Science Speaks: Conversations on Health Podcast - Keeping Pace on Virus Evolution

Keeping Pace with Virus Evolution with Lester Frei and Dr Beichen Gao

[Vicky Edkins]

Hello, welcome to the Science Speaks: Conversations on Health podcast. This podcast series has been developed by the Basel Research Centre for Child Health in Switzerland to share some of the great work that is taking place and to look at what it means to be a scientist within the various different disciplines that are represented by the Centre. In this episode, we are delighted to be talking to Lester Frei and Dr Beichen Gao.

Lester is currently a PhD student in Professor Sai Reddy's lab for Systems and Synthetic Immunology at ETH Basel. Beichen was previously a PhD student in this same lab and now works as a data scientist at a spin-off company from this lab called EngImmune. Lester and Beichen, thank you very much for joining us.

[Dr Beichen Gao]

Thanks for having us.

[Lester Frei]

Thanks for having us.

[Vicky Edkins]

So in this podcast, we're going to be discussing protein engineering, a technique that is central to the work that both Beichen and Lester carry out in the development of antibody therapies for a range of diseases.

Beichen, perhaps you could start by explaining what protein engineering is and what it achieves.

[Dr Beichen Gao]

Yeah, a good starting place to think about proteins is they're like little Lego blocks that's everywhere in nature that has specific functions that are crucial to life. And what we know about proteins is they're built from these 20 amino acids and things in nature. And with these amino acids, how they're built, it really dictates their final shape, their function.

And actually the goal of protein engineering is taking what we know about the shapes and functions of proteins and really pushing that to really create new proteins that might have different functions or actually improve certain proteins to sort of work better. So, you know, for example, usually a lot of the things we look at is binding. So proteins, if you think about it like a certain pocket, again with the Lego piece idea, you can make a square pocket that could basically accept, I don't know, rectangles, squares, some other shapes. But what we can do then is to go back by changing the amino acids, really turn that pocket into maybe a circle and then that'll accept something else. And so that way we've engineered a new protein that can work in another biological process.

[Vicky Edkins]

So when you talk about protein engineering and the kind of reshaping of the proteins to perform different functions, you mentioned amino acids. What is it that you change about the kind of configuration of the amino acids in order to carry out protein engineering?

[Dr Beichen Gao]

Yeah, so actually that's a good question. What we normally aim to do is by changing in different amino acids. So let's say it's always based on a certain sequence, right? So we can have A, B, C, D, which is what's seen in nature. But instead, you know, maybe we can create a different function by making it A, B, E, F, as sort of two new amino acids. And by doing that, we can actually create these changes in the final protein to do what we want.

[Vicky Edkins]

Okay, great. So how was this technique developed? Kind of what is the history of this? When was it invented? And what's happened since that first inception of the idea?

[Dr Beichen Gao]

Yeah, I mean, Lester, you can...

[Lester Frei]

Yeah. So protein engineering really started, I would say, about around 35 years ago.

And so what we haven't touched on with proteins is that they are encoded by the DNA of a cell. So what this means is that a cell, like all the cells in our body, they will just read their own DNA and then produce the corresponding proteins based on this. And this is important because it's not possible to directly sequence proteins.

So what we do in protein engineering is we rely on sequencing the DNA and thereby then reconstructing the protein sequence based on this information. And so really how it started out, protein engineering, is people came up with very clever ways on how to link the DNA sequence and the protein sequence. And based on this idea, people have developed several different ways of doing this.

And this is just like in general, it's called different display platforms. So when we talk later on about display platforms, what it really means is it's just a way to link the DNA information and the protein information together. And yeah, I mean, over the past like 35 years, obviously there has been a ton of innovation in this field.

Also, very crucially in 2018, the Nobel Prize in chemistry was partly awarded to a display platform. And yeah, to highlight how this works and what we actually mean by the different display platforms, let me just give you an example with yeast display, what we mostly work on in the lab.

So first of all, for yeast display, or I guess for any display platform, we construct a library. And what we mean by library, it's just a collection of different DNA pieces, which encode for different proteins. So like what Beichen mentioned before with the simple example of, you have one protein that's like ABCD, and then a second protein is ABEF. And just like the sum of all those different variants is what makes the library.

