Antibodies are proteins produced by our immune system that neutralize or help destroy abnormal cells and foreign agents, like bacteria and viruses. However, their utility extends beyond our bodies’ defense system. Antibodies can also be engineered in the lab to be used as therapies. Today, over 170 antibodies have been approved as medicines to treat a wide range of diseases including cancers, autoimmune diseases, infectious diseases and more. In this episode, co-host Maria Wilson chats with guests Yan Wu, VP and Senior Fellow, Antibody Engineering, and Paul Carter, Genentech Fellow, Antibody Engineering, to discuss all things antibodies! Learn about the history of therapeutic antibodies, how advances in antibody engineering are creating new classes of medicines, and the promising role of artificial intelligence in designing antibodies from scratch and optimizing their therapeutic activity.
If you would prefer to read a transcript of this episode, please click here.
SUBSCRIBE BELOW TO CATCH EACH EPISODE
If you want to learn more about the groundbreaking science happening in our labs, click here. To learn more about the jobs in our research and early development group, click here.
Transcript of Two Scientists Walk Into A Bar: “Engineering Therapeutic Antibodies” with Yan Wu and Paul Carter
Maria: I’m Maria Wilson.
Danielle: And I’m Danielle Mandikian.
Maria: And we are scientists. We. Love. Science.
Danielle: Yeah, we do. So, when we aren’t doing it, the next best thing is to talk about science! And what’s really awesome is we’re surrounded by some of the most brilliant minds in research!
Maria: We are going to step away from the labs today to talk to other scientists about the cool stuff they are thinking about, working on, and imagining…
Danielle: …as well as how some of these discoveries just might lead to new medicines. So, grab your favorite drink, get ready to unlock your science brain and join us for Two Scientists Walk into a Bar…
Maria: The show for scientists, science geeks and the people who love them!
Maria: So we have a new episode coming out about antibodies, and we're asking people: what is an antibody?
Employee responses:
So it's a molecule that is produced by our immune cells to fight off pathogens.
I will say that it is something that really helps your body create the right defense system against any disease, anything that is attacking your body.
Yeah. So, there's a monoclonal antibody and then there's bispecific. But I think in essence, an antibody is something that's a part of the immune system that elicits a response.
We know, but we are not sure about the absolute correct answer. [Laughs]
I can tell you how to get that antibody to the patient, but… [Laughs]
An antibody is a protein that is made by B cells that serves all kinds of functions in the immune system but can be used in drug development as a tool to treat diseases.
Maria: And if you could design your ideal antibody, what would it do?
Employee responses:
I mean a dream antibody would cure a disease, or nearly cure a disease. Right?
I would like it to cure any type of cancer.
I guess to target the disease cells or the disease without harming the individual you're treating.
I think an antibody that is customized to the individual or even can adapt to the individual's needs and the environment. I mean maybe that's where we're heading in the future.
Maria: I'm delighted to be here today with Paul Carter and Yan Wu to talk about antibodies and antibody engineering. So Paul, what exactly is antibody engineering?
Paul: [Laughs] Apart from having as much fun as possible, it means that we try to redesign antibodies to do things, maybe the things which they do already, and we try to improve upon them. We try to get them to do things which they've never actually done before in nature.
Maria: So how did we decide and figure out how to use antibodies as medicines?
Paul: Well, this is something which has been a long time in the coming, really going back to the mid-1970s with our – just pioneer work of Kohler and Milstein inventing hybridoma technology. And so, it was thought at that time that it would be a fairly, you know, simple kind of leap to go from being able to make antibodies to actually develop them as therapeutics. In reality, it took us – as a field – more than 20 years. And it was a bumpy road!
Maria: A bumpy road indeed. But before we keep going down the road, I'd like to get a quick definition. Remind me, a hybridoma is when you were able to take the cell from the mouse or the human that makes the antibodies – the B cell – and fuse it with another cell, so that it could make them in a dish. Am I correct?
Paul: That's right.
