Founder Stories: John Desjarlais, Optimization & Application of Fc Containing Bispecific Antibodies

First of all, thanks, Paul, and thanks, Mike, for inviting us to speak at the conference. Just a little bit of background of ZenCore and, you know, just to set up the, you know, the discussion of bispecific antibodies. So, you know, most of you know ZenCore as, you know, kind of one of the leading Fc engineering companies. The idea there was to, you know, move beyond sort of the standard palette of tools you could use to, you know, by targeting different antigens with the antibodies and to add a new dimension by engineering the Fc domain. But if you think about it, those are really bispecific antibodies, right? They're engaging important immune receptors like FcγRIIIa or FcγRIIb, and we've been able to use those tools to develop some very interesting molecules. Xmab5574 is now known as Morphosis-208, which is in Phase 2 for lymphoma. Xmab5871 is a Xencor program in Phase 2 studies now for autoimmune diseases. And these two are a nice comparison. They're both anti-CD19s, exactly the same FAB domains, they just have different Fc substitutions. One recruits activating Fc gamma receptors, the other one recruits inhibitory Fc gamma receptor. But you can only play that game so long with Fc gamma receptors, so what about targeting other types of antigens? And, you know, we're an Fc engineering company, so we thought, well, we can also engineer Fc heterodimers to facilitate the production of bispecifics. So just again, a couple more slides. So we've got a lot of licensees of our various Fc technologies at various stages of clinical development. And we're developing our own internal as well as partnered pipeline with partners like Morphosis and Amgen. Amgen, we recently partnered our CD38 bispecific with, and you'll see progress in these programs throughout time. So this is not quite as comprehensive as Paul's slide. This is something I put together for our board a couple of years ago. In fact, some of these things may be out of date, so please don't come after me later because I didn't get your format right. It just shows you basically how things are evolving over time. And I think really that the general trend that we're all seeing is there's the first-generation bispecifics like the BiTE and the first-generation DART, which are very potent molecules. They've shown very impressive clinical activities. On the other hand, one of the main challenges there is just the short half-life, right? And how you deliver the antibody. And so there's been kind of a general shift towards the bottom of the slide, in particular kind of toward the bottom right, towards Fc-containing bispecific antibodies. And then the other challenge that we all face with this is just manufacturing, right? These are not always easy molecules to make. And so that was our goal in trying to make these bispecifics is come up with our own version of this is something that was very easy to produce. So our lead format is this thing on the right-hand side in the middle. And the idea is that, you know, we engineered CH3 substitutions to facilitate heterodimer formation. You know, we didn't want to be too cocky about that. We thought, well, we may not get 100% heterodimers. So we also built in the easy ability to separate these things by ion exchange. So either side contains a couple of charge substitutions to push the PI of the homodimers on either side versus the heterodimer. And then when we put it all together, we've got a really easy-to-produce bispecific system. My favorite thing about this, at least with respect to something like a CD3 bispecific platform, is it's plug and play. So we spent a lot of time optimizing the single-chain FV that targets CD3. And then the rest of it, you just plug in your FAB domain without reforming anything in the FAB. You just take a parent antibody off the shelf, out of an old patent, expired patent, you pop it in, you've got a bispecific antibody to work with. And we'll talk a little bit later in the talk about the idea of potency tuning to maximize therapeutic index. So just looking at a little bit of the components that go into making this technology, this is really just a, right now, just an idealized test system. This is literally just an FC. Showing you the behavior of different components of the technology. So on the top, all we have in here are the PI engineering substitutions that push the PI either way, and of course if you take an otherwise native Fc from an Ig1 and you produce, you know, transfect in equal ratios either side, you're going to get 3 different species. You're going to get about 50% of the heterodimer that you'd love to have versus 25% of either homodimer. So the point of the top is just to show that we did build in this really easy way to separate out these different species, just by standard ion exchange chromatography methods. But then we discovered our own CH3 substitutions, 2 on either side of the CH3 domain. And you can see again in this test system, we get about 96% of the preferred bispecific heterodimer. We also, you know, we went through several iterations of this with the real goal of making something that's essentially native-like in terms of stability. So we were doing lots of melting curves, trying different variants to get what we liked in terms of our final properties. And then the other important factor here is since we've got a single-chain FV targeting CD3, we want that to be stable and robust to manufacturing and robust to in vivo situations. So we spent, I think, about 6 months— Greg