Meet our expert: Michael Bukwich

Can you give us a brief overview of your academic journey and what inspired you to pursue this field?

I took a bit of a circuitous route to get here. To put it generously, I was a rebellious teenager. I skipped school most days and dropped out once I turned sixteen. I worked as a roofer for the next few years. It was dirty, back-breaking work. One morning I noticed one of the older guys moaning in pain on the way up the ladder and I realised that this was how he felt every day, aching before we even began doing the day’s heavy lifting. That was my cue to start thinking about long terms plans and find a less brutal career. I went home and searched “how to go to college without a high school diploma” and found a route via community college. Over the next three years I spent my days on the roof and nights in the classroom, sometimes sitting in class covered in dirt and fiberglass when overtime demanded that I wouldn’t have time to shower beforehand. During that time, I became fascinated by the questions about consciousness I encountered in my intro philosophy class, though I was discouraged to notice that philosophers are essentially limited to contemplation from the scope of their armchairs. I wanted to think but also to do, and neuroscience revealed itself as that path. I quit roofing, transferred to NYU to study neuroscience, and never looked back. After that, I did a quick 8-year PhD at Harvard and eventually landed my ideal postdoc position with Neil Burgess and Caswell Barry here at UCL. Next, I intend to become a PI, and I will pursue that aim relentlessly.

What are the current research projects you are working on?

Our team is interested in how neural circuits represent maps of the world, and we’ve been developing some unique navigation experiments in virtual reality to probe this. Because the configuration of a virtual space is not constrained by the laws of physics the same way that actual space is, we can construct virtual mazes in which the geometry manifests like something out of an M.C. Escher piece. By decoupling the physical structure of space from the latent structure of experience, we can tease apart more precisely what is being mapped in neural activity. This level of control will help us uncover fundamental principles of how brains form “cognitive maps”.

Insofar as it doesn’t interfere with the scientific merit, I can build these virtual worlds however I want. So, I have fun with it like a delightful little art project. My mice are essentially playing video games, so why not build one of their worlds out of Super Mario pixel art? Or hey I’ve had some ideas inspired by 3Blue1Brown, so then I build another as an homage to the YouTube channel. Stuff like that. My work will be many things, but it will never be boring.

Rodent virtual reality seems like a fascinating tool. Could you paint a picture of how it’s used and its impact on your research findings?

We train mice to run on top of a foam ball surrounded by computer monitors displaying a virtual environment, plus ceiling projectors displaying visuals along the floors. The foam ball is effectively free floating via compressed air blowing up from underneath it, so they can run on it with minimal friction. As mice run, we measure the spin of the ball and convert that into a ‘velocity’ input to simulate movement through the virtual worlds.

Our VR rigs are unique from those more commonly used in neuroscience labs because Neil designed them to create a more realistic immersion. In traditional mouse VR setups, mice are harnessed on top of the ball in a way that keeps them facing a particular ‘forward’ direction. Our rigs incorporate a ball bearing, which allows the mice to turn and run in any direction they choose. This makes a critical difference for simulating realistic navigation behaviour. When recording brain activity in our VR, we observe grid cells1 — neurons that display remarkable hexagonal symmetry in their firing patterns2 — whereas traditional mouse VR setups fail to elicit these 2D grids. Given that these grid patterns are a hallmark of spatial navigation, it is compelling evidence that our mice are treating the virtual environments as real spaces.

Could you share any breakthroughs or interesting findings that have come out of your lab recently?

My labmate, Eleanor Spens, recently published impressive theoretical work3 on the neural basis of imagination as the interplay between two brain areas — the hippocampus and neocortex. She models the hippocampus as a modern Hopfield network, which stores episodic memories from one-time experiences (as we do when forming memories of events). After this initial learning, the network continually reactivates that learned representation, simulating a pattern of brain activity known as ’replay’. Modelling the neocortex, a second neural network known as a variational autoencoder, gradually learns from observing those hippocampal replays and forms a ‘schema’, which is essentially a map of the abstract relations between components of event memories. A key feature of the autoencoder is that it can translate between a holistic representation and its individual components, for example being able to switch perspectives back and forth between a forest and its trees. After that mapping is learned, the combined network can generate (imagine) representations of never-before-seen forests.

Can you speak about a challenge you’ve faced during your research journey and how you overcame it?

I found a critical mistake in a computational model I had built just as we were getting ready to submit a paper. It was gut-wrenching. Fixing the mistake negated a spurious result to the point that we would need to change the title. I was struggling to hold back tears as I emailed my team to let them know how badly I had messed up.

How I got through it will sound cliché, but it helped. As best as I could, I tried to follow the advice of talking to myself the way I would if I was speaking to a friend who had made that mistake. So instead of berating myself for carelessness, as would normally be my inclination, I tried to remind myself that it was a good thing I found it then, rather than after publication, and that fixing the mistake would bring us closer to ground truth. It didn’t make it easy, but it made it more bearable. In the end, we still wrote a respectable paper, albeit a different one than I previously envisioned.

In what ways do you see your research translating into real-world applications or therapies?

To be honest, I do not know, and I am fully comfortable with that uncertainty. The brain circuits our current study is designed to focus on are known to degrade in Alzheimer’s disease and, more recently, have been implicated in schizophrenia. Perhaps our work will provide some useful insights that could eventually help to diagnose or treat one of those ailments? But that’s speculation. With research, and in particular brain research, it can be tempting to infer a project’s value based on what we imagine to be its potential near-term applicability. However, that misses a bigger picture.

Let’s say your car breaks down. If you bring it to a good mechanic, typically they would be able to repair it for you. Why? Because we understand how automobiles work. But neurosurgeons have no such luxury when it comes to the brain. Instead, they are tasked with operating on this spectacularly complex piece of machinery without a sufficient blueprint for how it functions in healthy individuals. If we want to effectively treat neurological disorders and help people live healthier, happier lives, we need to understand as much as we can about brain function. So that is my job: solve the brain.

Further information:

1)     Chen, G., King, J.A., Lu, Y., Cacucci, F., Burgess, N. Spatial cell firing during virtual navigation of open arenas by head-restrained mice, eLife 7, e34589 (2018)

2)     Moser, E., Roudi, Y., Witter, M. et al. Grid cells and cortical representation. Nat Rev Neurosci 15, 466–481 (2014)

3)     Spens, E., Burgess, N. A generative model of memory construction and consolidation. Nat Hum Behav 8, 526–543 (2024)