Image-Activated Cell Sorting with Keisuke Goda

Welcome to the Axial Podcast. Axial is an early-stage investment firm based in San Francisco. We partner with great founders and inventors investing in early-stage life science companies, often when they are no more than an idea. Axial is fanatical about helping the rare inventor who is compelled to build their own enduring business. Okay, we're recording, Keisuke. So, okay, I'm really excited to have this conversation today just to tell your story, and really learn more about your research. Uh, it was great to meet you in Berkeley last week and take a walk. Uh, I think it's been 10 years since you were back in Berkeley. Yeah, but I think it's a really exciting conversation about your journey as a physicist to a biologist and your really incredible research around image-activated cell sorting. But maybe to start off this, like, podcast and conversation, you can just, like, briefly introduce yourself. Where do you want to start?

Okay, I can, uh, introduce myself, probably, I don't know, back to my, uh, childhood. Okay.

Yeah.

All right, so my name is Keisuke Goda, and I was born and raised in Japan. Actually, I came from Sapporo, where the beer and sushi are the best. And then I went to high school in Japan and then came to the US And I spent 15 years in total for my undergraduate education, graduate education, and postdoc research, and so sort of a junior research scientist work as well, and then returned to Japan. So I spent my undergraduate years at UC Berkeley in physics. I was doing neutrino detection. Research, and then I moved to MIT for my PhD work where I was working on the development of LIGO detectors. LIGO stands for Laser Interferometer Gravitational-Wave Observatory for the purpose of detecting gravitational waves from black holes or neutron stars, and that received the Physics Nobel Prize in 2017. My former boss, Rainer Weiss, received the Nobel Prize. And in LIGO, I was engaged in the instrumentation. So basically, the LIGO is, it's a pure machinery of precision measurement. So I was working on the development of precision measurement sensor or detector. And then after I left LIGO, I joined UCLA as a postdoc where I was doing high-speed imaging and spectroscopy and also combined the high-speed imaging spectroscopy with microfluidics for biomedical applications, and still based on precision measurement instrumentation. And then that's when I sort of came up with a nice direction of combining high-speed imaging and microfluidics because there's a great synergy between these two. Their time scales are quite similar for, you know, high-throughput image-based analysis. And then in 2012, I joined University of Tokyo as a professor and started my own group. And then I worked on interdisciplinary research based on microfluidics, photonics, nanotechnology, computational analytics, and so on and so forth. And then I got some large grant from the Japanese government to start the development of image-activated cell sorting technology. And then I sort of, you know, combined my expertise with other people's expertise to demonstrate image activity sorting. So that was published in 2018 in Cell, and then since then we continue to exploit the new application based on that technology.

Awesome. Yeah, you're definitely— so what makes you such a compelling inventor is you, the last decade, you, you know, led the field of image-based cell sorting. And created like the standard, the best device in my opinion. And there's a lot of— there's also a group in EMBL in Stanford I like. I think you probably know that group as well. I like them as well. A guy named Daniel, he's good. But your group and your research and the consortium you built out too is definitely setting the standard in image-based cell sorting. But maybe to back it all up, we can talk about being a physicist to a biologist. And there's a long history of physicists going to biology, right? Uh, Luria, Schrödinger, so on and so forth. And so how was that transition? You're at MIT in grad school studying on the LEGO project. What made you want to do a postdoc in biology? All right, like, what was the big— it was a big leap of faith. What was the logic there?

Uh, yeah, yeah. So I was I have been interested in both biology and physics, and back at MIT, yeah, I was working on LIGO, but at the same time I collaborated with some researchers in the other labs in the physics department. Professor Michael Faux doing Raman analysis, Raman cytometry. Using like intensive lasers for analyzing cells and molecules in a label-free manner. So, I sort of brought the great tradition of ligo technology to the field of cytometry, and then I collaborated, and then after, you know, then the After finishing my PhD at MIT, the detector, LIGO detectors were kind of complete, kind of ready for detecting gravitational waves. So I felt that there wasn't much to do left for me, at least from the point of view of the instrumentation. And then, you know, at the time actually I was at Caltech because this LIGO project was a huge collaboration between MIT and Caltech. I spent the last year and a half at Caltech as an MIT graduate student, and then it was kind of natural for me to move to UCLA because both UCLA and Caltech are in LA. And UCLA is quite strong in interdisciplinary research between biology and physics, so I joined Balam Gyaladis group at UCLA, and also collaborate with Dino DiCarlo, who is an expert in microfluidics. So we, so, you know, we came up with great ideas and demonstrated them.

