Scientist Stories: William Schief, Structure Assisted Vaccine Discovery for Infectious Diseases

Okay, I think that people are all reassembling here. I think we're off to a terrific start. And now we're going to really look at how can we use the advances in science and technology, the new understandings on a more fundamental level to really think about the next generation of vaccines. How can we combine our advances in understanding underlying mechanisms of disease with advances in understanding human immunology? And we are very fortunate this morning to have 5 really terrific speakers. Speakers, one who is a Michelson Award winner at the end, and 4 other speakers who will be really looking at different important disease categories and opportunities for the development of next-generation vaccines to address them. So I guess I should also introduce myself. I'm Peggy Hamburg. I, at the moment, have two titles: Foreign Secretary for the National Academy of Medicine, what many of you may have known as the Institute of Medicine, and also this year I'm President of the American Association for the Advancement of Science. But probably of more relevance to today's discussions, for 6 years in the last administration I served as the Commissioner of the Food and Drug Administration. And I also have known and worked with Wayne since I was just out of my medical residency training when we were both together at NIAID working on HIV/AIDS research. So let us turn now to our program, and I will try to be as good a taskmaster as Marie Paul was. But our first speaker is William Schief, Professor at the Department of Immunology and Microbiology at the Scripps Research Institute. And he serves there as Director for Vaccine Design at the International AIDS Vaccine Initiative as well. Is that right? He does it bicoastally. And he's an Associate Member of the Ragoon Institute of MGH, MIT, and Harvard. So he wears more hats than I do. He's also a founder and board member of Compuvax and is now working on computation-guided and structure-based design of immunogens and immunization regimens with the goal of inducing broadly neutralizing antibodies against HIV and other pathogens. Today he's going to talk to us about this germline targeting and structure-assisted vaccine discovery work for infectious diseases. So where did he go? There he is. Thank you.

All right. Thank you so much. It's great to be here. I'm going to— yeah, so in our work with our many hats, we're trying to learn more about the human immune system and then use what we learn and what other people learn to design vaccines that kind of manipulate the immune system as much as possible to elicit the antibody response that we want, very specific antibody responses. And we believe this kind of thing is probably necessary to make an HIV vaccine, and it may be very helpful to make a universal flu vaccine and some other vaccines. But I'm going to focus on HIV today. So, you know, we don't really hear about it in the US and in Europe, but, you know, Wayne knows and Peggy knows HIV is really still an incredibly serious problem. People are dying all the time. People are getting infected all the time. Ebola, the Ebola outbreak, you know, there was a big, there was a lot of press about this Ebola outbreak when it's terrible that 11,000 people died, but you know, during the same time frame, there's a million people dying every year of HIV, and there are 6,000 new infections every single day. So, and that's nearly a death sentence 'cause if you're in Africa and you're getting infected, you're not getting drugs and it's nearly a death sentence. So we really need to solve this problem, and that's what we spend most of our time on trying to do. So we're trying to make a vaccine for HIV that elicits neutralizing antibodies because neutralizing antibodies are the main correlate of protection in almost all licensed human vaccines. This is the envelope protein on the surface of the virus, and it's the target of neutralizing antibodies. The blue structures are glycans that are flickering around. They have lots of degrees of freedom, and they're moving all the time. The red parts of the structure are highly variable in sequence and length, and so if you're going to induce an antibody that's broadly cross-reactive to different strains, that antibody shouldn't bind to anything that's red. And then the parts in yellow are the parts that we've colored. They're actually less than 80% conserved, and the different positions, the variability from one strain to the other is not highly correlated, so it's even very difficult for antibodies to hold on to too much yellow. So there's not a lot of real estate for an antibody to grab onto and be able to be broadly neutralizing. And that's the central problem of making an HIV vaccine. It's possible to induce neutralizing antibodies against one strain. That can be done. You can get potent responses. But they're very, very narrow, and they wouldn't be useful in a human vaccine because there's 60 million strains circulating worldwide. So the trick is to induce broadly neutralizing antibodies. Broadly neutralizing antibodies have been discovered in natural infection. Dennis Burton and other people have— so these are— and this slide illustrates where the different main classes of them bind. This is again the HIV envelope trimer. So there's some that bind at the top, only one Fab per trimer. There's some that bind at the top of each GP120, you get 3 per spike. There's some that directly compete with CD4, the human receptor, and they bind to the CD4 binding site. And I'm going to focus the rest of my talk on those.

