Scientist Stories: Neil King, Design of Self-Assembling Protein Nanomaterials for Next-Gen Vaccines

Our next speaker, Neil King, joins us from the Institute of Protein Design, a spin-out of David Baker's lab at the University of Washington. Neil has spent his time and efforts looking at computational methods to design proteins for self-assembly, and specifically looking at self-assembly of proteins for vaccines. Neil? All right. Thanks a lot, Dan. I'd like to thank the organizers for inviting me to be here. So as Dan said, I'll talk to you about the work we've been doing lately designing novel self-assembling proteins. You know, we heard from Bill earlier today and also just heard from John about the importance of the presentation of antigen in a repetitive array. And that's the idea that's really been driving this work. Vaccine design is undergoing really a technology-driven revolution. As new technologies have come online, they've allowed us to do new things, design new types of vaccines, hopefully move towards truly rational design of vaccines. This is even a little bit outdated. We've heard about a lot of wonderful new technologies today. One technology again is the presentation of antigen in a repetitive or symmetric array. It's been known for a long time now that if you take just an antigenic fragment of a protein, often that's insufficiently immunogenic to form an effective vaccine on its own. If you take that same antigenic fragment and present it in a repetitive array, you get a much more robust immune response. It's thought that this is primarily driven through the clustering of B cell receptors on the B cell surface, although trafficking of that particular antigen is probably also playing a role. So recently, self-assembling proteins have been explored as a very promising platform. For antigen presentation. And a lot of work has focused on the use of naturally occurring self-assembling proteins, such as virus-like particles, or more structural proteins like ferritin. And this is work from the VRC a few years ago, where hemagglutinin, a very complex antigen, a viral glycoprotein, was genetically fused to the subunit of ferritin, a 24-subunit protein octahedron, to create these kind of spiky particles, right, that present multiple copies of hemagglutinin to B cells. And this indeed, as had been shown previously, led to robust increases in humoral immune responses. And so there are a lot of reasons that I think self-assembling proteins are a good antigen presentation platform. As shown here, you can seamlessly integrate antigen via just genetic fusion. It resembles viruses, as John was really driving home earlier today. I think there are lots of reasons that this is a good platform. But, you know, what we're doing at the Institute for Protein Design is designing new proteins so that we don't have to rely on just what we find in nature. And so this is kind of a hyperbolic example of building an alarm clock out of stuff that's just laying around your house. If you go around and look at what you have and you make an alarm clock, you're going to come up with some ridiculous Rube Goldberg contraption, right? But we're human beings. We make technologies. And to make technologies, you build the parts that you need to make the thing that you want. And so if you were going to make an alarm clock today, right, you end up with an Apple Watch because you make integrated circuits and all the pieces that you need. And you end up with something that's much simpler, more robust, more controllable, right. And so we're trying to do the same thing with proteins. Can we take proteins through a technological revolution where we can actually build what we want and do it in a rational way? And so I'm going to talk to you about two parts of this and two parts of the work that we're doing. So the first is the methods that we've developed for designing new self-assembling protein nanomaterials. And then the second half will focus on the use of these scaffolds for next-generation vaccines. We're obviously pursuing these as vehicles for the delivery of biologics as well. And if I don't run out of time, we'll talk briefly about that at the end. So this is a graphical depiction of the computational design process that I developed as a postdoc in David's lab. And it consists of two fundamental steps. So there's a docking step and then a design step. And docking is really taking protein building blocks and figuring out how they fit together. So we're building these things the same way that nature does, which is through a combination of symmetry and noncovalent protein-protein interactions. So if you think of microtubules, helical symmetry, tubulin dimers interact noncovalently. So we're picking symmetry groups— sorry, the one shown here is icosahedral point group symmetry, which in this case we're going to build using a combination of 5-fold and 3-fold rotational symmetry elements. We go and we select protein building blocks that have a symmetry that matches those in our target architecture. And then we take those