Well, yeah, I think I grew up in California, so I'm sure you loved living in LA. Traffic stinks, but the weather was fantastic. Yeah. And then Yeah, I think you said Jalali is the expert. Like, I think— is he the chairman of bioengineering at UCLA now or something like that?

Yeah, I think he retired.

Yeah, I retired. Oh yeah. And then Dino DiCarlo is also very, very fascinating and just like the best at what he does as well. Uh, great. And so then, okay, you're UCLA and then you moved to Tokyo to become a professor. Um, that's also very interesting, right? Like, I have a good friend, China, and he was in Japan and then went to college in America. Okay, very rare thing to do, uh, like very rare. So it's very rare for you to go to Berkeley from Japan originally. What made you want to go back to Tokyo? Was it family? Was it, you know, something else? But like, to go spend 15 years in America, yeah, you know, to go back home, was it something you've always wanted to do, or something that kind of came up along the way?

Well, sort of, I was interested in doing different things, you know, that, you know, sort of, uh, contribute with other people. So, you know, after spending 15 years in the US, I kind of wanted a change. So I, you know, I came from Japan, so, you know, I kind of, I felt that maybe I lost some connection with Japan, so maybe it's time to regain the connection. So I went back to Japan.

I love it. Yeah. I mean, you've grown such a big group. You told me 53 people in your group. That's amazing.

That's right.

Yeah, a lot of people. Okay, so, okay, you're in University of Tokyo, um, how did you— I'm always fascinated, especially professors, like establishing your laboratory. You know, your laboratory is an amalgamation of different people, engineers, biologists. How did you set the vision of a lab, and then how did you like manage such a diverse set of people? Because that definitely is a challenge for a lot of new labs, you know, especially to try to like bring different fields together, getting different expertise, and then, you know, setting the basis of the lab. And was the— oh, was the— and then, yeah, and then was the lab always focused on like image-activated cell sorting, or did that come up over time?

Well, I was interested in combining different fields or different areas of research or different methods, different expertise, And image-activated cell sorting is one of them. And of course, we, you know, my group has 53 people, but it's not like all the 53 people are working on image-activated cell sorting. Actually, about 10 people are working on that, and then the other 43 people are working on different projects. And there are lots of projects because out of 53 people, about 40 people are students. And each student has his or her own project. So there are lots of many research projects ongoing and within like several research programs. And I have been interested in interdisciplinary research always, and currently I'm in the chemistry department. You know, chemistry is sort of the interface between physics and biology, so it's actually the right place to do interdisciplinary research. And students here are normally talented and are willing to learn new things and then also exploit the new applications based on new technologies. So it's great. And also I have adjunct positions at UCLA and Wuhan University, so, you know, I do some student exchange in my own research group, but with different countries. So that's quite advantageous.

That's awesome. I mean, you definitely have— maybe show me pictures of your lab and kind of the sorter you have, and it's like such a complicated machine, and you have all these different like experts. One thing you kind of— your research on image-activated cell sorting really relies on machine learning. And then also on engineering and biology. You know, how did you, at least initially when you set the lab up, how did you get those people to join your lab? Like, what was it like? How did you get non-biologists excited about biology? You know, how did you— that's something that's a challenge for, I think, everybody in the field is how do you get people who are experts in AI into biology? And you've had success. On the academic side with your research? So what was that like? Any tactics, any shortcuts, or any tricks you've had?

Students are good. Students are fast learners, so they are really willing to learn new materials very quickly. I'm an older person already, but I'm not good at learning new things anymore. But the students are pretty young, like 18s, 19s, 20s. They're quite good. And I encourage them to explore new things based on new expertise and new knowledge. So that's what I do. And of course, not everyone is good at, say, machine learning or biology, but we work in a team, interdisciplinary team. So one person is good at computational things, another student is good at, say, biology, another student is good at photonics or instrumentation, and then we combine their ideas and methods to do something quite innovative. So I should say that not everyone is good at doing everything altogether. We work in a team.

