Scientist-to-Scientist Series: The discovery of therapeutic antibody cocktails

0:00:01.0 Dan Bedinger: Hi, I'm Dan Bedinger, the application science team lead from Carterra, and I'm here with Tom Yuan from Twist Biosciences. Tom, I wanna congratulate you on the recent publication in antibody therapeutics, and for all the great work you and your team are doing discovering antibody cocktails to Ebola. So Tom, can you tell me a little bit about your role at Twist Bioscience?

0:00:27.0 Tom Yuan: Yeah, so I am currently at Twist Bioscience in the Twist Biopharmaceutical. So a lot of my focus is on the discovery antibodies and the alchemization of existing antibodies. Like you mentioned, so for this particular publication, one of the workflows that we really wanted to highlight is, how do we harness the high throughput of twist synthetic biology platform with the high throughput PR of the Carterra OSA? This publication was first started by Jasmina and Aaron at Twist to really highlight how the two workflows really synergize with each other.

0:01:14.4 DB: Great. Is there anything else you'd like to say to introduce the paper before we go on to the first figure, or...

0:01:21.3 TY: Right. So this paper originates from the [0:01:25.3] ____ publication where there's a large number of neutralizing antibodies for Ebola glycoprotein. So what we're doing here is not the discovery of new antibodies. What we're doing here is the verification and clarification of the epitope bins, the competitive epitope bins of this set of existing anti-Ebola antibodies. What we're able to do at Twist, however, is that we can synthesize each one of these antibodies starting from just the protein sequence so we don't need any existing material. The only physical material that we need is the antigen itself, so this is the soluble Ebola glycoprotein antigen.

0:02:12.8 DB: It's very interesting. So can you walk us through a little bit about the Twist discovery technology and the workflow for this project in particular. How you were able to, in 28 days, get through this first set of characterization round and then follow that very rapidly up, I think, another eight days with the second tier analysis to even refine that further.

0:02:36.6 TY: Right. So because this is not on the outside any discovery project, we already had existing antibodies sequences to start with. We took those sequences, the antibody sequences, and reformatted them into our internal expression vector. At Twist we have our own internal expression vectors, but we're also able to take customer expression vectors and synthesize it in those vectors, as well. So in this case, we're inputting the antibody sequences originally discovered against Ebola and reformatting them and putting them into the synthesis pipeline. So that large 14-day span incorporates the algo printing, it incorporates the cloning of these assembled algo's into the full-length of VH and VL to express these antibodies.

0:03:34.7 TY: And it also incorporates the cloning into the final expression vector and the delivery of the DNA. So those 14 days encapsulates that entire process. So when we actually submitted this, we submitted this in the exact same way that a customer would submit the sequence via the ecommerce platform. Once we have those DNAs in hand, then we typically do a small 1 ml expression for all of these antibodies, and we like to automate as much as this is possible, so we automate these on applicant systems and they undergo a protein age verification. Once the antibodies are expressed and purified, which takes about four days, we set up a first pass binning. And the idea of the first pass binning is that, because you have so many antibodies and you have a single antigen to competitively bin against, the size of your matrix increases obviously for every single antibody that you encode in it. So the first pass binning, what that allows you to do is that you test for blocking of every single antibody against every other antibody for the same antigen.

0:04:50.2 TY: Because you have a larger and larger subset of that, it allows you to find the immuno-dominant epitopes. It also allows you to find extremely rare epitopes, as well, without any existing benchmark controls or known antibody binders to begin with. In this paper we specifically chose this antigen and these sets of antibodies because they were well-defined controls, but then the workflow of defining these epitopes is completely agnostic of having those benchmark antibodies available to begin with. So this entire process, including the expression and purification of the antibodies itself, these are routine workflows that we utilize at Twist for our both internal discovery projects and our client discovery projects. And the purpose of that first pass bin, as you've shown here, and according to the second figure...

0:05:52.7 DB: Yeah. It's figure two.

