Posted by Noah T. Ditto
During the COVID-19 pandemic, new technologies have enabled the biopharmaceutical industry to develop therapeutic antibodies fast.
Andrew Bradbury PhD, Chief Scientific Officer of Specifica, will discuss their next-generation antibody library platform that provides robust antibody diversity and delivers superior therapeutics. Andrew discusses how Carterra’s high throughput kinetics and epitope binning technologies enable the rapid discovery of SARS-CoV-2 neutralizing antibodies.
Noah T. Ditto, Technical Product Manager at Carterra will discuss how high throughput Surface Plasmon Resonance (HT-SPR) facilitates a paradigm shift in antibody screening and characterization—now discover therapeutic candidates in days vs. months.
0:00:07.4 Suzie Valdez: Hello everyone and welcome to today's webinar, Developing High Quality Therapeutic Antibodies Straight From Selections in Record Time. I'm Suzie Valdez of Labroots and I will be your moderator for today's event. Today's educational web seminar is presented by Labroots and brought to you by Carterra. To learn more, visit them at carterra-bio.com. Now, we encourage you to participate today by submitting any questions you may have during the presentation. To do so, simply type them into the Ask a Question box and click Send. We'll answer as many questions as we have time for at the end of the presentation. You may also submit any technical issues here as well if you are having trouble seeing or hearing the presentation. I'd now like to welcome our speakers, Dr. Andrew Bradbury, Chief Scientific Officer at Specifica and Noah Ditto, the Technical Product Manager at Carterra. Andrew and Noah, welcome to both of you. You may now begin your presentation. [pause]
0:01:19.0 Andrew Bradbury: Hi, my name's Andrew Bradbury and I'm the Chief Scientific Officer and co-founder of Specifica. Today, I'd like to tell you about generating high quality therapeutic antibodies directly from selections from the Specifica Generation 3 Platform. So this is work that I've been wanting to do for a long time but was unable to do in the academic space and have been able to do within the company, within Specifica. And so our library platform is based on well-behaved clinical antibodies. They are substantially liability-free, and I'll explain how we do that. We have a very high functional diversity and finally, we provide our libraries exclusively, and the reason we do this is because there's a famous case where Human Genome Sciences and Genentech both selected antibodies against the same target from the same library and ended up with the same antibody, and both parties tried to patent that antibody. And Genentech was seven years earlier, and so, got the patent.
0:02:25.2 AB: Now, the result of this design, library design, is that we end up with superior therapeutic antibody leads, so over 80% of the tested antibodies have no biophysical liability, over 60% have affinities less than 10 nanomolar, and about 20% is subnanomolar, and we get between 500 and 5000 different clonotypes per target. So in other words, we're getting very, very high diversities, good quality in terms of developability and high affinities.
0:02:56.9 AB: So how do we make this library? So the way... Before we get there, a couple of definitions. First definition is sequence liability... A sequence known to create potential problems, and I've given a few examples there. A sequence liability is theoretical. It may not always be a problem within all sequence contexts, but it does tend to be a problem when found within CDRs and that's because the CDR is a solvent-exposed, and so these sequences are available for modification. Now, biophysical liability on the other hand is a liability that can be physically measured. For example, polyreactivity aggregation and so on. And these are liabilities you can measure in vitro and they often are a proxy for in vivo problems. So an antibody that has aggregation may then have a reduced half-life, may cause an immune response, you may get anti-drug antibodies and so on.
0:03:51.7 AB: So what we did was we started off with the best therapeutic scaffolds and our reasoning is to start with antibodies that have actually been in patients, have been injected into people and behave well within that context rather than starting with antibodies that are either very common or commonly used or antibodies that are
[0:04:10.6] ____ generalized. And so we isolate natural CDRs by next-gen sequencing or PCR, we then eliminate those CDRs that contain sequence liabilities, and embed those within the scaffolds. And so what this results in is a number of different scaffolds, each one of which is a sub-library containing many, many different CDRs, each of which is natural and has been purged of all liabilities. So note that, here, we have no diversity. All the diversity within the CDRs is coming from natural CDRs. There is no synthetic diversity within a CDR. The diversity is coming from the combination or the combinatorials of putting different CDRs together in large numbers.
0:04:56.4 AB: So how do we choose the scaffold? Well, we relied on this paper by
[0:05:00.0] ____ et al, which went through all the therapeutic antibodies of various stages. And in the rows, you have different antibodies; in the columns, you have different biophysical assays. And if you assess these based on their properties, then red flags, so in other words, biophysical measurements on individual antibodies indicated in red, represent biophysical liabilities that are defined as the 10% worse values among the approved drugs for each assay. And so if you look at this table, what you'll see is a population of about 40 antibodies at the top that have no biophysical assays and then the number of biophysical red flags, sorry, increases as you go down the table.