Moore led the way on this this, basically making libraries, really just a lot of brute force screening, looking for, you know, weird amino acids in the structure, things that might lead to better stability of the single-chain FE. So you can see on the lower left, we started out around 59 degrees melting temperature, and after a couple of rounds of stability optimization, we got up to about 72 degrees. So our goal had been to at least match the 70-degree melting temperature of the CH2 domain. You can see in the middle on the bottom, this is, you know, it's nicely humanized, and of course we're particularly interested in having something that's cross-reactive with cynomolgus monkey to enable not only safety and pharmacology studies, but also some nice pharmacology studies in the monkeys. Okay, so when you put all this together, again, we've got the heterodimeric Fc, we've got the stabilized single-chain FV against CD3, and we got the plug-and-play FAB domain for your tumor-targeting domain. We also— I don't have it on the slide— we also have Fc knockout substitutions so you don't get Fc gamma receptor-mediated cross-linking of CD3 and so forth. So we put all this together. It essentially melts like a native antibody, and so that was a good sign, but of course you never really know. You know, you're like, great. We passed that gate, you know, how far can we get? And you'll see as I go through the talk, we get more and more validation of the robustness of this platform. This is, you know, we found out we can make nice high-yield stable cell lines with this. We just play around with a couple of ratios. Of course, it's a 3-chain system. You've got the one heavy chain that has the SCFV, the other standard heavy chain, and then the light chain. So play around with a couple of ratios. And then you select your clones, and we're typically getting, you know, 1+ grams per liter production yields. And again, you can see that you can easily separate out the preferred bispecific species. And ultimately, after you do the purification, you get a high purity bispecific by ion exchange and favorable solution properties. So we started to make a couple of leads. Our first was a CD3, CD3 CD123 by CD3 bispecific, then a CD20 bispecific and a CD38 bispecific. And any of these, you just pop them in, you get, you know, you use the appropriate target cell lines, tumor cell lines that have the appropriate antigen, you get nice potent in vitro, we call it redirected T cell cytotoxicity. And then just by a nice antigen-independent measure of the robustness of the molecule and just sort of intrinsic half-life, We do studies in black 6 mice to look at the half-life, and you could see we almost always get a half-life between 6 and 8 days in a mouse. So what about pharmacology? So we're lucky in that the first couple of targets we chose, they're all monkey cross-reactive against the antigens, and they're all nice cell populations. You can find nice surrogate populations in a healthy monkey to follow in terms of pharmacology. So a little bit background, CD123, most of you know, is a really hot target in AML and a couple of other neoplasms. In particular, in AML, it was discovered in the early 2000s that CD123, which is also the IL-3 receptor alpha, is actually expressed on the leukemic stem cell population as well as the leukemic blast population. And a beautiful study here showing progression survival curves, basically. Shows that if targeting this— or basically assaying this population, if you've got too many of these leukemic stem cells, then you have a far worse prognosis than if you have a lot fewer leukemic stem cells. So great target. We get nice in vitro pharmacology, nice half-life in mice. What happens when we go into xenomalous monkeys? This is just a single-dose study. We had to play around a little bit in terms of figuring out what are safe and effective doses, and it's active over a very wide dose range. The nice thing is basophils are a beautiful population to follow in terms of the synomolgous monkeys. You can gate them on HLA-DR and Fc epsilon R1, along with a bunch of lineage-negative gates, and you get this nice clean population followed by flow cytometry. They are very bright for CD123. We give a single dose of either 10 micrograms per kilogram or 100 micrograms per kilogram., and you see basically within about 15 minutes you've just wiped out the basophil population. And that lasts, that's pretty sustained for almost a week. So we see this as something where we could dose weekly in the AML setting. We also, fewer time points, we're able to look at bone marrow basophils, and once again we're essentially crushing that population. But you do see recovery ultimately because your antibody goes away and you've got intact hematopoiesis and the cells recover. We see all the hallmarks. This is starting to get pretty perfunctory when we do these studies of the T cell redistribution. So if you look in the first couple of days after dosing, here we're counting the CD4 T cells. They disappear, right? They're not depleted, they just, they redistribute, they marginate, whatever they're doing. I'm not sure anybody really knows, but it always happens when you've got an active targeting bispecific. We see with the very small number of cells you can actually count, you can count enough and acid them, and you can see nice activation markers like CD69 or CD25. I'm getting a phone call, that's nice. And then, you know, likewise on the CD8 positive— it's probably my kids calling from school or something. And then, you know, likewise on the other