How does the brain abstract knowledge from the world? Neuroscientists have been studying model organisms to explore this question and thanks to advances in genetic tools in mice and rats, the circuit level mechanisms of how the brain drives behaviour are starting to be decoded.

But modelling cognitive mechanisms in rodents is difficult. We know there are many things that humans can do that mice cannot. And so we asked SWC Seminar speakers about the key challenges. Here are their nine insights.

1. Studying language and higher-order thinking

“Language is the most obvious one. We love our rats and think they’re smart, but they are clearly not people! I think rats are great for modelling basic learning and memory processes, basic decision-making processes and they have rich social lives as they live together in groups. They are great for auditory and somatosensory work too. But if we want to study language, higher-order thinking, mathematical thinking, we won’t able to get close to those in rats.

You can get hints of understanding language from songbirds and other animals, but it is very much a human thing in its most extreme version. I’m grateful that I have a colleague, Eddie Chang, who is one of the world leaders on studying language in humans and it is amazing what they have figured out in terms of its organisation. This is something that really needs to be studied in people.” Loren Frank, UCSF

“We are getting much better at training animals and looking into animals’ brains. But sometimes we shouldn’t try to get an animal to do the exact same thing that you would ask a human to do and instead come up with a different behavioural task that an animal might do better.

I haven’t seen too many strong limitations to the kind of work one can do with laboratory animals, maybe with the exception of language because language is not something that animals have. But it also depends on what aspect of language you want to study. If you want to study very complex syntax of vocalisations, there are plenty of animals that have complex vocalisations. If you want to study how language can be used to express memories, then you could study an animal that expresses memory some other way rather than using language.

I’m very optimistic about the things that we can do with animals in the lab and I think we are only starting to scratch the surface.” Dmitriy Aronov, Columbia University

“There are always debates about what is special about humans. Historically, people have made claims about how certain functions are unique to humans – language, for example. But depending on how you define language, a number of other animals use similar communication means that check most of the boxes of how language is defined. The same is true for tool use. Certain animals can be trained to use tools.

So what is really unique about humans? That’s debatable. Are humans simply a more complex version of “lower” animals or are there qualitative differences in terms of brain functions? This research is ongoing.

Recently, there was a number of papers that came out from the BRAIN Initiative Cell Census Network. This is a large international team project funded largely by the National Institute of Health, trying to understand the organisation of brains at the single cell level, focusing on identifying cell types. Different neurons can be grouped into specific cell types and this project tries to catalogue every single cell type in a number of different animal models. The most recent set of papers focused on primates, a lot of them on humans.

An approach like this is beginning to uncover a lot of similarities between the human and rodent brain, but also the uniqueness of each. There are some functional properties of common-looking cell types that might be unique to humans. At the mechanistic level, our understanding is becoming more fine-grained to understand what might be special about the human brain. To me, that is very exciting.” Takaki Komiyama, University of California San Diego

2. Understanding social cognition and learning

“The social observation of learning is difficult in rodents as it is hard to know what they are tracking about their conspecifics. For example, in primates we can track gaze movements of the eyes but in rodents it is more difficult to know how they implement this. The social level is really hard to model and get a mechanistic handle over.” Ahmed El Hady, Max Planck Institute of Animal Behaviour

“While rodents are useful for studying many aspects of cognition, some higher-order cognitive abilities like complex problem-solving and abstract thinking might be harder to replicate across species. Researching social cognition – understanding others’ thoughts and making social decisions – may also be challenging due to their relative simplicity compared to species with more advanced social behaviours.” Kanaka Rajan, Harvard University

3. Modelling psychiatric symptoms

“I think more about psychiatric symptoms as opposed to cognitive mechanisms. There are certain symptoms that we see in the clinic that I think are harder to model, although people have tried. Psychotic symptoms are a clear example of something that we usually know only because people tell us what they are hearing or experiencing. We can’t ask the rodents to tell us in the same way. So it’s harder to be sure that what is happening is related to psychosis. So, I think it’s a tough nut to crack, unfortunately. But recent work by Katharina Schmack and others has got me excited that there might be a way forward.

