In 2016, three neuroscientists wrote a commentary article arguing that, to truly understand the brain, neuroscience needed to change. From that paper, the International Brain Laboratory (IBL) was born. The IBL, now a collaboration between 22 labs across the world, is unique in biology.

The IBL is modeled on physics collaborations, like the ATLAS experiment at CERN, where thousands of scientists work together on a common problem, sharing data and resources during the process. This was in response to the main criticism that the paper's authors, Zachary Mainen, Michael Häusser and Alexandre Pouget, had about existing neuroscience collaborations: labs came together to discuss generalities, but all the experiments were done separately. They wanted to create a collaboration in which scientists worked together throughout the process, even though their labs may be distributed all over the globe. 

The IBL decided to focus on one brain function only: decision-making. Decision-making engages the whole brain, since it requires using both input from the senses and information about previous experiences. If someone is thinking about bringing a sweater when they go out, they will use their senses to determine whether it looks and feels cold outside, but they might also remember that, yesterday, they were cold without a sweater.

For its first published (in pre-print form) experiment, seven labs of the 22 collaborating in the IBL tested 101 mice on their decision-making ability. The mice saw a black and white grating either to their right or to their left. They then had to twist a little Lego wheel to move the grating to the middle. By rewarding them with sugary water whenever they did the task correctly, the mice gradually learned. It is easy for them to decide which way to twist the wheel if the grating has a high contrast, because it stands out compared to the background of their visual field. However, the mice were also presented with a more ambiguously-patterned grating not easily distinguishable from the background, so the decision of which way to turn the wheel was more difficult. In some cases, the grating was even indistinguishable from the background. Between all seven labs –which were spread across three countries – the mice completed this task three million times.

Now, using the same test, different labs that are part of the IBL will study how different parts of the brain react. The ultimate goal of the collaboration is to combine the data that the participating labs collect from different brain regions to determine exactly what is going on in the mice's brain during this decision-making task. However, to be able to do this, they need to make sure that the mice have the same behavior during the test regardless of the lab they are in. And mice are notoriously unreliable, behaving differently depending on the sex of the experimenter, how inbred the mice are, or the laboratory environment.

The IBL aimed to make the training of the mice as comparable between labs as they could. They produced schematics for the setup to be used, which meant that everybody used the exact same setup. They also created guidelines on how the animals had to be trained. At first the animals were always given an easy task, where the stimulus (the grating) was clearly visible, and were generously rewarded. As they got better at the task, the stimulus became more ambiguous, meaning it was harder for the mice to distinguish the grating from the background. In addition they got less of a reward each time they did the task, to encourage them to perform the task more often within a single session. Although individual mice learned at different rates, some taking as long as 59 days, all of the participating labs successfully trained their mice to perform the task in the end. 

To make sure that the mice were indeed showing the same behavior in different labs, the researchers trained a classifier, a computer algorithm that determines what category (in this case, the laboratory) an object belongs to. The intent was for the classifier to determine which laboratory each mouse came from based on their behavioral traits, such as how well they performed the task, and how much their performance changed as the grating became more ambiguous. But the classifier was unable to distinguish between the labs based just on these traits, which suggests that the mice have the same behavior in the task regardless of what lab they are trained in – a good sign for the scientists, who wanted to ensure the reproducibility of these results.

The simple grating task allowed the researchers to look at basic decision-making, but they also wanted to test whether the mice could do a task that relied on their experience. They tested this by including a bias condition in the task. Without the bias, the grating had the same chance of appearing to the left or right of the mouse, so if the mouse randomly guessed, it had a 50% chance of getting it right. With the bias, the stimulus appeared on one side more often, so now if the mouse had to guess, it had a much higher chance of receiving a reward if it chose the biased side.  Each side was biased in blocks, which switched during the task, so the mouse had draw on its previous experiences to determine during the task what the biased side was. It turned out that in all laboratories, mice used the same tactic to solve this problem: they repeated the same choice. So if they chose left on the last trial, they would probably choose left on the next. 

By training the mice to reliably and reproducibly perform a decision-making task, the IBL has now laid the foundation for their future experiments, which will involve recording the brain activity while the mouse performs the task. Each lab will focus on their own specific brain region, but the IBL is set up so that they can easily share their data and combine their expertise. 

In the midst of the reproducibility crisis, this work shows that it is possible to reproduce behavioral results with the right approach. Of course, this is a unique case, where all labs have access to the exact setup blueprints, enough money to build it, and code to run the training program for the mice – but it shows the potential impact of effective scientific collaborations within neuroscience research.