0:05:47.5 AB: So what we did was we chose five antibodies at the top representing genetic diversity that had no red flags and one antibody at the bottom, Sirukumab, which was chosen to be a bad antibody, specifically for the assays that we're going to do. Which are the library scaffolds? So I've indicated them here and you can see that we have VH diversity, VH germline diversity 1, 2, 3 and 4, and then we have VL diversity as well, the kappa 1, 2, 3 and 4, and we also have V-lambda. You'll note that in the last column, we have the levels of immunogenicity and on the whole, all these antibodies have relatively low immunogenicity.
0:06:32.0 AB: The first strategy, again, quickly, we take the CDRs, all coming from real antibodies, we
[0:06:40.7] ____, which are synthesized for all but HCDR3. They're extracted from our proprietary data sets. We eliminate errors and sequence-based liabilities and then synthesize with DNA in arrays. For HCDR3, we used natural HCDR3s directly amplified from about a billion B cells and from at least 10 donors, and these will, because we're not doing any pre-screening, contain some liabilities. So if you look at the liabilities we're eliminating, and this is VH, we have a list here, and you can see that in CDR1, eliminating all those liabilities reduces the number of CDRs available by about 60%, and that's about 70% in CDR2.
0:07:25.6 AB: Now, one of the nice things about using natural replicated diversity is that it flattens abundance differences. So if you do next-gen sequencing on natural source, then you can see that the abundance range goes from... Is 300,000 to 1, going from the commonest CDR, which is the germline, down to the least common, which is usually a multiply mutated CDR. Compare this to the natural replicated CDR abundance range, which are about 60 to 1, about tenfold of which is due to that steep drop-off, which is due to various errors in sequencing, PCR, or in oligo synthesis. So the distribution of the oligonucleotide based diversity is about 5000 times better than natural and we use this for all CDRs but CDR3 of a heavy chain and other CDRs from other libraries show similar sorts of distributions. Here are some examples of CDRs with sequence liabilities that would have been eliminated and you can see deamidation, isomerization,
[0:08:36.2] ____ oscillation, unpaired cysteines and so on.
0:08:41.5 AB: Here we have the progression and improvement in our HCDR3 diversity, and so in yellow, at the bottom, you see what we call our Gen 1 Library. This is a library that we produced or I produced in my academic life about 20 years ago. I thought it had a diversity of about just under 10 to the eighth. It turned out the true HCDR3 diversity was in fact three million, so about 34 less than expected. Then in green is a library we made from a single donor, we just published that in Communications Biology, and you can see that from a single donor, we're increasing the diversity and that actually continues significantly above that. In gray, we have a Gen 2 Library. So Gen 2 is where we take the four VHs, the four VLs and we combine them together. This is a library made from 15 donors. You can see that we have a diversity in the HCDR3 of just about 0.6 times 10 to the eighth. And then finally, at the top, you have the two Gen 3 Libraries that we're demonstrating on this figure, which have a diversity of about 200 million in the HCDR3.
0:09:48.9 AB: We've now made, altogether, almost 70 libraries, and so we do quality control on them. We're getting very good at making these libraries. We understand where the problems are, where the bottlenecks are, and have improved our platform. So we can now make a library in about a couple of months or so. Testing it and quality controlling it takes a bit longer. So no clonal dominance in the individual libraries as shown in this table here. You can see that the most abundant HCDR3 in any single library is just over 0.6% and in the combined libraries, is about a quarter of a percent, and the most abundant VH when you take all the CDRs into account is no more than 0.05%, and this goes down tenfold when you look at the libraries altogether.
0:10:33.1 AB: So great design but how does it work, obviously, and are we getting good affinities and good diversities? And so the specific Gen 3 Platform starts with phage display. We do a couple of rounds on phage. We then transfer over to yeast in a single-step conversion, and this allows us to fine-tune affinities, specificity, expression levels and so on. And then what we do is we pick random clones and sequence them, and we also do next-gen sequencing and apply machine learning to identify different clusters, and this allows us to inform in both directions.
0:11:08.2 AB: So by virtue of the cluster identification following next-gen sequencing, we can see which random clones that have been picked fall into different clusters and then examine them further and we can also identify those clones that are being more common, both in random clone picking and next-gen sequencing. This shows a selection sort of profile for selection against GM-CSF. Each dot represents a yeast displaying between 30 and 100,000 of the same antibody, has some copies of the same antibody. On the X-axis, we have the single-chain display level, on the Y-axis, the amount of antigen bound.
0:11:47.4 AB: And so what we're interested in is dots found in that upper right quadrant. And you can see that when we go down to a concentration of target of about 1 nanomolar, we still see over 40% of yeast are binding the target, indicating they have pretty good affinities. When we sequence these using PacBio, and the reason we use PacBio, it allows us to couple VH to VL genes, we identified different clusters which are indicated in the columns... Of heavy chain, sorry, in the columns, light chains in the rows. And what you can see is that when we read about 20,000 reads from different concentrations, the number of clusters we identify is between 800 and about 3 and a half thousand.