Yeah, you have to set the culture and, you know, the vision properly. I mean, I'm sure you're really busy, so I know you are really busy, So maybe we can transition towards image-activated cell sorting and the history of it. You have various publications over the last decade in the field. You've really set the standard. Maybe you can talk about maybe starting UCLA or maybe at Tokyo. What was the initial basis of the project? What was the germ of the idea? What was the initial premise for that technology?

Yeah, sure. Well, before, long before image-activated cell sorting, there was FACS, fluorescence-activated cell sorting, developed by Professor Huttenberg at Stanford University. So that fluorescence-based sorting does not have spatial matrix, so it's not possible to identify and sort cells based on special phenotype. So that's one of the major downsides of FACS. And I was thinking, like, how to incorporate imaging capability or microscopy into FACS so that we can isolate, we can identify and isolate spatially unique cells from a large heterogeneous population of cells. So that was, I don't know, like Probably around 2007 to 2009, something like that. And yeah, at UCLA, I thought about it when I was doing high-speed imaging. And, you know, high-speed imaging, as I said, has a great synergy with microfluidics because their time scales are quite similar. So I was combining them, and then I I thought, then I learned facts, and then, oh, maybe it's a good idea to do sorting based on imaging so that we could, you know, sort cells that nobody has sorted before. So it's a great idea. So then, then later on I realized, oh, that requires real-time computational analytics because identification and sorting need to be done simultaneously. So that requires a lot of computational methods such as deep learning and, you know, like high-speed signal processing and image processing and so on, so forth. So, and then, so I learned a lot of basics about, you know, FPGA, CPU, and GPU, and in the electrical engineering department at UCLA. And then after a few years, you know, I came up with like a blueprint of the technology, and then I got, you know, forces from people to demonstrate it. So that's what happened.

Okay, this is like 10, 15 years ago, and like neural networks were very rudimentary. Especially the packages and data processing, so expensive. So you said this blueprint, like 2009, 2010. What were the challenges? You had this idea and what were the challenges to even get it to work initially?

The main challenge was interdisciplinarity, basically combining the different methods and different techniques from different disciplines. So that's quite challenging, not just technically but also culturally, because people in different fields speak different languages. So I needed to understand what they're talking about and also trying to combine those people from different disciplines so that they can work together in a collaborative manner.

Awesome. And so I remember TensorFlow, this machine learning package, was open sourced in 2015 of September. Yeah. And in my opinion, that led to a massive set of a lot of progress, at least on the machine learning side. A lot of people make models pretty efficiently. What was the breakthrough as you made this project? Like, what were the years that you made a lot of progress, and what enabled that progress? Was it, you know, better models? Was it cheaper data processing? Was it, you know, some microfluidic chip got, you know, more accessible? But between 2007 to '09 to present day, you know, you think about the journey of image-activated cell sorting, What were the key moments that made it possible?

I think the key moment was I recruited the right people, the right people who know how to do it, how to demonstrate it, how to sort of implement the blueprint into the real machine. So I led an interdisciplinary team of about 50 people. And so they worked quite well. And so, you know, I was a leader of the interdisciplinary, so I was not actually doing the experiments. I didn't do the optical alignment and coding and anything, but basically I told them to collaborate with each other and also to come up with solutions. To realize that image accuracy. So that's, that's what it is.

Cool. Okay, I did not know that. I'm learning something new. So Keisuke is just an expert leader. You really were able to get people together, inspire them, and find talented people and then get them to work together. Yeah. Okay, that makes a lot of sense. But I would imagine, like, you know, easier AI model last 5 years has made it pretty, you know, lower the barriers for people to, you know, because we talk more about image-activated cell sorting, but like you alluded to, is that lag between ID and sorting. And, you know, if you ID something and need to like send it to some processor and data center, it's too late. And so you think of a catch that cell and sort it like, you know, in like really, really short amount of time. So maybe we can talk about current state of the art of image-activated cell sorting., you know, in your laboratory, maybe other labs, you know, what's the kind of the cutting edge of sorting cells by image? You know, what kind of cells can we sort? How many cells can we sort per second? What's the lag time so far? I think it must be in like milliseconds or like maybe even smaller than that. Maybe you can talk about like kind of what's the, what's the cutting edge today of image-based cell sorting?

