Special Report on Cancer: Anticipating antibodies
Digging deeper into repertoires means tapping into Plan B (cell)
By Randall C Willis
As little as a decade ago, sitting through any session of the AACR or ASCO conferences, one could not help but be struck by the activity and potential for immunotherapies. What started as a trickle of monoclonal antibody (mAb) therapies in the late 1990s became a steady flow in the 2000s and a torrent in the 2010s.
But even as more biotherapeutics were approved and more companies explored the space, the benchmarks of success rose almost as quickly as the price tags associated with treatment. And while the development of products like rituximab and trastuzumab could never be described as easy, many consider those first few antibody targets the low-hanging fruit of immuno-oncology.
In the push to enhance the precision of treatment or to develop follow-on biotherapies when the initial molecules cease to work, researchers have had to explore increasingly intractable targets that often represent increasingly smaller patient populations. The result of that extra work and those smaller markets is an increasing price tag that, in some cases, threatens to shift the cost-benefit analysis.
Unable to change the patient population, researchers are faced with having to increase the efficiency of antibody discovery and development, working to focus as much energy as possible on only those candidates with the best potential for success.
“I’ll be very honest and say it’s hard for me really to understand how we’re going to address costs, primarily because the discovery of these drugs is not simple and it’s not getting any simpler,” admits John Proctor, senior vice president of marketing at Berkeley Lights.
“For a long time, it was all about monoclonal antibodies, and there have been some amazing mAbs discovered,” he continues, adding that although people continue to explore mAb discovery, it is often with an eye toward combining antibodies as bi- or trispecific molecules.
That, he points out, has introduced a level of complexity that is only now being realized. It is simple math.
“If you need two or potentially three antibodies to make one drug, then you need two to three times the number of campaigns than you did previously,” he says. “And, on top of that number of campaigns, you also have to be able to assess the diversity of antibodies generated against the particular target very quickly.”
For AbCellera CEO Carl Hansen, that perception of complexity has been an issue since the earliest days of the antibody therapeutics.
“If you roll back 30 years, therapeutic antibodies were not even on anyone’s radar,” he recounts. “In fact, when people first started doing them, the pharmaceutical industry dismissed it as being overly complex.
“Starting with some blockbusters in the 1990s, that field has grown into what is now probably north of [a] $120-billion market and has been consistently, for decades, the fastest-growing class of therapeutics.”
From his earliest days as a professor at the University of British Columbia (UBC), however, Hansen has monitored the development of the field, seeing opportunities in computation, genomics and microfluidics to address biomedical challenges.
“One of the dynamics we saw as the field has progressed was that pharmaceutical companies were being increasingly pushed toward targets that had proven to be difficult using conventional technology,” he says.
For Hansen, the best way to address these challenges was not to head to the bench to re-engineer what Mother Nature had perfected over 350 million years. Rather, he saw a need for methods that produced a better immune response and then to better screen that response to find the best molecules.
Another factor, he notes, was the birth of the biotech industry and the growth of specialized fields such as immuno-oncology, which has significantly layered on complexity to an already challenging task.
“These companies either have a new angle on biology or they have technologies that typically require some elements of antibody to make them work,” Hansen explains. “It could be CAR T. It could be bispecifics. It could be antibody-drug conjugates.”
It was in recognizing the changing landscape that Hansen’s academic interests took a more commercial turn.
“In 2012, we recognized that we could wrap those technologies together to make a best-in-world platform for searching deeply into natural immune systems to find antibodies that had the properties that made them suitable for development as therapeutics,” he recounts.
At the core of this advance was a focus on single-cell analysis.
“Single-cell analysis was a theme that I worked on for over a decade,” Hansen says. “And this was an opportunity to apply single-cell analysis to what is by far one of the most interesting things in biology: adaptive immunity. It is also one where the connection to a real problem in the industry—finding the next generation of drugs—allowed us to build a thriving business.”
Part of seizing on this opportunity meant seriously rethinking the technologies and methods that got the field to this point. Central to that was wondering if the limitations of hybridoma technology—a founding method of antibody discovery and development—had surpassed its benefits.