0:05:55.3 TY: Yeah, figure two, it's showing the basic format of running these epitope bins. So in this case, we're using a pre-mixed format for the epitope bins. We will directly couple the ligand antibody onto the chip and then pre-mix the secondary antibody or the aniline antibody at five-fold molar access with the Ebola EBOV glycoprotein antigen in solution. And then we'll measure for either blocking or non-blocking of that interaction with the ligand coupled antibody. So like I mentioned, for the first pass, it's completely agnostic of having any specific benchmarks, but what we're doing with this first pass is that we're generating sort of our own internal benchmarks that we'd like to use in the second pass binning. One thing to note here is that the first pass binning, because we have so many, and we do want to collect the most data as possible, not all the antibodies have expressed the same level.

0:07:01.5 TY: But we do get enough anti-reagent to process most of them, and we're not, for example, we're not individually normalizing the constitutions of every single antibody that's being expressed here, we're doing sort of bulk process where we'll dilute every antibody, we'll measure the constitution of every antibody, but we'll dilute these antibodies so that we reach a 5-fold molar access compared to the EBOV antigen itself. During the actual processing step for the epitope binning, we will double-check and QC the actual interaction of these antibodies with the EBOV antigen itself to make sure that we're fully measuring functional binding of that Ebola antigen with the coupled antibody, and that way we can be sure that any blockade that we measure during the epitope binning is actually a true blockade and it's not, say your antibody being degraded over the regeneration cycles or your antigen not being fully formed. So that's also why you see in figure C that it's not a one-to-one matrix where you have... It's more ligands than they were analyzed, and that's because we did have to remove some of those sections because we weren't able to fully recapture active binding in a portion of those antibodies.

0:08:29.7 DB: Yeah, and if they weren't high concentration, you couldn't necessarily assume that they'd fully inactivate the trimer, right? So you had to rely on those from the ligand side. Yeah, it's still very impressive that in this sort of single pass experiment, you were able to group more than 200 clones into these seven communities, quite a bit of resolution in throughput, for this type of assay I think.

0:08:55.0 TY: Yeah, that's what's exciting about running this in such a high manners, just by the power of re-testing the same ligands over and over and over again, that you can really define, or confirm that if it falls into immunodominant epitope that you have this big block, but if it's a rare epitope then that you're constantly getting non-blocking, except in that one specific matrix.

0:09:20.6 DB: Right. So maybe we can zoom in a little bit on Figure 3. So this is the second pass experiment where you focused in a bit more on a smaller number of clones. Can you describe sort of the experimental approach or what was intended with this optimized experiment?

0:09:40.9 TY: Right. So the idea behind the second pass epitope then is, like I mentioned in the first pass epitope binning in it's a quick and dirty assay. We obviously do go through the data and make sure that we see functional binding and blockade, but in the second pass binning, the idea is if we don't have existing benchmarks, we can self-select our own benchmark antibodies, what we call them pathway antibodies from that first set of epitope binning. And the way we select that is if we select at least one or two antibodies from each epitope bin or community block, community that we see in that first pass bin, and then we've confirmed that it has very good binding to both the glycoprotein itself, as well. In this case, we're also throwing in known structural benchmarks for EBOV glycoprotein just to see if they correspond to the initial bins that were assigned in the literature.

0:10:49.2 TY: So for the second set the epitope bin, we scale up the antibodies and the scale up allows us to collect more antibody and more region, and we're also normalizing inner part plates so that we have less loss of specific antibodies that don't express well, and we make sure that one normalized plate of all the antibodies have the same concentration and then we repeat the epitope bin. This let's us do two things. This let's re-confirm that the binning is able to be repeated and also that the rare epitopes that we do find still show the same blocking behavior as we had seen previously as well, so that's kind of the reason why we do the second pass bin.

0:11:41.1 DB: Great, so let's go to figure four now. So you do a great job in the paper of describing and comparing the SPR binning to the facts binning results. So can you tell us a little bit about how you view those two techniques and how the assay scale and fit into your discovery workflow typically?

