing fibers that fire. The longer puffs corresponded to more climbing fibers sending signals to their Purkinje cells.”

Their study could help explain how practice makes perfect, even when errors are imperceptibly small.

“If you felt a short puff and a long puff, you might not be able to say which one was which, but Purkinje cells can tell the difference,” Medina said. “The difference between a ‘very good’ and an ‘awesome’ tennis serve rests on being able to distinguish errors even as tiny as that.” 

Memories of Errors Foster Faster Learning

Using a deceptively simple set of experiments, researchers at Johns Hopkins have learned why people learn an identical or similar task faster the second, third and subsequent time around. The reason: They are aided not only by memories of how to perform the task, but also by memories of the errors made the first time.

“In learning a new motor task, there appear to be two processes happening at once,” says Reza Shadmehr, Ph.D., a professor in the Department of Biomedical Engineering at the Johns Hopkins University School of Medicine. “One is the learning of the motor commands in the task, and the other is critiquing the learning, much the way a ‘coach’ behaves. Learning the next similar task goes faster, because the coach knows which errors are most worthy of attention. In effect, this second process leaves a memory of the errors that were experienced during the training, so the re-experience of those errors makes the learning go faster.”

Shadmehr says scientists who study motor control — how the brain pilots body movement — have long known that as people perform a task, like opening a door, their brains note small differences between how they expected the door to move and how it actually moved, and they use this information to perform the task more smoothly next time. Those small differences are scientifically termed “prediction errors,” and the process of learning from them is largely unconscious.

The surprise finding in the current study, described in Science Express on Aug. 14, is that not only do such errors train the brain to better perform a specific task, but they also teach it how to learn faster from errors, even when those errors are encountered in a completely different task. In this way, the brain can generalize from one task to another by keeping a memory of the errors.

To study errors and learning, Shadmehr’s team put volunteers in front of a joystick that was under a screen. Volunteers couldn’t see the joystick, but it was represented on the screen as a blue dot. A target was represented by a red dot, and as volunteers moved the joystick toward it, the blue dot could be programmed to move slightly off-kilter from where they pointed it, creating an error. Participants then adjusted their movement to compensate for the off-kilter movement and, after a few more trials, smoothly guided the joystick to its target.

In the study, the movement of the blue dot was rotated to the left or the right by larger or smaller amounts until it was a full 30 degrees off from the joystick’s movement. The research team found that volunteers responded more quickly to smaller errors that pushed them consistently in one direction and less to larger errors and those that went in the opposite direction of other feedback. “They learned to give the frequent errors more weight as learning cues, while discounting those that seemed like flukes,” says David Herzfeld, a graduate student in Shadmehr’s laboratory who led the study.

The results also have given Shadmehr a new perspective on his after-work tennis hobby. “I’m much better in my second five minutes of playing tennis than in my first five minutes, and I always assumed that was because my muscles had warmed up,” he says. “But now I wonder if warming up is really a chance for our brains to re-experience error.”

“This study represents a significant step toward understanding how we learn a motor skill,” says Daofen Chen, Ph.D., a program director at the National Institute of Neurological Disorders and Stroke. “The results may improve movement rehabilitation strategies for the many who have suffered strokes and other neuromotor injuries.”

The next step in the research, Shadmehr says, will be to find out which part of the brain is responsible for the “coaching” job of assigning weight to different types of error.

Fruit flies ‘think’ before they act

Oxford University neuroscientists have shown that fruit flies take longer to make more difficult decisions.

In experiments asking fruit flies to distinguish between ever closer concentrations of an odour, the researchers found that the flies don’t act instinctively or impulsively. Instead they appear to accumulate information before committing to a choice.

Gathering information before making a decision has been considered a sign of higher intelligence, like that shown by primates and humans.

'Freedom of action from automatic impulses is considered a hallmark of cognition or intelligence,' says Professor Gero Miesenböck, in whose laboratory the new research was performed. 'What our findings show is that fruit flies have a surprising mental capacity that has previously been unrecognised.'

The researchers also showed that the gene FoxP, active in a small set of around 200 neurons, is involved in the decision-making process in the fruit fly brain.

The team reports its findings in the journal Science. The group was funded by the Wellcome Trust, the Gatsby Charitable Foundation, the US National Institutes of Health and the Oxford Martin School.

The researchers observed Drosophila fruit flies make a choice between two concentrations of an odour presented to them from opposite ends of a narrow chamber, having been trained to avoid one concentration.

When the odour concentrations were very different and easy to tell apart, the flies made quick decisions and almost always moved to the correct end of the chamber.

When the odour concentrations were very close and difficult to distinguish, the flies took much longer to make a decision, and they made more mistakes.

The researchers found that mathematical models developed to describe the mechanisms of decision making in humans and primates also matched the behaviour of the fruit flies.

The scientists discovered that fruit flies with mutations in a gene called FoxP took longer than normal flies to make decisions when odours were difficult to distinguish – they became indecisive.

