Clearing the Way:
A Conversation with Tatsuya C. Murakami

For Tatsuya C. Murakami, PhD, research associate in the Fisher Center Lab at The Rockefeller University, understanding the brain begins with being able to see it properly. Working at the intersection of neuroscience and imaging technology, Dr. Murakami has developed a powerful new method for clearing brain tissue—literally, making the brain transparent—which allows researchers to study the organization of specialized cell types across space, as well as how that organization is affected by Alzheimer’s disease.

In this conversation, Dr. Murakami reflects on how such innovations have shaped his career, the growing role of artificial intelligence in large-scale image analysis, and his hope that seeing deeper into the brain may unlock new ways of understanding it, better detection strategies, and more effective treatments for neurodegenerative disease.

How did you come to study the brain?

I think my desire to be a scientist goes all the way back to my childhood. I’m from Japan, and my father is a Shinto priest, so I grew up around shrines and a very particular religious atmosphere. That might sound a bit boring for a child, but in my case it actually made me curious. I saw adults all around me taking the existence of gods and other spiritual practices very seriously, and I was fascinated by this ability to believe in things that do not physically exist in front of us. It made me feel that something special—something very deep and distinctive—must be happening inside the human brain. And I wanted to understand what it was.

When I went to university I decided to study bioengineering, because I was interested in how engineering principles could be applied to work with cells and tissues and create something useful for human life. That became the foundation of my scientific training—but all along, I also kept thinking about using what I was learning to explore the brain and the deeper question of what is happening inside of it. That eventually pushed me to pursue a PhD, where I focused on the development of new technology for neuroscientific research.

Can you tell us more about your research?

My research began with an interest in a group of techniques known as tissue clearing. During my PhD, I learned how to chemically modify opaque brain tissue so that it becomes transparent. At that time, this was still not a very mature field—it was possible to make tissue transparent, but not to visualize specific molecular information inside of it. It was especially difficult to observe things like RNA or peptides without compromising the spatial organization of the tissue. So that’s the problem that I set out to solve.

I continued to work on this as a postdoctoral researcher, focusing on the visualization of messenger RNA, which is one of the clearest molecular signatures of cell identity. Different cell types express different messenger RNAs, so if you’re able to mark particular messenger RNA profiles, that allows you to identify the corresponding cell types and observe how they are arranged inside the brain. That was extremely important, because the brain contains an enormous variety of cell types, and to understand brain function we must be able to see how those cells are distributed and how they work together.

How did that work lead you to investigate Alzheimer’s disease?

Alzheimer’s became a natural focus for me because it’s the most prominent neurodegenerative disorder, and it presents exactly the kind of problem that my technology is designed to address. We’ve known for a long time that as Alzheimer’s progresses, brain tissue atrophies. But we don’t really understand which cell types are affected first by the disease, which ones are resistant, and how those differences are distributed spatially throughout the brain.

In my current work at the Fisher Center Lab, I’m applying my tissue clearing technique to donated brains from Alzheimer’s patients. This allows us to observe cells in their original anatomical context, which preserves the spatial organization of tissues while visualizing cell populations in detail, at single-cell resolution, in the actual structure of the brain. The fact that we’re able to do this makes Alzheimer’s a very compelling application of my methodology, because seeing how degeneration unfolds in space matters a great deal for understanding the basic mechanism of the disease.

Does that also make Alzheimer’s a good testing ground for this method?

Yes—another reason for focusing on Alzheimer’s is practical. In order to continue refining this method, we need access to high-quality brain tissue samples, which obviously are difficult to obtain in the case of rare diseases. Almost everyone who lives long enough faces the risk of dementia, unfortunately, and Alzheimer’s is the most common form of dementia. So, brain banks have substantial collections of brain tissue donated by Alzheimer’s patients—which gives us a real chance to further improve our technique through serious, large-scale research.

For me, there is also a personal element at the heart of this work. My grandfather had Alzheimer’s disease, and my wife’s grandmother also has dementia. In Japan, people have some of the longest lifespans in the world, and so this is a widespread reality. Alzheimer’s is one of those diseases that touch everyone’s life, directly or indirectly. And the longer people live, the more common it is destined to become. So, to me, solving or even just improving the situation around dementia is one of the highest priorities in neuroscience and medicine today.

Has the advent of artificial intelligence (AI) made a big difference for your work?

Definitely—AI has made large-scale quantitative analysis possible in a way that would otherwise be unrealistic. What we now call AI became popular years ago as deep learning, but it was unable to process datasets as large as ours, comprising terabytes of three-dimensional images. Now, through the Fisher AI Platform at Rockefeller, we have access to GPUs powerful enough to handle those volumes—and that has made a huge difference for our image processing and analysis workflows.

The main bottleneck for this kind of research is that a brain image is extremely hard to quantify, because it captures millions of cells. But AI is really precisely able to detect the position of every cell. A key feature of the technology in this regard is what is known as pretraining—basically, the ability to have the software learn the general patterns it has to work with before applying it to our particular datasets. That way, it can be used very efficiently to analyze our brain images. This approach last year helped us complete a big study that was recently published in Neuron [read the article here], which shows that AI does not just offer technical convenience, but a way to do research that the whole field recognizes as scientifically important.

Where do you think this research will go next?

Our current model is focused on segmentation—that is, locating and identifying cells in these complex three-dimensional images of brain tissue—but I think that the future of AI in this work is much broader. One possible direction would be to build a more general foundation-type model that could handle multiple image-analysis tasks, like removing blurriness or tracing neuronal fibers as they extend through bodily tissues.

Another direction we could take is to make our models more compact and efficient, so they can be used by other labs with less advanced infrastructure. I think both of these things are very important. Large infrastructure like the Fisher AI Platform allows us to develop more advanced AI models for biological research. But for these tools to drive discovery forward, I believe it’s crucial that they become accessible to as many researchers as possible.

Do you foresee any limitations to the expansion of AI in the field?

I said AI is a great tool, but I don’t think it’s a miracle technology. It works well for analyzing data, but its usefulness depends on how that data is gathered, and so on the quality of many other factors, like the accuracy of imaging techniques, how molecules are labelled, how researchers formulate their questions, and—of course—how we interpret data to come up with answers. When all those pieces are in place, AI becomes extraordinarily powerful. But we’re only at the beginning of seeing what it can do for this kind of neuroscience.

The biggest limitation to this type of research, to my mind, is the fact that we cannot study the living brain in as much detail as we want. We work with postmortem samples, which means that we cannot see how Alzheimer’s disease progresses in real time—all we have access to is static snapshots of something that was once dynamic. So, one of the most important things in the future will be finding ways to reconstruct the history of the disease from those snapshots—something else might be able to use AI for someday. As a researcher, that is what I ultimately hope to achieve: not just building better tools, but a clearer path toward understanding, prevention, and better treatments for patients.