How Do Our Brains Make Decisions? The International Brain Laboratory Is Closing In on Answers

Our lives are a constant string of decisions. Whether we’re choosing what to wear, what to eat for lunch or what route to take home from work, our brains are constantly engaged in decision-making processes that define our daily lives. Just how our brains make these countless choices has flummoxed neuroscientists for decades, leaving them with many more questions than answers. Uncovering how our brains make decisions could provide key insights into the choices that define us and how to intervene when impaired decision-making co-occurs with neuropsychiatric disorders. A collaboration of 19 laboratories called the International Brain Laboratory (IBL) thinks they might finally have the data neuroscientists need to uncover some answers.

Like most members of the collaboration, IBL founding member Alexandre Pouget has long dreamed of a future for neuroscience marked by large-scale collaboration and data collection. Pouget is one of many neuroscientists who study how neural activity in our brain reflects the decisions we make. Decision-making, like most topics in neuroscience, has typically been studied by individual labs, each investigating a small number of brain areas while animals perform different tasks.

“I found this really frustrating as a theoretician,” says Pouget, a professor in the University of Geneva’s Department of Basic Neurosciences. He believed that to understand the neural underpinnings of decision-making, experimentalists and theorists would have to work together to generate big datasets with standardized tasks and recording methods.

In 2017, Pouget joined several other like-minded neuroscientists to form the IBL, a collaboration that now consists of 80 neuroscientists and is co-funded by grants from the Simons Foundation, the Wellcome Trust and the National Institutes of Health.

“A group of us realized that we were in a position to collectively tap into problems that were too large for the scale of a single lab,” says Anne Churchland, a professor of neurobiology at the University of California, Los Angeles, and a founding member of the IBL. “If we worked together, we had a chance to generate a dataset that was unique and that no single lab could possibly produce.”

The IBL recently accomplished its ambitious primary goal: to record neural activity from the entire brain as mice perform a single decision-making task. Their success in producing this publicly available dataset, along with several other new resources, is already changing the way we think about the neuroscience of decision-making. With the power afforded by whole-brain recordings, the group has demonstrated that decision processes are reflected simultaneously across many parts of the brain rather than being relegated to separate structures, as was previously hypothesized. Decision processes aren’t the only thing that the IBL has shown should be global: In its success, the IBL demonstrates that international collaboration between scientists is critical to moving research forward.

Fittingly, the team began their study with a decision to make: Which decision-making task should they study? They settled on a behavioral task that was simple enough to replicate across labs while still being complex enough to prompt interesting questions about the neural basis of decision-making.

In the task, mice are placed in front of a display with access to a movable wheel similar to the wheel of a ship. During each trial, a wavelike pattern called a Gabor is shown on either the left or the right side of the display. The mice are trained to turn the wheel to move the Gabor toward the center of the screen (for example, by turning the wheel right when the Gabor is on the left). The team controlled the task’s difficulty by changing the Gabor’s contrast. When contrast was low, it was harder for the mice to see the stimulus, requiring them to sometimes guess which way to turn the wheel.

The first major challenge was training mice to perform the same task in several labs spread out over multiple continents. In fact, Pouget remembers that many of his colleagues thought it would be impossible. “The collaboration spent a lot of time making sure that we could replicate the behavior before we did any recordings,” he says. The team had to control every aspect of behavioral training, from the equipment that was being used to train the mice to what kind of food the mice were given each day. After hundreds of emails, messages and video calls, the team solidified a procedure that enabled them to successfully train mice across all the affiliated labs. With that done, they were ready to move on to their next major challenge: recording neural activity across the entire brains of the mice performing the task.

Postdocs and research scientists filled in the brainwide activity map as the mice performed the task by recording from various brain regions using electrodes called Neuropixels probes. Neuropixels probes terminate in shanks covered with a checkerboard of thousands of recording sites that can simultaneously record hundreds of neurons across several regions of the mouse brain. As with the task, the team tightly controlled the specifics of the neural recordings across all sessions and labs. In each animal, one probe targeted the same brain area every time, while others were directed elsewhere in the brain. The team chose the repeated location because it included several brain areas that were already known to contain signals related to decision-making behaviors, such as visual area A, the dentate gyrus, CA1, and the thalamic nuclei LP and PO.

