2024 年 21 巻 Supplemental2 号 論文ID: e212012
Established in 1960, the Biophysical Society of Japan (BSJ) has promoted the combination of physical and biological sciences since the early age of biophysics. The International Union of Pure and Applied Biophysics (IUPAB), founded in 1961, held its 21st congress this year in Kyoto, Japan, joint with the 62nd annual meeting of BSJ. On this occasion, the editorial teams of two journals published by BSJ - SEIBUTSU BUTSURI (Biophysics in Japanese) and Biophysics and Physicobiology - planned to bury a “time capsule”, which could be looked back on by future biophysicists. On the day just before the congress, we invited seven prominent biophysics researchers to a round-table discussion to foresee the future of biophysics and to leave messages for young scholars and future generations as well. We hope this article retains the fervor of the discussion.
For those who open this after 50 years: Let us see whether their prophecies below have been fulfilled ...(Y.T.)
Tamiki Komatsuzaki (Chair, a former vice president for Biophysical Society of Japan (BSJ), Editor-in-Chief of SEIBUTSU BUTSURI, Japanese magazine published by BSJ, former Vice Editor-in-Chief of Biophysics and Physicobiology (Biophys. Physicobiol.), an international online journal, published by BSJ):
As a representative of biophysical society of Japan, I really appreciate your participation for this round table. We would like to frankly discuss the future of biological physics, including the questions of what kinds of subjects will be accomplished or newly born.
Tamiki Komatsuzaki:
Prof. Fumio Oosawa, who launched the Biophysical Society of Japan and received Nature Awards for Mentoring in Science in 2009 for his research on the thermodynamics of actin polymerization and the force generation of molecular motors with mentoring many outstanding researchers, left us a message. He emphasized that research topics in biological physics should be of interest to both biology and in physics, not just one field. Thus, the ideal research target should be something new and related to both biology and physics. However, the interplay between these disciplines is likely to evolve over the coming decades. With this in mind, we would like to share our thoughts and imagination for the future.
To start, let me introduce myself. I am Tamiki Komatsuzaki from Hokkaido University. My research background is mainly theoretical biophysics such as extracting energy landscape and state space network from single molecule time series.
Ikuko Fujiwara (Vice-Chair, Editor of SEIBUTSU BUTSURI):
Hi. I’m from Nagaoka University of Technology. My research target is actin polymerization and depolymerization as Dr. Fumio Oosawa studied. Recently I have also been working on bacterial actin, MreB.
Akira Kakugo (Vice Editor-in-Chief of SEIBUTSU BUTSURI):
I’m from Kyoto University. I’ve been working on active matter systems using biomolecular motors, studying how they operate, organize, and how their assembly produce functions. Welcome to Kyoto.
Akira Kitamura (Editorial Board member of Biophys. Physicobiol.):
I’m from Hokkaido University. My background is the elucidation of proteostasis, for example, protein misfolding and condensation using advanced microscopy.
Yuichi Togashi (Vice Editor-in-Chief of SEIBUTSU BUTSURI and Editorial Board member of Biophys. Physicobiol.):
I’m a theoretical biologist working at Ritsumeikan University, near Kyoto.
Tamiki Komatsuzaki:
Okay, then, I would like to ask our panelists to introduce themselves and share their thoughts on the future of biological physics. However, please feel free not to stick strictly to this question, because spontaneous discussions are more welcome.
Shuichi Onami:
I’m currently at the Center for Biosystems Dynamics Research (BDR) in Kobe. BDR is one of research institutes of RIKEN in Japan.
I’m not solely a biophysicist. I graduated from veterinary medical school at the University of Tokyo with a background in mathematics and physics. When I was a junior, I found biology intriguing and chose veterinary medicine to gain experience with large animals. However, I realized that large animals were too complex to understand biologically. I then shifted my focus to studying C. elegans, a smaller organism, thinking it would be easier to understand and model mathematically. Yet even this system proved to be complex and challenging to grasp within a few years. Thus, I moved to the National Institute of Genetics in Mishima (Shizuoka Prefecture) to further study the molecular biology and genetics of C. elegans and completed my PhD there. I then joined Sony Computer Science Laboratory to study systems biology at the laboratory of Dr. Hiroaki Kitano, one of the founders of systems biology. In his lab, I simulated and measured the 4D dynamics of C. elegans embryogenesis using image processing. Afterward, I started my own lab at RIKEN, where I focus on developing imaging and computational methods, with an emphasis on data-driven modeling. While C. elegans is a key system in my research, I am also interested in larger systems, including mice and human organoid.
Tamiki Komatsuzaki:
Shuichi also addressed Waddington landscape for C. elegans. In biophysics community especially theoretician such as Gerhard, often people have studied energy landscape. Shuichi also explored the Waddington landscape from the data scientific viewpoint. It should be challenging to capture such from the experimental data of biological systems. Shuichi, do you have any comments on that?
Shuichi Onami:
We all have genetic information like, “the child of a frog is a frog” or we can also say “The apple doesn’t fall far from the tree” and “Like father, like son”. A human produces this kind of human shape (by gesturing to point himself), and a dog well produces dog shape. There are some kinds of deterministic process in their developments. On the other hand, each person has his/her own different looking or different way of thinking. That is, we also have some stochastic source or freedom inside the process. How we can visualize this deterministic process involving some stochastic freedom is the aim of such landscape depicted by Waddington. The landscape consists of many up- and down-hills. Suppose I put a ball on a hill-top and then the ball goes down. The complicated hill shape may generate some stochastic movements along the path from the hill-top to the bottom, which is somehow deterministic. But the rolling ball may also have some probability to switch and go to different pathways.
We can also have similar imagination, not only in developments, but also in cell differentiation. Cell has some stochastic nature. I’m more interested in that... this kind of landscape is defined by the underlying genetic information. You may know the most famous Waddington landscape is that where a ball is falling down. Another famous landscape is that supported by the underlying genetic information with some knots. So, when the connections of knots to each location of the landscape change, they result in a change of the landscape. Then, the ball may go to the different path on its landscape.
Tamiki Komatsuzaki:
For example, in the next 50 years, connections between genes and phenotypes in terms of landscape would be completely unveiled or still not?
Shuichi Onami:
This is a very important and difficult question. Would you mean complete understanding?
Tamiki Komatsuzaki:
Surely, I exaggerated, but I would like to hear your expectation for the future.
Shuichi Onami:
Well, developmental systems are very complicated systems. So, the way of our understanding should be a bit different from ….
Takeharu Nagai (a former vice president for BSJ):
What’s the definition of “complicated”?
Shuichi Onami:
That is, so many factors exist in a biological system. It’s stochastic in nature to some degree and many different situations or environments surrounding the system. For example, even if we want to describe it with usual mathematical equations it is impossible…, because so far, it’s out of our ability including computational power or the equation ability to represent the whole. Thus, such kinds of classical way of understanding largely depends on the use of the terms of the basic physics or related subjects. I think to deal with the whole biological system is yet difficult in general. So, we need other expressions to understand our biological systems. I am still looking for what the best way is for understanding our biological systems.
Takeharu Nagai:
What kind of equations? What do we need during these 50 years?
Shuichi Onami:
Well, let’s imagine how an airplane works. Airplane is a complex system.
Takeharu Nagai:
What is the definition of complex systems?
Shuichi Onami:
Let’s define complex systems as those composed of many parts that can be expressed by many mathematical equations. For instance, we can fly from Tokyo to New York using an automatic navigation system. However, we do not need to understand a set of the systems to travel. Rather, it is very hard to rationalize how the whole system works properly. A black box system that deep learning architecture uses in machine learning is similar. If we put some inputs to the architecture, we get some answers at a certain level.
So, the importance may be how we can understand the complex systems, perhaps, we may do with some kinds of main or principal equations or the package of many equations. It is mostly like a black box, but with some capability of prediction. One way of understanding (biological) systems is to predict something and currently this may be the only way. But I don’t know in the future. Different from the artificial system like aircraft, we don’t know the complete set of equations that composes the system. Biological systems are something like black box systems but somehow transparent.
Takeharu Nagai:
But even if we would have such equations, each parameter should be required.
Shuichi Onami:
I think that to best approximate the actual behavior, the number of equations can vary, and the variables do not need to be the same.
Ikuko Fujiwara:
In terms of the number of parameters, may I ask Gerhard’s opinion, because he showed a nice model of 400 molecules packed with water on his website? So, it’s a completely different scale of the life.
Gerhard Hummer:
Let me start with my background and what led me to such complex systems. I was trained as a physicist, but already early on I worked on systems with a strong biological component and then transitioned into biophysics. I studied in Vienna and then in Göttingen in Germany at the Max Planck Institute of Biophysical Chemistry. From there I moved to Los Alamos National Labs for my post-doc and my first independent group and then to the National Institutes of Health. For the last 11 years, I’m in Frankfurt, Germany at the Max Planck Institute of Biophysics. For me, the most exciting development in the last 25, 30 years has been integrating physical thinking into the biosciences and leveraging computational power to describe complex biomolecular systems. I work at the molecular scale mostly, using computational tools to predict system properties. I expect significant developments in the next 10 to 20 years with ever more computation at nearly all levels. This includes a detailed description of molecular interactions and motions at organellar and cellular scales, and an integration of these into multi-scale models. Advances in computer science and artificial intelligence will allow us to predict and facilitate exchange across scales from the molecular to ecosystem levels.
