Approaching bispecific antibody engineering

Dr Michael Fiebig, Chief Scientific Officer, Absolute Antibody offers insight for those starting with a new bispecific antibody project who may feel a little daunted by the wealth of formats out there. Over the past 10-plus years, there has been a dramatic rise in the exploitation of bispecific antibodies with over 100 currently in clinical development and three that have obtained regulatory approval1. Bispecific antibodies are unique because they offer simultaneous engagement of two distinct targets within the same antibody. Knob-into-hole (KIH) bispecific antibodies play an important role in drug development, as this pioneering format enables heavy-chain heterodimerisation and the creation of an IgG-like antibody with different targets on each arm. We are seeing bispecific antibodies being used in a variety of applications, from cancer targeting, to immune modulation and infectious disease. Particularly, in the oncology realm, bispecific antibodies have been developed to act as trans co-engagers, meaning they bind to different cells, or cis co-engagers, which bind to targets on the same cell. Bispecifics are one of the three main types of cancer immunetherapeutics approved for clinical use, the other two being monoclonal immune-checkpoint blocker antibodies and genetically engineered T cells expressing chimeric antigen receptors. Following the approval of blinatumomab, a T cell redirecting antibody commercially available as Blincyto, a plethora of approaches have been developed for immunomodulation. An excellent review by Blanco et al. covers many of these methods2. There are countless options when it comes to selecting a bispecific antibody, so it can be confusing to know where to start. In the following, I am going to discuss some background on bispecific antibody design, as well as some go-to designs for getting started with bispecifics. I will also discuss some additional topics, such as binder choice and bispecific murine surrogate molecules which can form part of the development process of a clinical bispecific antibody. Creating a bispecific antibody In theory, the concept of creating a bispecific is simple. You combine the antigen targeting two monoclonal antibodies into one single molecule to develop an extremely potent drug. However, in practice, there are many different ways to combine two binders into one, leaving drug developers with another level of discovery to contend with. When done successfully, bispecific antibodies offer drug developers additional layers of optimisation and fine-tuning of lead compounds, making bispecific antibodies an increasingly important class of therapeutics. However, it is often more challenging than expected to create your desired bispecific antibody. When you mix two heavy chains and two light chains together, it results in random pairing and a highly heterogenous mix of species, most of which are not the correctly formed bispecific. Additionally, mispairing is possible for the heavy chains, as well as for the light chains from each antigen-binding fragment (Fab). Even if ‘correctly folded’ material is obtained, it might not have the right properties for the biological question at hand. For example, the valency of binding, or how many antigen binding sites exist for each target, might not be achieved. This challenge has resulted in a vast range of ingenious solutions to create each desired bispecific format. Our own Periodic Table of Antibodies contains at least 40 different designs for bispecific antibodies, and even this greatly underestimates the number of designs that have been reported in the literature, which is now likely nearing 1003. Which bispecific design is the best? One of the questions asked most frequently is ’which bispecific design is the best?’. Unfortunately, there is no straightforward answer. The optimal design will depend on a range of things, including target antigens, their location (are they soluble or on the cell surface), concentration in the solution or density on the cell surface, specific epitopes the antibodies bind to on the respective antigens, desired half-life, mechanism of action (blocking, neutralising, cell killing by antibody-dependent cell-mediated cytotoxicity (ADCC)/complement-dependent cytotoxicity (CDC), T cell engagement by CD3 or co-stimulatory receptors, natural killer (NK) cell engagement, etc), and more. Due to the complexity of each research project and the results you are trying to achieve, there is no one size fits all bispecific design. When starting a bispecific project, it is important to understand several criteria and as much as possible tailor your bispecific to your particular application. Although many of the parameters may be unknown, the design phase is best addressed by contemplating a few simple questions: What is your desired half-life? Short, medium or long? Do you want Fc effector function? Such as ADCC or CDC? Have you considered the impact of avidity? The first two questions determine the presence or absence of the Fc domain and any mutations that may need to be incorporated. Although a long half-life is often desirable, this is not always the case. Assuming