“We wanted to have one population of neurons that was repeated every time so we could start to look at that in comparison to the other brainwide map sites,” says Noam Roth, a former IBL postdoc who contributed data to the brainwide map. “We ended up with this amazing dataset that had all these repeated site recordings where we could really look at what reproducible electrophysiology looks like.” Having access to repeated recordings in a few brain regions was essential to thoroughly evaluating the validity of the brainwide data being pooled across many labs.

Six years of work later, including during the COVID-19 pandemic, the IBL has assembled a nearly complete whole-brain map of neural activity in the mouse brain during their decision-making task. The resulting dataset is one of a kind, covering 279 brain areas (accounting for about 94 percent of the mouse brain’s volume) recorded across 139 mice in 12 labs using 699 Neuropixels probe insertions.

Few neural datasets rival the size of the one produced by the IBL, and those that do were collected by large research institutes based in the same location. “Standardizing within your own lab is challenging in and of itself, but doing that across about a dozen labs that are also geographically distributed is really an impressive feat,” says Saskia de Vries, a neurophysiologist and associate director of data and outreach at the Allen Institute who led a team that created another large-scale neural dataset recorded in mice.

The IBL team says its success largely stems from meticulous attention to detail, development of robust tools, and a commitment to clear documentation, which ensured that every lab could carry out the experiments identically. The team even sorted out challenges such as how to purchase compatible computers when labs in different countries use different power plugs.

The IBL has committed to sharing its tools with the larger neuroscience community, especially since many existing neural analysis tools don’t work for datasets this large.

“There are a lot of tools that the IBL developed in the process of collecting this data which we are now taking advantage of and using heavily within the Allen Institute,” says Josh Siegle, a senior scientist at the Allen Institute who has worked on similar large-scale neural datasets through the Allen Institute’s MindScope program. “For example, the Pinpoint application for planning where to insert electrodes is something that we use on a daily basis.”

Thanks to its careful planning and powerful tools, the IBL has nearly finished its brainwide map. The collaboration is now using the data to answer important scientific questions. One of the IBL’s first papers asks a question big enough to match the enormous scale of the data: Where in the brain are our prior expectations about the world represented?

When making decisions, we often consider what we already know. For example, when we’re buying something, a one-star review for a company we already like might be less of a deterrent than a one-star review for a company we know nothing about. This known information about the probability of something occurring is called the “prior,” and neuroscientists have long wanted to know where in the brain it is represented. Some, like Pouget, believed that it must be represented in higher-level brain areas, which are thought to carry out sophisticated cognitive functions, but not in sensory areas. “It would be crazy to inject a prior in the sensory data, because then you start to believe your own expectations,” he says. However, others argued that for information about the prior to be used throughout the brain, it would need to be represented in many brain areas.

The best way to resolve this controversy would be to look at neural activity across the entire brain in the context of a single task, and the IBL’s brainwide map is the perfect tool to enable this. The team leveraged both the brainwide map and another dataset containing neural activity recorded across the cortex while mice performed the same task. They took advantage of an aspect of the IBL’s task that Pouget calls the “prior twist.” As mice performed the task, the most likely location of the stimulus switched between the left side of the screen and the right. This meant that the mice should develop a prior, or expectation, that the stimulus would appear on a particular side. As the experimenters varied the visibility of the stimulus, the mice could use this prior to improve their performance on trials where it was difficult or impossible to see the stimulus by guessing based on which side it usually appeared on.

When the results came in, Pouget realized that his guess as to where the prior resides in the brain may not capture the full story. “We found evidence that the prior percolates everywhere in the brain,” says Pouget. “It’s most likely computed in the prefrontal cortex, and then it’s communicated to the rest of the brain, all the way down to the LGN” (a structure early in the visual pathway). Faced with the mounting evidence that priors are represented all over the brain, Pouget says that he’s changed his thinking about the neural basis of decision-making, and that his lab is already working on a series of new projects that follow up on these results.