Overall, I’m becoming more and more optimistic about types of behaviours or internal states that we can manipulate and understand in rodents. Something as complicated as empathy, or other types of social behaviours, are amenable to investigation even in rodents. We may never be certain about our interpretation, but that’s true regardless of what we are studying. I think we’re getting better and better at it, so I’m optimistic.” Neir Eshel, Stanford University

“You have to think about what the model is for. In my opinion, it is not a very productive way forward to try to mimic things in a non-human that you might be interested in understanding in a human. But what you can do is develop hypotheses and ideas about how things might work in the human brain and use a non-human neuronal system as a model to test those ideas rather than mimic and passively observe.

For example, if you read the cognitive behavioural therapy (CBT) literature, there is an idea that when you undergo CBT your brain learns from that training how to deal with future circumstances that don’t need to resemble what the training was. You have this generalised learning to learn ability and that can be therapeutic as synaptic plasticity happens and the brain changes in some way that makes you more capable of engaging with the world.

CBT is based on this notion, but it is very difficult to test that idea in the human brain, as you would have to measure synaptic plasticity and neural responses and so forth. So instead, we took mice and we trained them on what we thought was really important about CBT. One of the features of CBT is that you are trained to ignore misleading information. And so we looked at what happened in the entorhinal-hippocampal circuit when the mouse was trained to learn something in the presence of misleading information.

We had explicit predictions including a generalised ability to learn future information and synaptic plasticity changes (not everywhere in the brain and not related to encoding information that was trained itself) related to the efficiency of the neural network for processing information. We could measure those things and we published a paper in Nature at the end of 2021 (Chung et al.) that described that if you train animals to learn something in the context of distraction, then you could describe these changes and we could identify the parts of the neural circuits that caused these changes.

This is an example of the utility of an animal model as it allows us to test hypotheses about things that are important that we are unable to test in humans. However, trying to model schizophrenia or depression in animals seems to me foolish because we don’t even know what schizophrenia or depression is. Imagine going to a sculptor and asking them to model a made-up word, you couldn’t expect a useful outcome from that effort.” André Fenton, New York University

4. Interpreting complex behaviours

“I think the simpler the brain, the less complex behaviour that the animal displays. That is an advantage and also a problem. The advantage is that we have more tools and studying these simpler brains is easier. In mice, we have access to an array of genetic and molecular tools and the overall cost of running experiments is lower. But on the other hand, the behaviours are also simpler and less intuitive to study compared to species with more complex and cognitive behaviours.

Having said that, I do still believe that mice are valuable in studying complex behaviours. We just have to be careful and make sure that we have controls, interpret the behaviour in a way that’s rigorous, and have the right interpretation of the behaviour. The animal may perform the behaviour, but it may not be due to the reason we think. In fact, it can often be even more challenging to interpret behaviours in non-rodent species.” Weizhe Hong, University of California, Los Angeles (UCLA)

5. Examining episodic memory

“It depends on how much you reduce the cognitive operation. We can study a lot of the higher order cognitive behaviours – we can see when rodents are attending to something, for example, or when they’ve learned or remembered something. But it’s always a reduced version of those things. Learning and memory are the most interesting to me. They create a sense of self and consciousness. I think it will be very hard to examine episodic memory at the same resolution as that of humans.” Andrew Alexander, University of California Santa Barbara

6. Learning multiple things simultaneously

“Certainly, learning multiple things at the same time is harder for rodents. As primates, we are able to learn multiple motor tasks very easily without interference. Such a task would be stressful for rodents to learn! Rats are really good at spatial navigation and burrowing, among other ecologically relevant tasks. But when it comes to reaching for and grasping and manipulating an object, they definitely have a limit. It is not clear to me that rodents engage in visually guided reaching.

Primates have a visual basis for everything, but it does not seem that rats use vision to guide reaching and grasping. They seem to rely on olfaction for detecting objects. However, they do get quite skilled at picking up and grasping objects with practice. However, if we change the object size or location, they need time to adjust. This is very different from primates.