a long half-life is required, then utilising the naturally long serum persistence of the Fc domain makes sense over other options such as albumin fusion, albumin binding domains or PEGylation. The presence of the Fc domain will typically lead to engagement with Fc receptors and complement protein (C1q), but this can be modified through well-known mutations if ADCC/CDC are not required. Sunders gives an excellent overview of commonly used mutations and their function4. Using an Fc domain opens additional opportunities for Fc engineering. As already mentioned, there are a wealth of well-known mutations to reduce or abolish ADCC/CDC and using such mutations have been the dominant approach in bispecific antibody generation. However, in recent years, it is also possible to enhance ADCC/CDC with specific mutations in the Fc to harness its ability to engage with effector cells and complement, though this could be argued to go beyond a purely bispecific mode of action. Perhaps more common in bispecific antibody constructs currently under evaluation clinically are mutations to increase half-life by increasing affinity to the neonatal Fc receptor (FcRn), which mediates antibody recycling. Using an Fc in a bispecific antibody construct can serve as a canvas on which a number of fine-tuned characteristics can be drawn. However, care should be taken as not all mutations readily stack on top of each other without affecting their efficacy. The third question can be more challenging to consider. It is aimed at understanding how many ‘arms’ of the antibody you want binding to each antigen. Although people often think of a bispecific as the classic Y-shaped IgG, one Fab arm binding antigen A and the other binding antigen B, this does not have to be the case. This kind of design is known as a 1:1 binder, but you can also generate 2:1 and 2:2 binders (Figure 1). For some targets, more binding arms may be better to increase avidity. However, for other targets, this can be detrimental. This is particularly true for CD3e, where it is now widely accepted that over-engagement5 or very strong binding6 of CD3e leads to increased systemic toxicity. Moderate binding can be achieved by just having one binding arm or by using a CD3e binding antibody with only a modest affinity7, 8, 9. Therefore, in the literature you will often see researchers working with 1:1 or 2:1 bispecifics for T cell recruitment. Especially in the case of tumour targeting bispecifics, increasing the valency of the anti-tumour associated antigen (TAA) binder can be used to improve avidity. For formats with multiple specificities, this is a combination of all the antigen-binding portions of the molecule. For clarity with multispecific formats, an additional value is given in parenthesis representing the breakdown of valencies for the different specificities. For instance, 4 (2:1:1) would represent a trispecific format with two binding sites to antigen A, one binding site to antigen B, and one binding site to antigen C. Indeed, increasing the valency, whilst opting for a lower affinity of the anti-TAA binder is an elegant way of improving specificity for high-TAA expressing cells10. Answering these questions can help guide you to toward a narrower set of designs options, but the landscape is still complex. To truly find the most optimal design, one would test all the options, however, in reality this can often take more time, effort and money than most budgets allow. Go-to designs to start a bispecific project For researchers taking their first steps into bispecifics, it can be helpful to simplify the complexity down to a narrow range of options using a small panel of recommended designs. When we are working with clients, we tend to use the four designs shown in Figure 1 to represent our “go to” formats for bispecifics. All of these designs are IP-free, enabling complete freedom to operate (though IP may exist on particular target combinations). They each utilise a single chain Fv for one of the specificities to avoid issues associated with light chain shuffling, and they use a knob-into-hole platform to promote Fc heterodimerisation where this is required. Knob-into-hole (KIH) approaches are successful in promoting heterodimer formation at yields generally over 85%. This approach uses Fc engineering to introduce complementarity by inserting a ‘knob’, which increases side-chain volume, on the first Fc chain while engineering a matching “hole”, which minimises side-chain volume, on the second Fc chain. As a result, high titres of the desired bispecific can be obtained. The knob-into-hole approach has been cited for years and is noted in many publications, the first one being Ridgway et al., 199611. All four recommended formats contain an Fc domain since the majority of drug developers desire a long half-life for their drug and tend to incorporate some form of Fc silencing to reduce Fc gamma receptor binding and avoid off-target trispecific binding. The two 2:1 formats offer additional options in terms of simultaneous accessibility of both antigens and distance of engagement of both antigens. This affects the proximity with which, for example, two cells can be brought