populations. And, you know, we also do see cytokine release. We found that IL-6 tends to be the most informative in terms of the balance between activity and toxicity, and so that's another measure of the pharmacology. And of course, we can go into mouse model. You know, we can debate all day the relative merits of a monkey versus a mouse model. I like to have both. Particularly in this setting, I like having the mouse model because an immunocompromised mouse might actually look something like a chemotherapy-treated human patient in terms of their T cells. So our version of the model is we start with NSG mice. We have luciferase-labeled RPMI— I'm sorry, this is KG1As, which is a CD123 bright AML cell line. We give those IV, and then you just inject them. The cells go away. You can't see them for a couple of weeks. They ultimately show up in the femurs and the spine, so it really is like a myeloma. It localized to the bone marrow. And then around day 22, in this particular study, we basically wait until we have highly visible tumors by the luminescence. We give them both the drug and the human PBMCs at the same time. And you can see the control groups, basically that tumor just keeps growing. If you just give them PBMCs, it keeps growing for a while. Ultimately, you do start to get an allogeneic effect of those PBMCs against the RPMIs. But of course, if you give the bispecific, we were hoping to have a dose response here, but we kind of overshot. We went down to 30 micrograms per kilogram. Started as high as 1 mg per kg. We got no dose response because it's a very potent antibody. But you essentially eradicate the tumors. And this tells me a lot, right? We count about 40 T cells per microliter in these mice, which is— that's compared to about 1,000 T cells per microliter in a healthy human patient. The T cells and the disease start out in different places, right? So you've got disease established in the bone marrow, you inject the T cells IV, they find their way there by guidance of the bispecific antibody, and they can eradicate the disease. So I think that's very promising for these, even when you've got lower T cell counts. All right, moving on to CD20, a system everybody knows knows very well. Again, we plugged in an anti-CD20 targeting FAB domain, same exact anti-CD3 domain. We saw— you saw this data earlier, nice potent in vitro killing and about a 7-day half-life in the mice. We put those into the cynomolgus monkeys. Here we were able to dose a little bit higher, so we've got a— we only have PK data for the highest dose because we had a really poor PK assay at the time. So we got about a 5-day half-life at the 3 mg/kg dose. But, you know, for what it's worth, we're seeing a nice healthy half-life. You could see that either the 0.3 or the 3 mg/kg dose pretty much wipes out peripheral B-cell populations. You're seeing maybe a tiny bit of recovery by day 30 with the lower, the 0.3 mg/kg dose. And then importantly, we're able to look at the bone marrow and the lymph node B-cells. And of course, anybody that's worked with these tissues knows it's hard to get absolute counts that you can trust. So we play the same game other people have, which is to look at the B-cell to T-cell ratio in these samples. So you can see, you start out with a ratio of around 1 in the bone marrow, and essentially you go down to a very, very low B-cells versus T-cells. Likewise, in the lymph node, you start with a ratio of somewhere between 0.5 and 1. And again, you absolutely wipe out those lymph node B cells. For reference, if you look up some of Genentech's studies on rituximab in synomolgous monkeys, if you pump in multiple 10 mg/kg doses of rituximab, you'll go down about half of where you started in terms of the lymph node B cell population. So these bispecifics are incredibly potent. They can get into the deeper tissues and really make things happen. And again, you know, it's the same kind of stuff that you see on the T cell. You know, T cells redistribute, you see activation, so on and so forth. Okay, so we're developing the CD123 and the CD20. Both of these are expected to enter clinical development next year, first in humans starting next year. We partnered off the CD38 program with Amgen. And then, you know, this may be the most important slide of the talk. This is where we are in terms of manufacturing, which is one of the more difficult things with these bispecifics. But I'll say with this platform, it's actually been smooth sailing all the way. So we've got our process and product validated at GMP scale for the 14045, which is the CD123 bispecific. We did a 200-liter scale run, 100-liter working volume. We got 2.2 grams per liter. It was a standard 3-step purification process. We've now got 1-month drug substance stability data, which is unchanged at 10 mcg/mL, 5 degrees centigrade. Which means we've got, after a month, we've still got 100% bispecific by capillary electro— by, I'm sorry, CEX, and then 99% main peak by SEC. We're seeing incredibly similar, in fact, our upstream and downstream process were identical for the CD20 bispecific. Same yields, in fact, at around 2.2 grams per liter, kind of a crazy coincidence. And we saw very similar things with the CD38 program, but now of course that's in Amgen's hands. And we're, you know, early data suggesting that the process is pretty similar for multiple programs. In fact, we love the platform because we make dozens of these a week. We can test a lot of hypotheses with this. And so you can see a whole host of different systems here. We've made numerous other CD3 as well as non-CD3 