I expect dramatic changes in my research area regarding data sources. Right now, experiments are carefully designed and thoroughly analyzed by researchers, but I expect increased automation in experiments, with AI-systems steering the machines and analyzing the results. In the future, we may only be involved in the design, maintenance and final analysis. This vision is already evident in some imaging-based research areas.
Kumiko Hayashi:
Now, I think prediction is more important, although Onami-san said that equation is very important to understand cells. Definitely prediction can be an answer to understand the cells using the equations.
Shuichi Onami:
Yeah, right. Predictability is more important, rather than focusing too much on equations.
Kumiko Hayashi:
I really like the word “prediction”.
Gerhard Hummer:
One important aspect for me is the ability to make testable predictions. Additionally, understanding involves “inquirability” and transferability. We are not just able to make predictions for C. elegans with your beautiful model, but there are elements that we can extract from the model and carry over to entirely different organisms.
Rong Li:
That’s kind of how biology has always been operating on using model systems to come up with fundamental principles applicable across many different phylum species. And with some understanding of that, we can make predictions.
Tamiki Komatsuzaki:
I understand that what you try to say is in biology, of course, hierarchy or common principle and some of the principles might not be transferable, but some can be transferable.
Rong Li:
I think a lot of them can be.
Tamiki Komatsuzaki:
Yeah, through transferable principle we may discuss what we can understand and predict in some systematic fashion.
Jerelle Joseph:
One point I would like to raise pertaining to machine learning and AI, and the development of high throughput approaches is that we can make predictions about biology without understanding the underlying physics. Importantly, there are ongoing efforts to develop machine learning models that are physics-informed and interpretable. This is a step in the right direction, because there must be both predicting biology and understanding the underlying physics that is governing these predictions.
Tamiki Komatsuzaki:
For example, in 50 years, if we focus on computer simulations, how can we better understand what will happen at the molecular level?
Gerhard Hummer:
This also goes to Jerelle’s point. It is often difficult for a human to understand molecular systems, because they are so complex. Computer simulations represent a significant advance in this regard but lead to new challenges. We increasingly have the power to describe the systems, quantify them, and make predictions, but our understanding may still be limited. We hope to have AI systems, such as large language models, to interact with these complex simulation models, to phrase questions in a way that allows us to inquire about these increasingly complex models and assist humans in investigating them. However, biological systems are highly complex.
So, I think it would be audacious of us to expect one person to grasp it all. Therefore, I think it’s important for us not to give up and to seek the power of inquiry and interaction with ever more complex models.
Tamiki Komatsuzaki:
Jerelle, because you also perform computer research and target more complicated system, would you tell your opinion with introducing yourself?
Jerelle Joseph:
My background is in chemistry and mathematics. I began my training in the Caribbean, then completed my PhD, postdoc and a junior research fellowship at the University of Cambridge. I initially focused on protein folding and now largely study membrane-less organelles, also known as biomolecular condensates. I use molecular simulations to understand spatiotemporal organization inside cells. In January 2023, I started my research group at Princeton, focusing on condensate biophysics to understand how these structures emerge and their resulting material properties and functions.
In our field, physics plays a big role in biology. Ideas from polymer physics have been used now to understand condensate formation and phase separation inside cells, and how proteins and nucleic acids interact to form such dynamic structures that lead to emergent properties and functions.
Jerelle Joseph:
We’ve made a lot of progress using concepts from equilibrium thermodynamics to explain these types of systems. However, I believe that there is new physics that is needed to explain these complex behaviors. And that’s why I think the field of biophysics is very important for asking and answering these types of questions. One of the great promises of biomolecular condensates is the fact that you can form these organelles inside cells that are very dynamic, that are not delimited by membranes and the fact that you can engineer them to create new functions inside cells.
Additionally, there have been convincing results that these condensates play a role in the progression of cancers and neurodegenerative diseases and so developing drugs to target condensates is an important research direction.
Overall, there is a lot of potential for engineering new functions via condensates or prevent unwanted functions, and that is something that I can see 20 years down the line, 50 years down the line being done very well. The intersection between synthetic biology and engineering is a common theme in our field. But this goes back to our earlier discussion of understanding these systems very well.
For example, to prevent some function within the cell, classically one might resort to targeting a particular transcription factor. Targeting biomolecular condensates will likely require a fundamentally different approach. So, understanding these complex systems is crucial.
Kumiko Hayashi:
I have a question about, so you told us that new physics is needed to understand globally the liquid-liquid phase separation. Do you think even thermodynamics is not useful for liquid-liquid phase separation, or what? Because I am a physicist and curious about what for you is new physics.
Jerelle Joseph:
I definitely think it’s useful because there’s been a lot of predictions that we’ve made using concepts from equilibrium thermodynamics that hold up when you actually do experiments. And one can think of these as being in local equilibrium in different parts of the cell. And so, equilibrium thermodynamics can get us very far. But if you zoom out more and you think about the cell as this non-equilibrium system, approaches grounded in non-equilibrium physics, can prove to be very useful.
Tamiki Komatsuzaki:
Yeah, but I’m also curious about what kind of new physics might be born. We say interplay with biology. Kumiko, can you share your thoughts with us?
Kumiko Hayashi:
Really? I didn’t prepare for anything. I’m Kumiko Hayashi. Last year I moved to the University of Tokyo. Now, I belong to the University of Tokyo, the Institute for Solid State Physics. It’s very solid, but this institute recently changed its direction. They would like to include biophysics, as well as solid state physics. I’m very interested in participating in it. My background is, first, I learned non-equilibrium statistical physics. Thus, I asked you about new physics. But after I got a degree, I changed my major from theoretical physics to experiment.
You said that new physics is important, but I thought non-equilibrium statistical physics is very nice. But at the same time, it’s not so useful. ”Useful” is a very strong word. I mean, practically, I can’t find any good application to society, and I felt the limit of that area and then so I changed it from theory to experiment because I would like to contribute more to society at the same time and indeed I could do it. So, I worked for more than 10 years, I worked with motor proteins experiment with non-equilibrium statistical physics. But indeed, new physics, probably non-equilibrium statistical physics is newer than equilibrium thermodynamics, but still, I can’t find how, we can’t find out the way to use it to contribute to the society.
I asked you, do you think so indeed? So, the non-equilibrium physics is helpful for computation or real society and so on. And, still, it’s important. I understand, but still, I can’t find out the good way to apply. So, to which area we should apply non-equilibrium physics? Still, I’m searching for that. And then I moved to the institute for solid-state physics. I think that I’m searching for a new area. We have many kinds of applied physics techniques. And finally, I moved to the institute, because I am really curious to apply these real applied physics techniques to biomolecules, biological materials because it’s quite new and nobody knows how to use these kind of big, physics science.
Tamiki Komatsuzaki:
The definition of new physics depends on each individual. Measurements are also very important for new physics, but what we can go beyond the measurements? What actually biologists imagine from the measurements is much, much beyond my scope from physics viewpoint. In this sense, the remaining three persons really work there from biological insights. And I would like to shift the subject to biological questions. As such kind of the measurements, what measurements can actually explore new things in 50 years. I want to ask Takeharu first.
Takeharu Nagai:
Right, let me give you a bit of background about myself. I started my research career back in 1991. I was working with Xenopus frogs, so I was primarily focused on developmental biology and embryology. Have you heard of the Spemann organizer?
Hans Spemann and his student, Hilde Mangold, conducted the famous experiment where they transplanted the dorsal blastopore lip of a gastrula-stage embryo onto the ventral side of another embryo. What they found was fascinating—the transplanted tissue was able to induce the formation of a secondary body axis, complete with a fully formed head. To study the molecular mechanisms behind this, when I was an undergraduate, I carried out what’s known as the animal cap assay. I used proteins like activin, BMP2/4, and bFGF. I also performed mRNA microinjection into early-stage embryos to test the secondary axis-inducing activity of these factors—activin, BMP2/4, and Wnt8. Interestingly, among all of them, only Wnt showed a very strong ability to induce a secondary body axis. Even more remarkable is that a very low concentration of Wnt8, at pM level, was enough to induce a fully developed secondary head, eye, and body. At that time, it had been already known about some of the key components involved in Wnt signaling—Dishevelled, GSK-3β, and β-catenin—but our understanding of how these molecules interacted was pretty basic. We just had a few arrows connecting them on a diagram.
Takeharu Nagai:
In the signal cascade maps, there wasn’t any information about the actual number of molecules involved. It made me wonder how many molecules are actually present in a cell or an embryo. Even now, if you look at textbooks or research papers, you’ll find plenty of signaling cascade maps, but none of them include the exact number of molecules involved. But there is one thing we do know. That’s the number of genes. Typically, there are two copies of each gene, one inherited from the paternal side and the other from the maternal side.
That means there are only two copies of each gene in a single cell. And yet, these genes are controlled so precisely and robustly that they can direct the formation of a head, an eye, a hand, or even a foot. So, my question is: how can such a small number of molecules regulate or operate such complex multicellular systems? Let me give you another example. Do you know how big bacteria are? They’re about one cubic micrometer, which is roughly one femtoliter in volume. Now, the pH of bacteria is around 7.4, and from that, we can easily calculate the number of free protons. At pH 7.4, in one femtoliter, the number of free protons is only about 20. Just 20! The thing is, concentration is a continuous concept, typically used when you’re dealing with an almost infinite number of molecules, like Avogadro’s number. But here, we’re talking about just 20, which is a discrete number. So, can we really use the concept of concentration when we’re dealing with such a small, discrete quantity of molecules?