together. In particular, when comparing activities of the 1:1 or 2:2 architectures, being able to select similar arrangements of binding domains, but with ’one less’ or ‘one more’, is a useful process in optimising the final drug candidate. Expand your toolbox with trispecific antibodies and non-antibody domains If additional targeting is desired beyond two antigens, then three becomes a logical next step. Building on the architectures shown in Figure 1, it is relatively simple to create even more complex biologics. You can make use of Fc heterodimerisation domains, like the KIH mutations, to add N- or C-terminal fusions. Or you can append Fab light chains with additional binding domains to open a wealth of opportunities for researchers wanting to explore more complex biology. It is prudent to build up to these more complex constructs bit-by-bit to identify components that lead to expression or stability liabilities early on in the development process. Starting off with a solid bispecific core antibody can not only avoid undesirable physical characteristics but can also facilitate screening of the ideal method for incorporation of the third specificity. Antibody and Fc fusion protein combinations may be appealing to you if are looking for an antibody which encompasses molecules containing antibody and non-antibody affinity reagents. The suggested go-to format architectures lend themselves to a variety of antibody and Fc fusion combinations after initial development and can of course be developed even further. In particular, engineering opportunities arise with regards to the heterodimeric non-antibody fusion proteins used. Single domain binders The light chain shuffling problem requires the generation of at least one single domain binder on any construct. There are light chain heterodimerisation options, such as approaches using common light chains or CrossMab, where one Fab region’s CH1 and CL1 are swapped around leading to selective pairing of that Fab’s heavy and light chain. However, these options are either still under patent protection or require specialised binder-selection platforms. Single-chain Fv fragments are a convenient option for turning binders based on heavy and light chains into a single-domain binder. However, while it might be relatively easy to obtain scFvs from human antibodies or display libraries, it is often conveniently forgotten that this becomes slightly more complicated when using sources such as mice. There are antibody engineering approaches available to overcome this issue, but it certainly is not as straightforward as the process is often perceived. Identified in 1993, camelid single-domain VHH (also known as nanobodies), derived from heavy-chain only antibodies of camelids, are a great source of naturally occurring single-domain binders12. They are easy to engineer into constructs and can even be daisy-chained into multimeric binders. They have excellent production characteristics and generally show high thermal stability. Other sources of single-domain binders can be found in shark IgNAR antibodies13, which also show promise in diagnostic and therapeutic applications. Additionally, there are various entirely synthetic nanobody libraries available and in development. With the wealth of options and the excitement that drives the start of most antibody engineering campaigns, it is easy to forget that some engineering approaches are still under patent, such as methods to solve the light chain problem (e.g. CrossMab from Roche, CR3 from Golay et al. 201614), methods for improved Fc silencing (e.g. LALAPG from Roche, mAbsolve’s platform) or half-life extension (e.g. YTE from Medimmune, LS from Xencor). One may find that the expiration dates of particular patents are within a short enough timeframe so that any development work planned is going to take longer anyway. In other cases, there are benefits to working with a particular technology warrant attempting to obtain a license directly from the inventors. This can be costly but is often worth the investment. Mouse surrogates for antibody evaluation Model organisms such as mice offer a great opportunity to test out target combinations and bispecific architectures in various well-characterised cancer models. However, while antibodies targeting many mouse homologues of clinically interesting proteins already exist, simply using human Fc regions for engineering surrogate molecules may lead to the drawback of creating molecules that are immunogenic in mice and will lead to adverse reactions upon repeated administration. Translating findings from mice to predictions about human biology already must be viewed with care, as there are many differences in both target expression and biology. However, anything that can be done to reduce further variables will improve the value of these findings and help make better predictions where this is appropriate. While increasing numbers of humanised mouse models are available for some cancers that help to minimise challenges in the transition from mouse to human biology15, it has to be noted that the immune responses in humanised mice can suffer from both over- and under-activity compared to natural scenarios