Today.

And there are others, even some that bind down near the membrane. They just are— they occur in natural infection very inefficiently, only in a few percent of infected individuals, and it takes a long time. They're highly mutated, so in order to recapitulate that by a vaccine is going to be a very difficult challenge. And we think that one of the major hurdles is that if you take one of these broadly neutralizing antibodies and you look at the sequence and you revert all the recognizable mutations back to human germline genes, and then you make that germline-reverted antibody, it usually doesn't have any appreciable affinity for native HIV proteins. So vaccines so far that have been tested in humans that have just used wild-type HIV gp120 and soon to be a wild-type sort of HIV trimer from one strain, we think that they're very unlikely to trigger the beginning, to initiate the induction of a broadly neutralizing antibody because they don't bind to the right B cells. So our approach really involves first designing a priming immunogen that has affinity, appreciable affinity, for B-cell precursors that we know have potential to develop into broadly neutralizing antibodies, and then design a sequence of boosting or shepherding immunogens that will shepherd the somatic hypermutation and elicit their maturation required of the B-cell population to make broadly neutralizing antibodies at the end of the day. Day. And that is sort of what I just said, so I'm going to skip past it. Now, in one mouse model, we have achieved this. We set out for a broadly neutralizing antibody that binds to the top of GP120 called PGT121. If you make a germline-reverted version of PGT121, it doesn't bind to GP120. We made trimers, HIV envelope trimers. We engineered them to have affinity for the germline versions of PGT121. And we designed shepherding immunogens. And that was all reported in this study where you start with a germline targeting immunogen, and then you can imagine giving this guy as a first boost and this guy as a second boost. And we proposed series— different sequences that might induce broadly neutralizing antibodies. And Michelle Nusenzweig's lab made a knock-in mouse that had the germline-reverted B cells for this broadly neutralizing antibody. And they delivered sequences that we had designed, and they showed that if you started with our germline-targeting priming immunogen called 10Mu, and then you delivered a sequence of the shepherding immunogens that were progressively more native, and you ended with a cocktail of native trimers, you actually induced broadly neutralizing antibodies in that mouse model. And I don't have time to go through the whole study, but they were— they had mutations that were characteristic of the original human bNAb, so it was proof of principle that actually what many people regard as a crazy and way too impractical strategy, it showed at least in principle you can do this. And in this case it took at least 5 shots to do it. But, and many people say, why don't you just go do that, just do that in humans? I mean, it worked in the mouse, right? And we would like to do that, but this was a lower bar. This mouse model was much more easy than would be in a human. And so in this, it was a homozygous knock-in mouse, so all the B cells had precursors that we were looking for. There were no off-target competing B cells, but there will be in humans. So there was no competition from other B cells. The PGT121 precursor frequency in humans is extremely low. We don't actually know what it is, but we think it's very low. And in this particular case, although we could prime the B cells in this mouse model, we've had trouble with this particular class of bNabs we can't get the affinity, we can't, we're having trouble designing the immunogen to have a better affinity, which we think would be important to prime humans. So although this showed proof of principle, this one, this particular project is still a work in progress. We've actually switched to another precursor and designed a new germline targeting immunogen, and that's going to be, that's being made for testing in humans, and the process is starting now. So we should have a human clinical trial on that in a couple of years. But I'm going to switch gears to a project that's further along at the CD4 binding site where we are starting a human clinical trial that just opened for enrollment on Monday, in fact. And so this particular class of bNabs are called VRC01 class bNabs, and they have two key issues that make them all in the same class. They use a human VH1/2 gene, and all the light chains use a 5-amino acid CDL3 loop. And that allows them to fit into the CD4 binding site and compete with and block CD4 binding, essentially. And I won't take you through it, but we've engineered a mini protein that presents the CD4 binding site and has a lot of mutations in the CD4 binding site that enable the mini protein to bind to germline-reverted versions of VRC01 class antibodies, and also naive B cells that have VRC01 class precursors on them. And we've arrayed that mini-domain on a nanoparticle. So we have— this is about a 