I don’t think we can really define concentration if we want to fully understand what’s happening chemically inside small cells or organelles. Because when we’re dealing with such small spaces, like in a cell or an organelle, the concept of concentration breaks down. That’s why, even as an undergraduate, I felt we needed a new technology to monitor and measure the number of individual molecules in living systems. But, even today, it’s extremely difficult to accurately quantify the number of individual molecules inside a living cell.
Takeharu Nagai:
I hope that someone will develop such a technique in the near future. Besides that, another question I have is about the heterogeneity of molecules and cells. It’s something we still need to explore more deeply. We often use HeLa cells in our experiments. Up until recently, we assumed that each HeLa cell had the same properties. But recent studies by many cell biologists have shown that even HeLa cells exhibit differences—what we call heterogeneity. And now I’m wondering if each protein molecule also has some level of heterogeneity. Actually, I came across a paper that described something quite unique.
It was a 2017 paper published in PNAS. The authors measured the movement velocity of the kinesin molecular motor on microtubules in a Drosophila embryo, using single-molecule measurements. They measured the movement of over 500 kinesin molecules and plotted their velocities on a graph. The average velocity was around 500 nanometers per second, but a few kinesin molecules showed velocities over 2000 nanometers per second. Even though they were all the same kinesin molecular motors, the velocities were completely different. It suggests that there’s heterogeneity, even among the same type of molecules.
Kumiko Hayashi:
Indeed?
Takeharu Nagai:
Indeed.
Kumiko Hayashi:
But I think it’s there because cargo sizes are quite different, so it doesn’t come from heterogeneity of kinesin, but I think the cargo size, vesicle, so probably might be a heterogeneity of vesicles.
Takeharu Nagai:
Might be.
Gerhard Hummer:
Or heterogeneity in the surrounding environment.
Takeharu Nagai:
Yes, the surrounding environment could play a role. In other point of view, I think most biologists tend to focus on the mean value when trying to understand the molecular dynamics of kinesin motors or other protein functions. We also need to consider factors like cargo size, environmental differences, and so on. But most single-molecule analyses are done on a cover glass. The environment on a cover glass isn’t quite the same as what happens inside a living cell. A lot of studies have measured when ATP is hydrolyzed or when myosin takes a step. But the actual situation in our muscles is completely different. We need to measure the entire myosin movement along a single actin filament in muscle to really understand what’s happening. Otherwise, we can’t really understand why we can move so smoothly (by gesturing with his arm). There are many myosin molecules even on a single actin filament, and if each myosin hydrolyzed ATP independently or randomly, we wouldn’t be able to contract our muscles smoothly. So, there must be some mechanism where ATP is hydrolyzed cooperatively. To understand phenomena like that, we’d need to measure it. But the problem is, there’s no technology available right now that can do that.
Tamiki Komatsuzaki:
So, you mean within the 50 years, people can access that information on how collectively myosin’s motions happen together to create macroscopic motions?
Takeharu Nagai:
I think so.
Kumiko Hayashi:
Inside body?
Takeharu Nagai:
Inside body.
Gerhard Hummer:
I think this information is already coming. Electron tomography, for instance, is already dissecting the structures of individual elements of tissue down to the molecular scale. But to your interesting point of each protein or each molecule being a bit individual; I think the language of post-translational modification is one way to give a certain level of individuality to different proteins, different compartments, different microenvironments. However, I think that this aspect has not yet been studied much with high spatial and temporal resolution.
Ikuko Fujiwara:
In terms of the personality of individual molecules, I’d like to ask Feng’s opinion. Would you feel to manipulate the personality of the cell using CRISPR-Cas9 system?
Feng Zhang:
I think so. One of my interests has been to manipulate biological processes and developing tools that will allow you to recognize specific signatures in biological systems and then be able to make changes. If you look at DNA, RNA, protein, or even polysaccharides, these are all biological polymers. They usually have building blocks that you assemble into a specific order. So, developing ways to be able to recognize specific signatures of biological polymers, and using these recognition systems enables us to actuate these molecules. That’s something what we are working on.
In the DNA and RNA area, in the CRISPR-Cas system, or zinc finger proteins, or transcription activator like a TAL effector, these are kind of biological molecules that you can customize to recognize arbitrary combination of polymers. This means that you can design a CRISPR protein with a guide RNA to recognize the DNA or RNA sequences, to recognize protein polymers, especially unfolded primary polypeptide molecules.
If you wanted to count things inside a cell, you could use this as a way to count. Once you can control to bind and to recognize specific polymers, you can bring modulators. With DNA, you can cut the DNA or modify the DNA to change its epigenetic state. You can also visualize DNA by green fluorescent proteins to a specific genomic locus and see how it moves around in the cell as the cell stage changes or as the genome gets reprogram for a different function.
With proteins, you can try to localize by changing either the half-life of proteins or to bring things over to visualize the location of these proteins, or to induce aggregation or assembly of protein complexes. Certainly, you can also make perturbations even across polymer types such as between DNA and protein interaction, or RNA and protein interaction. Finally, polysaccharides are also important and abundant biological polymers that have information, including our current ability to decode these polysaccharides is not as advanced as they are for DNA or RNA or protein.
Developing these tools to recognize polysaccharides will help us to understand glycobiology. So then, how do you find these things? I think over the next 50 years, we’re going to find out.
There will be lots of biological diversity. Nature has almost anything that you can think of in biology. Nature has created some version of it. As people map biological diversity by sequencing everything around us, from microbes to animals to plants to any other organism, the genomes are getting decoded. That means we know what are the genes that encode protein systems in nature.
Using what we were talking about earlier with AI and kind of smarter information processing systems, we can predict the structures, we can predict interactions, we can predict functions. By applying that at a very large scale, today there are probably about 9 billion unique proteins that have been identified. This means we can eliminate redundancy of proteins that are 98% similar and disregard those. So, what are truly unique proteins? If you analyze them across 9 billion using these newer methods, we can predict the structure, predict interaction, predict how they may bind to different substrates. I think we’re going to find a lot of very interesting and probably useful systems that can help us monitor and perturb biology. That would not happen over the next six years, but you will see progress in the next 10, 15, 20 years. And then, of course, there will be new methods and way more data accumulating. I think that will be accelerated and be very exciting.
Ikuko Fujiwara:
My mind has been a bit hooked to Kumiko’s word, like how can we find a useful way to use the biology and biophysics that can contribute to the society?
Feng Zhang:
After people really will end up doing with understanding a system, then they will start an engineer system.
Feng Zhang:
People will try to modify it or to create new systems. That has some type of utility. The most obvious one is to improve health to treat disease and to make people live healthier and longer. Outside of health, applications in biomaterials with climate issues are becoming more urgently demanded. Biology plays a larger role in the environment and in climate.
In the long run, we’re going to other planets like interplanetary travel, exploration and colonization, even interstellar. When we go to another planet, we’re going to need to establish biological environments in these new places. Biophysics will play a huge and essential role in that, because many of these foreign heterogeneous environments require diversity for earth biology to thrive. Recently, it was published that one of the largest data sets about how human biology is affected under microgravity in space travel. Researchers collected biomarkers and biometric information, and what it has shown is that even transient exposure to microgravity can have a profound negative impact on human biology.
Tamiki Komatsuzaki:
Really?
Feng Zhang:
Even in outer space for a few days there are significant changes to the immune systems, causing the eye to vision and probably many other aspects of biology. This is very much a biophysical problem in microgravity. In the next 50 years we’re going to get to Mars and so understanding how that’s going to change us. The research will take linear development and then it’s exponential after that. It’s going to be changing very fast, and biophysics will take us there.
Shuichi Onami:
So far, the history of biology is too much focused on the cell or subcellular. The living system is not limited to the cellular system. Young scientists are more curious about the level of global. That is not coming from the society pressure. Recently, I have a feeling that we have been too much focused on the cell.
Rong Li:
I absolutely disagree with that. The cell is really the smallest yet the most complete self-replicating unit of life. And all the complexity principles we talk about are encompassed within the life of the cell. Of course, many cells come together to build tissues and organs. But it’s very much linked to the inner workings of the cell. In fact, a lot of our understanding still has to be geared towards finding the genetic information to how molecular assemblies work together, how cells interact with each other, how forces are transmitted across cell membrane from cell to cell to organize tissue, to generate organ functions.
Tamiki Komatsuzaki:
Maybe it’s a good time to ask Rong to talk her thought with a self-introduction. We have been discussing how the system can adapt, and you have been curious about how eukaryotes can adapt to environments. We don’t know exactly how the biological system has some potentials to adapt. And also, that’s beyond my imagination, and I would ask you to explain your background and your questions. What would be happened in the future?
Rong Li:
It’s a tall order to really predict what’s new in the next 50 years, but I would say in my own career, a lot of the new ideas, new thinking, or new findings were completely unexpected. I’ve been running a lab for 30 years, and I think every 10 years, there’s something that I really learned that profoundly affected how I look at biology.
I studied molecular biophysics and biochemistry in my undergraduate at Yale, and then I went to UCSF to do my PhD. At that time, in the field of cell biology, there’s a huge excitement about mitosis and cell cycle regulation. At that time, several giants in that field were at UCSF. So, I got really excited about that area.