depending on the experiments performed16. The use of surrogate bispecific antibodies in mice has traditionally been hampered by the lack of KIH-style heterodimerisation platforms, as the KIH mutations used in humans do not readily transfer to mouse antibodies. Using immunodeficient or humanised mice to avoid the immunogenicity opens up even more questions about how immunomodulatory activities of bispecifics should be interpreted. More recently, mouse heterodimerisation platforms have been described and made commercially available to researchers looking for off-the shelf ‘classics’ or to fast-track their proof-of-concept campaigns by combining their binders into bispecific antibody cassettes. The availability of such tools not only helps to identify new target combinations, but also helps to investigate the immunological consequences of targeting specific immune axes with bispecific antibodies. The more we understand about these pathways, the wider the pool of mechanisms is from which we can draw to design the next generation of bispecifc or trispecific antibodies. Embarking on your bispecific antibody project If you are starting with a new bispecific antibody project and feel a little daunted by the wealth of formats out there, I hope that this gives you a starting point to begin thinking about different approaches. Bispecific antibodies are an exciting class of therapeutics and specifically KIH bispecific antibodies are playing an important role in drug development. I am looking forward to seeing the new and exciting formats still to be designed. Figure 1. Illustrations of our four recommended bispecific formats. Each of these designs is IP-free, utilises a single chain Fv and the knob-into-hole platform. About the author Dr Michael Fiebig is the Chief Scientific Officer at Absolute Antibody. Fiebig studied Biochemistry before obtaining his doctoral degree from University of Oxford. He joined Absolute Antibody in 2014 to develop the recombinant antibody catalog and many underlying technologies, before transitioning into the role of Chief Scientific Officer in 2021. References Suurs, Frans V et al. “A review of bispecific antibodies and antibody constructs in oncology and clinical challenges.” Pharmacology & therapeutics 201 (2019): 103-119. doi:10.1016/j.pharmthera.2019.04.006 Blanco B, Dommínguez-Alonso C, Alvarez-Vallina L. “Bispecific immunomodulatory antibodies for cancer immunotherapy.” Clinical Cancer Research (2021). Brinkmann, Ulrich, and Roland E Kontermann. “The making of bispecific antibodies.” mAbs 9,2 (2017): 182-212. doi:10.1080/19420862.2016.1268307 Saunders, Kevin O. “Conceptual Approaches to Modulating Antibody Effector Functions and Circulation Half-Life.” Frontiers in immunology 10 1296. 7 Jun. 2019, doi:10.3389/fimmu.2019.01296 Perruche, Sylvain et al. “Lethal effect of CD3-specific antibody in mice deficient in TGF-beta1 by uncontrolled flu-like syndrome.” Journal of immunology (Baltimore, Md. : 1950) 183,2 (2009): 953-61. doi:10.4049/jimmunol.0804076 Mandikian, Danielle et al. “Relative Target Affinities of T-Cell-Dependent Bispecific Antibodies Determine Biodistribution in a Solid Tumor Mouse Model.” Molecular cancer therapeutics 17,4 (2018): 776-785. doi:10.1158/1535-7163.MCT-17-0657 Chatenoud, Lucienne, and Jeffrey A Bluestone. “CD3-specific antibodies: a portal to the treatment of autoimmunity.” Nature reviews. Immunology 7,8 (2007): 622-32. doi:10.1038/nri2134 Roosnek, E, and A Lanzavecchia. “Triggering T cells by otherwise inert hybrid anti-CD3/antitumor antibodies requires encounter with the specific target cell.” The Journal of experimental medicine 170,1 (1989): 297-302. doi:10.1084/jem.170.1.297 Van Wauwe, J P et al. “OKT3: a monoclonal anti-human T lymphocyte antibody with potent mitogenic properties.” Journal of immunology (Baltimore, Md. : 1950) 124,6 (1980): 2708-13. Moore, Gregory L et al. “A robust heterodimeric Fc platform engineered for efficient development of bispecific antibodies of multiple formats.” Methods (San Diego, Calif.) 154 (2019): 38-50. doi:10.1016/j.ymeth.2018.10.006 Ridgway, J B et al. “'Knobs-into-holes' engineering of antibody CH3 domains for heavy chain heterodimerization.” Protein engineering 9,7 (1996): 617-21. doi:10.1093/protein/9.7.617 Hamers-Casterman, C et al. “Naturally occurring antibodies devoid of light chains.” Nature 363,6428 (1993): 446-8. doi:10.1038/363446a0 Matz, Hanover, and Helen Dooley. “Shark IgNAR-derived binding domains as potential diagnostic and therapeutic agents.” Developmental and comparative immunology 90 (2019): 100-107. doi:10.1016/j.dci.2018.09.007 Golay, Josée et al. “Design and Validation of a Novel Generic Platform for the Production of Tetravalent IgG1-like Bispecific Antibodies.” Journal of immunology (Baltimore, Md. : 1950) 196,7 (2016): 3199-211. doi:10.4049/jimmunol.1501592 Chen, Qingfeng et al. “Cancer Immunotherapies and Humanized Mouse Drug Testing Platforms.” Translational oncology 12,7 (2019): 987-995. doi:10.1016/j.tranon.2019.04.020 Shultz, Leonard D et al. “Humanized mice for immune system investigation: progress, promise and challenges.” Nature reviews. Immunology 12,11 (2012): 786-98. doi:10.1038/nri3311