30-nanometer diameter nanoparticle that presents 60 copies of our mini-domain. The mini-domain is called engineered outer domain, EOD, and it's germline-targeting version 8, and it's a 16-mer. So we call it EOD-GTA-16-mer. And I'm just going to take you through the data that we have on this. So in analogous to the study that I just showed you about the other BNAB, we tested this nanoparticle in a knock-in mouse. This was actually a little bit of a higher bar because it only had the heavy chain knocked in, so it was using mouse light chains. And the question was, could the immunogen prime naive precursors that had a VH1/2 heavy chain with mouse light chains that had— they had to select for light chains that had a rare 5-amino acid CDR L3. It only occurs about 0.1% of mouse B cells, so that was the bar that we were trying to reach, and it did work. We published that in Science in 2015. The frequency of those precursors was on the higher side, higher than it will be in humans, and the affinity was a little bit on the higher side than it will be in humans. But still, you know, we and everybody else was impressed that this worked. So it was a good first step. Then we went to a more challenging mouse model, Kymab mouse, which is a human Ig transgenic mouse model where basically the mice make human antibodies with incredibly diverse repertoire with mouse Fc domains. And in that particular case, for technical reasons, the frequency of VRC01 class precursors that we were trying to prime was extremely low, 1 in about 100 million, we reported. And the affinity was lower than it was in the knock-in mouse. This is more like what the affinity will be in humans. But actually the precursor frequency we thought was even lower than it is in humans, and we published that. But we could still prime in that situation, so we thought it was very promising. For the human case. If we could prime when they're that rare, maybe it'll work in humans. We also know that you can take our mini-domain and you can sort naive human B cells. You can do flow sorting and you can pull out BRC01 class precursors, and you pull them out at a frequency about 1 in a million, and you make those antibodies, and their geomean affinity is about 3 micromolar. So that is the target population that we want to activate in humans when we immunize them, and at least we know we can fish them out by sorting.. So we think that's a really important test for any sort of germline-targeting immunogen or any immunogen that is seeking to specifically launch a specific class of B-cell maturation in a human being. You should be able to fish out those B-cells, the starter B-cells, if you will. And so now we're conducting a human clinical trial to see how does this play out in humans. And for the rest of the talk, I'm going to walk you through a study that we published in January that even more, in a more refined way, gets at the rules of immunogenicity and tries to forecast what might happen in a human being. And so in this particular study, our colleague David Damasi made a knock-in mouse that had both the heavy and light chains of germline-reverted VRC01. So the B cells in that mouse are exactly a precursor that we know exactly what the affinity is. We actually have a crystal structure of that BCR bound to our antigen. So he made that knock-in mouse, and then working with Shane Crotty, who's an HVP PI, an important PI, or I don't know, he's an important member of the HVP, Shane's lab did adoptive transfer studies where they took B cells from this knock-in mouse and they diluted them into wild-type mouse at day -1, and then they immunized with our construct at day 0, and then characterized what happens after a single shot over time after that. And the reason that they wanted to dilute into a wild-type mouse is that Shane's lab had shown by sorting naive human B cells that the frequency of VRC1-class precursors that bind to our antigen in humans is about— in human naive B cells is about 1 in a million. So he wanted to dilute the target, these precursor B cells, down to 1 in a million in a mouse and see, well, can you— does the thing prime in that situation? The one caveat in this study was that this particular knock-in, this sequence, VRC01 germline-reverted sequence, binds incredibly tightly to our immunogen that's going to go into humans. It binds with an affinity of 9 picomolar, which is not the affinity that our immunogen binds to true naive VRC01 class precursors in humans. That range of affinity is 100 nanomolar to 60 nanomolar. 