bispecifics. But, you know, as you start moving beyond some of the initial, you know, easy HEMONC targets where it's easy to look at synopharmacology, you start to ask new questions. And that'll be kind of the rest of the talk is how do we think about assaying these things and what can we find out about these things in an in vivo situation. So the first example was we made a PSMA bispecific. Again, you know, you just plug in an old anti-PSMA antibody or maybe we humanized it first. We see nice, potent in vitro killing. The question is, now when we put this into a monkey, how do we know if the thing is working? Right? So we set up this cool study with a very important negative control. We made RSV, this is basically Synagis, you know, coupled with CD3, the same exact CD3 domain. We dosed that at 3 mg per kg, right, a pretty high dose. We figured it's going to be pretty innocuous, we can dose it pretty high. Pretty high, and then we dosed the PSMA right where our sweet spot was for the other molecules, around 30 micrograms per kilogram. We put them both in a monkey. We weren't sure that we'd see anything in terms of— we weren't sure what to look at. We tried to look at the prostate, but, you know, that's not so easy. But what you do see, which is remarkable, is with the 3 mg/kg dose of the RSV bispecific, essentially nothing happens. You don't redistribute any T cells. You don't see any CD4 activation, and you don't see any cytokine release. On the other hand, you put in the PSMA bispecific, you see nice T-cell distribution, you see nice activation of the T-cells, and you see a nice level of IL-6 release. And in fact, the level of IL-6 we saw for this maps almost perfectly on the curve we've developed with about 100 different monkeys in multiple studies for multiple programs. So this tells us, okay, we are engaging target, something active is happening here, and we're dosing appropriately. Another thing that we've— question we've had to ask as we work with different targets is, you know, are we in the right place in terms of the potency of the molecule, right? So, and maybe we can play around with the potency to improve the therapeutic window. So of course, cytokine release syndrome is a common problem with both CAR-T and CD3 bispecific approaches. Our hypothesis was that we could improve the therapeutic window by modulating the potency, and the ideal potency is likely to depend on the target. And so, you know, putting it more simply, you know, you're— there's sort of a rock and a hard place you can be at if your format is too potent. If it's too potent and you give too high a dose, you get too much cytokine release and you literally kill monkeys, and you don't want that to happen to humans. On the other hand, if you have to dose it very low and you've got any appreciable antigen sink, You put in your molecule, you see a little bit of pharmacology, and then it's gone, and you're done, right? So how can you navigate these two extremes and come up with a solution to that? So Greg thought, well, we could play around with CD3 affinity, and he started making a library of our CD3 anti-SCFv, came up with a couple of different affinities covering a very wide range. I'm not going to tell you what the KDs were, but I can tell you that the CD3 high is essentially like a normal antibody. The CD3 low is pretty much way down with like a TCR-type affinity. Right? We've got the high, intermediate, and low. This is in the CD20 system, just as a nice test system. You can see that the cytotoxicity goes as predicted with the affinities. And then also interestingly, if you look at the T-cell activation, note with the CD3 low we can get 100% depletion of the target cell population, but we're actually getting lower activation of the T-cells. And that's kind of hinting at what we're looking for here is let's keep the cytotoxicity but not activate the T cells quite as much. So we did another monkey study with the CD20 system. And, you know, you have to kind of take some guesses at doses because you know you ought to be able to dose higher with the lower potency versions. So we dosed at 3 mg per kg for the CD3 low, 3 mg per kg for the CD3 intermediate, and then 0.3 mg per kg for the CD3 high. And note, despite the 10-fold higher dose of the lower affinity versions, we actually have— this is a log scale on the IL-6— we have significantly lower IL-6 being produced. On the other hand, because we can dose much higher, even though it's a less potent molecule and less active over time, it stays around a lot longer above the antigen sink, and so you can see very similar B-cell depletion pharmacology. What you do see here is that the CD3 low in this particular system might just be too low. We're not getting quite the same amount of B-cell depletion. Depletion, at least quite the sustained amount of B-cell depletion. But I've, you know, I'm just showing you one plot here. I've stared at these day after day comparing the 3 mg/kg dose of the intermediate versus the 0.3 mg/kg dose, and they look identical in terms of pharmacology. And so we can get away with a higher dose and lower cytokine release. And I'll end now. I'll just say there's other formats we can play with. A lot of different ways. Once you have a nice stable CFP, you can work with a nice solid Fc heterodimer and an easy way to purify. And then the final slide is just, you know, we're continuing to develop these bispecifics, multiple additional CD3 bispecifics, also playing around with checkpoint bispecifics. And hopefully you'll see a lot more from us in the years to come. Thanks for your attention.