In fact, the IBL’s dataset suggests that many other computations relevant to decision-making may also be represented diffusely across the brain. They’ve found that important signals like motor action and reward outcome are strongly and globally represented. Even visual representations, which start in traditional visual areas, percolate to other brain regions that represent other relevant variables like choice. These insights challenge the traditional view that distinct brain regions are involved in separate parts of the decision-making process and suggest that a whole-brain view may be necessary to unlock how the brain makes decisions.

The IBL is already eager to apply its approach to even more scientific questions. “We built something great, and we developed the ability to make large-scale measurements that no one else can make,” Churchland says. “I think what we need to do is to take advantage of that ability to pivot and apply it to a new problem.” Scientists might use the data to study addiction, how different individuals approach the same decision, or how nodes of a neural network interact. There is widespread excitement about the future within the IBL and agreement that the next steps will require a balance between large-scale and small-scale approaches.

The IBL isn’t just a success for its data. The collaboration is also a first-of-its-kind blueprint for large-scale international neuroscience collaboration. The group’s governing practices stem from the ideals of sociocracy, a commitment to giving all members a voice in the collaboration’s ideals and practices. “When we make sure that everyone has a voice, it can really change the tenor and scientific direction a lot,” says Churchland. “I think the vision of having a group where everyone can speak is a beautiful vision, and we made steps toward that, but we need more work to really put that into place.” The IBL’s members still consider their governing practices a work in progress, but they hope they can act as a starting point for groups interested in setting up similar collaborations.

How does such a big group give a voice to all its members? Important decisions about how the IBL operates, like those about authorship on papers, are first considered by a smaller group of stakeholders composed of IBL members. This group talks to other members and works together to come up with solutions. Once they have a proposal, the general assembly (composed of all the principal investigators, or PIs, plus an elected trainee representative and an elected staff representative) can vote “yes,” “yes with concerns” or “objection.” Any objections will prevent a policy from being enacted. Every decision can’t include everyone, so the IBL is organized into smaller subgroups, called working groups, that take the lead on different projects. Working groups are led by at least one PI along with a postdoc, who assists in guiding the scientific direction of the group. For postdocs, this is an opportunity to hone their leadership skills, says Roth. “It’s taught me a lot about how to work in bigger groups in science and how to write a paper with 10 people, which is different than the traditional Ph.D. or postdoc experience where you write a paper with your PI,” she says.

Protecting the professional interests of trainees has been an especially difficult challenge for the IBL. During a traditional postdoctoral position, the postdoc will publish a handful of papers as the first author — something required for earning a faculty position. Because of the collaborative and large-scale nature of the IBL’s main projects, a lot of the postdocs’ time is spent on projects that won’t lead to first-author publications, and finding the balance between contributing to bigger projects and working on personal projects was difficult for some postdocs.

“We’ve tried really hard to take credit assignment seriously and write contribution statements that are explicit and really show what people have done,” says Hannah Bayer, executive director of the IBL. “But I think we’re still in a world where people are counting first-author papers. Even though I think people learn leadership skills and teamwork skills that will make them better PIs later on, there’s not yet total recognition of that.”

Overall, the IBL has accomplished more than many people thought possible a decade ago. The data, released freely to the scientific community, is sparking new research papers and enabling new lines of scientific inquiry. The IBL’s approach demonstrates that the neuroscience of decision-making should be considered a global endeavor, whether in terms of brain activity or scientific collaboration. For the members of the IBL, it has also been a unique opportunity to connect with other researchers in the field. “I learned so much from the other PIs — scientific things and things about how to run my lab better. Getting to know the students and postdocs was totally awesome,” says Churchland, echoing a sentiment shared across the IBL. “That has really been a great joy.” Wherever their research goes next, the IBL has laid the foundation for Pouget’s dream of a more collaborative future for neuroscience.