60 micromolar. So it's a range and it's lower affinity. So what we wanted to do is not only immunize in these adoptive transfer systems with our EOD-GT8 60-mer, we had earlier versions. We had GT1, GT2, GT3, GT4, and GT5. And those guys, they're not good enough to put in humans. They have— their affinity for human naive precursors is not measurable. But they are very analogous to EOD-GT8 60-mer in structure. And they have lower affinity for the B cells that are in this mouse model. So we could kind of— this is all what I just told you. These are— this is the range of affinities of human naive precursors binding to our GT8 protein. And we also have GT5, GT2, and GT1 that have lower affinities for the knock-in B cell that sort of span the range of affinities that GT8 has for the naive human cells. So we thought, okay, we're going to do adoptive transfer immunizations. We have 60 mers of GT1, GT2, and GT5 that represent the very low affinity range of what GT8 is going to encounter in humans, a middle affinity range that GT8 is going to encounter in humans, and a sort of a higher affinity range that GT8 is going to encounter in humans. We wanted to know, well, do they all work, or does only the high affinity ones work? What happens? And, you know, we really did not know. We had some expectations. Patients, but we didn't know, you know, what was going to happen. So this is sort of getting at fundamental rules of immunogenicity and trying to figure out what could happen in humans to a vaccine. And so they did these— Shane's lab with Robert Abbott did these beautiful experiments where they did lots of different adoptive transfers at different frequencies. They transferred them so that there would be 1 in 1,000, or 1 in 10,000, or 1 in 100,000, or 1 in 1 million.. And this 1 in a million is the physiological range. And they did that with nanoparticles, 60-mers of GT1, GT2, and GT5. And they looked in this particular graph at day 8. They looked in germinal centers in these mice, and they said, what percent of all germinal center B cells are derived from the transfer? Because the transferred B cells are CD45.2, and the the host B cells, the mouse that you're diluting them into, are CD45.1. So you can distinguish them very clearly by FACS. And so this graph is showing you what the percent of GC B cells are. And you can see at very high frequency, when the transferred B cells, when there's lots of them, they're 1 in 1,000, it doesn't matter if you're low affinity or middle affinity or high affinity, you get germinal center responses at day 8 in all cases. But if you go to the physiologically relevant range of 1 in a million,, then only the higher affinity case could they actually detect significantly above background transferred B cells participating in germinal center response. So that's— we concluded that precursor frequency and affinity are limiting for germinal center entry at day 8. And if you don't get in germinal centers at day 8, you're not going to get in later, basically. So the middle and low affinity antigens in this particular case, even though they're pretty fancy nanoparticles, they're basically duds. So that is a really— that was a shock actually to us that the 14 micromolar guy didn't work. Okay, and then we wanted to know, okay, so that was for nanoparticles, what happens with a monomer? You know, do the same— if we immunize just with the EOD-GT8 mini protein not on a nanoparticle, what happens? And in that case, we immunized with GT8 monomer that has 9 picomolar affinity, and for the B cells transferred in the system, GT5 and GT2, you know. And at high frequency, they sort of all— well, at least GT8 worked very well and GT5 worked a little bit. But again, at the physiologically relevant precursor frequency of 1 in a million, only the GT8 monomer that has ridiculously high affinity was able to initiate germinal center reactions. So that shows you that not only does the affinity matter, but the multimeric state is really important, right? And this— I mean, you are really hardly ever going to have a vaccine going into a human that's a monomeric protein that has 9 picomolar affinity for a naive human B cell of the kind that you want. That's just not going to happen. So this shows you that if you want to turn on the range of B cells that you want, you at least— you probably need to have a multimer. So that was all day 8 data. And just to show you, I want to go back really briefly. So when we were doing this study, this was day 8. It took a very long time for the lab to, you know, they had to figure out how to do these assays, calibrate the system. They did a lot. This is a lot of experiments. So, you know, months and months and months later, we're having these meetings and they're showing us this data. And I'm looking at this and we're saying, okay, at least we can prime the high affinity case, at least with GT5 we can prime the high affinity case, and GT8 does have this level of affinity for an appreciable number of VRC1 class precursors in humans. So it forecasts that maybe GT8 will be able to activate some of the right B cells in humans. But at day 8, it was a very wimpy response. I mean, this dashed line is their limit of detection, meaning if they repeatedly do things with negative controls, they get things bouncing around under this level. This is just above the level of detection, and we were thinking, what happens if it dies out? What happens if they die down? If the BRC01 class precursors don't adequately compete with wild-type mouse B cells, this might fade, and that would tell us that the human clinical trial that we spent 5 years and like $10 million planning is going to fizzle, and we can see it ahead of time. So we were very nervous, and they looked in the kinetics, and so This was the day number