Antibody discovery may be driven by the interrogation of two primary sources:
(1) in vivo (animal-derived B cells) or (2) in vitro (library-derived antibody display platforms). Evotec’s strategy for the optimal path to lead candidates is to offer access to both, in vivo and in vitro sources of antibodies for discovery, coupled with the exploitation of state-of-the-
art technologies to ensure success for a broad range of targets and disease states. In addition, if needed, selected lead candidates can be further optimised using powerful computational platforms such as our proprietary AbacusTM in silico tool suite to enhance productivity, manufacturability, and formulation stability.

AUTOMATED, HIGH-THROUGHPUT HYBRIDOMA PLATFORM

For the in vivo antibody discovery, we are harnessing the power of the immune system to generate antibodies, combining the traditional hybridoma technology with automated devices for highthroughput clone selection, screening as well as recombinant expression and purification.

As source for best-in-class human antibodies, we are using the ATX-Gx™ platform, a suite of immunocompetent transgenic mice from our collaboration partner, Alloy Therapeutics. Antigens are first injected into the host to elicit the expansion of antigen-specific B cells. After the humoral response has been mounted, splenocytes are harvested and the antibody-producing B cells are fused with a highly proliferative, immortal myeloma cell line via strategies such as electroporation or polyethylene glycol (PEG) treatment. Subsequent clonal selection results in a single-cell hybrid known as hybridoma. The B cells confer antibody production capability, while the myeloma cells enable hybridomas to divide indefinitely and to grow well in cell culture. A hybridoma cell line secrets only one antibody type, effectively ensuring an infinite supply of antibodies selective for a single epitope, which are also known as monoclonals.

Hybridoma-based mAb generation and screening

Streamlined workflow from immunisation to recombinant mAb

At Evotec, we have shortened and simplified this process of hybridoma establishment by coupling the traditional technology to an automated clonal selection of hybridoma colonies from methylcellulose-based, semi-solid medium using the cell separation robot CellCelector™. Thereby generated monoclonal hybridoma colonies are screened using highthroughput screening technologies, such as the iQue® Advanced Flow Cytometry Platform that enables rapid, high-content, multiplexed analysis of cells and beads in suspension. Once promising hits are identified, a next-generation sequencing protocol is used to obtain the VH/VL DNA sequence information of these potential lead antibodies. Availability of the immunoglobulin sequence information is a pre-requisite for the recombinant production of the mAbs. Additionally, it allows for the in silico analysis (using our proprietary AbacusTM software tool) of antibody sequence information to rank candidates in terms of diversity, clades, germline background and pairing, potential post-translational modifications (PTM), missing or inserted residue errors, potential immunogenicity, isotype, subtype, and can drive the engineering modifications necessary to repair or modify antibody sequences. After sequencing, selected VH/VL genes are synthesised and subjected to a transient, high-throughput expression and purification platform that allows the small-scale production of up to several hundred mAbs in parallel. The high quality material can then be used for further downstream characterisation, such as functional activities that allow the selection of potential lead candidates.

J.HAL – A NOVEL HUMANOID ANTIBODY LIBRARY

Human-like antibodies designed and sequence-biased in silico for superior efficiency and developability features For the in vitro antibody discovery, Just – Evotec Biologics exploits artificial intelligence (AI) and machine learning (ML) to generate novel, humanoid antibody sequences that both represent natural repertoires and are biased towards desirable features.

Antibody-based biotherapeutic discovery is optimised by enabling efficacy, epitope diversity, and suitable developability during the discovery process. High costs and long development times present key challenges in the global accessibility of monoclonal antibody therapeutics. In vivo discovery methods deliver therapeutically relevant antibodies, but often with limitations in epitope coverage and with no selective pressure toward developability. To enable broad target and epitope engagement, focused efficacy, and a bias toward developability, we have developed an Antibody-GAN (Generative Adversarial Network), a new synthetic approach to designing a novel class of antibody therapeutics which we term humanoid antibodies.