And so yeah, so we then for yeast display, we put this library into the cell. And then we just use the natural machinery of the yeast cell to produce the protein variants. And then from this pool, we fish out variants with novel and improved functions.

So I guess back to Beichen's example, like if we want to have a new variant that instead of binds a square pocket, binds a round pocket, we can isolate this from this pool. And then ultimately, we just get out the DNA, sequence it, and then we know which protein does the job.

[Vicky Edkins]

What was the very first application of this technique? And can you also talk about kind of where we've gone from there, some of the other applications that this technique is used for to kind of really provide some context.

[Lester Frei]

So one of the early applications, and that's still used today is to find novel therapeutics. So one really big therapeutic that was developed early on using display is an antibody called Humira. And it's used for to treat rheumatoid arthritis.

And this is like a blockbuster drug. Like in 2021, there was 1 million prescriptions. And it was so groundbreaking for treating rheumatoid arthritis, that actually that was in part the reason why the Nobel Prize was awarded for this technique.

And so that was like back in the early 90s when that happened. Still today, a lot of display platforms are used to find novel antibody drugs.

[Dr Beichen Gao]

Yeah, I mean, we see these in not only antibodies, but for example, I mean, even other industries like food also actually use a lot of protein engineering that people might not have expected, right? Synthesising new, well, this sometimes sounds a bit bad, but synthesising new chemicals to use in our food. But, you know, of course, for things that are perhaps a bit more natural and potentially less harmful and more effective.

So yeah, there's a lot of times where protein engineering is used, but we don't really see it or are aware of it.

[Vicky Edkins]

We don’t necessarily know that it’s there.

Can we go back to antibody engineering? Can you just kind of briefly provide an explanation of what an antibody is and what it does?

[Lester Frei]

So antibodies are a very crucial part of our immune system. Like just naturally, they already float in our body. And they're these proteins that are really geared towards binding targets.

So for example, if you have like, say you have a flu infection, then your body will recognise this and it will produce antibodies that effectively bind this virus and then thereby neutralise it. So it's naturally used by the body to defend itself from external agents.

And what people then did is basically, or I guess in antibody engineering, we just take this framework of an antibody and we just re-engineer it so it binds like novel targets that are not already naturally bound by the body.

[Vicky Edkins]

So you talked a little bit about using protein engineering for like food. You also talked about it a little bit for antibody engineering, particularly in relation to treating rheumatoid arthritis. Are there any other kind of uses that kind of spring to mind that you could talk about to give us a bit more of an idea about the way in which this technique has had implications for people's lives?

[Dr Beichen Gao]

Yeah, I think for me, one of the things that often come to mind is sort of new antibiotics, because this is also one of the sort of pressing issues that we have in the world with the need to discover sort of novel antibiotics that can counter what we're seeing with antibiotic resistance and a lot of diseases that we're having right now. So again, a lot of these novel antibiotics themselves are just little proteins. So they again have different shapes and their goal is whether if you wanted it to enter the bacterial cells and sort of destroy a particular mechanism that's really crucial for their survival and growth, we're engineering that.

There's also some that will also specifically target certain bacteria. So right, one of the things you have when you take these heavy, broad, we call them broad-spectrum antibiotics is you take it, they go into the gut, they basically kill everything that's in there. So you have sort of all of these side effects that are quite unpleasant.

But the goal of engineering these new antibiotics is they can go in and only kill the bad ones that you don't want and sort of keep the good ones there alive and thriving. And again, that requires this engineering for that specificity there.

[Lester Frei]

And yeah, another I guess everyday topic would be allergy to cats. So the way people who are allergic to cats is because cats secrete a immunogenic factor in their saliva. And so when they clean themselves, what happens is they spread it to the fur.

And so what people have done is they have engineered antibodies that target those factors that elicit the allergic response and thereby, again, with the shape, they made an antibody that specifically recognises and binds the shape. And so what people did is just they mix it into the cat food. And so ultimately, it neutralises this allergic component and thereby making the cats less allergic to people.

[Vicky Edkins]

That's amazing.

[Dr Beichen Gao]

Like it's a blockbuster cat food.