Yan: So, there's a lot of antibody discovery technologies. So originally, the main one for many years are hybridoma. So, the hybridoma technology is really the foundation to where we are using therapeutic antibodies. So hybridoma, usually you take the B cells from the mouse, you know, beginning with from the mouse spleens or lymph nodes and fuse with immortal myeloid cells and basically form a hybridoma that continually secrete antibodies. And these antibodies are monochromal antibodies, so the one single specificities.
Maria: Right, because an organism will make multiple antibodies. If you just get antibodies from the plasma of a mouse, even if it's against one protein, they'll recognize multiple different what we would call epitopes, right, of the protein. But the hybridoma meant you got one single clone of an antibody that you knew exactly what it bound to. And drugs are monoclonal antibodies.
Yan: Yes.
Maria: What was the first one that was ever used?
Paul: The first one was a molecule approved in 1986.
Maria: Yeah.
Paul: It wasn't very successful, and there was a period of about more than 10 years where lots of different groups would take mouse antibodies into clinical trials, particularly in cancer. Our result was really, pretty dismal, that these were recognized as foreign proteins –
Maria: Yeah.
Paul: – by people. People mounted an immune response which neutralized them. And they basically just didn't do anything. So that was kind of a low point in the field.
Maria: I'm remembering some of my own antibody background now because what you expect your antibody to do is target the immune system to kill the thing it attaches to, right? So that's why – you can probably explain this better – that in cancer, an antibody against a cancer cell was supposed to bring, was it called ADCC?
Paul: Yeah.
Yan: Yes.
Paul: Yeah, that's one of the mechanisms. So, yeah, one strategy with antibodies for cancer is exactly that – to try to recruit immune cells to kill the tumor cells. And that doesn't work very well with mouse antibodies and human cells.
Yan: Basically, can say that it is about variety. Now for the simplify is, let's say, block –
Maria: Yeah.
Yan: – flag and deliver.
Maria: Okay.
Yan: So “block” will use the specificity of antibody to recognize a unique antigen in the cell surface or secreted protein that block ligand-receptor bindings. Right? So there's another way is “flag” it to say this is highly expressed in the cancer cell. So you bring the antibody there and you basically eradicate the cancer cells. And “deliver” is really using the antibody to deliver your payload to the site of action.
Maria: So when we think about an antibody in simple terms, we always draw it as looking like a Y, don't we? And the two tops of the Y, those are the parts of the antibody that recognize the antigen. And in a typical native antibody, those two tops of the Y both recognize the same thing that we call an epitope. But I know that we've been developing and there are drugs now that are called bispecific antibodies, which are really, really cool. So you make an antibody that would never exist in nature that's got one arm of the Y recognizes one epitope, and another arm of the Y recognizes another epitope. So you can bring two things in the body in proximity to each other which wouldn't normally be in proximity together with antibodies. Is that right?
Paul: That's right. So you can, you know, like – say binding a tumor cell and an immune cell.
Maria: Yeah. Yeah.
Paul: – like a T cell.
Maria: And then the other thing that I think is really cool is the ADCs – the antibody drug conjugates – where you, I think you stick something onto an antibody and it takes it to a cell. Right?
Paul: It's a seductively simple idea to take an antibody, arm it with something which is, you know, exceedingly potent and you couldn't use on its own. But the devil is in the details of the so-called payload, the so-called linker, how you attach it.
Maria: Yes, yes.
Paul: You know? How many of these you stick on your antibody. And as a field, it kind of collectively took us many kind of iterations to get that to the point where it actually works.
Yan: Yeah. ADC has been working very well, but it's still a challenging field.
Maria: What are the challenges?
Yan: The challenge is really the TI – therapeutic index.
Maria: Okay.
Yan: Because you think you put the antibody, have a magic payload and it can cure cancer. But there is a toxicity.
Paul: Yeah, I think ADC is sort of an example of a technology which – I mean, it took a very long time to develop. But it sort of went through cycles where there was no interest in the field, or very limited interest, to maybe just pendulum maybe swung too much in the other direction to kind of irrational exuberance about they were going to solve everything. And then you know, so you go through these –
Maria: Sort of pit of despair –
Paul: [Laughs] Yes.