building blocks and we move them around in 3-dimensional space in the computer until we identify configurations that give us what we consider to be designable structures. And designable simply means that there's a new, a de novo protein-protein interface between the blue protein and the gray protein upon which we mutate the amino acid sequence and derive a new amino acid sequence that's predicted to be low in energy, when it adopts that structure. So at the end of the day, what the computer gives you is hundreds of thousands of hypotheses in two parts, a pair of amino acid sequences and a three-dimensional structure that that pair is predicted to form. And so then you go into the lab and you test this and you see how accurate your predictions were. And so these are some materials that we've recently designed. These are 120 subunit assemblies the size of small viruses. And on the left, we're comparing averages from electron micrographs to projections calculated from the computational design model.. And you can see that they match quite closely. And then if you come over here, we're showing two crystal structures. We've got lots of crystal structures of these things. And they tend to have an angstrom or less of deviation from what we made up in the computer. And that's over the entire 120-subunit complex. So these are two megadolpin structures in which we've predictively positioned every single atom. And that gives you the ability, again, to design structures that are tailored to specific applications. So, you can make up structures that have big pores or small pores. They can be 16 nanometers, 17 nanometers, 18 nanometers, whatever you want. You can have termini on the outside for fusion. You can have termini on the inside for packaging. Right? You're not just relying on ferritin. If ferritin doesn't work for you, you're out of luck. Right? And so, one of the other distinguishing features of these materials is that they're what we call two-component nanomaterials, meaning that they're— they comprise two distinct protein subunits that coassemble. And one of the things that this allows you to do is control their assembly in vitro by expressing and purifying the components separately. And then you simply mix them in aqueous solution at a 1-to-1 ratio, and they spontaneously zip up to the design structure. So that allows you to package things inside, put things on the outside at different ratios. You have control over this process. Which again, with homomeric particles like ferritin or the lysine synthase, you lack that control. So, you know, I'll tell you in a minute about what we've been doing for multivalent antigen presentation and vaccine design here, but, you know, to get to know the platform a bit better and to really with an eye towards manufacturing, we wanted to ask questions like, hey, what do you get out of this in vitro assembly process? Do you get materials that are complete? Is it really 120 subunits or are there holes in these things? Because that's going to be important for manufacturing. And licensing. You know, how robust is this process to slight inaccuracies in this 1-to-1 component ratio? What are the kinetics? You know, how stable are they? Things like this. And so Adam Morgacki in my group has been doing some pretty detailed analyses of the products that come out of this in vitro assembly process. So regarding stability, we find that these things tend to be really quite stable and often much more stable than the constituent building blocks. So these are two different particles called i5-350 and i5-340.1. And each of them you'll see has kind of a weak link building block. So the gray pentamer here, the blue trimer here are from mesophile organisms and they melt at normal temperatures. The other component in each case is from a hyperthermophilic organism and is very stable. When you assemble these into the particles, the particles are also quite stable and you can see stabilizations of, you know, 30 to 40 degrees Celsius. °C for the weaker component upon assembly into the particle. That's, we think, mainly driven by the very, very large amount of hydrophobic surface area that's buried upon assembly of these particles. So stability is a very important characteristic of vaccines, and so we think this is favorable for the use of this platform for multivalent antigen presentation. The kinetics of assembly are, I would say, fairly rapid. So at double-digit micromolar concentrations, assembly is half complete within a few seconds. And it's highly protein concentration dependent, as you would expect for this mini-body structure. The extent of assembly is also concentration dependent, again, as you would expect. And you get curves that kind of mirror those from natural virus capsids, where you have an appreciable amount of residual components left at very low concentration. But at some point, you get this crossover where the particle starts to dominate. And so, you know, we took advantage of the two-component nature of this system to really explore how robust this process is to changes