[Vicky Edkins]

So we've talked about lots of different uses for protein engineering. And I think we're going to focus more in on antibody engineering for the remainder of our discussion because I believe that is a technique that you use in your lab and in your research.

Can you provide a bit more of an explanation of what that technique actually involves? You know, what is the method? What is it that you actually do in your lab when you're engineering proteins for the purposes of antibody production?

[Dr Beichen Gao]

Yeah, so in our lab, I mean, we've got quite a few projects always ongoing that's around immunoengineering, as we call it, or synthetic immunology. I think one of the good ones that comes to mind is, for example, we had a previous PhD student in the lab as well, Dr. Derek Mason, if he's listening, hello. So there, what he took is a really popular antibody that's used in breast cancer called Herceptin and basically wanted to explore, okay, what can we do with, you know, protein engineering platforms, for example, like yeast display that Lester introduced to sort of find based on Herceptin to make it better.

And basically, how we would do that is, all right, so then we're talking about the libraries, we can start with Herceptin, we can make these sort of really big libraries where we have, you know, mutations across this sort of recognition part in the antibody, right? So we start to build again, like the square pockets, the triangle pockets, the circle pockets, something.

And, you know, sometimes these libraries are huge, they can be in the millions, the billions, if not more, right? And having these libraries in hand, what we can do is then we go and we call them, select them, right? So one of the things we wanted to do was to have really improved specificity. So Herceptin targets HER2, it's a different receptor, that's really a major target in breast cancer, right? So we actually just combine our library that we have with the HER2 protein. And again, we can fish out what's really, really specific, and we leave those unspecific ones behind.

And then we can improve it sort of next steps of looking for maybe those that are stable under heat, certain other conditions, we can then also test that with, you know, that library. So we're coming from trillions, maybe we get down into 100 million. And then from that, we get down into the thousands, and sort of, we could engineer it that way, experimentally.

[Vicky Edkins]

What does that process of kind of getting down the number of different possibilities involve? How do you do that?

[Lester Frei]

Yeah, I mean, there's different ways of doing this. But I guess that the easiest way to think about this is we call it sorting. It's really like you interrogate every single variant on their ability to, in this case, like with Herceptin, to bind the cancer component. And thereby, it's a very effective way to separate the two populations of the ones that do bind versus the ones that do not bind.

[Vicky Edkins]

So is it an experimental process or a kind of computational process?

[Lester Frei]

So, I guess this stage is like all purely experimental. But yeah, I mean, then in the next step, it's really where the computational part comes in. So when we have those, like the sequence, the variants that still do bind, then that's really the point of where the whole machine learning and the AI comes in.

Because what Beichen mentioned before, it's like you have those enormous libraries. And really these days, the limitation in the whole protein engineering field is that you can only look at so many individual variants. It's like you are limited, even though the whole sorting process, it is very efficient.

But there is just an upper limit to what you can like feasibly do experimentally. And yeah, that's really the point where you take the information you gathered from the experimental part, and you take it and you feed it into the machine learning models.

[Dr Beichen Gao]

Yeah, machine learning is kind of this broad term that's getting thrown around in I mean, every discipline where we're using machine learning. But basically, what it comes down to is it's where we're creating these algorithms, some of them are really complex, some of them are quite simple.

But the goal is, right, I've got 10 different objects, 10 different sequences that we have for proteins, five of them are good, five of them are bad, right? We're going to challenge the machine learning algorithm to say, all right, can you figure out what are sort of these patterns that's in the sequences that might be more associated with sort of, in our case, binding than the non-binding proteins?

[Vicky Edkins]

So you're basically kind of feeding, correct me if I'm wrong, you're feeding your machine learning algorithms with some data, they use that data to learn. And therefore, they can then predict which of your kind of combinations or structures of proteins are going to perform the function that you want them to perform in the best possible way. Does that kind of summarise it?

[Lester Frei]

Yeah, it kind of loops back to what I mentioned earlier about the, just the limitations of how many individual variants you can look at experimentally. It kind of bridges that gap, because the machine learning can then predict like millions and billions and trillions of variants that you just cannot, it's not feasible to actually experimentally screen. So you kind of, you try to overcome this, like this limitation.