Aaron Winters and colleagues at Amgen Research and UBC offered their take on hybridomas in a recent paper.
The efficiency of immortalizing antibody-secreting cells through fusion with myeloma cells is quite low, they suggested. Even with optimized electrofusion protocols, as few as one in 5,000 input B cells manages to not only survive the fusion, but also become immortalized and secrete antibody.
“Additionally, hybridoma methods generally require extensive cell culture, which is labor intensive and dependent on mitosis, further slowing development timelines,” the authors continued.
This challenge, they argued, can lead to the identification of low-affinity antibodies, and often requires multiple rounds of resource-intensive affinity maturation to generate potent molecules.
In some ways, Hansen suggests, those limitations were acceptable in the early days because of the targets that researchers were tackling.
“Thirty years ago, that was fine,” he remarks. “There were easy targets, there wasn’t a lot of competition, and all you needed to find was any old antibody that happened to bind the target and block it.”
Hansen adds that another bonus was that it was pretty easy to mount immune responses against the chosen targets.
“Hybridoma, for that reason, has been the cornerstone from which we have had the blockbusters that we do today,” he continues.
“One of the big advantages of hybridoma is that it allows you to go after antibodies from natural immune responses, as compared to synthetic approaches—e.g., doing yeast or phage display—and they have been much more successful in getting through to the clinic,” he recalls. “I think 80 percent or more of the antibodies that have been approved have come from immunizations or from natural sources.”
However, Hansen is quick to highlight the inherent limitations of what he calls the Franken-cell approach and the loss of more than 99 percent of what was available in the animal starting material.
Implementing plan B
“If you care about diversity, which is the thing that we really emphasize in our program, it is important to have a large number of antibodies so that you’re not just picking any antibodies that work, but rather you are picking the one that is most potent and has the properties that make it most quickly and easily developable into a drug,” Hansen notes. “You want to make sure that you cast as wide a net as possible.”
“We are in a numbers game,” adds Marian Rehak, vice president of research and development at Sphere Fluidics. “If you want to screen the whole repertoire, you need the technology that allows you to do that.”
With such challenges in mind, Winters and colleagues heralded the advent of microfluidic and microencapsulation technologies that permit the direct identification and characterization of antibody-secreting B cells.
“These micro tools eliminate the need for immortalization, are species-agnostic, allow high-throughput sampling as well as multiparameter phenotyping of the input cells, have reduced reagent consumption compared to hybridoma and display technologies, and maintain the ability to retain the native VH and VL pairings of the original antibody,” they noted.
Earlier this year, Rehak and colleagues at Sphere Fluidics and UCB described the application of Cyto-Mine in both antibody discovery and cell line development. In this proof-of-concept experiment, antibody-secreting CHO cells were monitored rather than hybridomas or B cells.
As the authors described, the process is basically broken down into four steps.
Initially, using a biocompatible surfactant, both cells and assay reagents are encapsulated with culture medium into picodroplets. These picodroplets, which can carry anywhere from zero to one to a few dozen cells as required, are then incubated to permit protein production and secretion.
Using a FRET-based assay to detect IgG production, picodroplet fluorescence is monitored and positive cells are collected and stored in a chilled microchamber or dispensed into a collection device, such as single cells into microtiter wells. Negative picodroplets, meanwhile, are diverted to waste.
The assay can be customized by the selection of appropriate FRET-based detection probes specific for the protein of interest, such as IgG, or for antigen-specific IgG using labeled antigen assays, Rehak and colleagues explained.
The picodroplets can then be interrogated a second time and verified for monoclonality.
“From all the experiments shown, there was no apparent difference in cell outgrowth rate between Cyto-Mine and manual [limiting dilution cloning],” the authors noted. “These data suggest that Cyto-Mine is a gentle and cell-friendly cloning technology.”
In this experiment, the picodroplets were approximately 450 pL in volume, which the authors suggested was about five to six orders of magnitude smaller than volumes used in assays done in a 96-well plate.
“This means that in the same time, the concentration of secreted antibodies from a single cell can be 5-6 orders of magnitude higher in picodroplets than in conventional vessels,” they explained. In a typical Cyto-Mine instrument run, between 100,000 and 40 million cells can be screened.