0:12:33.0 AB: Now, you may ask, how different are these clusters one from another? And that's illustrated here. Looking at the Levenshtein distance when we compare all CDRs one to another, you can see... Or the HCDR3, sorry, one to another, you can see that the peak difference is about 10 to 11 amino acids. So when we say that two clusters are different, they are actually very different indeed, in that the average difference is about 10 to 11 amino acids. Now, instead of looking just at the HCDR3, you look at the full compliment of CDR, so that's all six CDRs concatenated together. Now, you see that the average difference is between 20 and 40 amino acids. So when we say that we have two clusters, these clusters are not the same. They're not differing by just one or two amino acids, these are very different antibodies.
0:13:24.5 AB: We found that different targets prefer different sub-libraries, as illustrated here, so GM-CSF prefers the VH1 and V-lambda library, interferon prefers the VH1 and V-kappa 1 library, and IL-6 prefers the VH3/V kappa 2 library. However, although targets may prefer particular sub-libraries, here's an illustration where we did a selection against the mixed libraries. You can see we're getting very good population directly after phage, and that is dominated by sub-libraries C and D. As you can see, both of these after phage, we're getting good populations. However, if we do individual selections against from the other three libraries, A, B and E, you see that we start off with less abundant populations after phage, but by the end of the two rounds of sorting at 2 nanomolar, you can see that we're getting very good populations here in the upper right quadrant for all of them. So essentially, we've found that, usually, we can select antibodies from all sub-libraries but there are some sub-libraries that are preferred by different antigens.
0:14:32.4 AB: We find that most sequence liabilities are eliminated from antibodies coming from the Gen 3 Platform. And here's a comparison between antibodies that were selected against the target using the Gen 1 Library where you can see that only less than 10% of antibodies have no liabilities whatsoever and about three-quarters have between two and four liabilities. When we look at the Generation 3 on the other hand, we find that three-quarters of the antibodies coming out in this particular case have no liabilities. These are biophysical... Sorry, these are sequence liabilities and about 20... Most of the rest are found in the HCDR3 only and 80% of those are single liabilities.
0:15:15.2 AB: When we do high-throughput SPR using the Carterra LSA, you can see that we get very high affinity antibodies as indicated by these SPR plots. And when we plot them out on an iso-affinity plot, you can see that, directly from selection, we're getting about 40% of antibodies between 1 and 10 nanomolar and 20% which are subnanomolar. And what we've done here is we've selected against a bunch of different targets.
0:15:42.3 AB: Now, it was very difficult being in the antibody business not to be responsive to the pandemic and select antibodies against SARS-CoV-2. And we, of course, weren't the only ones to have this idea and in fact, I would argue that what we've had with the pandemic is a global experiment that has been focused on a single target. It's like a big competition in the likes of which we've never seen before and everybody who has anything to do with antibodies have selected or generated antibodies against the spike protein, and it allows us to compare different platforms for their antibody properties.
0:16:19.7 AB: So what we did was we selected antibodies against different forms of the S protein using either the Generation 3 Platform, as indicated in the top row, or two libraries made from a convalescent burner with very, very high titers either using the kappa chain or the lambda. And what you can see is that the number of yeast after going through the selection from the Gen 3 Library that are binding at 1 nanomolar is a little bit less but very similar to what you're seeing with the Immune-Kappa Library and much better than what you're seeing with the Immune-Lambda Library.
0:17:11.4 AB: The protocol we were following was to take the Phage Library to do two rounds of phage selection, three FACS guided sorts, take that output, convert it directly into IgGs, sequence 96 clones, we identified 23 different sequences, and then went to antibody lab validation. As you can see on the right in the ELISA above and in the flow plot below, most of the antibodies that we generated against S1 recognized the RBD as indicated in these assays.
0:17:49.1 AB: When we did a kinetic study on these antibodies using the LSA, what we have here is the selected antibodies, 23 of them, with affinities determined against the RBD in a blue circle and against the trimer in the blue triangle, and these are to be compared with antibodies selected from the Scripps from the Burton Lab. And again, yellow circles against RBD, yellow triangles against trimer. And there's one antibody which is able to cross-neutralize both SARS-CoV-1 and CoV-2 which is in red. What you can see is that the affinities of the antibodies that we've tested here are pretty similar to those coming out from the convalescent patients used by the Burton Lab at Scripps, and in some cases, perhaps even better.
