Cell scientist to watch – Gokul Upadhyayula Free

Gokul Upadhyayula

What inspired you to become a scientist?

In my case, it was purely an accident. When I was growing up, I really wanted to study computer science. The dot-com crash happened as I was finishing high school, and the majority of the Silicon Valley companies were playing musical chairs with their jobs, including my dad's. At that point, I felt nudged towards the medical profession. My bachelor's degree was in biological sciences with an emphasis on computer science, and all throughout undergrad, I was aiming to go to either pharmacy school or medical school. But because my family immigrated to the United States when I was 13, I was on an H visa, which meant that I had to show proof of funds to be admitted. Not being in a privileged state to do so, I was trying to figure out what to do next. A friend introduced me to a professor at the University of California (UC), Riverside, in my hometown, and he offered me a position in his lab. That's how I ended up on the route to a PhD. I happily fell into the role of a scientist, and to be honest, I can't imagine myself doing anything else.

What first interested you in studying applied engineering in microscopy for your PhD?

I was a PhD student in the Department of Bioengineering, and my PhD advisor had a background in chemistry. Most of the work that I did was geared towards understanding the kinetics of charge transfer over ridiculously short timescales – femtoseconds to milliseconds. The application of this work was in the context of, for example, polymers for solar voltaics or light-absorbing materials like dyes. It wasn't really until the end of my PhD that I got into microscopy. I started building microscopes because I found a burnt-out, fire sprinkler-damaged microscope body in a hallway, and I was able to convince my advisor to give me a couple thousand dollars to buy the parts to build a reflection-interference contrast microscope. The purpose behind this was to understand the dynamics of protein–protein interactions in the presence of mechanical forces at the single-molecule level, and this type of microscope allows us to visualize these interactions with nanometer scale precision.

For your postdoctoral work, you worked with Tomas Kirchhausen's and Eric Betzig's groups, using lattice light-sheet microscopy (LLSM) to image subcellular and molecular dynamics. What inspired this shift toward cell biology?

I met Tom Kirchhausen at a Gordon Conference where I presented the work that I did with the reflection-interference contrast microscope. At that point, I didn't have any in-depth knowledge in cell biology and I had never even heard of clathrin! I saw the talk that Tom gave and decided that I wanted to learn biology because I realized I could combine every element that I had affinity for – image analysis, math and computer science – to bring quantitative meaning to biological observations. I think I was drawn to that kind of work because I was able to actually visualize what I was studying. About a year into my postdoc, before Eric Betzig's group published their paper on LLSM, we decided to build the first one outside of Janelia Research Campus. At that point, I had never cultured cells before. Tom brought me into his office and said “I know you can do microscopy, but I'm worried that you cannot do biology.” So he gave me a five-minute tutorial on how to plate cells. Once I had learned, I took the trip to Eric's lab. It was the first time I'd been to Janelia, and it was the first time in my life that I realized what it would be like if my time was the most limiting resource. Throughout my academic journey up to that point, the limiting factor had always been resources or equipment. At Janelia, anything I needed was at my disposal. I got to work on the lattice light-sheet microscope for about two weeks to put the finishing touches on the build, test it, benchmark it, and learn the operations and alignments. Then we brought it back to Boston! Looking back, my PhD and postdoc advisors both gave me amazing opportunities and took huge risks by taking on somebody that didn't know much about the field they work in, and I credit them both with changing the trajectory of my career.

How has LLSM evolved since you started working with it?

Light-sheet technology primarily uses two orthogonal objectives to illuminate only a thin slice of the sample, allowing great precision in what you illuminate and where you collect signal from to enable observations in as close to physiological environments as possible. Light-sheet microscopy was revolutionary for studying intact transparent tissues – for example, developing zebrafish embryos – but initially only at cellular resolution. Conventional light-sheet microscopes used Gaussian-type light sheets that allowed control over the thickness of the sheet at the expense of the field of view. Eric Betzig's previous work on the lattice theory made him realize that he could create longer light sheets that were still thin enough to study subcellular details in three dimensions at near isotropic resolution, using non-diffracting beams. This was a key innovation that let light-sheet technology truly go from cellular to subcellular resolution.