Thus, not only does the microfluidic system offer increased throughput at up to single-cell resolution, but it also offers significantly reduced reagent costs.
Rehak also points out that Cyto-Mine can be used to monitor cell-surface protein expression and even flag issues related to antibody misfolding and aggregation.
Winters and colleagues had similar experiences working with Berkeley Lights’ Beacon platform in a process the Amgen team described as NanOBlast.
“Amgen was one of the earliest customers of Berkeley Lights, and they have been super-supportive of our work in developing applications for biopharma, both in antibody discovery and in cell line development,” says Berkeley Lights' Proctor.
Rather than sequester B cells and reagents into picodroplets, the Beacon system uses microfluidics to move cells, reagents, beads or other objects through the channels of a culturing microchip, and then optoelectronics to dispense single cells and reagents into any of hundreds or thousands of one-nanoliter reaction chambers called nanopens.
It is within these nanopens that the B cells produce antibodies, which can be assayed for IgG secretion or antigen-specific IgGs. Beyond these two basic assays, however, the system also allows researchers to perform more functional assays such as competitive binding, cell binding and ligand-receptor blocking. The key, according to Proctor, is to reveal functional characteristics as quickly as possible.
“If you’re able to ask basic questions, like 'does this antibody bind my target,' that’s informative,” he states. “But if you then have to re-express and do all of this analytical characterization downstream on, say, 1,000 antibodies because you weren’t able to assess function to find out you only have 100 on the backend, you’ve probably invested months of work and hundreds of thousands of dollars trying to answer that question.
“Because we can do these functional assays on-chip up front during the primary screen, we can find all 1,000 or however many hits would be target binders, but we could then tell you a priori that there’s only actually 100 functional ones.”
This effect was highlighted in a recent application note where 33,377 mouse plasma B cells were screened for binding to PD-L1 beads, resulting in 598 positives. These were then screened for binding to PD-L1 expressed on the surface of CHO cells, reducing the positive hits to 273. A subsequent ligand-receptor blocking assay demonstrated that 46 leads not only bound PD-L1, but also blocked the interaction between fluorescently tagged PD-1 and PD-L1.
Thus, performing these assays within the same chip reduced the deeper characterization effort from 600 potential leads to 46.
Key to the platform and to maintaining cell viability are the optofluidics.
“At a very high level, we are using broad-spectrum light to activate a series of optical switches on a siliconized chip,” explains Proctor. “When we use light on that chip, the switches are essentially able to turn on and off.”
When that switch is on, he continues, it creates a dielectric force that essentially repels an object, whether it’s a microbead or a cell.
“If you just draw a box around it, so that there’s a force on all four sides, then you can move the box and the cell or object stays inside the box, and you can direct it wherever you would like on the chip,” he adds.
This progress in high-throughput B cell analysis, however, doesn’t mean that hybridomas have been completely abandoned.
“We are seeing movement from hybridoma to B cells, but recently, we have also seen movement back to hybridomas,” says Rehak.
And many biopharma companies have long used and extensively validated hybridoma approaches.
Last year, Scott Dessain and colleagues at the Lankenau Institute for Medical Research, FDA’s Center for Biologics Evaluation and Research, and Children’s Hospital of Pennsylvania acknowledged the opportunities still afforded from the technically straightforward methods.
“They produce full-length, glycosylated mAbs that maintain their original heavy chain:light chain pairings without the need for recombinant gene expression,” the authors explained.
“However, their major shortcoming is that mAbs are secreted into the cell culture medium, so that hybridomas must be maintained in oligoclonal pools while their secreted mAbs are analyzed separately,” they acknowledged. “This impedes the discovery of rare mAbs because it imposes practical limits on the numbers of cells that can be analyzed, and is a disadvantage compared to yeast display methods, in which mAbs are expressed on the cell surface and can be screened for antigen binding in bulk culture.”
Rather than abandon hybridomas, however, Dessain and colleagues looked for ways to mimic the cell surface expression capabilities of yeast display within the hybridoma screen. The result is the platform On-Cell mAb Screening (OCMS).