or perturbations in subunit stoichiometry. So, you know, you can assemble these things with, you know, 0.25 to 1 of the A to the B component, 1 to 1 A to B, 2 to 1 A to B. And then if you run them over a sizing column, you can say, okay, how much nanoparticle do I get out of this? And how much residual component do I have left over. And when you do this, what you find is that you get this linear increase in particle up to 1:1, and then it flattens out. And this is indicative of a cooperative process that's yielding complete assemblies. It doesn't appear that there's any partial assemblies here. The minimum near 1:1 for the residual components also supports that. So then we threw a slew of biochemical and biophysical assays at these things to see if we could detect any difference in the assemblies that we generated at different subunit stoichiometries, right? If there are holes, we would expect that particle to maybe be less stable. We didn't find that to be the case. So no matter, no matter what the subunit stoichiometry, the thermal melt by CD looks identical. There's an enzymatic activity assay that we can use where you incubate the things at elevated temperature for an hour, and then you cool them back down and measure activity. And they all drop off at the exact same temperature, which is the same as the melting temperature. We've been unable to detect using any assay that we have differences between the materials assembled at different subunit stoichiometries, which again speaks to cooperativity and completeness. But we really wanted a definitive answer, and so we got in touch with Albert Heck, who does native mass spectrometry of very large assemblies, and they were able to shoot these things and they found that indeed we're only getting 120 subunit assemblies. There's nothing else in there. This is data from a 1-to-1. In vitro assembly reaction, but you see the same thing at a 1:3 ratio of the subunits. You see the excess B subunit come through, and then you have 120 subunits, and that's it. There's nothing else in there. And so from a manufacturing perspective, we're happy about this because that means that any process that we might develop could be robust to slight perturbations or inaccuracies in our subunit stoichiometry. So, okay, vaccines. So the approach that we favor for multivalent antigen display in the platform, we've done it a few different ways, but the approach that we favor is genetic fusion of your antigen to a nanoparticle subunit. Right, so again, seamless integration of your antigen. You know exactly how many copies there are. There's no conjugation here where it's not 100% and you're uncertain about how much antigen is in there. So we usually, we've been doing a lot of work with viral glycoprotein antigens. Which are often trimeric. So we match the symmetry of the nanoparticle subunit to that of the antigen. And again, this is an advantage of being able to design these things, right, as you can tailor it to display a specific antigen. And we produce that in whatever expression host is appropriate for these glycoprotein antigens that's secreted from mammalian cells. The other component we produce in whatever expression host is most convenient. E. coli is cheap and robust, and so that's often what we use. You purify those, you mix them, and then you get the 120 subunit particles. So this is a particle displaying 20 copies of the pre-fusion stabilized RSV F protein, DSCAP1, from the VRC. So these are electron micrographs or single particle reconstructions of particles that we've made displaying many different antigens. So this is again DSCAP1, this is full-length influenza hemagglutinin, we've done stabilized stems as well and monomeric. Receptor binding domains. These are HIV envelope trimers, which are my least favorite protein in the entire world. Very complex proteins to work with, and the particles have proven very robust to display of these proteins. We've put a large number of other proteins on the outside of these particles, and I don't think we've failed yet, actually. I think everything we've tried, the particle has tolerated, which is a clear differentiation from things like bacteriophage VLPs where you can display small epitopes but not large complex antigens like this. We really think that this is a robust platform. One of the interesting things, so I'm going to talk a little bit more about the RSV particles because those are the ones that are furthest along. DSCF1 again is this pre-fusion stabilized version of the F protein and locking the protein down into the pre-fusion conformation is absolutely critical for eliciting neutralizing antibodies. The protein wants to pop to the post-fusion conformation, but that's very poorly immunogenic. You don't get a lot of neutralizing antibodies from that. So stability of the pre-fusion conformation is critical for success as a vaccine. And what we found is that the stability of the pre-fusion conformation is actually further enhanced when it's genetically fused to our nanoparticle subunit. And again, this subunit