[Vicky Edkins]

So you were both part of a team of researchers that received funding from the BRCCH back in 2020 to carry out research in response to the COVID-19 pandemic. And I bring this up because I know that this project that you're involved in used protein engineering in kind of an attempt to respond to this new virus. Can you explain what the kind of the rationale or the need for that research project was? You know, what is it specifically about the COVID-19 pandemic that we didn't know, the problem that, you know, we had, that required you to carry out that research?

[Lester Frei]

I mean, I feel like there's nothing necessarily unique about the COVID-19 pandemic. It's a lot of, the virus has a lot of other characteristics, like with other viruses, like it mutates super fast. I guess what's really novel about it is was that it's like such a global, that it was such a global phenomena.

And yeah, so I guess one big problem was that, you know, the virus mutates super fast. And especially also that the thing is, if you, if you have a new antibody therapy that you want to use to protect the populations at risk, what happens is if you use this antibody, it will just favour the virus to mutate around it. So then to just evade this antibody. Which made it so that all the antibodies that were approved, really super quickly again, lost their efficacy.

[Dr Beichen Gao]

Yeah, it was kind of like playing that. What's it called? The Red Queen hypothesis?

[Lester Frei]

Basically, that you always have to run to stay at the same place. Like you always, you have to continuously put in effort just to remain at the same place. Otherwise, you would fall behind.

[Dr Beichen Gao]

Yeah. So with COVID-19, at the start, we didn't have any immunity against it. And that's why it was so bad.

But then as we developed an immune response, the virus would then start to mutate even faster to escape our immune response, right? And yeah, like Lester said, we started to see all of these rapid mutations start to emerge. It was also the first time that we started to really have the tools and the amount of data to sort of, we call it surveillance, right?

To watch it mutate throughout the months, throughout the years and throughout all the different, you know, countries. And yeah, the main question there was, okay, as it's continuing to evolve and escape our immune response, then what can we do if we're looking for antibody-based therapies? How can we make sure that we're finding ones that won't be escaped immediately?

Some ones that won't, you know, lose efficacy in six months' time, which then we need to find a new one and start to develop it again and produce it and get it out into the public. And then by that time, maybe it's mutated again and that's also rendered useless. So yeah, it's kind of playing that game.

[Vicky Edkins]

And that's where your project came in. I guess.

[Dr Beichen Gao]

Yeah.

[Vicky Edkins]

So how did your project respond to that need? What was it doing that was, you know, unique or different? Or, you know, you were given these capabilities for the first time in a pandemic because of the capabilities you had around protein engineering. How were you able to tackle that problem?

[Dr Beichen Gao]

I think a good way to describe it actually is we've basically flipped what we normally did on its head. So normally what we would always do is we have a single target, like again this HER2 for Herceptin, and we sort of make a library of antibodies and try to figure out what's going to bind better against that single target. But instead here, we're basically taking a set of antibodies that have shown clinical efficacy. At the time, some that were basically a bit more experimental, and then we made a library of sort of the SARS-CoV-2, the COVID-19 virus' spike protein on its own. So not the live virus or anything, but just that spike protein and the specific part of it that's targeted by antibodies.

[Lester Frei]

So yeah, we just took the main component of the virus and we just put it on yeast. And then we went from there. We used the method that I explained earlier with yeast display.

[Dr Beichen Gao]

And from that, right, we could get a subset of variants that escape antibody A. We have another set of variants with mutations that potentially escapes antibody B, C, D, right? So after a while, we get this sort of data set of, okay, which mutations are potentially going to cause escape against maybe one, two, or a bunch of different antibodies.

Right. And this, I think, was kind of the new thing that we did because at the time, there were other groups that had made sort of, again, yeast display libraries of sort of the spike protein. But what they've reported mainly are just the effects of single mutations, right?

So let's say if we have a single amino acid that's changed. And they did find some crucial ones that really like with only a single mutation, one of the early antibodies would basically completely lose its efficacy. There was like over 100-time reduction in its sort of neutralisation and its protective abilities.

But the question was, all right, as we start to get more and more mutations, what's going to happen, right? Because it's not always going to be a single mutation that causes escape, but perhaps a combination of three that really causes a change in its shape that then causes a lot of escape, right? And what we could do with our data set was to potentially see that, which I think was really cool.