0:18:48.0 AB: When you compare the affinities against the RBD and the avidities generated against the trimer, there's usually an increase as you'd expect and you can see that in the raw numbers on the left and then in the change in the trimer avidity versus the RBD affinity on the right. And basically, the average increase in affinity is about sixfold. However, there are a couple of antibodies here that, in fact, decreased their binding affinities upon, when tested against the trimer.
0:19:23.5 AB: We also used the LSA to do some antibody binning and the most common bin was the same as the most common bin identified by Dennis Burton, a bin we call 3b. This recognizes the RBD and is the most potent neutralizing bin. In addition, we have a bin called 4b which splits and otherwise, a bin that Dennis had identified into two by distinguishing between the bindings of two antibodies. And then there was a bin... One of Dennis's antibodies and bin 1b, which is that antibody I mentioned before, which is cross-neutralizing SARS-CoV-1 and SARS-CoV-2. When we tested all these 23 antibodies for neutralization in a pseudo-viral neutralization assay, you can see the curves on the left and the IC50s on the right, the best is 1.7 ng/mL, and that compares very well with the two antibodies at the bottom coming from convalescent patients, 48.6 and 6.4 ng/mL. We tested the top 10 for live virus neutralization. The results are indicated here, and the results are actually slightly better here. So the best again, same antibody was 1.3 ng/mL, and there's quite a few actually, there's four antibodies total with IC50s less than 2 ng/mL. Now, interestingly, there was poor correlation between the affinity of the antibodies recognizing the RBD or the trimer with the IC50s in live virus as illustrated here. And this applies both to our antibodies, as well as the antibodies that were isolated in convalescent patients.
0:21:20.7 AB: We took these 10 antibodies, also tested them against the UK and the South African variants, or those variants found in those countries originally. And as you can see 1, 2, 3, 6 of the antibodies are basically as potent against these variants as they are against the original wild type with some of the others showing reduced activity against some of these variants.
0:21:48.0 AB: Now, as I mentioned earlier, this is a bit of a global experiment going on here. These were some of the first papers that came out, the top top papers with antibodies on the whole generated from convalescent patients. And if you... And these are all portraying IC50s in various ways, either as curves or as dots. And if you plot our best antibody against... On the same scale as all these, you can see that on the whole, our best antibody coming from a naive library here is tracking with... Usually with the best antibody identified in each of these different publications. There are a couple of exceptions, so publications three from Zoster et al, were about twofold less, and the publication by Hanson et al were also about two fold less as well.
0:22:46.0 AB: Now, if you take each of those publications and all the other publications that I've been able to find that have been peer reviewed, and you identify the top... The best antibody, either for affinity or for IC50 against pseudo-virus or against live virus... I portrayed that here in this particular slide with a black circle, and you compare all the antibodies in the published record, in the peer reviewed published record, against the best 10 antibodies coming from these 23 that we originally isolated, what you can see is that the selected antibodies coming from the Specifica library indicated as selected antibodies compare very well with antibodies coming from immune sources and much better than those coming from naive libraries.
0:23:34.5 AB: That's true for affinity. It's also true for pseudo-virus neutralization. You can see that all our antibodies here are on the right-hand side are better than the best other antibodies coming from naive libraries, and are very similar to those found from immune sources. And that is also true for live virus neutralization.
0:23:55.2 AB: So what it seems just from taking those very small collection of antibodies and testing them in these various assays is that we've essentially been able to replicate the results of antibody generation from immune sources from a naive library.
0:24:12.1 AB: Now, as I mentioned, we sequenced 96 clones, we identified 23 sequences, these fell into 12 clusters and tested them. Those are the results I've just shown you, but we wanted to go beyond that. And so what we did was next-gen sequencing analysis of three yeast binding populations as illustrated here. And after we did our machine learning, we ended up with a total of just over 600 different clusters. And these overlapped to various extents, depending upon the... Upon the sequencing and the overlaps are all defined at the next-gen sequencing level. Now, some of these are...
0:24:48.7 AB: Have been generated specifically for specific targets, which is why some of the overlaps are not as large as you might expect. So we took 178 of these additional clones representing 70 additional clusters, synthesized these, expressed them and further analyzed them. And so this resulted in a total... When we tested these antibodies, this resulted in an additional 82... Sorry, additional... 70 additional clusters.
0:25:19.2 AB: So now when we tested the affinities again, using the Carterra LSA, and compared those to the antibodies that we had previously identified, which are indicated in pink salmony sorts of circles, these new NGS clones are in the blue circles. The previously identified antibodies coming from convalescent patients are in red triangles. What you can see, interestingly, is that there is a whole population that's jammed up against the far left, where essentially, what we're seeing is a flat-lined off rates. And these are really beyond the capture kinetics of the Carterra LSA and indeed any SPR platform.
0:26:03.3 AB: And these are illustrated here. You can see these flat lines indicated in the arrowed affinities, and the affinities here, rather than giving a specific value, they're all being indicated as less than a certain amount.