The next logical evolution that I was part of was combining LLSM with adaptive optics. In 2018 we published a paper (Liu et al., 2018; https://doi.org/10.1126/science.aaq1392) that achieved the same level of detail in space and time that is typically possible in isolated cells plated on a piece of glass, but inside living, intact, transparent tissues like organoids or zebrafish embryos. If you don't need your sample to be alive, you can also combine LLSM with expansion microscopy (ExLLSM), which we published on in 2019 (Gao et al., 2019; https://doi.org/10.1126/science.aau8302). After that, we met a serious bottleneck in improving the working distance of the microscope. To deal with this limitation, Rui (Ruixuan Gao) and I came up with a new method called photochemical sectioning, which we posted on bioRxiv recently (Wang et al., 2024; https://doi.org/10.1101/2024.08.01.605857). The concept here is that if you modify the conventional polymer crosslinkers used for embedding samples in expansion microscopy to include photocleavable moieties, you can mill the sample, similar to focused ion beam-scanning electron microscopy but for an optical microscope. We can now effectively image as deep into the sample as possible by basically using a ‘lightsaber’ to etch away everything that has already been imaged. For me, this is the most exciting application of ExLLSM, because we were able to generate what is probably one of the largest super-resolution fluorescence volumes ever collected: a volume of two mouse olfactory bulbs totaling about a petabyte of raw data – nearly two petabytes if you include processing intermediates and the final stitch. Now we can image truly nanoscopic detail in tissues at the scale of a cubic centimeter. I would encourage readers to check out the supplemental movies in our preprint, because the data we generate with these types of microscopes is so immense that it's difficult to communicate.

What do you enjoy most about imaging cell architecture and dynamics?

There's no other area of science that will let you observe the dynamic nature of life at the level of detail that bioimaging can and let you see things that you've only been able to piece together in your imagination. We know a lot about how life works, but so much of it has come from biochemistry and genetics using pulverized cells and tissues. The scientists that came before us have provided all these tidbits of how things work, but being able to watch life in action is mind blowing. Imaging is extremely powerful in the sense that it truly tests whether you understand a mechanism correctly. For example, by watching a process in the context of an entire tissue, we can understand how everything ties together. Back in 2016, I collaborated on a project imaging the endolymphatic duct and sac in zebrafish embryos. We were trying to observe the role of the endolymphatic sac in regulating pressure in the developing ear. In one field of view, we monitored endocytic events and cell migration for 11 different cell types. Our movies unexpectedly got completely photobombed by crawling neutrophils, and this blew my mind because I had never seen cells moving inside a living organism in that level of detail before. It was these kinds of movies that made me passionate about wanting to do biology and dive deep into understanding aspects of these mechanisms.

In 2019, you started your own group and were appointed Scientific Director of the Advanced BioImaging Center (ABC) at UC Berkeley. What is your day-to-day like?

The type of work that I most want to do is really interdisciplinary – combining physics and engineering with computational biology. In that context, I don't really fit into the traditional academic pipeline, because I think that there are very few institutions in the world that truly understand the value of that type of work. Around the time I was working with Eric on adaptive optical LLSM and ExLLSM, he moved to Berkeley, and he had a vision for building an advanced bioimaging center there, similar to the Advanced Imaging Center at Janelia but with a key focus on data analysis solutions. I was really interested in being a part of that, so everything lined up for me to apply to Berkeley. I joined as an Assistant Professor in Residence with a focus on building up the ABC.

