Life is full of learning and then re-learning as it gets more detailed. One day you log into an app with just a password, then the next day you also need to text you a code. One day you microwave your favorite lunch in the oven for six minutes straight, but then the packaging changes and you have to cook it for three minutes, stir, and then heat it up for another three minutes. Our brains need a way to keep up.
A new study led by neuroscientists at MIT’s Beckware Institute for Learning and Memory has revealed some of the circuits that help the mammalian brain learn to add steps.
in Nature Connections The scientists reported that when they changed the rules of a task, requiring mice to adapt from performing just one step to performing two steps, a pair of regions on the surface of the brain, or cortex, collaborated to update this understanding and change the mice’s behavior to fit the new regime.
The anterior cingulate cortex (ACC) appears to recognize when mice were not doing enough and update cells in the motor cortex (M2) to fine-tune task behaviour.
“I started this project about 7 or 8 years ago when I wanted to study the decision-making process,” said Daigo Takeuchi, a researcher at the University of Tokyo who led the work as a postdoc in the RIKEN-MIT Laboratory of Neurogenetics at The Picower Institute that he directs. Senior author and Professor Picower Susumu Tonegawa.
New studies have found a role for M2. I wanted to study the effect of upstream circuits on this.”
The second step kicks in
Takeuchi and Tonegawa traced the neural circuit connections that led to M2 and found that many of them originated in the ACC. They began to see the role of the ACC in directing M2 sequencing decisions when they infused genetic manipulations into ACC cells that allowed them to suppress their activity. This ‘geochemical’ inactivation of the ACC had a very specific effect.
When the rules of the task changed so that instead of having to stick their nose into only one nostril to get a small reward, the rats had to poke their nose into a series of two nostrils, the rodents with silent ACCs took longer to perceive the rule change.
Compared with mice with normal ACC activity, they failed for a longer time to recognize that a second poke was necessary. However, the mice had no problem going from two backward steps to just one step, regardless of whether their ACC was silenced.
When the scientists chemically silenced the tips of the ACC cells in M2, they obtained the same results as silencing the ACC in general. They also silenced other areas of the cortex, but doing so did not affect the mice’s ability to notice and adapt to a base switch.
Together, these manipulations confirmed that it is the connections of the ACC with M2 that help mice notice and adapt to the one-step change.
But what is the effect of ACC in M2? Takeuchi and colleagues measured the electrical activity of cells in M2 while mice were playing a rule-changing game. They found that many cells were specifically activated by different task rules (that is, from one or two steps). When they silenced the ACC, though, they suppressed this selective rule.
Within M2 Takeuchi and the team also observed groups of neurons that responded preferentially to positive outcomes (rewarding for doing the task correctly) and negative outcomes (not getting rewarded for doing the task wrongly).
They found that when they silenced the ACC, this actually increased the activity of neurons encoding negative outcomes during negative feedback, especially in the first 10 to 20 rounds after changing the rules from one step to two steps. This was strongly correlated with the timing, or ‘epoch’, of the rats’ worst performance.
“It appears likely that the age-specific disturbance of the animals’ second-choice performance is related to the over-enhancement of negatively-activated neurons due to ACC silencing,” they wrote in the study.
The team further confirmed that the reactions, or outcomes, are important using a different technique for silencing the ACC. By engineering the ACC neurons so that they are suppressed by flashes of light (a technique called “optogenetics”) they can precisely control when the ACC is disconnected.
They found that if they did so after the mice made an incorrect choice when the rules switched from one poke to two, they could cause the mice to continue to make mistakes. The optogenetic silencing of the ACC after the mice had made the correct choice did not undermine their subsequent behaviour.
“These results indicate that ACC neurons process the error feedback information after a second false response and use this information to fine-tune the animal’s sequential selection responses in subsequent experiments,” they wrote.
The evidence painted a clear picture: When the mice needed to notice that an additional step was now needed, the ACC’s job was to learn from negative feedback and signal M2 to take the second step. If the ACC is not available when feedback is provided, it is clear that M2 cells stressing negative results will become particularly active and mice will fail to perform the required second step for a while before finally catching up.
Why might decreased ACC activity somehow increase the activity of M2-negative coding cells? Takeuchi posits that what the ACC is actually doing is stimulating the inhibitory cells in M2 that normally modulate the activity of those cells. With reduced ACC activity, M2 cells encoding negative results experience less inhibition.
The behavioral consequence, according to his theories, is that mice therefore require more evidence than they should for a rule change. Takeuchi acknowledged that the mechanism is not entirely clear, but it appears that the mice need more time to experience feedback on the results than making the right decision to take the second step before they become convinced they are on the right track to do so.
Takeuchi said that while the results show the cycle necessary to adapt to a base change that requires more steps in the process, it also raises some interesting new questions. Is there another circuit to note when a multi-step process becomes a one-step process? If so, is this circuit integrated with the circuit discussed in this study? And if the threshold model is the correct model, how exactly does it work?
The implications are not only for understanding the neural basis of natural sequential decisions, but may also relate to AI applications ranging from playing games or industrial work, each of which can involve tasks with multiple steps.
About this research in Neuroscience News
author: press office
Contact: Press Office – MI
picture: Image credited to Tonegawa Lab / MIT Picower Institute
original search: open access.
“Belt motor circuits update rule representations for sequential choice decisions” by Daigo Takeuchi et al. Nature Connections
Belt drive circuits update base representations for sequential choice decisions
The anterior cingulate cortex mediates the elastic update of animal selection responses when rules change in the environment. However, how the anterior cingulate cortex enters the motor cortex to reorganize representations of rules and generate the required motor outputs remains unclear.
Here, we demonstrate that chemical silencing of terminal projections of cingulate cortical neurons in the secondary motor cortex in rats disrupts selection performance in experiments immediately following rule switches, suggesting that these inputs are necessary to update base representations of choice decisions stored in the motor cortex. Indeed, silencing of the cingulate cortex reduces the base selectivity of secondary motor cortical neurons.
Furthermore, optogenetic silencing of target cortical cingulate neurons temporarily for error trials immediately after switching rules exacerbates errors in subsequent trials.
These results indicate that the cingulate cortex monitors behavioral errors and updates rule representations in the motor cortex, revealing a critical role for cingulate circuits in adaptive choice behaviors.