0:26:19.3 AB: So now, when we take all these antibodies and we again compare them to the published record for affinity, then one thing that's clear is that we have this off-rate traffic jam here of antibodies that are beyond capture kinetics capabilities. And so we're going to take these antibodies and we're going to test them now for their affinities using the Conexer which is the only platform that really is able to get down to the low... Very low picomolar and fentamolar levels. It is also however true that... That's a misprint it should say, 31 antibodies... We have 31 antibodies with affinities less than a hundred picomolar. In fact, it's 94 antibodies with affinities less than one nanomolar. And again, these are all coming from the naive library which represents remarkable results without any immunization.
0:27:14.8 AB: The last thing I'd like to talk about is developability. As you remember, we generated the CDRs, and purged them of all liabilities all sequence-based liabilities. So the question is, what happens when you make real antibodies, do these have no biophysical liabilities? So what we did was we took single chains, we converted them into IgGs, we then tested them for these different assays, these different properties using these different assays, and the antibodies we tested were Sirukumab, which as you remember, is that known poor developability antibody, the parental scaffold which provided us with a baseline and four clones from each library. So here is an example of size exclusion chromatography of a single clone coming from library A. You see here aberrant peaks for Sirukumab which is probably interacting with the columnar matrix in a strange way. You also see a broader distribution representing an aggregate shoulder for Sirukumab. We also see additional aggregation, which is particularly pronounced for Sirukumab over at lower retention times.
0:28:28.2 AB: And then finally, you see this single peak with four overlapping plots where the parental and the clone which have been untreated are in the lighter colours, and the parental and the clone which have been frozen and thawed 10 times are in dark blue for the parental and orange for the clone. And you can see that these figures, these plots essentially overlap, indicating that the antibodies are essentially as good as the parentals from which they were derived. Here is four more plots from the same series, and again, you see that the plots are essentially overlapping, indicating again, that the clones derived from a particular parental antibody are behaving almost as well as the... Or as well. And you'll see in a second, sometimes better, than the parental from which they were derived. So we did a number of other measurements, including AC signs, SGAC signs, poly-specificity, HEK titer.
0:29:31.5 AB: So what we have in this plot is for each set, is the parental is indicated in red, the clones are indicated in black. If a clone is within the therapeutic limits, then it's black, if it's better than the therapeutic limit, it's blue and it's beyond the blue line, whichever direction that is. And if it's in the other direction so it's worse than therapeutic limit, then it's in orange. And so what you can see, and this represents two standard deviations from the parental mean, and what you can see first, which is quite striking, is that all the clones that we have derived from these different parental scaffolds essentially cluster with the parentals themselves. However, there are a few examples where the clones coming out are better, for example, in the AC signs, there are four antibodies there that are clearly better than the parental from which they were derived. A couple of other antibodies, a few antibodies with better HEK titer's, but there are also a couple of antibodies which have worse TMs as indicated in the TM plot.
0:30:41.7 AB: So now, when you look at this in the aggregate, and say how many antibodies have biophysical liabilities as measured, as described previously? And what we have is over 80% of tested antibodies have no biophysical liabilities, 16 out of 19. And only three antibodies have a single biophysical liability. And we have not identified an antibody with more than one biophysical liability yet. And so conclusions is that the Generation 3 Antibody Library platform performs... Generates a very high proportion of drug-like antibodies, 84% of tested antibodies have no measured by physical liabilities, the rest have only one, we have very high affinity antibodies generated routinely.
0:31:32.4 AB: And we were particularly struck by the high affinities of the antibodies we were able to pull out from the... Against, sorry, against the spike protein. We have very high antibody diversity, usually between 500 and 5000 clusters or clonotypes, depending on the target and the sort concentration. And as I mentioned, the SARS-CoV-2 spike protein selections yield antibodies that are better than most immune antibodies generated from convalescent patients and within two fold of the very best antibodies that have been described to date.
0:32:03.8 AB: We have also found that if you do next-gen sequencing and cluster, then you can increase your cluster diversity by up to 50 fold, and the amount that you increase depends on how many Sanger clones you sequence at the beginning. And so just to turn quickly to the people that did the work, and so the libraries were constructed by André Teixeira in collaboration with Frank Erasmus, who did a lot of bioinformatic work to define the CDRs and how we were building them.
0:32:39.0 AB: Sara D'Angelo built the generation two libraries and is the CTO of Specifica and keeps the whole lab running and Fortunata and Camilla are doing the selections. This work was done in collaboration with Rutgers, in particular, the lab of Abe Pinter who did some of the pseudo-virus neutralization assays and also with Scripps, Dennis Burton and Deli Huang who did the live virus neutralizations. And with that, I would like to thank you. You can visit us at specifica.bio, send me an email at firstname.lastname@example.org if you have any further questions. We supply libraries, we do selections, we are happy to help you with your antibody discovery campaigns. And so with that, I'll stop and I'll pass over to Noah.