happens to be one of those that's from a thermophilic organism, so it's extremely stable. So if you look at the retention of binding by pre-fusion-specific monoclonal antibodies after incubation at elevated temperature, what you can see is that DSCav1, just the trimeric antigen on its own, dies after an hour at 70 degrees Celsius. You lose all binding of this pre-fusion-specific antibody, whereas the trimeric nanoparticle subunit with DSCav1 attached retains half of its D25 binding. The particle is artificially inflated in this system because it's multivalent, right? So a single pre-fusion DSCAP1 on the surface of the particle will give you a signal here. So I really, I put more weight on the red bar here. At 80 degrees, pretty much everything dies. But this stabilization is really quite substantial. We see the same thing. So we've interrogated this using 3 orthogonal assays. This is intrinsic tryptophan fluorescence. Where we're doing guanidine melts, kind of classical protein biochemistry. And you can see this early transition— sorry, this early transition here and here for DSCAP1 that appears to be absent when it's fused to our nanoparticle subunit, again suggesting this stabilization. And then, of course, what you really want is immunogenicity. And so here, this is an experiment in mice. We've obtained similar results in NHPs. I don't have the data in this talk. These are binding antibodies on the left, specific to the pre-fusion form of F. These are neutralizing antibodies in the middle. And what we've done here is we've actually titrated up the density of the antigen on the particle surface. In the in vitro assembly reaction, you can dope in unmodified nanoparticle trimer subunits, and you can basically make partial valency particles. And you can see that immunogenicity correlates with the density of the antigen on the exterior surface, mirroring what John was describing earlier. And the other important thing to notice here, so two other key points. At full valency, we're seeing a tenfold increase in neutralizing antibody titers over dScap1, which for indications like maternal immunization could be really important. And then the other thing to notice is that the increase in binding antibody titers is lower. So it's about 3 or 4-fold, tenfold here. And what that means is we're getting a higher proportion of neutralizing antibodies to total binding antibodies, so the quality quality of the response is also improved. We're investigating exactly why that's the case. It could be geometrically driven, focusing on potent neutralizing epitopes near the trimer apex, but we're not totally sure yet. What we do know is that the increase in immunogenicity is strictly dependent upon particle assembly. So if you take just the trimeric component to which we fuse the antigen and mix in a non-assembling version of the pentamer, all the T cell epitopes are present, everything is there, the thing just doesn't assemble, you see no change from immunization with just trimeric DSCAP1 alone. So it's particle formation that's really driving the increase in assembly, or in immunogenicity, sorry. So we've been analyzing some of the cellular responses to this antigen. And one interesting thing that we found is that T follicular helper cells and germinal center B cells these are total cell counts, not antigen-specific, but the total number of these cells after immunization goes up, and that appears to be, again, driven by the particle. So the particle alone without any antigen drives a large number of these cells. When you put the antigen on there, it's even higher, and in both cases, it's higher than just trimeric DSCAP1. And then another thing we've looked at is anti-scaffold responses, right? So this is a foreign protein nanoparticle. That's underlying the antigen, and so it is absolutely going to be immunogenic. You're going to get antibodies against it. The question is, does that matter? So we did a couple experiments here where we immunized with just i5-350, just the nanoparticle alone, and then we interrogated the anti-i5-350 binding titers against the particle itself and then the particle displaying DSCAP1 at various valencies. You can see there's this slight decrease in binding titers as you put the antigen on the outside of the particle. You're tempted to think that that's physical masking of the underlying particle by the displayed antigen, and that is not the case. So if you immunize with these different versions of the particles with various valencies of the antigen on the outside, and then come back and measure binding titers against just naked IFI-350, you see no difference whatsoever. So this doesn't appear to be a physical masking effect. It may be an immunodominance effect, where dScap1 is dominant over the underlying particle. And then the other thing we wanted to look at was, does preexisting immunity adversely affect the antigen-specific response? So here, these are mice immunized with just the particle. And then these are mice that were preimmunized twice with naked nanoparticle, no antigen