Okay, the current, uh, say the throughput, the best throughput or event rate is about a couple thousand cells per second. So to demonstrate it, we need high-speed signal processing, of course, for real-time, you know, image acquisition, image construction, image analysis with and without convolutional neural network, and also decision-making. To activate the cell sorter. So that's needed, and that has to be done within, you know, milliseconds or maybe on the order of 10 milliseconds or so, because cells keep flowing continuously in the microfluidic channel. So we need to sort the right cell at the right timing.

That's That's a key.

Yeah.

Awesome. Yeah, I think also just like, you know, then becomes the complication of the flow rate of the cells, trying to ensure it's like, you know, consistent flow, you know, the image certain blocks.

Yeah.

So there's a lot of moving parts to that. You know, everybody show me some pictures of your device. It kind of looks like a Rube Goldberg machine. What are these complicated things? There's a little thing there and it's all these different pieces to just make it all work. What do you see as the key drivers for growth, like to increase throughput, to increase lower error rates? Is it better devices or is it better modeling? Is it better neural networks?

Yeah, better neural network, of course. But the thing is that this technology is based on image processing or image acquisition. And luckily, you know, we live in the era of smartphone technology. The smartphone technology is a great technology driver. So, you know, image sensor, image analytics, and all technology is evolving quite rapidly. And all we do is to sort of exploit and also bring the new technology, the latest technology from the industry to scientific problems.

Cool. And so, like, if you had to make like a, like, 5 years from now, do you see Thermo getting to tens of thousands of cells per second, 100,000 cells per second? I think FACS can do how much? Is it tens of thousands, FACS? Or yeah. Yeah, do you see image-based cell sorting reaching FACS parity next 5 years or so, or maybe it takes a little longer than that?

So in FACS and also in image-activated cell sorting, there's always a trade-off between throughput, purity, and yield, and yield is sort of success rate. Depending on the application, sometimes we need higher throughput for, you know, identifying and sorting extremely rare cells or purity. Some, some people want higher purity, like, like stem cell applications. They want— they don't want like diseased cells or cancerous cells or like undifferentiated stem cells before injection back into the human body. So the high purity is needed. And yeah, so depending on the application, so we, after the demonstration of the image acquisition itself, so that we kind of stopped improving the specifications of the machine, and then instead we sort of focused our efforts onto the application development. And we collaborated with many biologists and medical researchers to develop new applications to actually increase the body of the technology.

Great transition. I think now we kind of set the basis of what image-activated cell sorting is. We can talk about applications of it because it definitely is like a new tool, and I think next 5 to 10 years we're going to find more and more use cases for it. Um, maybe you can talk at a high level. You know, my opinion, I think, talk about this, I think the highest impact place to work on is cell therapies, but there's also other areas, but maybe you can set, set the, at least the initial tone, like what kind of, what do you see the main applications of image-activated cell sorting?

So there are many potential applications, uh, I can tell you a few. Of course, one of them is cell therapy, and for example, in my group, we do a T cell sorting, T cell identification and sorting. Based on immunological synapse. Immunological synapse is a kind of a biomarker of immune ability, immune response to pathogens. So the immunological synapse gets localized in the T cell, and we identify it by image acquisition. So to identify the cell T cells have higher ability of immune response. Yeah, so that's what we do. So that's good for high-throughput screening before injection back into the human body in CAR-T cell therapy and many, you know, like induced pluripotent stem cell therapy and so on. So that's one of them. Another application is yeast-based analysis, and yeast is a good species and model organism for microbiology research, and as well as in food science. Yeast is good for developing wine, sake, beer, bread, and miso, and soy sauce and stuff. And there's a great correlation between the morphology of yeast cells and the taste of these drinks, and also the foods as well.

So you can show me a cool picture of yeast and sake, and then yeast in like beer, and you can see the pictures of the yeast, and it definitely is like pretty significant morphological differences. So something I didn't even know about, or it's pretty, pretty cool.