0:33:34.8 Noah Ditto: Hi, everyone. My name is Noah, and I am gonna be talking about The Carterra LSA platform today, and how the Carterra LSA is changing the paradigm of therapeutic discovery and characterization.
0:33:54.5 ND: So SPR is a technology that's existed for quite some time in pharmaceutical drug discovery. On the LSA, what really makes it unique, is the novel microfluidic scheme that the LSA employs to introduce sample to the surface. So on the LSA, we have two fluidic modes that both address a single chip that's docked inside the instrument. One is considered multi-channel mode, where we have 96 channels all flowing simultaneously to the surface, and the other is single-channel mode, where we have a single injection being presented across an array of up to 384cc.
0:34:32.7 ND: So the interplay between these two modes allows the user to create an array of up to 384 proteins on the surface, and then come in and inject single analytes across this array and measure 384 interactions simultaneously. This really takes SPR to another level where we're measuring interactions on a scale that are 100 times faster than what was previously done on other platforms, and 10% of the time and using around 1% of the sample. So this is a significant leap in technology in the field of SPR and generally labels free binding, which again, historically has been a little bit on the lower throughput end due to maybe fluidics that didn't really fully enable the technology to screen at the levels that the LSA does.
0:35:25.8 ND: One other aspect of the LSA that's very unique is that when we are introducing an analyte onto the surface, we do it in a bi-directional flow pattern, and what I mean by that is we have a single fixed volume of sample that's introduced onto the surface, and we flow in one direction and rapidly flow back in the other direction, bi-directional is the way to describe it, and that enables, one, a very easy assay set-up because we're not trying to do complex calculations of how much contact time and volume we're gonna have in the assay. 0:36:04.2 ND: But it also means that you can really extend your contact times quite long relative to what typically has been done in the SPR field, so we're getting maybe 30-minute contact times or more. We can also do an on-ship enrichment in the assay where we're pulling out material for an extended period of time during this bi-directional flow phase, and this allows us to really get excellent assay sensitivities that weren't often achieved in SPR because we are doing this on chip enrichment via bi-directional flow.
0:36:39.5 ND: So there's 10 main workflows that users gravitate towards on the Carterra LSA, Andrew highlighted them, one being kinetic, so screening for kinetic affinities, and the other being epitope binning, where we're competing antibodies and ascertaining their epitope clusters. So this first example I'm gonna show here is a kinetic screening approach that's very common to the LSA. So in this case, we're using the example of CAR SVFCs, but in reality, any sort of therapeutic modality could be plugged in here. The assay set-up begins by arraying the 384 clones on the bio-sensing chips surface using our multi-channel device.
0:37:24.1 ND: And then, we switch over to our single-channel device and then inject a titration series tumor receptor, in this case across the SVFC array. So we see increasing binding signals. We fit those on and off rates, and from those on and off rates, derive kinetic affinity values for the panel. And what's really cool is within this assay, you can also introduce, in this case, receptors because that's the analytes we're flowing, additional receptors to check for specificity, which is very much relevant to CAR SVFCs, for example. And so if you have the ability then within one assay workflow to not only derive kinetics but also ascertain specificity for the clones in one go, and this is really what investigators are after, truly high throughput kinetics. 0:38:16.5 ND: So this is kind of a snapshot of 384 kinetic affinity measurements all done in a single experiment, using only about seven micrograms total of antigen, so extremely low antigen consumption. Again, due to that ability of injecting one sample across 384 species simultaneously. And it's an eight-point titration series, so fairly rigorous, we're getting a lot of good resolution here of the kinetics. And you can see within this, there's certain different highlighting for each kinetic tile within this.
0:38:48.7 ND: That's the software auto-flagging lower signal, there are complex hitting signals. And the software does this to sort of guide the user and highlight any pieces of data that might need more investigation or particularly pieces of data where reporting of rate constants would be appropriate. And this 384 array is only a third of the maximum capacity that can be screened in a single unattended run. The instrument does hold three 384 well plates, so there is the ability to do three times this level of screening in a single experiment.
0:39:22.9 ND: So that could be 1152 unique affinities or possibly a smaller number of that using different antigens against maybe the same array of 384. And that experiment would run overnight, so you can imagine just in a single overnight run, you would have 1152 affinities, a really good resolution.