And what I worked on was a new concept at that time, where and when cells undergo the mitotic process. If something goes wrong, somehow cells can sense it, and then reverse control the cell cycle to stop the progression. So, it’s a concept pointed out by Leland Hartwell as “checkpoint”. And I was working with a young professor at the time, Andrew Murray. We decided to look for, to understand how mitosis can also have such a checkpoint.
The approach we took was a genetic approach. We decided to see which gene, if mutated, would disrupt this function. We identified three genes. And later on, more genes were found that mediate this checkpoint control of mitosis. After that, I became very interested in morphogenesis. So, I went to Berkeley to study actin polymerization. I was more interested in inside the cell how actin cytoskeleton architecture and dynamics come through in a spatial manner, especially in a polarized cell.
How do cells control one kind of actin structure at one end of the cell, and other types at the other end? And then, when I went to Harvard to start my own lab in 1994, I continued to work to study that. But one problem that really changed how I look at biology was that biology cannot just be explained by linear genetic pathways. At that time, we were looking at cell polarization.
The paradigm was “you have a cue”, such as gradient of chemoattractant or a mark on the cell that causes the spatial cue to activate a set of GTPases which regulate actin polymerization to organize the cytoskeleton in a polarized manner. It’s a linear hierarchical pathway that was dominating the thinking in the field. When we expose neutrophils to a chemoattractant gradient, the cells will polarize towards the gradient in general. But it’s also well known that if you just give neutrophils a uniform concentration of chemoattractant, every cell will polarize just in random directions. This suggests that the ability to polarize is intrinsic to the cells. It doesn’t actually depend on a spatially preexisting asymmetric cue.
So, I became very interested in how cells can self-organize, break symmetry and become polarized. And that was the first time when I worked with mathematicians to use mathematical models to try to understand how a spontaneous polarization can happen through feedbacks between signaling mechanisms, small GTPase signaling, and cytoskeleton assembly, and then cytoskeletal-based localization of GTPases. So, it’s a very simple model, but it could bring about spontaneous polarization in the absence of any external cue.
Kumiko Hayashi:
Excuse me, is it the model of a chemical reaction?
Rong Li:
Yeah. It is based on chemical reactions but is actually a stochastic model. The main idea is there are stochastic distributions of a GTPase, called CDC42, on the membrane, which can activate stochastically. When CDC42 was activated, it would start to generate cables of actin. CDC42 is stored inside the cell on secretory vesicles, and these vesicles will travel around along the actin filaments and then deposit more CDC42 on the membrane where actin is nucleated. So, you have a feedback, self-enhancing mechanism to amplify a stochastic asymmetry. This simple model was very eye-opening, and we wanted to continuously understand the system by building more quantitative models.
And again, I learned something really new about how modeling and experiments can work together. One very interesting conversation with the mathematicians who we were working with was about something our first model didn’t really deal with. They said, “Once CDC42 gets transported to the plasma membrane, and gets deposited. What happens to it? Does it just stay there?”
And in our first model, we just assumed it just stays there. I said, “Actually, I don’t know. It probably would diffuse on the membrane.” It really drove us to go and actually measure membrane diffusion and using live-cell imaging. When we measured Cdc42 diffusion, it actually diffused really fast. With such fast diffusion, the model stopped working because once you deposit the molecule, it diffuses away. Finally, we found that the protein actually gets recycled, through endocytosis and then transported back. So now you have this continuous system of targeting, turnover and recycling. The kinetics of these subprocesses have to be matched to generate polarized morphology and with specific characteristics. That was a very useful experience to learn.
Tamiki Komatsuzaki:
I can imagine the recycling process. If the time scale of the recycling process is very slow, the system can forget where polarity is. So, in this sense, the recycling process should have some timescale.
Rong Li:
Right, exactly. Because cells will actually grow in that direction, it also defines the shape of the polarized domain, which also determines the shape of the growth as well. And it is determined by the kinetics of the recycling, as you exactly predict. So, I thought it was a really good example of biological functions and behavior of biological systems are controlled not by genes but also by physics.
The second important lesson I learned was during the second 10 years. I moved my lab to the Stowers Institute, a brand-new institute that gave each scientist a lot of resources at that time. And I got really interested in a phenomenon. My lab was studying how cells divide with a contractile ring of actin myosin that contracts and pinches a cell into two. We were studying this motor protein called myosin II, which is evolutionarily conserved from yeast to human, using the same motor protein. We deleted the gene encoding myosin II in yeast. It’s actually classified as an essential gene. Because if you delete it, most cells should stop dividing. But what was really unexpected was that my student deleted the gene, and she just left a plate of cells on the bench. She didn’t throw the plate away. And then there were some really tiny colonies that came up eventually.
Takeharu Nagai:
Was that just contamination?
Rong Li:
No, and then we picked those colonies and spread each onto a new plate and some bigger colonies grew up. Within about 10 passages, we have colonies that grow and divide as beautifully as the wild type. And yet there’s still no myosin II gene. So, these cells have evolved in front of our eyes in the absence of this evolutionary conserved motor protein.
We got really curious. We wanted to see what happened to these cells. Eventually we found that they came up with several different very bizarre ways to divide that are very different from normal cells. But even more surprising was, there was no mutation. It’s simply by changing the relative chromosome copy numbers, in other words, they became aneuploid. So, yeast cells have 16 chromosomes. Euploid cells have the same number of each of the chromosomes. Most cells are Euploid, just like cells in our own body.
You know diploid cells have two copies of each of the chromosomes, haploids have one. So, the myosin mutant cells evolved just by changing their chromosome copy numbers, which affects the stoichiometry of genes. And that created brand-new functionality. So, there was a set of work that we started to do to understand that. But that also led me to get very interested in the ability of biological systems to change and adapt and evolve.
People have coined this term evolvability. So that was one question that I think is still very important to understand. What is the design principle for biological systems? If we think about it, it’s not like engineered systems where there’s a blueprint, and it’s built from there. But our systems evolved over billions of years of evolution, and it’s tinkered at each step.
Thus, the design principle is potentially quite different from human engineered systems. And secondly, how can biological systems balance to accomplish two fundamental features? One is robustness. Our biological system is robust, which can maintain stability even with changes in environments, noise and stochasticity. On the other hand, biological systems can change and adapt when there are environmental fluctuations.
So how can a biological system accommodate two seemingly opposite properties?
Kumiko Hayashi:
So, you mean stability and...
Rong Li:
And change, and the ability to evolve to adapt.
Tamiki Komatsuzaki:
Robustness and stability are in the same meaning?
Rong Li:
Yeah, in some ways. On the strict definition, they’re probably not the same. However, on the superficial level, probably, from theories we define them somewhat differently with mathematical formulation.
Tamiki Komatsuzaki:
Stability is more familiarized in mathematics, but robustness is not so well defined, I think. So, it’s a really open question. How can evolvability and robustness be competed to each other?
Rong Li:
Yeah, in evolutionary biology, robustness is defined as the ability to maintain functional optima when there are changes in genetic makeup or accumulation of mutations. If the system functionality drops sharply, then there’s no robustness. But if there’s a nice functionality plateau to accommodate a large degree of changes. That’s defined as robustness.
Tamiki Komatsuzaki:
Mathematically, stability is well-defined as minimum of, say, some cost function, once we may have an equation to describe the system in question, whose degrees of freedom or variables are fixed. But robustness, that is, what we may imagine in biology, may accept dynamic changes of degrees of freedom. That is, some relevant degrees of freedom to represent the system can change. There, some variables newly arise from the outer environment and the others may disappear in time to time. But nevertheless, some function preserves even at such moments. So, this could be ‘robustness’ some people try to differentiate from the concept of stability.
Gerhard Hummer:
This is a really powerful, beautiful example. I think it connects to the question that you all got us thinking about: the “new physics”. What is it? Where should we seek it? Because I think that the traditional physicist’s approach has been reductionist and, in a way, thinking about complexity arising from simple rules that can be identified and written down. As you just said, I would expect new physics to emerge by embracing this type of complexity. A high level of compositional complexity and redundancy is difficult to capture with current physicist approaches and physics formulations. This is not normally dealt with.
Tamiki Komatsuzaki:
Yeah, maybe that’s what we should invent in the next 50 years.
Rong Li:
Maybe I can just quickly answer my take of your question about what biophysics can contribute to biology. Now I’m the head of the Mechanobiology Institute in Singapore, so it’s a pitch that I often have to make. I think physics and mechanics are very much at the heart of the next 50 years of biology, because, if you think about the previous 50 years, it was very much focused on the chemistry side of biological systems, identifying the genes and molecules, the genomes, the proteomes. These give us the components. But only having the components doesn’t tell us how life works. So, physics and mechanics are what’s missing between the components and biological functions.
(After moving the round table room to another eating room with Kyoto foods)
Tamiki Komatsuzaki:
Thank you very much again, this is the second round, we missed to hear the research background of Feng. Let’s start with asking Feng to hear about his background.
Takeharu Nagai:
I want to hear about the story of optogenetics!
Feng Zhang:
Wow, okay, my name is Feng Zhang. I’m a professor at MIT at the Broad Institute. I’ll start from when I first got interested in biology. I was born in China, and I moved to Iowa when I was 11 years old. When I was in middle school in Iowa, they had a biology class, and I thought it was very boring. It was memorizing different types of plants or leaves or dissecting frogs and identifying anatomical parts.
