When traversing an area with which we are only partially acquainted, we frequently depend on notable landmarks to assist in our navigation. Nonetheless, if we seek an office in a brick edifice, and there are numerous brick structures along our path, we might adopt a strategy like looking for the second building on a street, instead of depending on differentiating the building itself.

Until that uncertainty is clarified, we must keep in mind that there are several possibilities (or conjectures) regarding our position in relation to our objective. In a study involving mice, MIT neuroscientists have recently uncovered that these conjectures are explicitly represented in the brain through distinct patterns of neural activity.

This marks the first instance that neural activity patterns that encode simultaneous conjectures have been observed in the brain. The researchers discovered that these representations, which were noted in the brain’s retrosplenial cortex (RSC), not only encode conjectures but could also be utilized by the animals to select the appropriate direction to proceed.

“As far as we are aware, no one has demonstrated in a complex reasoning task that there exists an area in the association cortex that maintains two conjectures in mind and then employs one of those conjectures, once it acquires more information, to actually complete the task,” states Mark Harnett, an associate professor of brain and cognitive sciences, a member of MIT’s McGovern Institute for Brain Research, and the senior author of the study.

Jakob Voigts PhD ’17, a previous postdoc in Harnett’s lab and currently a group leader at the Howard Hughes Medical Institute Janelia Research Campus, is the principal author of the paper, which is published today in Nature Neuroscience.

Ambiguous landmarks

The RSC gathers information from the visual cortex, the hippocampal formation, and the anterior thalamus, integrating them to assist in navigation.

In a 2020 paper, Harnett’s lab revealed that the RSC employs both visual and spatial data to encode landmarks used in navigation. In that research, the scientists illustrated that neurons within the RSC of mice amalgamate visual information about the surrounding environment with spatial feedback about the mice’s own position along a track, enabling them to learn where to locate a reward based on the landmarks they observed.

In their latest study, the researchers aimed to explore further how the RSC utilizes spatial data and situational context to aid navigational decision-making. To achieve this, the researchers developed a significantly more intricate navigational task than what is typically utilized in mouse studies. They established a large, circular arena with 16 small openings, or ports, along the side walls. One of these openings would provide the mice with a reward when they inserted their nose through it. In the initial set of experiments, the researchers trained the mice to approach different reward ports indicated by dots of light on the floor that were visible only when the mice approached them.

After the mice mastered this relatively straightforward task, the researchers introduced a second dot. The two dots were consistently equidistant from each other and from the center of the arena. However, the mice now had to navigate to the port by the counterclockwise dot to receive the reward. Since the dots were identical and only became visible from close distances, the mice could never view both dots simultaneously and could not immediately identify which dot was which.

To tackle this task, the mice had to recall where they anticipated a dot would appear, combining their own body position, the direction they were traveling, and the path taken to determine which landmark is which. By measuring RSC activity as the mice neared the ambiguous landmarks, the researchers could ascertain whether the RSC encodes hypotheses about spatial location. The task was meticulously designed to necessitate the mice’s reliance on visual landmarks for rewards, rather than employing other strategies such as odor cues or dead reckoning.

“What is crucial about the behavior in this scenario is that mice need to retain information and subsequently utilize it to interpret forthcoming input,” states Voigts, who contributed to this study while a postdoc in Harnett’s lab. “It’s not merely about remembering something, but recollecting it in a manner that enables action.”

The researchers discovered that as the mice gathered information about which dot corresponded to which, clusters of RSC neurons exhibited distinct activity patterns for incomplete information. Each of these patterns seems to relate to a hypothesis regarding where the mouse believed it was in relation to the reward.

As the mice approached enough to discern which dot signaled the reward port, these patterns converged into the one that signifies the correct hypothesis. The findings imply that these patterns not only passively store hypotheses but can also be utilized to compute how to reach the correct location, according to the research team.

“We demonstrate that the RSC possesses the necessary information for utilizing this short-term memory to differentiate between the ambiguous landmarks. Furthermore, we illustrate that this type of hypothesis is encoded and processed in a manner that allows the RSC to apply it in solving the computation,” Voigts states.

Interconnected neurons

In reviewing their initial findings, Harnett and Voigts sought insights from MIT Professor Ila Fiete, who conducted a study approximately a decade ago using an artificial neural network to perform a similar navigational task.

That study, previously published on bioRxiv, showed that the neural network exhibited activity patterns that were conceptually akin to those observed in the animal studies conducted by Harnett’s lab. The neurons within the artificial neural network ultimately formed highly interconnected low-dimensional networks, similar to those of the RSC.

“That interconnectivity appears, in ways that we still do not fully comprehend, to be pivotal in how these dynamics emerge and are controlled. It is a vital characteristic of how the RSC retains these two hypotheses in mind concurrently,” states Harnett.

In his lab at Janelia, Voigts now aims to explore how other brain regions involved in navigation, such as the prefrontal cortex, are activated as mice navigate and forage in a more natural setting, without undergoing training on a specific task.

“We are investigating whether there are overarching principles governing how tasks are learned,” Voigts notes. “We have substantial knowledge in neuroscience regarding how brains function once an animal has mastered a task, yet in contrast, we know alarmingly little about how mice learn tasks or what they opt to learn when permitted to behave naturally.”

The research received funding, in part, from the National Institutes of Health, a Simons Center for the Social Brain at MIT postdoctoral fellowship, the National Institute of General Medical Sciences, and the Center for Brains, Minds, and Machines at MIT, supported by the National Science Foundation.