The researchers tracked down the activity of the FoxP gene to a small cluster of around 200 neurons out of the 200,000 neurons in the brain of a fruit fly. This implicates these neurons in the evidence-accumulation process the flies use before committing to a decision.

Dr Shamik DasGupta, the lead author of the study, explains: ‘Before a decision is made, brain circuits collect information like a bucket collects water. Once the accumulated information has risen to a certain level, the decision is triggered. When FoxP is defective, either the flow of information into the bucket is reduced to a trickle, or the bucket has sprung a leak.’

Fruit flies have one FoxP gene, while humans have four related FoxP genes. Human FoxP1 and FoxP2 have previously been associated with language and cognitive development. The genes have also been linked to the ability to learn fine movement sequences, such as playing the piano.

'We don't know why this gene pops up in such diverse mental processes as language, decision-making and motor learning,' says Professor Miesenböck. However, he speculates: 'One feature common to all of these processes is that they unfold over time. FoxP may be important for wiring the capacity to produce and process temporal sequences in the brain.'

Professor Miesenböck adds: ‘FoxP is not a “language gene”, a “decision-making gene”, even a “temporal-processing” or “intelligence” gene. Any such description would in all likelihood be wrong. What FoxP does give us is a tool to understand the brain circuits involved in these processes. It has already led us to a site in the brain that is important in decision-making.’

Bioengineer Studying How the Brain Controls Movement

A University of California, San Diego research team led by bioengineer Gert Cauwenberghs is working to understand how the brain circuitry controls how we move. The goal is to develop new technologies to help patients with Parkinson’s disease and other debilitating medical conditions navigate the world on their own. Their research is funded by the National Science Foundation’s Emerging Frontiers of Research and Innovation program.

"Parkinson’s disease is not just about one location in the brain that’s impaired. It’s the whole body. We look at the problems in a very holistic way, combine science and clinical aspects with engineering approaches for technology," explains Cauwenberghs, a professor at the Jacobs School of Engineering and co-director of the Institute for Neural Computation at UC San Diego. "We’re using advanced technology, but in a means that is more proactive in helping the brain to get around some of its problems—in this case, Parkinson’s disease—by working with the brain’s natural plasticity, in wiring connections between neurons in different ways."

Outcomes of this research are contributing to the system-level understanding of human-machine interactions, and motor learning and control in real world environments for humans, and are leading to the development of a new generation of wireless brain and body activity sensors and adaptive prosthetics devices. Besides advancing our knowledge of human-machine interactions and stimulating the engineering of new brain/body sensors and actuators, the work is directly influencing diverse areas in which humans are coupled with machines. These include brain-machine interfaces and telemanipulation.

Lining up our sights

Neurologists at LMU have studied the role of the vestibular system, which controls balance, in optimizing how we direct our gaze. The results could lead to more effective rehabilitation of patients with vestibular or cerebellar dysfunction.

When we shift the direction of our gaze, head and eye movements are normally highly coordinated with each other. Indeed, from the many possible combinations of speed and duration for such movements, the brain chooses the one that minimizes the error in reaching the intended line of sight. Dr. Nadine Lehnen, who heads a research group based at LMU’s Center for Vertigo and Balance Disorders, in collaboration with her colleague Dr. Murat Saglam and Professor Stefan Glasauer of the Center for Sensorimotor Diseases at LMU, have now published a paper in the latest issue of the journal of Brain which investigates the significance of the vestibular system for this optimization of motor coordination. The vestibular system in the brain is mainly responsible for the maintenance of balance and posture. The new work focused on subjects suffering from bilateral defects in the vestibular system (a complete vestibulopathy) or lesions in the cerebellum, which is functionally linked to it.

The authors of the new study had previously developed a mathematical model that enabled them to predict the horizontal movements of the head and eyes in response to the presentation of an off-center stimulus. “When subjected to repeated trials, healthy subjects are able to select the combination of eye and head movements that minimizes gaze shift variability,” says Glasauer. They unconsciously choose the set of movements associated with the least error in the endpoint. Moreover, they can do this even when wearing a helmet with weights attached, which alters the moment of inertia of the head.

Learning to find the endpoint

However, patients who show defects in the vestibular system or the cerebellum have greater difficulty in controlling the direction of gaze in response to changes in their environment. “It turns out that information relayed from the balance organs to the vestibular system is essential for the optimization of gaze shifts,” says Nadine Lehnen. Patients with complete bilateral vestibular loss are therefore unable to perform such shifts in the most efficient way. “In striking contrast, patients with cerebellar damage can, to a certain extent, learn to optimize certain parameters of head and eye movements, by adjusting the velocity of head movement, for instance,” says Glasauer.

"These results provide the first evidence that the vestibular system is critical for optimizing voluntary movements“, says Dr. Kathleen E. Cullen from McGill University in Montreal in a scientific commentary to the study appearing in the print issue of Brain. The new findings are of relevance for the rehabilitation of patients who have suffered damage to the cerebellum and patients with incomplete vestibulopathies. “We assume that gaze shift control in these patients can be enhanced by a rehabilitation training based on active head movements,” says Nadine Lehnen. Head movements provide the vestibular feedback which generates the sensorimotor error messages that underlie the ability to learn how to optimize the coordination of eye and head movements. Instead of trying to hold their heads steady, these patients should be encouraged to actively move their heads, when they shift their gaze.