“OCMS transiently captures and displays mAbs on the hybridoma surface, while preventing mAbs from binding to cells that do not secrete them,” the authors explained.
The system relies on an anchor-linker strategy, whereby an anchor protein—an anti-rabbit IgG tandem scFv—is expressed in the fusion partner cell line and is maintained in the hybridoma. This is complemented with a rabbit anti-human IgG antibody (RAH) linker.
When RAH is added to the culture, it binds to the surface of the hybridoma cells, where it captures antibodies secreted by the cell to which it is bound. Excess RAH in the culture medium acts as a competitor to prevent secreted antibodies from one cell binding to neighboring hybridoma cells.
“This provides specificity to the reaction, so that mAbs secreted by cells within a heterogeneous population can be analyzed individually in association with the cells that make them,” Dessain and colleagues suggested. “Cells expressing mAbs with desired features can be identified by fluorescence imaging techniques.”
By mixing cells with different binding properties, the researchers showed that a given mAb was only bound by the cells that secreted that mAb. Beyond fluorescence microscopy, the researchers also demonstrated the utility of their platform with flow cytometry.
The researchers also noted how the analogy with yeast display extended beyond antibody capture at the cell surface.
“High-throughput competitive binding, epitope complementation, and dissociation rate assays developed for yeast should be adaptable for screening mAbs expressed by OCMS hybridomas,” they proposed. “OCMS can also be used to assess mAb expression levels by individual cells in a heterogeneous population in real time, using either fluorescence imaging or flow cytometry.
“This feature should be useful to establish and monitor stable, high-expressing cell clones for master cell banks and bioreactor production runs.”
Earlier this year, the technology was licensed from Lankenau to be commercialized by new company OCMS Bio, for which Dessain serves as chief scientific officer.
Identifying and characterizing cells that produce antibodies against specific targets is still a long way from having something that will work as an immunotherapy, however. Given that an antibody for clinical treatment faces many different stresses and strains than one in its natural immune environment, other molecular facets must be explored to determine whether a given molecule can be developed into a therapeutic.
Developability and design
“The question of developability is an important one,” says Hansen.
Recognizing parallels with Lipinski’s Rules of Five in the small-molecule space, he suggests there are several metrics that they examine in the antibody space.
“In fact, there are many more rules that have to do with how well it expresses, solubility, predictions of PK, chemical instability and the like,” he offers.
In 2019, Charlotte Deane and colleagues at University of Oxford, MedImmune, Roche, GlaxoSmithKline (GSK) and UCB Pharma set out to computationally define developability guidelines for antibody profiling by correlating protein sequences of 242 post-Phase-1 therapeutics with their biophysical properties.
“Using the distributions of these properties, we built the Therapeutic Antibody Profiler (TAP), a computational tool that highlights antibodies with anomalous values compared with therapeutics,” the authors described. “TAP builds a downloadable structural model of an antibody variable domain sequence and tests it against guideline thresholds of five calculated measures likely to be linked to poor developability.”
Based on previous work, they focused their attention on the complementarity-determining regions (CDRs) and metrics involving the surface hydrophobicity, positive and negative charges of the CDRs, length of the CDRs, and asymmetry in the net heavy- and light-chain surface charges.
Once the researchers created a red-, amber- and green-flag labelling system, they tried to reproduce in silico the real-world developability experiences of clinical antibody candidates.
Using datasets supplied by MedImmune, the researchers examined the anti-NGF antibody MEDI-578, which showed minor aggregation issues that were significantly aggravated during affinity maturation to MED-1912.
“TAP assigns MEDI-578 an amber flag and MEDI-1912 a red flag—by a large margin—in the CDR vicinity PSH metric,” Deane and colleagues noted. “The paper describes how back-mutation of three hydrophobic residues in MEDI-1912 to those of MEDI-578 led to MEDI-1912STT, fixing the aggregation issue while maintaining potency. TAP assigns MEDI-1912STT no developability flags.”
They then examined issues with protein expression, starting with anti-IL13 candidate AB008, which offered no developability issues until it was affinity matured to AB001. Offering poor levels of expression, AB001 was further modified via sequence modification to AB001DDEN, which restored expression levels.