0:12:06.3 DB: Right. So what we have done here is obviously different. This is SPR binning, high throughput binning. Fax binning is a great technique. One of the downsides was that it's typically a few on many approach, so you will have, say, some known benchmark antibodies. It will test many other antibodies against that smaller group of those antibodies. The problem with that is that you don't get that matrix, heat map matrix that you can generate with high throughput antibody, where one every single antibody is tested against every other antibody in the same set. And that way you lose some resolution. What you see in this specific figure is that we've generated halogenic tree of these antibodies and looked at their epitope bin, that was assigned initially in the borno paper by facts, so the outer ring on this in figure A or sub-figure A is the assignment from the borno paper, you'll see that there is some sections that are white, so those were antibodies that were not able to be assigned to an epitope bin initially, and then on the inner ring is the assignments that we generated from our high throughput epitope binning, completely separate from their binning.

0:13:34.1 TY: And you'll see that in many cases, the vast majority of cases actually that the bins completely agree, there are some cases where there's a disagreement, but for the vast majority of these clones, they completely agree, and you'll also see that we are able to assign some of the more difficult to bin antibodies via this technique that were initially not assigned.

0:13:54.1 DB: That's great. So I was curious from reading this, were you surprised to see that you had mAbs from the same germline family appearing in different epitope bins? And also about how much CDR variation do you think you need to see for them to separate into different epitope clusters or communities?

0:14:15.2 TY: Right. That's a really good question. So we do note that with some commensal pathogens such as staph aureus and yellow fever virus, that in many cases, the neutralization seems very strongly germline-encoded. For an antigen like Ebola or a virus like Ebola, we don't tend to see that. And you can see that both in the epitope bins from the facts stated and also through the epitope bins from the SPR binning, as well, where you'll see very similar... At least, by sequence it looks very similar in terms of the antibodies being very similar, but they'll bind to capacity, different epitopes. And this is one of the reasons where it's always critical to test for the epitope itself, not just assume or not base your antibody discovery efforts just based on sequence diversity. So that's part of the power of this and part of the advantage of using workflow like this.

0:15:16.5 DB: That's a really good point. Let's see if we go to figure six. So can you talk a little bit about how the mechanism of action of these different antibodies varies and maybe just kind of give us your interpretation of this figure and what you found?

0:15:39.8 TY: Right, so in general the mechanism action, in terms of finding antibodies to successfully neutralize a viral pathogen, it's obviously very important to broaden your epitope diversity as much as possible. Part of the reason of this is that, for example, in say, the common cold or influenza, there are immuno-dominant epitopes that the virus has evolved to be highly immuno-dominant, so most antibodies their would generate for a specific epitope but that is highly mutated, year to year. So that's part of the reason why we don't have a single influenza vaccine that can work indefinitely, right? So in some cases, for example, for SARS-CoV-2, certain antibodies will only bind, say, a dumbed-down version, of the spike protein itself. Not all antibodies will be able to access both confirmations. For any antibodies that we discover, we obviously want to be able to hit as many epitopes as possible. Whether that's directly blocking, say, the binding domain that actually infects or whether you're binding to another domain and another epitope that's able to lock your antigen, whether it's Ebola GP or whether it's the SARS-CoV-1 S1 spike protein into a confirmation that is unable to confer infectivity. So being able to broaden your epitope diversity is really key to that.

0:17:32.7 DB: Yeah, there's quite a few surfaces that can provide potential neutralization activity for some of these viral constructs, so it's great that this model system had so many different mechanisms involved that are distributed around protein. It's a really great case for this type of epitope binning, I think. So Tom, we really wanna thank you for taking this time today with us to talk about this paper and describe your research on Ebola and antibody discovery in general. And so we know you guys are doing a lot of work also on the COVID-19 project, and we're rooting for your success on that. And yeah, so thanks for meeting with us today. Is there anything else you'd like to say?

0:18:30.8 TY: No, just thank you, Dan. Thank you, just to yourself and also Carterra for having me on. I really enjoyed working with the LSA system itself and with everyone at Carterra, and I think this whole concept of trying to... I think synthetic biology and the LSA platform really go well hand-in-hand. And I think we'll see more and more great data come out of it.

0:18:55.9 DB: Alright, well, thank you very much. Have a great day, Tom. Thanks for talking with us.

0:19:00.5 TY: Thank you.

0:19:00.6 DB: Bye.