Oh yeah, but so far it has been quite difficult to isolate yeast cells with particular morphological traits, right? So now we use this image-activated cell sorter to isolate those cells. That means we can control the taste of the wine or the food.

So that, that's interesting.

That has not been possible in the past.

Absolutely. Yeah. And you also— is there applications in cyanobacteria, if I remember correctly? And then some stuff in diagnostics as well. But what's kind of been the— and then maybe you can talk about potential applications, like any progress. I know you lead and you help organize a massive consortium around the world of, I think, 20 laboratories at least, probably more. Maybe you could talk about that consortium and, you know, how it was like to set it up and what they do, what they're doing now.

And yeah, sure. Yeah, so after the publication of that paper in Cell in 2018, many people got interested and they sent me emails about the potential collaboration, and then we work on that collaboration ideas, and then we started the collaboration. And then because many, you know, ideas came, so we started this sort of open innovation platform for collaborating with these people. And that was around the late 2018. And then unfortunately, COVID-19 came, and then we kind of had to stop. And then people couldn't come from overseas. So we had to do the experiments on our own, that, you know, based on their suggestions and advice. And also they sent cells and stuff, and we, of course, the kind of what we could do was kind of limited, but still we kind of managed to do some experiments. And then, you know, now that the pandemic is pretty much done, so we are resuming that the Open Innovation Platform. So more people are coming to my lab to do collaborative research.

Cool, that's really exciting. I think there's some cool use cases of profiling T cell exhaustion and trying to find new markers for it. And I think, like I said, I think Kristin McCall at Stanford, she does like the best research. I think there's probably a collaboration between you two that align. But maybe we can talk beyond image-activated cell sorting. So, you know, you think your lab is moving, like you said, you have majority of your laboratory working on some other things. So maybe we could discuss what those things are and what the future looks like for the GOTA lab. I think at a high level, you're trying to merge— your research has been trying to merge different tools together. And so maybe for the, you know, maybe for the first 10, 15 years, you've been merging microscopy with single cell, and you've developed all these tools to do that. And then you have sequencing too. So now in the next 10 years, it seems like you're trying to merge microscopy, single cell, and sequencing, right? So, you know, kind of any plans there? What's any projects ongoing? Any kind of things you can talk about? And we can always do an update conversation down the line and we can say, hey, how's it going? But you kind of, you're trying to merge this trinity of biological phenomena or something into one?

Yeah, so in my opinion, the biology is sort of split into three domains: flow cytometry at the population level, microscopy at the cell level, and sequencing at the gene level. And people talk about biology using three different languages. So I was interested in sort of unifying different domains of biology so that we could have a better understanding of biology, especially and temporarily. So image activity cell sorting is in a way, it's a machine that integrates flow cytometry for the statistical level analysis of cells and microscopy for the single cell analysis. So I sort of unified these two, and eventually I'm interested in combining all those three domains of biology by incorporating the final stage, the sequencing, so that we could have a bigger picture of biology by combining flow cytometry, microscopy, and sequencing. And because these things, these domains work on the different scales of time and space, I think human brains are not so good at sort of understanding them simultaneously or like in a comprehensive manner. So maybe then, you know, AI or, you know, computation techniques can be quite helpful for sort of interpreting the information from the different domains.

Cool. And so, like, what do you see the main challenges in terms of, like, you know, having an ability to identify a cell and sort it? Like, and then to sequence it, you need to be able to— I mean, you probably already sequence cells right now. You get— it's not like that hard, but like When you say sequencing data, is it just as simple as sorting a certain cell type and sequencing it, or is there something more special to it? Is it like trying to capture spatial information, spatial sequencing? Or when you really say sequencing, what do you mean? Is it sequencing proteins, or is it capturing the 3D genome, like Hi-C? When you really say sequencing, what are the challenges in terms of what you're Real Vision is beyond just like simply sequencing the genome itself, which I think you can do now.

Yeah, sequencing itself is not too difficult, and also it has been evolving quite rapidly by other people. So my job is to integrate the image acquisition software and sequencing machine. That's kind of challenging, technically challenging, because we need Sorted cells are kind of fragile. We need— and also they're contained in a microfluidic chip, so I need to make sure that I transport those sorted cells to the sequencing machine without losing any cells. Yeah, so because the loss of any cells will lead to like misinformation and misguided signal acquisition. So that's kind of— it could be a minor thing, but technically challenging.