Maria: – and then you come back out the other side.
Paul: Yeah, these cycles. [Laughs]
Yan: In science, in drug discovery, you have to start with the biology to really truly understand the pathway, the mechanism of action, the disease. Right? And also, you have to think about the clinical plan. See whether the molecule can be developed, but also combine with the technology innovations. So for us, you know, we basically – in the past, we try to make a lot of successful antibodies. Still, most of the drugs are still monoclonal antibodies. But right now, the field mostly is focused on make the undruggable targets druggable –
Maria: Yeah.
Yan: – or make the previous druggable targets with a totally new way.
Karen: Hey, Maria.
Maria: Hi!
Wellington: Hi, Maria.
Maria: Hello.
Wellington: [Laughs] So, I heard ADC in there, and I heard ADCC. What's the difference? What's going on?
Maria: Yes. I was confused by these terms many years ago when I first started to work in antibody drug development too. ADCC is a natural property of an antibody. It’s called antibody dependent cellular cytotoxicity. And so, when your body makes antibodies against, you know, an infected cell, that antibody binds to that cell, sends other immune cells to come kill that cell. So that's one of the things antibodies do. And when we use antibodies, traditional antibodies, to kill cancer cells, that's usually how they're working. It's like, yeah, the antibodies recognizing something on a cancer cell. And we're using the natural property of an antibody to – it's ADCC properties – to drug that target, to kill that cell. You can engineer out the ADCC properties if you just want an antibody to go and bind to something and block it and not have this immune system activating effect. So depending on what you want your antibody to do, you would engineer it to have ADCC, or not have ADCC. ADC stands for antibody drug conjugate, and that's when you tie a small molecule, usually a toxin, onto an antibody to deliver it to a cell. So it's a completely different thing that sounds quite similar.
Karen: So Maria, Yan was talking about undruggable targets. Is the word undruggable accurate? Is a target really not druggable?
Maria: Well, that's such a great question. I think it depends what you're trying to drug it with. I mean, for an antibody, the target does need to be extracellular, really. There's some places antibodies just don't go. But in the broader question of drug discovery, I think there are a lot of people now who are hopeful that there will be a modality that can tackle almost any drug target, and that we shouldn't be thinking about things as being undruggable. It's some, just some things are going to be harder than others.
Maria: So Wellington wanted me to ask this question, which is: So we talked about bispecific antibodies, and he was wondering, is this like, you know, disposable razors? Like we have two blades, can we get three blades? Can we get to seven blades? Is there like an extra strip on there? You know, what else can we do with antibodies?
Yan: I think we can make more. So monospecific is a basic. Now, we’re routinely making bispecific. We’re going to add in more modalities to it. So, we are working on trispecific. I don't think we're going to do seven-specific, right? But, we do have a variety of modalities. Do infusion proteins. We can bispecific. Then there's fusing another antibody, protein-like cytokine fused to the C-terminal of the antibody. Or we can link the modalities of antibody, you know, called beading a string.
Maria: Okay.
Yan: You can use your imagination. Of course, we have to have a follow-up with a technology that we can produce those complex molecules –
Maria: Yeah.
Yan: – in a single cell that becomes the drug in a vial.
Paul: You know, for sure you can link bispecifics, trispecifics. You can certainly add additional specificities. I think the – I don't know what the record is, but I think it's at least five. But the question that you have to ask yourself is “why?”