“TAP assigns no developability flags to AB008 but a red flag to AB001 and an amber flag to AB001DDEN for its CDR vicinity PNC metric, again red-flagging the candidate with prohibitive developability issues,” the researchers noted.
Pleased as Deane and colleagues were with their findings, they were quick to note the inherently limited scope of the TAP guidelines.
“For example, they will not detect sources of immunogenicity or more subtle mechanisms that lead to poor stability,” they suggested. “Nevertheless, we have shown that the TAP guidelines can selectively highlight antibodies with expression or aggregation issues.”
Beyond these biophysical developability issues, Hansen sees so much more arising from sequence analysis.
“The sequence base is so diverse that very often it is not simply a matter of looking for identified liabilities, but starting to be able to read the protein sequence to infer more complex properties that generally apply to antibodies,” he says. “That’s one of the things that we’re working on in our platform.”
Working with partners over the last few years, he continues, AbCellera has identified, sequenced, cloned and expressed hundreds of thousands of antibodies. That database allows the company to learn which features of antibodies are ones that make them more likely to express well or be more soluble, and to help with that question of developability.
“In the end, what it really comes down to for any target is that if you can do a much deeper search of the natural immune system and more effectively and efficiently harness the total diversity, then, as you take those hundreds or thousands of molecules through the various tests that help you predict which one will be a good drug, you can stand the attrition,” Hansen explains. “You can leave the ones that don’t have the right properties behind and still make sure that you have a robust pipeline that gets all the way to the end so that you have multiple leads that you can finally bring into clinical trials.”
The reduced candidates achieved with lower-throughput systems like hybridoma or other microfluidic systems, he contrasts, reduce your chances of finding antibodies with the right potency or developable properties.
“In the end, what that means is that you need to start circling back and doing protein engineering, and that leads to inefficiencies and delays in getting to the clinic,” he remarks.
Seeing the whole board
AbCellera recognized early on that it isn’t enough to have a screening platform, Hansen says. Rather, you have to work on all of the steps.
Thus, the company spent considerable time and effort learning how to generate antigens and get good immune responses, which form the input for the screening platform.
“If you don’t get the right input, it’s unlikely you’re going to find what you’re looking for in a drug,” he states.
“Once you’ve done that for many targets and you have a throughput like ours—we can easily screen through a million cells in an afternoon—then the challenge is no longer can I find an antibody against my target, but rather how can I get the most information content in that screen,” he suggests.
Those screens allow them to go from those thousands of hits down to a much smaller and manageable number—say, 100—that have the other properties important to turning a lead into a drug.
Hansen offers the example of a screen to determine not only if an antibody binds to the target, but also to look at cross-reactivity against eight different targets at once.
“That can help when you want to find an antibody you can test in a non-human primate or in other animals,” he explains. “It can also help when you want an antibody that hits one receptor but misses another isoform.”
“We can do experiments to select for antibodies that have higher affinity, that recognize certain epitopes, that block ligands,” he continues. “There is a lot of functional information that can be gained.”
Each piece of information advises the next step of moving through development, starting with cloning, moving through expression, and into further testing that is simply better performed in microtiter plates.
According to Hansen, “I don’t think you’re ever going to be able to do all of those at the single-cell level. I think you want to bring some number through so that you’re doing your tests in a rigorous way.”
In March, however, Berkeley Lights announced their effort to try to expand what was possible at the single-cell level, introducing Opto Cell Line Development 2.0.
In support of this effort, Jennitte Stevens and colleagues at Amgen Research recently used GFP- and RFP-expressing CHO cells to compare clonality assurance with the Beacon platform and industry-standard FACS-assisted cell deposition and limiting dilution seeding.
“When comparing between growing colonies, the Beacon cloning and confirmation process calls 94 percent of exported cultures as positive that they were clonally derived,” the authors noted. “This is compared to 45 percent for a FACS and 17 percent for a limiting dilution process.”
“Additionally, Beacon clones that have been selected for export into 96-well microtiter plates have a higher recovery rate (56 percent) (positive + negative / attempts) compared to FACS (24 percent) and limiting dilution (33 percent) in the same plate format,” they added.