But the turning point came when I was in seventh grade. The school had weekend classes where you can go and learn something different, like graphics design or Shakespeare or art, and drawing. They also provided one class called molecular biology. I had no idea how biology could be molecular, so, went to this class to see what it’s all about. In that class, the teacher taught us about DNA and the central dogma.
All of the modern advances are happening in biology. They also showed us a very exciting documentary. It’s called Jurassic Park (everybody laughs). Because the things depicted in that science fiction movie were tangible and could be related to what we were learning in classes, I got really interested in biology.
Tamiki Komatsuzaki:
So Jurassic Park in a sense promoted a young high school student to be a researcher. What’s an interesting story!
Feng Zhang:
So, in a way, it inspired me by showing how biology can become an engineering discipline rather than just cataloging information. That’s how I got interested in biology.
The teacher remembered our interests in biology, and when we started high school, he came to us and said, “there’s a really interesting opportunity”. The hospital in our city had a gene therapy lab, and they are taking volunteers to work there. So why don’t you go and apply to see if you can work in this gene therapy lab? So, I started working there after I turned 16, going every day after school starting from 1:30 p.m. to 7 p.m., I went there almost every day in the afternoon for three years. Then based on that experience I was really excited about bioengineering. And I went to college at Harvard and worked with a biophysics lab. I worked on solving the structure of a variant of hemagglutinin, HA, on the virus membrane. After working in that lab for a year, I then joined a brand-new professor, Xiaowei Zhuang. She developed STORM imaging.
Takeharu Nagai:
Oh, really? This is my first time to hear that.
Feng Zhang:
Yeah. I worked with Xiaowei for three years. After that I went to graduate school at Stanford and majored in chemistry, but I became very interested in neuroscience and I joined the lab of a brand-new professor, his name is Karl Deisseroth. Karl had just begun his lab. I joined his lab because he told me that he was both a practicing psychiatrist and a bioengineer. I also collaborated with a couple of professors in Germany, Georg Nagel and Ernst Bamberg. And so, we put a protein from the green algae Chlamydomonas reinhardtii called channel rhodopsin into neurons. It became the basis for a technology called optogenetics, which allows you to control neuronal activity with light. After we developed that first light-controlled activator, we started thinking about how, just like there are green fluorescent proteins and red fluorescent proteins and yellow fluorescent proteins, there may be other proteins that respond to different colors of light. So, we then looked into nature, and we found other opsin proteins that allowed us to shine other colors of light on neurons to then be able to inhibit activity.
Then we used halo-rhodopsin to allow yellow light inhibition of neuronal activity. And now there are many other opsins, building out a whole toolbox of optogenetics. When I was finishing my PhD, I thought the biggest problem for optogenetics was gene targeting. We had all these tools now, but to study the brain, we need to be able to control different circuits in the brain without perturbing other parts.
And the way to do that is to have genetic targeting so that we can insert channel rhodopsin or other opsins only into the neurons of a specific circuit. So that took me into this area of thinking about how you precisely find and manipulate a gene sequence in the genome.
That’s why I started to work on zinc-finger proteins, trying to engineer them to recognize DNA sequences. And then I developed TALEs, which are based on plant proteins that use modular building blocks to recognize specific DNA sequences, and this “code” that TALEs use was discovered in 2009. Because TALEs are modular, I thought they could be reprogrammed to target the human genome, so we turned them into a tool. But it was still very hard to use, because you had to rebuild the protein every time you wanted to target a different part of the genome. And then I learned about CRISPR and began to develop Cas9 for gene targeting, which is a lot easier to retarget because you don’t have to change the protein, just the guide RNA.
All my experience of seeing how nature has invented these powerful molecules has inspired us to mine nature, kind of doing treasure hunting to find useful molecules and develop technology.
Takeharu Nagai:
Do you remember when I visited your laboratory more than five years ago, maybe around 2016 or 17? And at that time, you showed me this big plastic container with a green solution in it. And you said it was green algae. I wonder if you published a paper using the algae.
Feng Zhang:
Yes, we did! That algae have a special type of transposable element that we discovered that transposes using an RNA to target the insertion site. We showed that you can reprogram that transpose to insert its DNA wherever you want, by just changing the RNA guide, and that is a very useful tool.
Takeharu Nagai:
How did you find such very unique organisms?
Feng Zhang:
We do genomic mining. What we did is we aggregated the sequences from all the different public databases. NCBI, JGI, there are several repositories in Japan, too. We put all the data together - about 10 billion unique proteins! And then using some of these newer methods, and also developing faster ways of clustering proteins, we can go into this data set and find interesting systems.
Tamiki Komatsuzaki:
So, what is your scheme to find new proteins from such a big database? It is mining, and it looks very difficult to infer how to pick up that one. How can you find it?
Feng Zhang:
This is true, so we use principles of evolution to help us find interesting systems. And one of one of these, at least in bacteria or prokaryotes, is the concept of guilt by association.
Ikuko Fujiwara:
Guilt?
Feng Zhang:
Guilt by association. In language, this expression means that if you hang around with people that have committed crimes, you are likely to be guilty of crimes too. In evolution, it means that in any genomic locus in bacteria the function of a given gene is likely to be similar to the function of the neighboring genes. For example, you usually find clusters of immune systems together in bacteria, and those are called defense islands. If it’s metabolic clusters, you will find metabolic islands.
Takeharu Nagai:
It’s like an operon?
Feng Zhang:
Yeah, there are clusters of operons. And if you have one operon for a defense system, chances are the next operon is also a defense system. And this is because most of these systems are passed back and forth between bacteria through horizontal gene transfer. And when they’re next to each other, they’re more likely to get co-transmitted. And so, from an evolutionary perspective, it’s more efficient. So, this is one principle. But there are many evolutionary principles that we use.
Tamiki Komatsuzaki:
That principle is basically found on the earth. So, you said in the first round even the immune system can change, when you put the system on the other environment by feeling different microgravity. So then, the clustering scenario might not work outside of the earth. In that case, such databases or knowledge are still transferable to different environment?
Feng Zhang:
I think what you will have to do is to begin with a function that you’re looking for. Whatever that function is. And then I think you have to engineer that function. So, this could be done through direct evolution, or nowadays you can use protein language models to rapidly diversify the sequence of a protein while conserving key structural domains. And then so the idea is that if you can go very far away in sequence space, you might be able to begin to go further away in functional space.
But even when we do that, the best way is probably just to have many, many cells send it out into the environment that you want it to function within. Then select for cells that survive and let them evolve even more improved functional proteins. But also, I think the other thing that’s interesting is probably collecting samples from outer space. And it’s very likely there’s microbes on these other planets. It’s this idea of panspermia, where there are likely other DNA-based organisms beyond Earth that evolved in those specific environments. NASA is sending a probe to Venus to collect samples, and there are many moons of Jupiter that have water under the frozen surface. So, there’s lots of interesting stuff. Another interesting thing I just learned, …
Takeharu Nagai:
Are you a friend of Elon Musk? (everyone laughs)
Feng Zhang:
In Ontario Province in Canada, there’s an old mine, an iron mine. And this mine goes very deep, about two miles deep. Very deep. A few years ago, they found a water reservoir two miles deep in the earth. This water separated from the major water system of Earth since 1.6 billion years ago! I’ve been thinking I should go there and get some samples and sequence it, because whatever evolved in that separated water supply, it’s probably going to have very interesting biology. And the biodiversity of Earth, outside of Earth, is going to give us a lot of useful knowledge.
Rong Li:
There are some expeditions collecting deep sea organisms.
Gerhard Hummer:
Right. In deep Earth, organisms may have critically slow rates of replication, powered by slow mineral decay.
Rong Li:
Yeah, there are definitely a lot of unknowns there.
Gerhard Hummer:
A friend at the Carnegie Institution once pointed out that now there’s almost certainly life on Mars because early space probes were apparently not sterilized.
Feng Zhang:
Very cool, that’s panspermia!
Takeharu Nagai:
As I mentioned earlier, Hayashi-san, I’d like to discuss the issue of efficiency. For example, we have solar panels, but their efficiency in converting light energy into electrical energy isn’t very high. But when we look at living organisms that have survived through evolution, their efficiency in energy conversion is much higher.
In mitochondria, the efficiency of electron transfer is nearly 100%. How did organisms achieve such high efficiency? I believe it’s a problem we need to solve within the next 50 years. Most physicists and engineering researchers aren’t very interested in biological phenomena. They’ve been focused on developing machines or technology independently of biology. but I think researchers should start learning from organisms. There must be a design principle in life that we can apply to other areas. So, one of the key directions for biophysics is to understand how organisms achieve such high efficiency. If we can figure that out, we might be able to achieve the same level of efficiency ourselves.
Kumiko Hayashi:
What if there are any creatures with low efficiency, from the past?
Takeharu Nagai:
Which creatures?
Kumiko Hayashi:
I don’t know. Nowadays, almost all creatures have high efficiency, but the opposite type might have existed many, many, many years ago.
Takeharu Nagai:
I don’t know much about creatures that lived more than 1,000 years ago, or even 1 million years ago. It feels like only something like Jurassic Park could bring those creatures back.
Kumiko Hayashi:
Would you think that only evolution can make some proteins efficient?
