On a recent afternoon at Johns Hopkins University, neuroscientist Jeremiah Cohen watched a thirsty mouse consider her options. To maximize her chances of a much needed drink, she had to discover by trial and error which of the two water spouts in front of her more reliably paid out. To make things even trickier, the odds were subject to change without warning at any time. Cohen, with graduate students Cooper Grossman and Bilal Bari, had designed this simple yet vexing challenge to explore a complex problem: how the brain responds to an unpredictably varying situation. The brain is constantly making predictions. We make decisions by predicting the outcome of different options. We learn by comparing the actual outcome, once it arrives, with the outcome we had predicted. When outcomes differ from our expectations, we revise our ideas about the world, so that we can make more accurate predictions in the future. But how quickly do we make these revisions? How many times does an unexpected event need to occur before we come to expect it? “Learning should be adapted to complement the environment,” Cohen says. An ideal learning strategy not only accounts for expected outcomes, but also for the likelihood that expectations will be met. In the case of a deadly viral pandemic, for example, if transmission rates have been steady over time, a week of unexpectedly low infection rates might justify loosening safety protocols. If infection rates have been consistently variable, however, it would be a mistake to infer that the situation has changed every time case rates unexpectedly rise or dip.
In a paper recently published in the journal Current Biology¹, Cohen and his students demonstrate that the brain accounts for this sort of predictable unpredictability, and how the chemical serotonin is involved. Serotonin is a neuromodulator, a class of chemical messengers that don’t directly cause neurons to fire, but serve to modify their activity. Neuromodulators are known to be important in the brain’s decision related predictions and calculations, and previous evidence had suggested that serotonin plays a role in learning under uncertainty. To determine what that role is, Cohen’s team needed to actually get inside a brain and measure what serotonin neurons were doing in real time - which is where the thirsty mice came in.
Recording neural activity in the brain of a mouse is relatively easy. The challenge is training the mouse to perform an interesting behavior - like learning in uncertain conditions - at the same time. “Behavioral task design tends to be iterative,” Cohen explains. “Version 1.0 is getting it to work. We then refine it based on the questions at hand- for this study, it was one- or two- year effort.” Happily, that effort paid off. By recording the activity of serotonin releasing neurons as mice performed the water spout task, the scientists discovered that these neurons keep track of expected predictability. Serotonin neurons fire in specific patterns that reflect how confident mice are in their predictions as they wait between one decision and the next, and send a quick signal of the level of certainty when the time comes to actually make a choice. Once the outcome occurs, serotonin neurons signal if it violated a strong expectation.
Preventing these neurons from firing revealed an active role in guiding learning. When serotonin neurons are silent, mice can still figure out when the odds at a spout have changed and update their decisions accordingly. However, their behavior no longer accounts for predicted unpredictability. Instead, they learn from unexpected events equally quickly in uncertain situations as in consistent ones. Could serotonin neurons be responsible for longer term changes in behavior- for example, how quickly we decide to update our travel plans in response to new Covid variants? It’s hard to say. Cohen’s mice, after all, are responding to second by second changes , not adapting over the course of days and weeks. Scientists are hoping to explore how this process may play out on longer time scales, but that is a more challenging problem to tackle. For now, they’re still uncertain.
1.Grossman CD, Bari BA, Cohen JY. Serotonin neurons modulate learning rate through uncertainty. Curr Biol. 2021 Dec 17:S0960-9822(21)01682-1. doi: 10.1016/j.cub.2021.12.006.