Whatever the method used, the ability to screen broader repertoires of cells earlier and more thoroughly, and failing or adjusting tempting candidates without expending as many resources, is sure to change the landscape of antibody development and immunotherapy more broadly from initial exploration to, perhaps, a patient’s bedside.
Going wide to go deep
To explore the greatest possible repertoire of antibody candidates or to find leads against heretofore intractable targets, there is growing interest in looking beyond the usual sources.
“The lion’s share of discovery is still done from rodents,” says AbCellera CEO Carl Hansen. “Of course, there are wildtype rodents, but there are now methods for taking antibodies from a rodent and then humanizing them into an antibody that looks like a human antibody.”
Looking to reduce this process even further, companies like Trianni have developed transgenic mouse lines that produce fully human antibodies directly.
For some targets or functional epitopes, however, evolutionary conservation between humans and mice can make it difficult to mount an immune response, explained Torben Gjetting and colleagues at Symphogen (now part of Servier) in 2019.
“One solution to overcome this limitation is to use divergent animal species that are evolutionarily more distant to mammals,” the authors wrote, introducing the chicken as one such species.
“Chickens may not only be able to raise antibodies against very conserved targets, but also against novel human functional epitopes that are masked in mice due to sequence conservation,” they continued. “Furthermore, antibodies against human targets generated in chicken are often cross-reactive to the mouse orthologous target.”
To test their thinking, the researchers generated a large antibody repertoire against the immune checkpoint protein PD1 in chickens. They then humanized the antibodies and compared the best candidates for PD1 binding affinity and functional activity to the commercial anti-PD1 immunotherapies pembrolizumab and nivolumab.
“The epitope of Sym021 was particularly interesting since it allowed for exceptionally strong cross-reactivity to both human, cynomolgus monkey, and mouse PD1,” the authors noted. “Moreover, Sym021 shows potent PD-L1 and PD-L2 blocking activity in reporter cell assays and was at least as efficient as nivolumab and pembrolizumab reference antibodies in enhancing T cell responses and cytokine production in vitro.”
The researchers also tested Sym021 for its ability to activate T cells in vivo and its impact on tumor growth in four mouse models.
“Sym021 treatment was found to induce significant tumor growth inhibition in several syngeneic tumor models and even complete tumor eradication in the Sa1N mouse fibrosarcoma model,” they reported.
Given these results, Sym021 was further subjected to toxicology testing in cynomolgus monkeys and is currently in a Phase 1 clinical trial as a monotherapy or in combination with anti-LAG-3 or anti-TIM-3 leads vs. solid tumor malignancies or lymphomas.
To the authors' knowledge, this was the first in-human study of a chicken-derived therapeutic antibody.
Alternatively, you may switch species to explore antibodies structurally distinct from those in humans, such as camelids.
“They have antibodies that have a single heavy chain, and that allows you to have a single polypeptide that can be generated within the animal’s immune system, which specifically recognizes the target and makes it very amenable to protein engineering,” Hansen explains.
“One of the big advantages of camelid antibodies is that they are modular and can be combined with simple linkers and protein engineering methods to make a variety of different molecular constructs that would otherwise be very challenging to produce and to manufacture,” he adds.
But even without changing species, the full antibody repertoire is spread across the multiple immune compartments—e.g., spleen, lymph nodes and bone marrow—which cannot be equally accessed by all methods.
For example, B cells from bone marrow have not been historically amenable to fusion with myeloma cells to form hybridomas, explains John Proctor, senior vice president of marketing at Berkeley Lights.
“They seem to be overly sensitive to the process and don’t survive,” he continues. “So, hybridoma for a long time has been limited to just splenocytes.”
Likewise, immunization and epitope localization can differ across the immune compartments and thereby produce a different immune response.
“Even just immunizing the same target over and over, you’ll see different immune responses,” Proctor adds. “By accessing all compartments, we’re able to scan that diversity every time to do the screen.”
Given the increasing complexity and demands of newer immunotherapy approaches, starting with the broadest array of options can only help to increase the chances of success.