Takeharu Nagai:
I guess so. Anyway, one of my research projects is focused on creating FRET-based biosensor using two fluorescent proteins. We’re trying to achieve almost 100% FRET efficiency between the two fluorescent proteins. I believe we can make it, even though it’s going to take a lot of trials. Creatures on Earth have gone through countless trials since the birth of the planet. That’s why so many plants, for example, have such highly efficient energy and electron transfer systems. And not only this. but the structural analyses will also help us develop these highly efficient, functional systems. By understanding the incredible structures that organisms use, like how they capture photons and transfer them to the reaction center. That knowledge can guide us in creating similar efficient systems.
Takeharu Nagai:
I honestly don’t think molecular biology can solve this. Only biophysics can solve it! (everyone laughs) Just deleting or depleting genes won’t give us the answers we need for something this complex.
Rong Li:
Well, I would probably say molecular biologists and physicists can work together to solve this.
Jerelle Joseph:
And even with thinking about molecular biologists and physicists collaborating with each other, there’s been a trend of engineers moving into biology. The field of bioengineering is very young. And there’s been a big rise in engineers thinking about how they can manipulate biology. Therefore, we’re going to see the reverse happening where engineers are now taking what they learned about biological systems and creating biological inspired materials.
Gerhard Hummer:
Let’s think about energy efficiency, which isn’t always high in biological systems. Evolution obviously is a strong driver to harvest more of any given energy source. However, other important factors are the abilities to repair and remove damage, and to regenerate. We often perceive materials as quasi-static, such as a solar cell on the roof. I think what is so different in soft materials is that they age rapidly, but living materials have an innate ability for self-repair. Endowing soft materials with repair capabilities is one of the big challenges, for instance to deal with unavoidable damage that comes with light absorption. In an ideal scenario, soft chemistry could then “survive” forever.
Shuichi Onami:
OK, so that means we must try to kind of collaborate with material techniques and try to put together the idea. So, I’m making a such kind of more sustainable materials in the future.
Tamiki Komatsuzaki:
In the past, I discussed evolution with a friend of mine at Stony Brook university. So, he discussed the evolution by the terminology of the so-called flux landscape. Flux landscape is an extension of free energy landscape into steady state nonequilibrium condition. But this means that, physically, that landscape is prepared (or preexists) for the evolutions. So somehow, this setting implies that all the consequences of the result of the evolution is what God has already known where to go. Because God, somehow, he has already recognized about the landscape. In order to discuss about the landscape, God requires samples of all detailed, possible pathways. Then, he can tell us this is a way to go or evolve.
So, such pictures require that someone should have already sampled the whole space and picked up all the pathways. I was strongly puzzled and against such landscape interpretation but after we discussed a lot, he told me “Tamiki, nevertheless, maybe the God is now inviting you as “come, come, come, this is the way to keep going.” I think a way to describe or rationalize evolution is a sort of language but still incomplete for all the questions that can be addressed.
Feng Zhang:
I think one thing that’s going to be very profound in the next 50 years is the integration of AI as a tool for doing research. And I think biophysics could really help to advance the development of robots in a biological context. Because within 50 years, I think we’re definitely going to have fully closed loop systems where AI is reading literature, synthesizing information, generally creating novel hypotheses. And then if the robotics are great enough, the AI can then instruct the robots to do the experiment to generate more data, and the data can be fed back right into this AI system to then generate more hypothesis. And then this closed loop system can move things forward a lot faster than we’re going to be able to.
Tamiki Komatsuzaki:
AI can generate the hypothesis? I think this is the ultimate final stage of what AI would do. But if AI can also generate hypothesis spontaneously or autonomously, then, what humankinds can do for the research?
Rong Li:
Go on vacation (everyone laughs).
Feng Zhang:
the AI is going to increase our productivity so much.
Ikuko Fujiwara:
However, recently I’m with young students and then now after the especially COVID-19, the students do not want to do repeat experiments. They think that one is good enough.
Takeharu Nagai:
We need to tell student that reproducibility in research is a very important point. To get reproducibility we must repeat the same experiment many times.
Ikuko Fujiwara:
What students say is that AI can predict what’s going on without experiments. Why do we have to think what AI can suggest?
Tamiki Komatsuzaki:
Well, I think it’s crucial for experimentalists not to believe AI too much. For example, image analysis would be very dangerous, especially for molecular optics.
A well-known simple example is that AI can recognize starfish classification like which date they were sampled, female or male, or what kinds. AI can perfectly do using imaging data. The whole image has some other information not necessarily relevant to starfish. For example, if somebody grabs a starfish when taking photos, the photo includes not only starfish but also fingers. Likewise, if some photos are taken under sunshine and the others are not, the photos can involve not only intrinsic information about the starfish but something else on environmental difference when the photos are taken. So, AI cannot differentiate such intrinsic or extrinsic information on starfish.
Thus, AI can use and pick any available information for the purpose we define in image recognition problem. So, if we do not know the underlying methodologies of AI and if AI provides a very nice, output and a high classification score as you expect, AI may not necessarily handle the classification problem in the right way along what you may expect. In some cases, maybe it’s very misleading (obviously this is not only image recognition but also spectral classification in experiments where instrumental noise, device dependence may also be made use by AI for the purpose). Actually, the question of how AI can change biophysics subjects was requested by other editorial members to ask you in this round table.
So, I would like to ask each person how AI can change your subject.
Takeharu Nagai:
Can AI increase efficiency?
Gerhard Hummer:
At a basic level, artificial intelligence relies on the ability to represent functions in very high dimensional spaces, with an enormous level of flexibility. But high dimension comes with the price of very limited data. As a result, there will be risks of interpolation and extrapolation, and overfitting, and so on. In my immediate area of research, we are already building closed learning cycles, where our simulations are coupled to an AI system that is essentially self-learning. The system makes predictions, runs new simulations, and tests the predictions. If the predictions are wrong, the system rejects the model, and if there’s enough evidence building up against the model, the system retrains the model.
All this is quite simple and basic, but it is changing the way my group conducts research projects now, because there is now a tool available to them where one can represent high-dimensional information on what I would call a mechanism, in our case the mechanism of a molecular reaction, in a mathematical form.
And again, with similarly very rudimentary approaches in molecular regression, the learned mechanism is reduced to mathematical equations. So, every now and then, the AI spits out formulas that the human, us, can inspect, and usually they look somewhat ugly, but at least it’s something to look at.
Combined with the powerful tools that are emerging, say foundational models, large language models, our models may not just use local information and learn on it, but integrate across scales. We have already seen examples of AI-driven experimentation. In our case, we think of a simulation as a computer experiment. With similar approaches, it would be possible in principle to interact with hardware, say if the microfluidics needs replacement.
To minimize risks, AI needs sophisticated tools to detect if one has left the region of confidence, if one is in an area where the model lacks predictive power. Looking forward, I expect AI to change the way we operate: from simply accumulating a large amount of data to serving as a powerful hypothesis generator. I think that even hints at new physics might be spit out, at least in hypothetical form.
Tamiki Komatsuzaki:
How do you think about it, Rong?
Rong Li:
I think I’ll be dead in 50 years (everybody laughs). Well, I mean, I think it’s a very powerful tool, for sure. Beyond that, I’m not sure. But I’m not an AI expert at all. I do know AI models, at least now, use data as training set that’s generated by humans. And if we give machine learning bad data, it will generate a bad model. And so, bad information can be propagated through AI, which then generates wrong information that is further propagated. So how do we guard against that?
Tamiki Komatsuzaki:
I think it’s a human obligation.
Rong Li:
We’ve seen AI being widely used in protein structure prediction, but it’s also known a lot of the predicted structures are wrong.
Tamiki Komatsuzaki:
One very remarkable AI is Alpha GO Zero. Alpha GO Zero has a very interesting story. Do you know Alpha GO, Alpha GO is the first algorithm that won against the world champion. The next one was so-called Alpha GO Zero. Alpha GO Zero has never ever used any information of what the AI has to do (no records GO players performed), only what the input data was the rule.
Suppose you have two computers. They start to compete with each other, but they are somehow baby because they only recognize the rule. They can understand only whether they won or lost. But they can coevolve, and they can find their own strategy to compete against each other. Then the interesting story is that Alpha GO Zero competed with Alpha GO. Do you know which one won?
Actually, Alpha GO lost against this spontaneously-coevolved Alpha GO Zero. This means that there exists a strong strategy to accomplish the task, that is, to win the game, which we humankind have never ever reached. So, this is a possible way the AI may influence us in right direction. The take home message here is that in the case of science can we define such a task with a simple well-defined rule? Only we can put the rule, then just we can ask the AI to solve, which may end up with a solution humankind has never imagined.
At the very beginning, they were just babies, and their actions in the play were very stochastic. But eventually, they could spontaneously find a new way of solving the task much more efficiently than humankinds.
That was a case of Alpha GO where the rule is well-defined. But in the case of our science, it’s very difficult to have such.
So then, can I ask you, Jerelle, about your thought of AI?
Jerelle Joseph:
I agree with many of the things that were said before. In our work, we spend a lot of time thinking about how we build computational models that can robustly predict the behavior of biomolecular condensates. And there’s been a lot of ways in which machine learning has been leveraged to develop these types of models. As well as recently, it’s been shown that if you can develop computational models that can recapitulate very well biomolecular condensates, then you can use them as a data generation tool for training neural networks to predict other biophysics.