Maria: [Laughs]
Paul: Every additional kind of binding site you add, whether it's binding the same molecule or a different one, adds complexity because you can now tune each one. You can make it higher affinity or lower affinity or anything in between. And so, there can be very good reasons for adding extra binding sites for being able to, say enhance the selective binding to say a tumor cell over a normal cell. But it can certainly be overdone. You can make something which is extremely complex to make. And I can certainly – from personal experience – I've sort of worked on things in my, you know, going back sort of decades where they were so complex. So, our so called, our immunoliposomes. And we got some very nice papers, which are very widely cited. But they were kind of essentially useless in terms of benefiting people. They're just so complex, so difficult to make. And so, it's shaped my thinking that you – in terms of how you develop drugs – that you have this kind of toolbox of technologies, and it's great to be, you know, pushing the envelope of as to what's possible. But it also – I think it's kind of wise to be kind of grounded in pragmatism to –
Maria: Yeah.
Paul: – make your drugs only as complex as they need to be to get the job done. And if you can do something with something simpler, other things being equal, that's probably going to be a preferred way to go.
Yan: Yes. In our experience, on the paper, you can design any molecule you want.
Maria: Yeah.
Yan: But the most successful drugs are usually the simple ones.
Maria: Paul, shall we get back to that bumpy road that you mentioned at the beginning?
Paul: So, let's talk about the, you know, the bumpy road. So we started working in antibodies in 1990. And I think the prevailing wisdom in many places – actually, quite arguably, most places – is that given this history of, you know, really no success with mouse antibodies, that people had kind of written off antibodies as ever being sort of useful. And so it was definitely an exercise in sledding uphill in terms of trying to kind of overcome the mindset that this would not work. And of course, we had no idea whether it would work either, but you know, we were just sort of young enough and pigheaded enough to pursue it anyway.
Maria: And philosophically, it should work. Right? I think you must have had that drive.
Paul: Or just feeling that, you know, with our advent of humanization – you know, here is a technology that addresses – has the potential to address the known limitations of mouse antibodies. Of course, you don't know for sure until you get in the clinic whether it actually will. And even if it does, there may be some other obstacles that you don't even know about yet.
Maria: It must have been amazing when you first saw – was it human PK data or something that told you that the antibodies were okay?
Paul: I would say the first – our moment was when we were humanizing our first antibody. And within a few months, we had made a humanized version that worked almost as well as the mouse antibody in the test tube, and then a few months later actually had a version that we felt was – based on our knowledge at the time – was as good as we could make it and was kind of worth our exploration in the clinic. That was kind of a very, very exciting few years.
Maria: Amazing!
Yan: Yeah, truly, you know? Antibody has been described as magic bullet a very long time ago, though the technology innovation really made it possible to be real therapeutics. You think about from hybridoma, mouse antibody, to chimera, to humanized antibody. Now we have actually ways to making fully human antibodies.
Maria: When we talk about design of molecules, I wanted to ask you a question that we are asking everybody this season on the podcast, which is about how AI and machine learning are applied to the work you do. And this feels like a perfect place to talk a little bit about how that's applied to antibody and protein drug design.
Paul: I mean, I think that's a great, great question. This almost kind of feels like a gold rush era in terms of applying AI to redesigning proteins. And for sure it will have an impact. It's starting to have an impact. Where exactly, where it's going to be most useful, I think it remains to be seen. I think the sort of holy grail is to be able to design our antibodies and proteins from scratch that bind at specific sites, specific places, and other proteins. That's a very difficult thing to do. I think there are perhaps simpler problems which are more, you know, tractable. And one of the things that we do in engineering antibodies is, to use a bit of jargon, would be sort of multi-parameter optimization, which is a fancy way of saying you want to maybe engineer an antibody to do several things at the same time. So, maybe have high affinity, have good biological activity, have other drug-like attributes. And so, I think there – AI has the, at least the potential, to be able to simultaneously solve to satisfy all of those things, whereas, you know, historically in the antibody field, we have tended to do one thing. So, we've tried to improve the affinity. And then, sometimes, you find out in improving the affinity, we've mucked something else up. And then we have to go back and fix it. And then we've mucked something else up, and we have to go back and fix that. And so rather than do things serially like that, if you have enough capability in AI in terms of our, you know, sort of models in principle and probably in practice over time, you'll be able to improve one property and maybe improve other properties at the same time – or at least not mess them up.