[Vicky Edkins]

So we've talked quite a lot about the amazing research that you did. And by the sounds of it, it was very impactful, made a huge difference to the COVID response. We haven't really looked at all about kind of who was involved in that. Obviously, you both were, but you know, who else is part of the team of researchers that carried out that research? Who else contributed to it?

[Dr Beichen Gao]

I mean, these were pretty big projects. We had, so for the latest publication on the computational part, we had some of the other PhDs in the lab. Jiami, we had some, actually a lot of master's students who have come by and helped us out at little steps along the way.

And on the experimental side, I mean, mostly Lester. Lester led a large part of it. But we also had a postdoc, Eddie, Dr. Eddie Irvine, who came in and helped out.

[Vicky Edkins]

And what about your kind of your background and your route to where you are now, Lester? You know, where did you start off and kind of what was your trajectory in order to end up as a PhD student in Sai Reddy's lab?

[Lester Frei]

So for me, I guess the trajectory really started back in high school. So I was really interested in natural sciences in general. Early on, I was thinking about going on to study chemistry. But then when I was like learning about molecular biology, I was really amazed by it. Like, I found it really interesting to learn about the molecular components that make up life.

And so, yeah, for me then, so after high school, I went to ETH to study biology first. And then later on, I did my master's in biotechnology here in the department in Basel. And yeah, so then I ended up doing my master's thesis in the lab. And I just stayed on after that.

[Vicky Edkins]

So it's a good place to be.

[Lester Frei]

Yeah.

[Vicky Edkins]

And what about you, Beichen? Kind of what's your story to where you got to now?

[Dr Beichen Gao]

Yeah, for me, my background is also more wet lab. So, you know, I went to uni back in Canada, London, Ontario - so second-best London in the world. And there I studied physiology and pharmacology because I was really curious about, already about drugs and how they worked in our bodies.

And then went on to do a master's in molecular biology. And then started working actually in a company where we were sort of developing and finding antibody therapies. And there I was using these display techniques.

So not yeast display, it was a different thing called phage display, which we won't get into, but it's similar. Same, same, but different. And kind of there, I basically saw, okay, in the process of doing the experiments. Yeah, okay, we would still sequence the, you know, the variants, the antibodies that we would select at the end. But I felt like we were just throwing data away.

And I was really curious about what we could do if we just took everything, right? Every step in the selection, we took the good, the bad, and maybe everything else in between. What if we could train these sort of machine learning models? And how much better it could be to help our discovery efforts.

So then kind of from there, you know, reached out to a few labs, ended up finding myself in Sai Reddy's lab, and kind of working on a few projects there that started off experimentally.

But then, you know, at the very end, I was pretty much just at the computer, doing all the machine learning models and all the computational components, which is just still good. That's exactly what I wanted.

[Vicky Edkins]

And in that, in the kind of the time between when you finished your master's and you were working, and you came to Sai Reddy's lab, kind of what, how did the space change? Because I think machine learning has grown as a discipline as, you know, its capabilities have got markedly bigger now. How have you seen that development? What's the scale of that?

[Dr Beichen Gao]

Oh, God. I mean, the way that I think about it, I think, for you, probably the same, Lester, is in protein engineering. If we go back to that, it's almost like pre-alpha fold and post-alpha fold.

So for anyone who's listening, who might have not heard about this, basically one of the biggest challenges in protein engineering is, yeah, we have all of these sequences. But one of the things that we really need is to know how these proteins look like, right? Because there is still that underlying structure that we could, you know, we could see, but it was really difficult to do because you had to sort of purify. You had to crystallise these proteins. You had to put them under these like crazy microscopes.

And it was always a challenge. You could really never predict protein structures. But then back in 2020, late 2020, you know, alpha fold came onto the scene. And basically it was this model that was trained on all protein sequences that we've ever sequenced and seen. Not only humans, from all species, plants, bacteria, animals, everything.

And what we found was it was probably the most accurate predictor of protein structure. And it was, I think at the time there were these competitions where most models would perform at like 40, 50% accuracy, if anything. And this one was like 85% accurate.