So, there is nice feedback that can happen where you use tools from machine learning to develop computational approaches, and in turn use the resulting models to train other machine learning-based approaches. A challenge that we’ve had in our work, where we aim to develop computational models by integrating information from experiments, is a lack of quantitative experimental data. With AI being used to automate experimental workflows, there will be a significant increase in data that can be used to build such physics-informed models.
The resulting approaches can provide biophysical insights and can also serve as tools in the engineering pipeline for manipulating biology. Something I’ve noticed is that ML-based models perform better when they’re based on a solid prior or physical insight. Intuition and understanding of biological systems are crucial when designing models because if you provide incorrect data, it will lead to inaccurate predictions. I always advocate for the need for humans to carefully consider how we use ML/AI tools and what we’re actually learning from them.
Rong Li:
I think the key here is intuition, and no one understands what human intuition is and whether that can be recapitulated by the machine systems.
Gerhard Hummer:
But isn’t it the difficulty that when we leave our area of competence, when it gets very difficult, we need a team? A single investigator having full control over a project is barely possible now in biology with its inherent complexity. Looking forward, I expect that the approach of relying only on our human intuition, which is resting strongly on experience and deep knowledge of the accumulated facts, will be pushed to the limit.
Kumiko Hayashi:
You mean that AI can do anything?
Gerhard Hummer:
Just looking around in my own institute, all areas of molecular and cell biology suddenly seem to collide. You need to know so many things already. This is just my microcosm. And, yes, AI may help us.
Rong Li:
Well, I mean, there’s nothing wrong with collaboration, right? With people working as teams. Most small problems were solved. What’s left unsolved in biology are these really complex problems and questions as well. I think right now, building interdisciplinary teams is really “a must” for future biology. But that doesn’t argue against using powerful tools such as AI. Intuition is important, but we all know that’s not the only important element of scientific exploration.
Tamiki Komatsuzaki:
Feng, you expect AI can even find CRISPR-Cas9?
Feng Zhang:
AI can help us to discover. I mean if you think about what human intuition or creativity are, scientists come up with this hypothesis or creative ideas. But if you ask anybody to come up with creative ideas, they can come up with something. And so, there are better creative ideas, less creative ideas. But each one is really what they call a hallucination in the AI field.
So, the good creative ideas are more accurate hallucinations. And then the less good creative ideas are less good hallucinations. And it’s all based on our own probabilistic understanding of some sort of real-world model. I think AI is doing the same. It’s a probabilistic model that is represented in AI space. And there are better and worse probability outcomes.
And probably there are things that we can learn from the way that our computational system works, that AI hasn’t implemented. The current AI models are very simple. They’re probably just like the neocortex. But there are a lot of aspects of the human brain beyond just the basic architecture of the cortical layers. For example, basic processes like sleep and oscillatory activities that get propagated through neural networks to help with consolidation, to help with garbage cleaning.
That isn’t implemented in the way that data is represented in AI models currently. And these are things that I think people are going to be starting to explore more and will help enhance AI models or AI systems so that they can learn better, remember better.
And the human brain doesn’t consume nearly as much energy as these GPU (Graphics processing unit)s consume.
Shuichi Onami:
Yeah.
Feng Zhang:
You can compute more than a GPU can.
Jerelle Joseph:
Is it about dimensions?
Kumiko Hayashi:
Yeah, it’s just right.
Feng Zhang:
Yeah, but these are things that people can also improve. So, it’s hard to know how fast it will go, but it will probably go pretty fast. Probably slower than what the AI experts say, but faster than what non-AI experts say.
Tamiki Komatsuzaki:
How about your opinions, Kumiko?
Kumiko Hayashi:
How can I express my feeling about this (AI) project? In daily life, of course it’s really useful. I often use chatGPT and recently my Python programming is always based on it. And I believe totally what chatGPT suggests to me. But when we consider our future, there is a possibility that AI may destroy human culture. I think it’s really difficult for me to answer about 50 years later, because I’m not so optimistic.
Tamiki Komatsuzaki:
Thinking about the distant future is interesting and challenging. But the importance is …the story is not just along the current grant applications, much more beyond. So that was the reason why we chose 50 years.
Ikuko Fujiwara:
Curiosity is the message for the next generation. I think that’s a key, especially for kids in high school now. To keep their motivation and being interested in the research field, they must have fun. So, fun and curiosity would be the keys to keep ourselves as a human.
Takeharu Nagai:
I believe we should consider updating the educational system in Japan.
Kumiko Hayashi:
How can we change it?
Takeharu Nagai:
So just getting knowledge isn’t enough to spark curiosity. Yeah, of course, gaining knowledge is important. But it’s not just about that—learning how to think outside common sense is crucial too. That’s where the current Japanese education system falls short.
Shuichi Onami:
If logical thinking is the only way to take the things moving, the computers should do better. Curious driven approach, is it the most efficient way to make progress on the current state?
That is one question. Many people argue about efficiency of energy, something for efficient way to develop a model, something like that. If we think about efficiency, ideally, AI should do better than human, or AI plus robotics. The curiosity driven may not always be the most efficient way of going. And I’m relatively optimistic about the feature in engaging curiosity driven part and AI-based efficient processing.
Takeharu Nagai:
Therefore, we have to change the theme from now to beyond AI.
Shuichi Onami:
How the human is intelligent enough to just utilize the new tool, that is, AI is just a new tool, and we will get the advantage of this new tool, and we will find out the difference to interest (curiosity).
Tamiki Komatsuzaki:
As far as I understand AI, it still cannot generate a program spontaneously from nothing.
Jerelle Joseph:
Maybe the opposite is also true. If you have AI to do all these painstaking tasks that you would normally spend hours on, maybe then we become even more ambitious and more curious, and we can think about harder problems to solve.
Feng Zhang:
I don’t think we’re ever going to finish science. I think, as we discovered, we’re going to have more questions. There will be harder problems, though, to solve.
Kumiko Hayashi:
But I have become very lazy after I started using ChatGPT, I become very lazy (everybody laughs).
Feng Zhang:
But you do other thing (everybody laughs).
Shuichi Onami:
You can be more creative.
Kumiko Hayashi:
Yeah, I can be creative, it’s true.
Feng Zhang:
Yeah, but if you think about it this way, say you run a simulation, and the parameters are unbounded for a simulation. There are going to be an infinite number of solutions. And what that really means is that whatever happens with AI or other things, time and future outcomes will always exist. And whatever it takes to exhaust that space.
Kumiko Hayashi:
If there is no AI, probably I would spend a lot of time to run Python programming, for example. But if so, I can save my time, possibly. Yeah, it’s true that I can be creative. But there is a dangerous way. I may read a comic book (everybody laughs). I worry about this kind of tendency for a human being, just to become lazy.
Feng Zhang:
No, I think humans are going to fight the reward system. It’s just about comparing whatever gives you more dopamine.
Is it really a comic book? Or is it making an even bigger discovery? with the help of AI, and it should be different, right? Some people are going to enjoy that little bit of dopamine burst, and that’s enough, and they just keep on getting small ones of dopamine. But then there are other people where this incremental amount of dopamine is just not enough. And they will go to drugs, right? And then there are other people who need to feel that through discovery, to really feel that adrenaline, dopamine. But this is why diversity is so beautiful.
Takeharu Nagai:
So, your tomorrow’s talk! This is a preview for us.
Tamiki Komatsuzaki:
Now I need to move to the next theme, the next generation. We may have a chance to create the next Feng. I’d like to ask you to leave some message to such generation. Can I ask you to start with, Feng? Your story was so impressive.
Feng Zhang:
Will there be some message for a future generation? I think we’ll have to keep them curious. You know, curiosity and just a desire to learn. It’s going to be really, really important. And, just tell young people to stay curious, stay hungry.
Tamiki Komatsuzaki:
How about you, Rong?
Rong Li:
I have two kids, one just finished college, one is in college, and what I want them to do is to chase their dreams. That’s how I can tell them.
Tamiki Komatsuzaki:
Is there any special promotion or recommendation to kids to try to get them, to let them have more curiosity in a field, a scientific field with kids?
Rong Li:
My son is very into environmental science. That’s his curiosity and his drive. My daughter is a budding young author. She’s on her third book. I can’t make her want to do science. So, they can have a satisfying life that contributes to society in their own ways.
Tamiki Komatsuzaki:
Thank you very much and then how about Takeharu?
Takeharu Nagai:
When I was young, I guess I was a bit strange. In high school, whenever I had a question, I always raised my hand. I never really felt hesitant about anything. So maybe people from other countries don’t quite understand the Japanese tendency to hesitate. Most students hesitate to ask questions because ‘not understanding’ is seen as ‘not good.’
So, in that environment, students can’t just say, “I don’t understand what you’re talking about, teacher.” Nobody can say that. That’s why 99% of students don’t raise their hand, even if they have a question or curiosity. What I want to say to future students is: “Don’t hesitate.”
That’s the most important lesson that should be included in Japanese education, from elementary school all the way through high school. I believe every student has curiosity about something, but they just don’t ask, “Why does this happen?”
Do you know why Potassium cyanide is dangerous for us? When I was in high school, I asked my biology teacher how certain toxic chemicals work as toxins, but he couldn’t answer right away. So, he reached out to some of his former lab members from his university days. Eventually, I got an answer from them, explaining, “Here’s how the chemical can block the function or activity of the human body.” Then I asked my biology teacher, “Why can plants turn light into glucose?” He couldn’t answer that either, so he asked his former lab members again.