0:39:49.6 ND: So kind of switching gears to the other main application on the LSA is antibody epitope binning. So similar to kinetics, we start by building out our array of species on the surface of 384 antibodies. In this case, it's most common to couple the antibodies directly to the surface using the multi-channel device, and then the assay commences as an exercise where we're injecting antibodies across the array and looking for the ability of antibodies to sandwich with one another in the presence of antigen. And from this collective sandwiching or lack of sandwiching signals that we see in the assay, we can cluster groups of antibodies together that share a similar epitope and visualize them as these proprietary network plots shown here. 0:40:38.0 ND: What's really powerful in the software is the ability then to bring in additional data in the form commonly of antibody attributes and layer this on top of these network plots. So in this particular case I'm showing here, there's different information on the specific background of each of these clones, but you can really imagine that any piece of data that you have describing these clones could be incorporated into the network plot shown here. For example, affinity or expression levels, developability, characteristics. There's an endless amount of information you can use to sort of color and describe these clones in the context still of their epitope, which is a fundamental characteristic of each one of them.
0:41:25.8 ND: And again, ultimately, what the investigators really want out of this is to do epitope binning faster and earlier in the development process. Epitope binning itself has been done at a much smaller scale for some time now, but again, it was often happening at a later stage of drug discovery where the candidate pool was so small that frankly, if lack of epitope diversity was identified, programs were challenged at what they could do at that point anyway. Here on the Carterra LSA, we're doing up to 384 by 384 matrices in a single experiment, so hundreds of antibodies being compared and grouped, and characterized, and this can happen very, very early in the discovery process to really drive effective decision-making and ultimately, reduce risk in the process by having a diverse pool of candidates progress forward.
0:42:16.8 ND: In terms of the outputs of binning... Or excuse me, the inputs of binning, it's only 5 micrograms of each antibody, so it's very amenable to early stage processes where small scale yields are available. And then ultimately, if you do a 384 by 384 binning exercise, you're gonna get over 147,000 interactions in that experiment, in a single experiment, so it's a huge amount of data using a very small amount of antibody.
0:42:51.7 ND: So I'm gonna switch gears a little bit and kind of talk about a real-world application, particularly in our new COVID world, about the development of a therapeutic antibody which utilized the attributes of the LSA, which I've just described to you, speed and screening for kinetics and epitope. So the example I'm gonna show here highlights Lilly's therapeutic antibody discovery effort, which was amazing because it was a 23-day effort from B cell screening to getting 187 clones screened and characterized and ultimately, down to 24 final candidates that were ready for pre-clinical assays and then ultimately, first in human studies. And of this pipeline here of effort, the LSA shined in the screening of 187 clones for epitope and kinetic affinities.
0:43:57.0 ND: So in Lilly's particular case, given the timelines they were working with, the LSA was really the only label-free option available to fully characterize epitope and affinity for these 187 clones. Not only did the team determine kinetic affinities and epitope clustering data for their candidates, they were also able to determine sub-domain binding, as well as ACE-2 blocking on the LSA. They really fully leveraged the platform to get the most data they could as possible out of their candidate pool. It really came through this pre-print publication that is currently in peer review right now. The authors really highlighted that it was the advanced discovery and characterization platforms which really gave them the leap they needed to set this record-breaking pace in identification of this therapeutic lead.
0:44:53.4 ND: And this is a snapshot of some of the data from that manuscript. So as I mentioned, affinity, so this is a nice affinity plot where we're looking at the relationship of on rate and off rate and looking at relatively these affinities shown here. Epitope clustering, so identifying how the antibodies relate to one another in the context of competitive epitope, domain binding by epitope cluster and ACE-2 blocking by epitope cluster. So in these two figures here, you see this data layering capability that we have in our LSA analysis software that allows you to bring in external sources of data, in this case, domain binding, and layer it in the context of epitope. And the same is done here for ACE-2, highlighting specific clones within these epitope clusters that show ACE-2 blocking. And interestingly, you can see this one particular cluster of antibodies having a very strong blocking profile of which, in both of these cases, included the lead candidate, CoV555. 0:46:02.6 ND: Well, I'd like to thank everybody for their time today and we can open things up for questions.
0:46:09.7 SV: And thank you, Andrew and Noah for that informative presentation. We will now start the live Q&A portion of the webinar. If you have any questions you'd like to ask, please do so now. Just click on that Ask a Question box located on the far left of your screen and we'll answer as many questions as we have time for. So let's get started. It looks like we already have some great questions coming in from our audience. Our first question is, how many antibody libraries have you made?
0:46:47.8 AB: Can you hear me? Yes. So, as I mentioned during my presentation, we've made up just under 70 different libraries. And the reason we make so many libraries is, first, because we've got... We can do it quite straightforwardly now, but also, we've found that you can't really test a new vector, for example, or a new selection strategy, unless you actually make a library and go through the whole process. We find that if you just make an antibody or two and then do the work, it gives you a false sense of security or pessimism. And so, in order to study the different elements when we make new things, we always make a new library and then test that.
0:47:33.6 SV: Thank you so much. Now, how long does it take to make an antibody library?