Yeah, I guess that's kind of the exploding point in the field. I mean, before that, it was still kind of exploratory. A lot of publications would still use it, right? It was really useful for a lot of engineering projects. But alpha fold was the turning point.

[Vicky Edkins]

And given how much it's grown in that time, Lester, what do you think, you know, the next big thing is? Where can you see it going? This is, I know, quite a difficult question to ask.

[Lester Frei]

I mean, there is a few really big fields, I feel. Like one of the big fields, I think will be some sort of library-on-library screening. So what we only, what we described here is with yeast and how pretty much all or almost all display platforms work is you have like say your yeast and you have your protein that you have in yeast. And then you use your breast cancer target, your autoimmune target. And then you use that to find novel proteins that bind to it.

But it's always just like many versus few problem. So like in yeast, you have tens of millions of variants. And then you have like one breast cancer target. And so one field of interest and active research is could we have systems where you have two libraries?

So say one is your antibody library and the other is your COVID library. And then you would do an all-by-all screening at the same time. So instead of having to do massive efforts to only one target at a time, you could test millions.

[Vicky Edkins]

So it's true to say that the field continues to evolve and develop?

[Lester Frei]

Yeah, definitely.

[Dr Beichen Gao]

Yeah. Not to mention, I think the other thing that I that immediately came to mind, again, on the experimental side is, is actually protein sequencing.

So one of the restrictions is we have to sequence everything, you know, genetically. So with the DNA. But one of sort of the new things that's currently being investigated is can we do the same but with the 20 amino acids?

So instead of the four bases that we see in DNA, we've got the 20 amino acids. And is there a way that we can directly feed in the protein and just have it read out what this would be? Because that would also solve a few bottleneck issues that we have in the field. And would sort of give us a lot more, a lot more data to work with.

[Vicky Edkins]

What difference do you think these developments are going to make for society at large?

[Lester Frei]

I would say like, there will be a lot, you'll have like, it probably will fuel like, drug discovery. So having more and better therapeutics. Also against really difficult targets. But also, you know, even like outside of, I guess, what we in our work focus on, there's a lot of other fields where it will have a huge impact, like for biofuel production, for greener and more sustainable processes, especially like industrial process, where being able to like substitute, like really toxic chemicals, with novel proteins, will also be like, you have, will just make the whole process more sustainable.

[Dr Beichen Gao]

Yeah. And it'll, it'll speed things up massively. I think, right, we're seeing, I mean, about a decade ago, we were still sequencing out of the libraries that we had constructed and, and evaluating clones individually. But now we're scaling up to be able to test out thousands. And then plus ML, we can, we can test out even more. Right. And I can only see it sort of improving from here, getting much, much faster. So in terms of engineering all of these new proteins.

[Lester Frei]

And I feel like in 50 years, people will look back at what we do now. And they will, they will wonder, how, how did we get anything done? Like, how is it possible to get anything, any new, like antibodies and any data? So yeah.

[Dr Beichen Gao]

What do you mean you had to do all of these rounds of sorting and  it took took you 6 weeks to get.

[Lester Frei]

It's the way we look back now at the 1970s and wonder, how, how is that possible? Like, how could they, how could they get anything done? But it worked.

[Vicky Edkins]

But without that, we probably wouldn't be where we are now. Beichen and Lester, thank you both very, very much for taking the time to talk with me. I really enjoyed our conversation. And it's been a real pleasure discussing your research today.

[Dr Beichen Gao]

Thank you.

[Lester Frei]

Thank you.

[Dr Beichen Gao]

Thanks for having us.

[Vicky Edkins]

Thank you for listening to the Science Speaks: Conversations on Health podcast. This podcast was produced by the Basel Research Center for Child Health in Switzerland and hosted by me, Vicky Edkins. Editing was carried out by Sebastian Schell at the University of Basel's New Media Centre. Thank you to our generous funder, Fondation Botnar, and our partner institutions, ETH Zurich and the University of Basel, for your ongoing support.

If you would like to learn more about what's happening at the BRCCH, you can visit our website at brc.ch for information on our upcoming events and to sign up for our newsletter. We're also on social media. Follow us on LinkedIn at the Basel Research Center for Child Health or on X, formerly Twitter, @BRC_CH.