These kinds of questions are really important for high school and junior high school students. But back then, most students didn’t ask them. That’s a problem in Japan, and maybe it’s something tied to Japanese culture that you don’t see in other countries.
What I really like about this committee, the Biophysical Society of Japan, is that there’s an environment where you can ask anything at the conference, even if the question isn’t that great. It’s such an important environment. But in Japanese high schools, there’s nothing like that. I always ask what might seem like a poor question at conferences, but I don’t hesitate. However, when I attend other conferences in Japan, the atmosphere is different. I’m really sorry, but this seems to be just a Japanese issue.
Ikuko Fujiwara:
Eventually, children will shut down if their questions annoy the teachers. So, I think what we have to teach our children is not ‘do SONTAKU’. How do I explain SONTAKU in English (surmise people’s thoughts)? That must be one of the messages from biophysicists.
Takeharu Nagai:
But AI might help overcome this issue, since it’s easy to ask questions without hesitation.
Ikuko Fujiwara:
Yes, that’s true.
Tamiki Komatsuzaki:
Can we hear your message, Gerhard?
Gerhard Hummer:
The first message is that I think science has never been more exciting than now. Because we’re increasingly thinking about the real world around us, say, how wood works, how muscle works. Up to now, we had to deal with real-world systems at a high level of abstraction. Earlier I mentioned self-healing materials. There’s an explosion of similar questions that are getting into reach.
The second message is to embrace complexity. Don’t try to be a reductionist from the outset. Of course, one will have to be eventually. But look at problems open-minded and try to be as holistic about the problem before narrowing it down.
The third message is that biophysics is maybe the first integrative science. Galvani’s work, for instance, led to the marriage of the physical world and the biological world. I think it has been incredibly fruitful for both sides. Indeed, many concepts in the physical sciences trace back to biological observations. Diffusion and Brownian motion, for instance, were first described as pollen flickering around.
For me, modern biophysics is no longer just a marriage of physics and biology, but a field that increasingly includes computer science, mathematics, and of course chemistry. Essentially, all sciences coming together in biophysics. Biophysics really epitomizes this new way of thinking about the real world, with all the nice and important things that have been said about staying curious and being willing to ask questions. I think young biophysicists will be well prepared to tackle challenging scientific problem and maybe even for societal problems, being cognizant of complexity and issues of stability.
My fourth message is, if you are in any way attracted to this field, even if you may not find a job in academia for whatever reason, which has often little to do with your accomplishments and more with random factors, I think you should be very well equipped to deal with scientific and societal challenges that go far beyond what you might say is the core area of the biophysicists.
Jerelle Joseph:
Well, I agree with a lot of the things that Gerhard mentioned, especially with the fact that within the field of biophysics, there are a lot of unanswered questions. Also, interacting with undergrads and the younger generation I believe that they are very curious. So, I won’t worry too much about remaining curious, instead one should focus on nurturing their curiosity.
From personal experience I find that, given how competitive our field is, one of the most important things has been having good mentors: persons who can champion you, encourage you and help nurture your curiosity. I always like to hear about people’s trajectories and success stories in biophysics and many people that I talk to have had very good mentors throughout their career. I think having great mentors is a huge game changer.
So, I always encourage students, rather than just focus on going to the best schools, thinking about who your mentors are and who are going to nurture you is going to be an important element in helping you to navigate this competitive environment.
Prof. Kumiko Hayashi, The University of Tokyo
Prof. Kumiko Hayashi is a leader of the Biophysical Measurement Group in the Institute for Solid State Physics of the University of Tokyo. In her lab, they develop techniques to precisely measure physical quantities such as force, velocity and energy for proteins and organelle inside cells, based on fluorescence microscopy and non-equilibrium physics. They develop microscopes (hardware) as well as analytical methods (software) using non-equilibrium statistical physics, information science and mathematics. Their aim is to understand cellular phenomena quantitatively by constructing theoretical models using the measured physical quantities. They hope such theories can contribute to the understanding of neurological disorders particularly.
Prof. Gerhard Hummer, Max Planck Institute of Biophysics in Frankfurt
Prof. Gerhard Hummer is a theoretical biophysicist affiliated to Max Planck Institute of Biophysics in Frankfurt. His goal is to develop detailed and quantitative descriptions of key biomolecular processes, including energy conversion, molecular transport, signal transduction, and enzymatic catalysis. In his group they develop, implement, and use a broad range of computational and theoretical methods that allow us to explore the structure, stability, dynamics, and molecular functions of biomolecules and their complexes. They use high-performance computers and work in close collaboration with experimental groups that employ a wide variety of tools, from x-ray crystallography and electron microscopy to single-molecule fluorescence and force spectroscopy. Their computational and theoretical studies aid in the interpretation of increasingly complex measurements and guide the design of future experiments.
Prof. Jerelle A. Joseph, Princeton University
Prof. Jerelle A. Joseph is an Assistant Professor of Chemical and Biological Engineering in the Department of Chemistry at Princeton University. The goal of her research is to determine the principles governing intracellular compartmentalization and to employ these rules for bioengineering. A leading aim of her research group is to determine the physicochemical factors that dictate the fate of biomolecular condensates and to engineer strategies for regulating their properties. Her ultimate aim is to expand the fundamental understanding of self-assembly in living systems and make impactful contributions to bioengineering applications.
Prof. Rong Li, National University of Singapore
Professor Rong Li came from Johns Hopkins University where she served as the Director of the Centre for Cell Dynamics in the Johns Hopkins School of Medicine. She was recruited to NUS in 2019 as the second Director of Mechanobiology Institute (MBI). Professor Li is a globally respected leader in the study of cellular dynamics and mechanics. Her interdisciplinary research integrates genetics, quantitative imaging, biophysical measurements, mathematical modelling, genomics and proteomics — to understand how eukaryotic cells transmit their genomes, adapt to the environment, and establish distinct organization to perform specialized functions. The diverse projects in Professor Rong Li’s lab contribute to two main research thrusts: cell and tissue aging; cellular and organismal adaptation. The insights gained will be applied to the development of new methods for prolonging healthy aging and the repair and regeneration of deteriorated functions, and for preventing cancer associated with chronic inflammatory diseases.
Prof. Takeharu Nagai, Osaka University
Takeharu Nagai is a professor at SANKEN, Osaka University. His group aims to elucidate ‘what is life?’ from the perspective of rare or minor biological components. Conventional scientific approaches generally focus on major components constituting the observed object, often excluding outliers significantly deviating from other data points. Therefore, there has been little accumulation of knowledge on rare or minor components. To tackle this issue, his group has been developing not only fluorescent and bioluminescent indicators for biological functions but also a state-of-the-art optical imaging system, including the trans-scale scope AMATERAS, which enables simultaneous observation of more than 1 million cells with subcellular spatial resolution. By combining these technologies, his group is now deciphering how a small number of elements (such as proteins, viruses, and cells) can create singularities in biological systems. In addition, his group is developing autonomous glowing plants that could be used in electricity-free lighting devices as a technological singularity, aiming for the realization of a next-generation super energy-efficient society.
Prof. Shuichi Onami, RIKEN
Prof. Shuichi Onami is the team leader of Laboratory for Developmental Dynamics, RIKEN BDR, Japan. The development of multicellular organisms is a spatially and temporally dynamic process. A single cell, the fertilized egg, divides many times to generate many functionally different cells, each of which is brought to a specific position to produce complex multicellular structures, i.e. organs and the body. An effective approach to such spatially and temporally dynamic processes is an approach that combines quantitative techniques with modeling and computer simulations. To understand the mechanism of organism development, he is developing mathematical models for developmental systems like the C. elegans embryo, mouse embryo and three-dimensional cell culture systems, by combining molecular cell biology and genome science with biophysics and computer science methods. He also recently worked on standardization of open image data formats and repositories. Nature Methods 18, 1440-1446 (2021).
Prof. Feng Zhang, MIT
Prof. Feng Zhang currently holds the James and Patricia Poitras Professorship in Neuroscience at the McGovern Institute for Brain Research and in the departments of Brain and Cognitive Sciences and Biological Engineering at the Massachusetts Institute of Technology. He also has appointments with the Broad Institute of MIT and Harvard (where he is a core member). He is most well-known for his central role in the development of optogenetics and CRISPR technologies. He explores and studies biological diversity to understand nature and discovers systems and processes that may be harnessed through bioengineering for human well-being.
It was a very exciting trial to dream together our possible futures and possible “phase transitions” in the way to do research in our biophysics, by gathering different disciplines who have been leading each subject in biophysics. We would like to express our deepest gratitude to all panelists and especially to foreign invited panelists, Feng, Rong, Gerhard, and Jerelle. They just arrived that day or the day before that day so that they were under a severe jet lag. The actual time to spend for this round table was four hours including time for dinner, longer than what we had scheduled. The total words we recorded were roughly 24943. IUPAB congress held in Japan before 2024 was in 1978, that is, 46 years ago. How could we imagine the current states of biophysics in 1978? We imagine that some would be within the imagination, but the others should be much more beyond what we could imagine. We are not sure, but we hope that this round table article could be “Jurassic Park movie for Prof. Feng Zhang” for some of our next future generations (T.K.).