0:47:41.0 AB: Again, I think I mentioned that during the talk and to make a library, it usually takes about a couple of months. When I say to make a library, in that particular case, we will actually make five sub-libraries to generate the whole platform. And the library, after we've made it, then takes another couple of months to do the QC, which involves next-gen sequencing and also selection against a panel of targets that we know about, and measuring the affinities and looking at the diversity of the outputs and so on. So, we're confident with each library. We know what the parameters are. It's interesting, we've now discovered that there are next-gen sequencing or diversity measurements that if we don't hit, we know that the library is not gonna work very well.
0:48:31.6 SV: Thank you so much, and again, I wanna thank our audience for these great questions coming in. Our next question, can you customize antibody libraries, and if so, how does one do that?
0:48:45.0 AB: So, we can do that actually at a number of different levels, and we've done that at a number of different levels. So, for example, we've had some partners who would have come to us with particular scaffolds that they like. They say that they've worked very well in their hands. So, can we make a library in their scaffolds? We've certainly... We can do that. Other people say, can you eliminate additional liabilities?
0:49:08.3 AB: And we can do that, too. We've recently had a partner that came and wanted a particular liability, which in the literature is indicated as being extremely rare, but this was something that they were worried about. We can also put back some of the liabilities. So, we've been very broad in our definition of a liability. So, if you look at the literature, there's lots of liabilities, we've identified some from in-house work, and as we make new oligo sets, we eliminate more and more liabilities, but some people don't like... Are a bit scared about eliminating so many liabilities, and so would ask us to put some of them back. So, we can customize at that level as well. And then the format, single chain fab, vector, we can use our vector. We can use somebody else's vector. We make phages display libraries, so we can also make [0:50:03.7] ____. 0:50:06.1 SV: Thank you so much. Now, we have time for a couple of more questions. Are the affinities routinely as good as you show?
0:50:14.9 AB: Yeah, actually, they are, and that's a good question. We've had problems. Our Generation 1 library, if we got a single-digit nanomolar affinity, we were overjoyed. And so, when we saw these affinities, we were rather surprised. We were planning to get the Carterra. We actually got it, we tested some antibodies on it, and we were sure that we were using it incorrectly, because the affinities were so high. So, we actually sent the antibodies that we were testing to Carterra, and they tested them for us and confirmed that the affinities were what we say, so, yeah. In general, if sometimes target specific peptides, for example, tend to be much less good, but on the whole, it's 20% sub-nanomolar and 40% between one and 10 on the whole. 0:51:10.9 SV: Thank you very much. We have time for one more question. Do you have any plans to market your SARS-CoV-2 antibodies?
0:51:22.7 AB: Well, we're in a glut. There's so many other antibodies out there. We'd love to market them. If anybody listening is interested, then please get in touch. Yeah. 0:51:35.1 SV: Andrew and Noah, thank you so much for your time today. Would you like to provide any closing remarks for our audience members before we go?
0:51:43.7 AB: Sure, actually, there's a couple of more questions, I can just address those quickly. Why do you think there's a poor correlation between IC50 and affinity when comparing RBD and trimer? I just don't know. I think it's something to do with orientation, with epitopes being hidden, but we're not the only ones to find that. I was speaking to Johanna Hansen at Regeneron recently and she said exactly the same thing. It's the difference between biology and biophysics, I think.
0:52:13.0 AB: Does the starting pool of clinical antibodies change depending on the target? Basically, whatever target we do, we get lots and lots of different binders as I've talked about. And just in closing, we're very collaborative, very happy to work with anybody that's interested in accessing our platform and go to our website, you'll see all the details there. I'd be happy to talk to you further.
0:52:37.2 SV: Thank you so much, Andrew. And Noah, would you like to provide any closing remarks?
0:52:41.8 ND: Yeah definitely, I'd just like to reiterate our thanks to Andrew for providing a great talk and really getting some great insight into how Specifica does things and really the unique ways in that they do them, and to thank our audience in general. We're really excited to talk about the LSA today and how it supported Specifica in these efforts. And please, anybody has any questions, additional questions that we couldn't get to today, please just reach out and we'll follow up with those as soon as we can. Thank you.
0:53:13.1 SV: And thank you, gentlemen, both of you, Andrew and Noah for your time today and for your important research. I also wanna thank Labroots and our sponsor Carterra for underwriting today's educational webcast. Audience, before we go, I wanna thank you for your interesting questions. Questions we did not have time for today and those submitted during the on-demand period will be addressed by the speaker via the contact information you provided at the time of registration. And best of all, this webcast can be viewed on demand. Labroots will alert you via email when it's available for replay. We encourage you to share that email with your colleagues who may have missed today's live event. Until next time, stay healthy and take care, everyone. Bye-bye for now.