And when you see— when you think about the impact of like being able to image a cell, be able to capture it, and then be able to sequence it, what do you think is that— what kind of— what do you think if you invent— if you create that tool, what do you think kind of new discoveries could be made? There's already been research using image-acquired cell sorting to, you know, discover new genes involved in some sort of cancer pathway or growth pathway. But if you can merge microscopy, cytometry, and sequencing together, you know, what kind of new areas of biology could be probed potentially?

My honest answer is I don't know, because we explore new sort of unexplored areas. So what I do is data-driven research, not hypothesis. Driven research. So I don't make hypotheses because, you know, there are a lot of smarter people out there who can make great hypotheses. So I'm not a biologist at all, so I don't make hypotheses. So, but I generate data, and data can be analyzed and cooked in combining different domains so that we can makes serendipitous discoveries, the kind of discovery that other people did not expect, even smarter people.

I think that's great. And so maybe to tie it all together, you know, you've been— you're a physicist, get into biology, you've invented all these tools. You know, if you had to step back and say, look about your career, look about what others have done, What do you see as the big long-term opportunities in genomics imaging, maybe even beyond biotech biology in general? We definitely live in a golden age from a technical perspective, right? There's new tools on sequencing, imaging, gene editing. It's amazing. Each week there's a new great paper out. And so what do you see in the next few decades? Where do you see biology going? Do you think there's some substantially, yeah, like what excites you?

Sure, yeah, I think each technology is quite mature now, like sequencing technology or imaging technology or flow cytometry technology and many other technologies are quite mature, but combination of these technologies to develop or to extract new domains of information is challenging. And also interpreting these, this kind of data on different scales of space and time, like millisecond to say hour, or micrometer to kilometer, you know, combining all those different like scales, different scales of space and time. Is quite challenging, and the interpretation of those, that kind of information, would probably lead to some unexpected discoveries.

Interesting. So like maybe trying to figure out how to translate all the different datasets you're producing into kind of something common, and then you might be able to like discover some of those things. I think it's really exciting, maybe to like kind of finalize things You know, you've also, with the incredible research you've done, you've increasingly got involved in spinouts. Yep. Um, and, you know, becoming an entrepreneur as well. How's that experience been? I mean, I know you're really early and then you have another decade to go, but what got you interested in starting companies? And, you know, and can you maybe discuss some of the companies and your experiences so far?

Sure. Uh, so I want to make sure that the technology is actually being used by people. So not just publishing papers. So, you know, people, you know, other scientists and other medical researchers or medical doctors, they take advantage of the new technology and do something innovative or, you know, that cure diseases and save patients. That's important. So, you know, translational science or translational medicine is sort of my secondary goal as well because it's kind of important. And then after the transition is done, then I come back to basic science and then start to work on something new.

What's been like the What's been like— has it been when you start off on that journey to start companies, we talk about languages, right? Translating the language of cell and DNA or something, sequencing and microscopy. Has it been more challenging to translate the language of science and business? How did you learn the language of business? Small thing. Has it been difficult? I think a lot of scientists in your position, they have an invention and they're trying to learn business. What kind of were your tactics to get kind of empowered and learn a bit more about startups and stuff?

I don't think there's a secret. I just make friends with business people and also make friends with medical doctors and also read magazines and newspapers. That's what I do. Yeah. Not on a daily basis.

Yeah. Okay. I think it's a great way to end it. Just make friends. I think I'm a true believer in making friends because it makes— I think, I think it's a whole lesson of your story is, right? Like you had a blueprint, you made a lot of friends, and you know, you all worked well together and you made some incredible machines. And I think the next decade, I, I, I know you're gonna do some really incredible work and I think I'm excited to read up your research as it comes out, but I know you're really busy. You have a hard stop. And so it's great having this conversation. Really appreciate it. And we'll do this a few years from now. We'll do an update because I'm sure you have a lot more to talk about. But Keisuke, I really appreciate it. This has been really awesome.

Thank you very much.