Yan: I think with the advance of computation, machine learning and artificial intelligence may be a game changer for antibody discovery and engineering. But it will be gradual work.
Maria: Yeah.
Yan: Because the machine learning and AI needs a lot of data.
Maria: I see.
Yan: Although we have over 100 approved drugs – antibodies on the market – but there are not that many!
Maria: Yeah.
Yan: So, we actually need to be working closely side-by-side with machine learning and AI scientists. A lot of industry too, including us, is using a model called “lab in the loop.” So basically, we're closely working with the machine learning scientists, with the bench scientists. Bench scientists provide as much data as possible and help the machine learning scientists train the model. And we also need to test in the wet lab of the design of the molecule to see whether achieve the goal. But I think in the future, it's looking very promising, you know. So, we need to continue to work on now. Everything takes time, too, for the technology to develop.
Paul: I think it's certainly an exciting time in the field.
Yan: Yeah.
Paul: And I work with kind of different disciplines. It's like kind of learning a foreign language, if you're not an AI person, trying to understand the rudiments of that. And it's also for the AI folk, you know, learning some of the, you know, acronyms and the ways of the wet lab world. It's kind of like, you're kind of trying to get people, or train people, so that they're bilingual. And so maybe they’re experts of one of those disciplines but have enough understanding of the other discipline so that you can ask good questions. And I completely agree with Yan the need to do this, you know, iteratively and to do that so-called “lab in the loop” as fast as possible, because that's how we're going to make progress. AI is definitely an endeavor where, other things being equal, the larger the datasets that you have, the more that is possible. So I think it feels like this early period is trying to learn the language of the other discipline and together figuring out what kinds of problems are most tractable because, you know, you're just kind of only limited by your imagination in terms of things that you can do. But there are some things which aren't really feasible at the moment because it just takes us too much time and effort. We just don't have enough data.
Yan: Yeah, it's exciting. I think the only way to be very successful is working very closely between the antibody engineers and the machine learning scientists – working as one team.
Maria: Yeah.
Karen: Hey Maria, they were talking about “lab in the loop.” What do you know about that?
Maria: So, the way I understand it is the idea that human physiology is incredibly complicated. The inner workings of the cells are incredibly complicated. And what we typically have done is kind of approached it piecemeal. Like here's this kinase cascade or here’s this receptor signaling pathway – let's drug it, let's look at it. And the idea of “lab in a loop” is that you generate an enormous amount of data – whether that's from cells or from treating animals or even from patients – and you crunch that using AI machine learning to see how the drug that you made manipulated the whole system, like in a way that you wouldn't be able to understand just by looking at things in isolation. And then you use that information to then help you feed back and design better molecules or choose better targets for drug discovery.
Wellington: Maria, when Paul is talking about the humanization of antibodies. What is he talking about?
Maria: Well so the first antibodies that were made to use as therapies were made from mouse B cells. So they had a mouse sequence. So when you would put them into a human being to use as a drug, the human would have an immune response against the mouse protein. So they had a lot of side effects that they didn't work very well. So because there was, at the time, no ability to make human antibodies from human B cells, you had to take this mouse protein and make it look so much like a human protein that the immune system of the patient wouldn't reject, wouldn't react to it essentially. So it was a way of replacing sequences in the antibodies so you don't lose any of its functionality, but it looks human enough that it is safe and well-tolerated.
Maria: It's interesting that there are only a few, a hundred or so approved drugs, which I hadn't thought about it that way. That does limit the dataset of what a successful antibody drug is. So what other limitations are there around the data availability for the work that we're doing with AI and machine learning?
Yan: I think the most challenge is actually good data.
Maria: Ah!
Yan: You really need good data to train the model. It's called garbage in, garbage out. You have a lot of data sequences. But you really need to say the antibody with the right sequence, with the right affinity, with the right binding site. So all these data – binding, pharmacokinetics and its developability – so these are all really need real data, high-quality data to train the model. So we actually work together to see – we already accumulated a lot of data over the years, but I think we actually need to generate more data, high-quality data. But also we need, as a policy, we need to ask the right question. So basically, after we get to know each other, we really ask what's the most fundamental question in antibody engineering that I would think machine learning and AI can contribute? In the short term, antibody optimization. But in the long term, it's more like de novo design.