The Antibody-GAN architecture utilises competing deep layer neural networks to learn and produce the features of the mature human antibody repertoire, including sequence characteristics and structure properties, allowing for the encoding of key properties of interest into diverse libraries for a feature-biased discovery platform. Our Antibody-GAN architecture (1) captures the complexity of the entire variable region of the standard human antibody sequence space, (2) provides a basis for generating novel antibodies that span a larger sequence diversity than is explored by standard in silico generative approaches, and (3) provides transfer learning (continued training of a model with a subset of data with specific desirable characteristics). This last method is of critical utility towards antibody discovery. It provides a method to bias the physical properties of the generated antibodies toward broader efficacy traits such as CDR lengths and surface properties, improved developability such as improved thermal and pH stability, and diverse chemical and biophysical properties.

A GAN operates by utilizing competing, or adversarial, neural networks, as shown in Figure 4. The discriminator network is trained using human sequences to recognise real versus fake, while the generator network generates fake sequences in an attempt to fool the discriminator and is trained by feedback from the discriminator. We filtered the training sequences to remove recent immunisation and infection bias, as well as early recombination sequences, to obtain a set of training sequences which represent broad and somatically hypermatured antibodies. Over training, the two networks get progressively better at their tasks. After full training, the Antibody-GAN generator is eventually able to produce fully human, novel antibody sequences for the germline for which the GAN was trained. We have utilized the Antibody-GAN to generate a discovery library we call J.HALSM, the Just Humanoid Antibody Library. This novel library contains an initial set of 25 sub-libraries across a variety of germline backgrounds, all containing both framework and CDR diversity representative of matured human antibodies. This initial library represents human repertoire response and is applicable to therapeutic antibody discovery. Additional library expansion is occurring through creation of sub-libraries focused on a broader range of efficacy and developability. Biasing toward longer and shorter HC-CDR3s, building out a range of surface properties to engage antigen via electrostatic or hydrophobic-driven binding, and expanding the representations of germline pairings, all while limiting the impact of potential immunogenicity provide enhanced library characteristics.

Biasing toward developability comes from GAN transfer learning using stable sequences obtained from ultra-high-throughput Fab library stressing. Improved conformational stability may impact many areas of process development such as increased titre and yield, decreased aggregation propensity, decreased particulation and sub-visible particle formation, improved yield from low pH viral inactivation, improved ability to withstand freeze-thaw cycles, improved room temperature stability, and has even been shown to improve serum stability, thereby extending half-life. The J.HALSM platform is also being extended to include single-domain VHH libraries to broaden functional epitope diversity as well as to provide a platform for multispecific modalities. 
These libraries are being physically realised as Fab-displayed phage to limit loss from conversion to full-length antibodies, to enable the measurement of biophysical properties at the library level, and to allow for library-level stressors to enable transfer learning. This Fab-based phage library is applicable to standard library panning methods for target binding antibody discovery, as shown in Figure 5.

Multiple rounds of panning with increasing stringency results in enriched binders which are then translated to a recombinant system for affinity and activity verification. Target specificity and species cross-reactivity screening is also performed as this stage of antibody panel refinement. The library sequences are also being built as a scFv-displayed yeast library to enable direct conversion to scFv modalities such as CAR-Ts and multi-specific platforms, as well as full-length yeast display to enable direct antibody discovery and small-scale production directly from the library screen output. The full-length display and secretion library lends itself to microfluidic discovery strategies, thus enabling direct interrogation of binding and activity during the discovery process, as well as direct biophysical characterization without the need for recombinant conversion. These workflow platforms are displayed in Figure 6.

To validate the utility of the initial J.HALSM suite of libraries, we chose a few targets of interest. The first target to be used in our antibody discovery platform was the SARS-CoV-2 wildtype variant Receptor Binding Domain (RBD) of the spike protein. Three rounds of phage panning resulted in the enrichment of 22 unique binders as determined by phage ELISA and DNA sequencing. Of these, eight were also found to bind to the B.1.1.7 variant spike protein. All enriched Fabs were transiently expressed in HEK293 cells as full IgG antibodies. Nine of these antibodies exhibited blocking activity of the spike protein to the ACE2 receptor in an in vitro functional assay. Another panel of enriched binders isolated from a second phage panning enrichment and NGS selection process are being screened in a similar fashion and have approximately doubled the number of active leads. Most of these antibodies will be tested in pseudovirus assays to assess their potential as therapeutics. The J.HALSM discovery platform is functionally diverse, developable, fully human, and under continual expansion.