My day-to-day at the ABC is effectively like running any other research lab. I'm still doing bench work, whether it's experiments, hands-on training people on the microscopes or working with collaborators to generate new reagents. I also spend a lot of time making figures and movies for papers. Being able to do a mix of activities is part of what I love about my job – no two days are exactly the same. I feel very fortunate that I get to direct my passion effectively. I have some added administrative responsibilities to make sure that everything at the ABC is up and running correctly, but I think that every PI has to deal with bureaucracy and administrative red tape in some form or fashion. As a PI, I try to emulate an environment similar to what I received as a postdoc – space to focus and protected time – for the staff and the collaborators at the ABC. That means efficiently dealing with disruptions like network outages or running out of data storage. Not only do I have to deal with the biology and wet lab instrumentation, I have to deal with computing hardware and infrastructure. Lots of little things can stack up!

What questions are your lab most interested in answering?

Thematically, our science is truly interdisciplinary. We develop new microscopes, new computational tools and new sample preparation methods, but the core focus of the lab – the biology – has always been collaborative. Collaborators with biological knowledge come to us and learn microscopy, while we want to understand how our tools can be applied in their field. A biological question that I'm personally motivated to answer, which I carried over from my time at Harvard Medical School, is how nuclear pores assemble at the end of mitosis. In a matter of minutes, roughly 2000 nuclear pores will reform following mitosis to create a gateway between the cytoplasm and the nucleus. Nuclear pores have been studied for quite some time using all manner of microscopy methods, but it wasn't until we used LLSM and high-speed imaging that we were able to discover that these pores don't completely break apart – pore components are actually retained within the fenestrations of the ER, and we hypothesize that they are used as a template to reassemble nuclear pores.

On the technology side, we're currently designing and building a next-generation multimodal optical scope with adaptive imaging correction (MOSAIC) that we internally call the ‘Swiss Army knife’ microscope because it combines the functionality of about ten different microscopes, including LLSM, adaptive optics and structured illumination microscopy. This setup enables imaging anything from bacteria, yeast and cultured cells to organoids, zebrafish, flies and tissue explants, and is by far one of the most complicated microscopes that we've ever designed and built. We are hoping to wrap up a publication about this microscope soon. Another goal is to essentially make some of the hardware built into our microscopes obsolete by using artificial intelligence (AI). We want to replace certain components like sensors with software that can use the Fourier spatial frequency fingerprints embedded in the sample itself. If we do that, we can, in theory, instantly upgrade any microscope to leverage adaptive optical corrections.

As you mentioned earlier, techniques on the cutting edge of bioimaging generate huge amounts of data. These datasets require advanced computational approaches to interpret – how are you tackling the analysis challenge?

People can shy away from using advancing imaging technology like LLSM because it generates so much data. For example, in a single day, we're able to generate tens of terabytes of data, but it takes about a year to analyze that data effectively to extract biologically meaningful insights. That is a huge disconnect and means that every iteration of your experimental process could take months to years. Although a lattice light-sheet microscope is a one-time investment, you also have to keep reinvesting in computing and storage resources because they become obsolete roughly every five years. Because of this, watching LLSM develop has been bittersweet because it hasn't had nearly the impact that I thought it would. However, I think we are missing out on answering questions that we don't even know to ask because this technology remains specialized and inaccessible, and we've reflected a lot on why that is.

That's why we are doubling down and shifting the main focus of our lab to develop computational tools that will make it practical to deal with the massive datasets generated by volumetric imaging. We are all-in on our vision to build a foundation model that is capable of dealing with four-dimensional (4D) datasets. We are thoroughly convinced that unless we solve the analysis problem, the type of work we do at the ABC will not be scalable. It's impractical for one person to become an expert in all areas of imaging, biology and analysis, so we need to effectively put the power in the hands of biologists to both ask and answer questions using these datasets within minutes to hours rather than months to years.