Maria: Yeah.
Yan: Right now, it's really need to work in the – still in the model at the “lab in the loop” with the computer design and wet lab.
Paul: I think something else that's important to really be rigorous in this area is when you're looking at AI designs, you know, compare them with human experts. And obviously depending on particular experiments, maybe you just want to, you know, randomly select, you know, related sequences. That will kind of keep us honest as to whether we're being successful or not. And so that, that I think is kind of an important paradigm in exploring to see what AI actually adds.
Maria: So being kind of rigorous about it to say, well I'm not going to assume that this is going to give me something better.
Paul: No, because otherwise if it works or doesn't work, maybe that just happened by dumb luck.
Yan: And right now, it's basically the data generation, training the model and validation.
Maria: Yeah.
Yan: The goal is eventually we improve the process and speed for discovery and engineering.
Maria: So before I let you guys go, I wanted to ask you, what are you excited about right now in the field of antibody drugs?
Paul: I can go first. We kind of talked about the, you know, the low hanging fruit analogy. And you think of, okay, what are the challenges now that we really don't know how to do, or can't do very well? And I would certainly include things like, you know, getting antibodies into the brain in useful amounts. We talked about sort of, gut delivery. Or you know, I get excited about targeted delivery, so trying to get antibodies to be functional where you want them. So active or binding or maybe other activities in the tumor, but not a lot of other places. And so I think there are lots of challenges there for the antibody engineers to continue to work on for many years to come.
Yan: I agree with Paul. Basically, we try to continue to solve the challenging disease problems. So we need to continue to do our technology development, with the new modalities, new ways of delivery. So for example, we really try to work on a new format modality of antibodies and probably combine the antibody large molecule, with the small molecules, with antibody, with the peptide – right? – to see how do we enhance the new mode of action. It's not just what antibody can do, it's go beyond what antibody can do. So I think there's a lot more we can do –
Maria: So pretty exciting time!
Yan: – with the continued technology development and the new modalities.
Maria: Fantastic. Fantastic. Thank you so much for sharing your enthusiasm for this really exciting field.
Yan: Thank you so much for having us.
Paul: Yeah, thank you.
Wellington: Maria, that was a great episode.
Maria: Thank you!
Wellington: What is your relationship to antibodies, antibody engineering, in your career?
Maria: Oh, so early on in my career, I led a group that was looking for novel drug targets for antibodies for metabolic diseases. So I used to work very closely with antibody engineering scientists. Like my job was to say, “hey, why don't we design an antibody against this, you know, cell surface receptor? Because maybe it will help improve fatty liver disease,” for example. And then I would get to work with the scientists on designing the antibodies, testing them, that type of thing. So, yeah, part of my career, a good three or four years, was spent doing work, leading teams, doing antibody drug discovery programs.
Karen: So Maria, you were asking this at the beginning of the episode, but I get to ask it to you now – what is your dream antibody?
Maria: Oh, that's such a great question. So I would love to see – I think going back to our episodes we've done on Alzheimer's disease, I think the idea that you could get an antibody that can get into the brain, that we can deliver in a nice, easy manner to a patient that can really stop the progression of that disease by clearing some of the proteins and things that we know are involved in Alzheimer's disease. So, I see a lot of promise for antibodies in the CNS disease area space. But there are so many exciting applications I think for antibodies going forward.
And that's our show! Thanks so much for listening. If you haven't already, rate our podcast, wherever you listen – it'll help new people find us. And be sure to subscribe. If you have a question about the show, you can contact us at [email protected] That's G-E-N-E dot com. And now for me, it's back to wrestling with data!
The name Two Scientists Walk Into A Bar is under license and used with permission from the Fleet Science Center.