for GT5, and fortunately it increases. So by day 20, the germinal center reaction is peaking, and now 15% of GCB cells are derived from the transfer. And it's really interesting, when you're doing things at a very low precursor frequency, they only transfer a few B cells into a mouse, and they do— I don't have time to show you the imaging, but they can look and they can count which— you know, they can see all the GCs and they can count which which GC has a transferred B cell and which doesn't. And it's only about 1/4 of all the germinal centers in the mice that have a VRC01-class precursor in them, like at day 8. And that doesn't change over time. It stays about 1/4. But what happens is in all of the germinal centers that had a VRC01-class transferred B cell in the beginning, those guys expand and win, and they become 90% of all the B cells in that individual germinal germinal center. And this 15% is counting those where they've won and they're 90% and all the guys where there isn't any, so it's just averaging over, if you can follow that. So that's probably what's going to happen in the human if this all works. There's going to be lots of germinal centers responding that don't have the right stuff, and some that do, and hopefully in those that do, our guys win. So that probably has implications. The other thing that we wanted to know is, well, if it's priming at day 8 and over time, how much mutation is getting. And we— Shane's lab sorted and got sequences from the B-cell receptors, and you can count total amino acid mutations and also VRC01 class mutations. These are mutations that we know we've seen them in VRC01 class B-Nabs before, so we know that they're productive. And you can see at day 8, there's not many mutations. At day 16, there's a little bit more. This is a 2D histogram, so if this— every— this is 14 there were 14 B cells that they found that had this combination of these two mutations. And then at day 36, you can see a lot more mutations. And actually, this is quite a good level of mutation from a single immunization, and the adjuvant was alum. So we were very encouraged by this. And, you know, this dashed line is the most you could get, because you can't have more VRC1 class mutations than all of them. So the last thing that we are looking for, and what we want in our human clinical clinical trial, in order to— if this prime is going to work in humans, what it really needs to do is it needs to lay down a population of memory B cells that can then be boosted or shepherded to increase the mutations and select more mutations that will produce a broadly neutralizing antibody. So we wanted to know in this mouse, do we get a memory population late that's VRC01, and has it— how many memory B cells do we have compared to how many naive B cells we started with? If it's only, you know, you started with 1 in a million and you're still at 1 in a million in the memory population, that may not bode so well for the boost. We'd like to see it expand. And in fact, that's exactly what they did find. This is— sorry. This is data from 16 mice at the 1 in a million case for GT5, and it's day— I think it's day 36 memory B cells. And they measured the percent of all B cells, and this represents a 30-fold expansion expansion in the number of B cells from naive B cells at the beginning to memory B cells output at the end. And that was really encouraging for our human clinical trial. So I'm just going to close with this slide. And this sort of gets to some— it's too early to call them fundamental rules of immunogenicity, but in this model they were the rules that we believe were controlling what was happening. We found that there was an affinity threshold for germinal center entry in competitive of fitness, and the threshold was somewhere between 500 nanomolar and 14 micromolar, and we take it as about 3 micromolar, okay? And that was for nanoparticle immunogens. In humans, we— what we know is the frequency of VRC01 class naive precursors that bind to our antigen better than this threshold, better than 3 micromolar, is about 1 in a million. So those are— if we literally take the conclusions of the study and transport them to humans, we still have a precursor of 1 in a million, precursor frequency of 1 in a million that we think can possibly be primed, you know, with this affinity of better than— of 3 micromolar or better. And that we have a lot of data for that. So this is a lot of different precursors, and here's the 3 micromolar. So all— we're saying all these guys are potential that could be primed. And what we think is that these human precursor B cells, they may respond to EOD-GT8 60-mer in humans. In a similar manner that these transferred B cells responded to GT516 in these adoptive transfer experiments. And so we're cautiously optimistic that our priming immunogen will do sustain VRC1-class germinal center responses and VRC1-class memory, perhaps with robust SHM like I showed you in the clinical trial that actually opened for enrollment on Monday. So one of the— So these thresholds, to make an HIV vaccine, we think we have to do this process not just for one class of bNAbs, but we need