Essentially, how do we build a model that we can put in the palm of a first-year graduate student, using large language models like ChatGPT that can interface with a visualization of a 4D dataset of, for example, a developing zebrafish embryo expressing fluorescently tagged transgenes, so that they can simply ask a series of questions about a biological process? This would require a vast diversity of 4D data and annotations that, as of right now, doesn't really exist anywhere; AlphaFold is probably the only example of a success story that generalized at scale in the life sciences. There are several groups, institutes and foundations that are interested in tackling this idea, and I think now is an opportune time. It's an ambitious project, to say the least, but we have to do something to change the status quo. Five years ago, I would have said that this was still in the realm of science fiction, but enough progress has been demonstrated in terms of AI model architecture, microscope technology and our ability to generate transgenic lines that we have every aspect of this covered. However, the initial upfront cost of this is immense – much larger than any single grant can cover – so we need to convince funders that this approach is the way forward, regardless of who achieves it, because it has the potential to transform the way we do science as a whole.

What challenges have you experienced in bringing together the ideal combination of engineering, biological and computational expertise at the ABC?

We've assembled a team of instrumentation scientists who were able to build two of our next-generation microscopes, and computational scientists who've built our processing pipelines, along with the computing, networking and storage infrastructure from scratch. We're able to handle the nearly four terabytes of data per hour that these microscopes each generate and pair collaborators up with the computational team. We've thrived as a small but interdisciplinary team, but this system is just not scalable given the resources currently available. The reason for that is there's a misalignment of incentives for biologists and computational scientists to collaborate at larger scales.

One of the biggest challenges can actually be finding lab members. Particularly because we're in the San Francisco Bay Area, it is hard to find high-caliber computational scientists who want to work in academia rather than in the tech industry. We have been able to recruit some world-class computational postdocs at the ABC, but they are ultimately interested in getting faculty positions in computational fields, and unfortunately collaborative work that applies existing tools to innovative questions in biology is not necessarily going to help with that. Some of the advancements I've mentioned, especially photochemical sectioning, wouldn't have been possible without collaborations between chemists, instrumentation experts and computational scientists. They're all co-first authors on these papers, but there's still stigma remaining in certain fields where co-first authorship doesn't get you recognized for your intellectual contributions because all that matters is where your name is in the list of authors. To remove that tension, we need to come up with a metric that makes sure hiring committees really value individual contributions rather than just the number of single first-author papers and demonstrates to younger scientists that the whole is greater than the sum of its parts. I think that would go a long way towards making these types of collaborations more fruitful.

What advice would you give to researchers aiming to get more involved in advanced bioimaging, either for biologists who want to use imaging to test a hypothesis or for those who are interested in a career in imaging?

My baseline advice for any researcher is to chase excellence – be persistent in your quest and always be curious. During my early career, in graduate school and as a postdoc, I never really considered becoming faculty. Even towards the end of my postdoc, a faculty position was definitely not on the top of my list of future directions. I just wanted to do cool science and do it well, and this is where I ended up. Maybe there are other places outside of academia where I could be just as effective, but what I value most is the intellectual freedom that academia affords. There are very few places that allow you the freedom to be who you need to be to solve the problems you're interested in.

For biologists interested in imaging, I would advise you to pick the best imaging tool that will get you a reliable answer to the hypothesis that you're testing. Put the hype aside – there is a lot of excitement around each new imaging method, but chances are that the hype will not necessarily help to answer your specific question, so choose an approach to base your work on that will stand the test of time. For those interested in careers in imaging: learn the fundamentals and understand how microscopy, optics and image analysis work deeper than knowing the buzzwords. Learn to always have a plan of analysis, because you can generate terabytes of data, but without a concrete plan on how to analyze datasets, you won't be able to bring out the meaning within them.

Finally, could you tell us an interesting fact about yourself that people wouldn't know by looking at your CV?

I'm an ordained minister in the State of California. I officiate weddings for friends and family.

Gokul Upadhyayula was interviewed by Amelia Glazier, Features & Reviews Editor for Journal of Cell Science. This piece has been edited and condensed with approval from the interviewee.