on the outside. So we generate a preexisting immunity against the underlying particle., and you see no deterioration in the antigen-specific response when you boost them with antigen-bearing particle. In fact, there's, if anything, a slight enhancement. And so the presence of antibodies against the particle doesn't seem to be a problem here. So that was with the RSV particle. That's kind of our lead. This is— we've demonstrated this in multiple different systems now. This is data from a transmission-blocking antigen for malaria, PFS25, which as a monomer is extremely poorly immunogenic. Most of the animals don't even respond to the antigen. When you put it on two different particles, you get several hundred-fold increases in binding antibodies. These are mirrored in functional antibody responses in a standard membrane-feeding assay where we're getting complete blocking of transmission. And so we've seen this in large animals, non-human primates, cows, across a variety of antigens. So things that we're doing now. So we're designing new nanoparticle platforms that are really specifically tailored to the display of specific antigens. And we've done this in several cases now and shown that you do get more robust particles out of that. We're co-displaying multiple antigens on the same particle in an attempt to activate cross-reactive B cells. So again, because of this in vitro assembly reaction, you can put into that reaction nanoparticle subunits that have different antigenic variants and make particles co-displaying those. And preliminary data is looking very promising there. We're designing next-generation nanoparticle platforms in which you can control the placement of functional elements on the particle. You can put this thing here, that one there. We are working on delivering these via genetic immunization, things like RNA vaccines, and that's going pretty well. We're packaging epitopes, T-help epitopes, immunomodulatory molecules, adjuvants on the inside and the outside of these particles. And then we're also designing them and evolving them, and I don't think I'm going to have time to talk about that, to target specific locations or cell types. So I'll skip through the evolution part and just thank the people who have been doing the work. So in my group, Dan Ellis, Brooke Viala, and Adam Orgacki did the bulk of the work that you're seeing here. It's extremely collaborative. We're working with a wide network of really excellent collaborators who are helping us evaluate these things. David Baker runs an amazing place at the Institute for Protein Design. It's a little shop of wonders. You should come by if you can. And thank all of our funders. Thank you for your attention. Is there any possibility that host proteins will interact with this highly multimer and create a more avid presentation of epitopes to the immune system? If you looked at the immune transcriptome before and after to see if there's other responses showing up. We haven't. It would be really fun to do so, and I think a lot of the technologies that we heard about today would be really fun apply here and see what other sorts of effects these particles are having. But we haven't looked at that at all. For influenza, it seems like this would be a really neat platform to mix in HA and NA. Yeah. Kind of recapitulate what the very nice structure looks like and see what— Yeah, absolutely. And I mean, from a manufacturing perspective, you can see advantages there as well, right? You can make a single drug product that that contains both antigens. So we're— yeah, yeah, yeah. So we're working on all of that. Yep. John? You really need to start keeping some of your vaccinated mice around for a year. Yeah. Your responses to the vaccine are going to be way exaggerated. Yeah. You can see more of a difference between the 1/3 loaded versus the 5/3 loaded. Yeah, we do need to keep them around. So we haven't done that yet. The furthest we've gone out was we did 6 months in mice and we did 4 months in non-human primates. And we see the same thing you see where you get this peak and then you get a decrease and then a plateau. And it follows the rank order of immunogenicity. But we need to go longer than that. We just haven't done it yet. One other thing we do know is that in non-human primates, the particles induce higher numbers of both circulating memory B cells and bone marrow resident long-lived plasma cells. [INAUDIBLE] Yes, yeah, and we're doing that now. We're not looking necessarily at— at retention of antigen. We're looking more at trafficking. And we're seeing some really interesting stuff. So the trafficking of the particles to the draining lymph node doesn't differ at all from soluble antigen, from trimeric antigen, which I thought was pretty interesting. But the localization within the lymph node is totally different. And it's really strikingly different. And we think that that's probably the basis for some of the enhanced immunogenicity. But we haven't looked yet at particle retention. Be a really cool thing to do. All right, thanks.