It is also very likely that rats have limits to cognitive flexibility. It is not that hard to make tasks that become really challenging for them to learn. We have this rule that if it takes more than 5 days for a rat to learn the task, it's probably too hard for them.” Karunesh Ganguly, University of California, San Francisco (UCSF) and the San Francisco VA Medical Center

7. Making sure the model organism is learning the cognitive process you want to study!

“Animals often have multiple ways to solve the task problem to get rewards, so you really need to train them in the correct way in order for them to use the strategy that you want them to learn. This takes a lot of experience as well as careful observation and control to make sure the animal is not learning something you do not want them to learn. Otherwise, the actual algorithm being used can be different from the cognitive process that you wanted to study!” Ning-long Xu, Chinese Academy of Sciences

8. Being aware of each species’ ethology

“I think the important thing is to be aware of each species’ ethology. For example, the primate brain has much expanded and elaborated lateral prefrontal and parietal regions, which may reflect their importance for complex visuomotor behaviour, like paying attention to things in the distance, then planning visually-guided actions. It could very well be the case that we cannot get at those kinds of processes using rodents.

So there are clearly going to be limitations, but what is interesting is how many building blocks might be common across species – how animals plan out strategies, decide whether to take actions and how long to persist with them – are going to have a lot of similarities across species.” Mark Walton, University of Oxford

“For a function such as vision, the sensory processing hierarchy of rodents is of course very different from that of primates, for instance. But the sort of decision making that I work on could actually be well understood in rodents, although there are certainly aspects of flexibility that might be missing. The wider story seems to be that the smarter we get in designing tasks for rodents, the smarter the rodents seem to be in terms of what they can pick up.

The work of people at SWC, like Tim Behrens, shows that beautifully. What would be considered a very complicated task for a monkey is actually something that rodents can learn well because we understand how to present the task to the rodents in a way that they can actually do it. Of course, there remains very important aspects of species specificity. However, for decision-making, we are in the fortunate position that animals have to be good at gaining rewards and avoiding punishments, because if they weren’t, then they wouldn’t be around for us to study.” Peter Dayan, Max Planck Institute for Biological Cybernetics

“There are a lot of good and important questions you can ask and answer in a mouse model, but there a lot of things that humans can do that mice cannot. At the same time, there are also questions you can ask in mice that are very difficult to ask in humans. In mice, you can perform genetic manipulations and use techniques like optogenetics or chemogenetics. You can also perform recordings from many individuals simultaneously. Some things are model dependent and one of the things that we try to do in our lab is to try to ask a broad array of questions using these different models.” Ziv Williams, Harvard University

“Using the rodent as a model for cognitive tasks makes a lot of sense as long as it is taken into account what behaviours a rodent would naturally do. One could argue that category-learning is a very complex task and needs at least primates to study. One could even say that category learning is related to semantic memory, and a true representation of a category requires the subject to have language. But, rodents have to categorise as well in their life, the basic principle of grouping items by perceptual features or by their function is relevant for every organism. In addition, aspects related to flexible categorization, such as the ability for switching rules and generalizing to novel situations, will be very useful in the natural environment that a rodent would finds itself in.

I think that rodents will use, in many cases, similar types of cognitive behaviours to solve problems that are similar to what we are used to in our everyday life. Still, there are limits to this, and we should always be aware that different species may solve problems in specific ways. In neuroscience, research focuses on rodents, zebrafish and fruit flies, as well as on monkeys and humans. However, there are so many other species between a rodent and a primate, which could still give the methodological advantages of rodents (relatively small brains and ability to breed or handle them easily) yet have somewhat larger brains, better cognitive abilities, and are possibly less impulsive – e.g. wait for delays tasks in a much more reliable way.

So, while I think that rodents can be used for studying several basic aspects of cognition, I also think that it’s very important to focus not only on rodents, or any specific organism, in systems neuroscience. Of course, tools are an important reason to use rodents, but there are also other species for which such tools exist or could be developed for, and which potentially allow for more elaborate behaviours to be studied.” Pieter Goltstein, Max Planck Institute for Biological Intelligence (Munich)

9. Modelling being in love!

“Luckily, I think that sleep is quite amenable to modelling. Because all animals sleep, I think there are general evolutionarily conserved mechanisms. In a way, sleep can be studied in almost any organism because there must be some general principles for why the brain enters this state. But of course, there are a lot of behaviours that cannot be modelled in a mouse – being in love, for instance.” Julia Harris, Francis Crick Institute and Incoming Group Leader at SWC