The question if patients with partial vestibulopathy can optimize gaze shift behavior by engaging in active head movements is now under investigation. This work forms part of a rehabilitation study which is being carried out at the Center for Vertigo and Balance Disorders at Munich University Hospitals, and is financed by the Federal Ministry for Education and Research.

How sleep helps brain learn motor task

Sleep helps the brain consolidate what we’ve learned, but scientists have struggled to determine what goes on in the brain to make that happen for different kinds of learned tasks. In a new study, researchers pinpoint the brainwave frequencies and brain region associated with sleep-enhanced learning of a sequential finger tapping task akin to typing, or playing piano.

You take your piano lesson, you go to sleep and when you wake up your fingers are better able to play that beautiful sequence of notes. How does sleep make that difference? A new study helps to explain what happens in your brain during those fateful, restful hours when motor learning takes hold.

"The mechanisms of memory consolidations regarding motor memory learning were still uncertain until now," said Masako Tamaki, a postdoctoral researcher at Brown University and lead author of the study that appears Aug. 21 in the Journal of Neuroscience. “We were trying to figure out which part of the brain is doing what during sleep, independent of what goes on during wakefulness. We were trying to figure out the specific role of sleep.”

In part because it employed three different kinds of brain scans, the research is the first to precisely quantify changes among certain brainwaves and the exact location of that changed brain activity in subjects as they slept after learning a sequential finger-tapping task. The task was a sequence of key punches that is cognitively akin to typing or playing the piano.

Specifically, the results of complex experiments performed at Massachusetts General Hospital and then analyzed at Brown show that the improved speed and accuracy volunteers showed on the task after a few hours sleep was significantly associated with changes in fast-sigma and delta brainwave oscillations in their supplementary motor area (SMA), a region on the top-middle of the brain. These specific brainwave changes in the SMA occurred during a particular phase of sleep known as “slow-wave” sleep.

Scientists have shown that sleep improves many kinds of learning, including the kind of sequential finger-tapping motor tasks addressed in the study, but they haven’t been sure about why or how. It’s an intensive activity for the brain to consolidate learning and so the brain may benefit from sleep perhaps because more energy is available or because distractions and new inputs are fewer, said study corresponding author Yuka Sasaki, a research associate professor in Brown’s Department of Cognitive, Linguistic & Psychological Sciences.

"Sleep is not just a waste of time," Sasaki said.

The extent of reorganization that the brain accomplishes during sleep is suggested by the distinct roles the two brainwave oscillations appear to play. The authors wrote that the delta oscillations appeared to govern the changes in the SMA’s connectivity with other areas of the cortex, while the fast-sigma oscillations appeared to pertain to changes within the SMA itself.

Meticulous measurements

Possible roles for fast-sigma and delta brainwaves and for the SMA had suggestive support in the literature before this study, but no one had obtained much proof in part because doing so requires a complex experimental protocol.

To make their findings, Sasaki, Tamaki and their team asked each of their 15 subjects to volunteer for the motor learning experiments. For the first three nights, nine subjects simply slept at whatever their preferred bedtime was while their brains were scanned both with magnetoencephalography (MEG), which measures the oscillations with precise timing, and polysomnography, which keeps track of sleep phase. By this time the researchers had good baseline measurements of their brain activity and subjects had become accustomed to sleeping in the lab.

On day 4, the subjects learned the finger-tapping task on their non-dominant hand (to purposely make it harder to learn). The subjects were then allowed to go to sleep for three hours and were again scanned with PSG and MEG. Then the researchers woke them up. An hour later they asked the subjects to perform the tapping task. As a control, six other subjects did not sleep after learning the task, but were also asked to perform it four hours after being trained. Those who slept did the task faster and more accurately than those who did not.

On day 5, the researchers scanned each volunteer with an magnetic resonance imaging machine, which maps brain anatomy, so that they could later see where the MEG oscillations they had observed were located in each subject’s brain.

In all, the experimenters tracked 5 different oscillation frequencies in eight brain regions (four distinct regions on each of the brain’s two sides). Sasaki said she expected the most significant activity to take place in the “M1” brain region, which governs motor control, but instead the significant changes occurred in the SMA on the opposite side of the trained hand.

What was especially important about the delta and fast-sigma oscillations was that they fit two key criteria with statistical significance: they changed substantially after subjects were trained in the task and the strength of that change correlated with the degree of the subject’s performance improvement on the task.

After performing the experiments, the team of Sasaki, Tamaki and co-author Takeo Watanabe moved from MGH to Brown, where they have set up a new sleep lab. They have since begun a project to further study how the brain consolidates learning. In this case they’re looking at visual learning tasks.

"Will we see similar effects?" Sasaki asked. "Would it be with similar frequency bands and a similar organization of neighboring brain areas?"

To find out, some volunteers will just have to sleep on it.