to do it for two or three, because none of them are 100% broad and you're going to have inefficiencies in inducing any one. So we're trying to do, trying to learn these rules and reapply them several times in parallel to make a vaccine that induces multiple different classes of bNAbs by priming the right precursor for each class and shepherding them them all sort of together. We're working on that. So these— this model study helps inform us if we're designing germline-targeting immunogens for another class of bNabs. We're aiming for 1 micromolar affinity for human naive precursors. We're not happy if it's only 100 micromolar because we think— we're not happy if it's only 14 micromolar because it might not work. And then the last thing I'll say is that, you know, with the HVP project, this mouse study and the other ones that we've done Once we complete the human trial, which has 48 people and is taking lots of samples over time after 1 or 2 shots, we can compare all the data from the human clinical trial to the data from our mouse model studies to find out which, if any, of the mouse models kind of predicted what happened in humans. And to the degree that they did predict, you know, that'll give us confidence in reusing those kinds of mouse models. And to the degree that they didn't predict, We've got to rejigger the mouse models to better forecast what's going to happen in humans. Because to make an HIV vaccine or to make other vaccines in this rational way, the more you can iterate in mice and kind of gain confidence that what you're going to eventually pay GMP for and deploy in humans, the more you can gain confidence that those— the responses will be what you want, the much further and much faster you can go. And so that's the end. And I would just— there's a lot of people who did work for this, obviously, and they're in blue. I mean, Joe Jardine was a student and postdoc in my lab who started the germline targeting project for BRCA1 antibodies. Po Su originally designed the EOD. He's now a faculty member at Stanford. Joe is at a private institute at Harvard. Dan Culp worked with Joe on the GT8 version, and he has his own lab in Philadelphia. Sergei Menes designed the nanoparticle., and he's working at IAVI. David Nemezi's lab did a lot of B-cell experiments. Dennis's lab and Devin Sock and Brian Briney are responsible for almost all of our sequencing, kind of a lot of stuff that Jim talked about. Colin in Shane's lab did all the naive B-cell sorting, and Robert Abbott did this beautiful transfer study that I just spent a lot of time on. So, and, you know, lots of funding, obviously, to do all this.

So thank you very much. Thank you. We have— you did go over, but lots to talk about, lots of important, exciting work, and we're like on the cusp of this work moving to the next stage, so that's very exciting. Couple questions.

Yes, in the back. Speaking of durability, do you have any indication that your nanoparticles can induce long-lived plasma cells? And are you thinking of using nanoparticles for your heterologous boost as well as your product? Yeah, so I didn't show it, but we do see, at least by ELISA, we see the titers staying for a long time after just one shot. So it does seem to be— I mean, it goes down a little, but it stays quite— yeah, they don't, they stay quite high. So, and that does present some challenges when you come in and boost. You need to be a little careful if you're gonna— if your boost is gonna access the memory B cells, you need it to be not so reactive to the existing serum antibodies, otherwise it might get whisked away. So we're dealing with that. But yes, we are using nanoparticles for our boosting immunogens, at least that we're focusing on doing that, because we see better— a lot of other stuff I didn't talk about. We see a lot better trafficking of these nanoparticles. And they also— we're using them as a conserved T help. So when you do sequential immunization with different immunogens, you do want to make sure that you've got T help epitopes that are preserved between them. And even if we're changing the antigen itself, if we use the same nanoparticle that has a lot of amino acids, we can use that for conserved T help.

Great. I think I saw another question. Yes.

So one thing we found useful to mimic the diversity in people is just to use the HMDP, which is a mouse diversity panel of about 80 different mouse strains. And that then reflects better what's going on in people. And you can go down to as low as maybe 10 strains. My question for you is, can you put your engineered B-cell receptor into various mouse strains and test your vaccine in different mouse strains?

I really like that idea. I know there are issues with— if you transfer B-cells from one kind of mouse to a different one, the B-cells can just get killed and it might be a problem. So I suspect that we might not be able to do it, but I need to talk to Shane to find out. To the degree that we could have— if we can make the recipient mouse incredibly much more diverse than just a black 6, it would be very interesting. Yeah, thank you.

Okay, thank you. Let's move on. Thank you very much.