FROM CONCEPT TO IND

Less expensive, faster, and more flexible ways of discovering, developing and manufacturing biotherapeutics

A perfect therapeutic antibody is, most of all, a new molecule leading to a significant improvement for patients. This can result from either its ability to reach a novel target, the fact that its mode-of-action differs from already available drugs, or simply because it provides significant advantages to an existing antibody, such as an improvement of its efficiency, a reduction of its costs, or a lowering of its side effects.

In the previous sections of this Drug Discovery Update issue, we had laid out the antibody discovery platforms we offer at Evotec to support the selection of such novel biotherapeutics. These capabilities can be accessed as stand-alone services on demand. However, in addition, through our extensive drug discovery know-how and experience, we can also offer seamlessly integrated antibody drug discovery capabilities. No matter where the project lies on the idea to-IND continuum and beyond, Evotec’s experience supports all activities from target identification through to IND submission and beyond.

Our clients and collaboration partners can make use of our industry-leading, long standing experience in a broad range of target classes, the breadth and depth of the extensive disease biology, the availability of biologyrelevant and mechanism-driven assays and models applied in a rational and efficient way. Evotec has in-depth disease expertise in anti-infectives, respiratory diseases, immunology, oncology, metabolic diseases and neurology. Depending on the therapeutic area and the target, Evotec employs relevant translational in vitro or ex vivo assays that assess desired functional effects of the candidate molecules. We develop suitable pharmacodynamic read-outs that either provide a target-proximal read-out, target engagement, or that quantify desired downstream effects. If available and desired, candidate molecules can also be tested in relevant disease models to assess efficacy in animal models. Evotec’s pre-clinical department offers the full range of in vitro and in vivo GLP and non-GLP pre-clinical evaluation studies to thoroughly assess the safety profile of the drug candidate. Our clients and collaboration partners can benefit from our extensive understanding of PK/PD relationship for human dose prediction. Finally, Evotec can perform Biomarker discovery and Biomarker studies to enable diagnosis, assessment of target engagement and pharmacodynamics effects, patient stratification or prediction of therapeutic success. In this context, Evotec offers a comprehensive proteomics and ligand binding assay platform to support Biomarker sciences from discovery through to the clinic.

All of this is paralleled by data-driven solutions to biologics design and manufacturing to ensure smooth transition into development and reduced risk of downstream attrition and delay. With the acquisition of Just Biotherapeutics in 2019, Evotec now provides all capabilities for the discovery and development of optimal biotherapeutics. With decades of industry experience, the goal is to utilise our in-house, integrated technology platform to design and manufacture biologics (J.DESIGN). Utilising J.DESIGN, we can accelerate development and provide superior manufacturing process control for higher quality molecules.

The discovery, development, and manufacturing of biologics is complex and expensive. As the number of biologics modalities increase to target specific biology directly, such as multi-specifics, antibody fusions, and multi-domain antibody-based constructs, the complexity and costs increase. To drive the technologies required to increase speed and capacity, decrease costs, while maintaining high quality, an integrated approach to discovery and development is crucial. This breaks down the silos in technology development, which allows for technology innovation across functions and is the basis for Just – Evotec Biologics’ technology platform, J.DESIGN.

At the core of J.DESIGN is a common data management system that centralises and integrates the highly complex data sets generated from the distinct activities involved with the discovery, development and manufacture of biologics – at Just – Evotec Biologics we refer to these as Discovery, Molecular Design, Process and Product Design, and Manufacturing Design – into a singular biologics design space. J.DISCOVERY™ contains the large, diverse, manufacturable, and developable AI-designed discovery libraries, including J.HALSM. Molecular design, J.MD™, provides analysis, humanisation, and optimisation of parental antibody sequences to enhance manufacturability and stability, saving time and costs during process development. J.MD™ uses an in-house suite of computational tools called Abacus™ in conjunction with structural tools that can assist in designing the best molecules and predict the best conditions for development. JP3® involves the design of the process and product, spanning cell line development, upstream bioreactor, downstream purification development, analytical method development, drug product and formulation development, as well as processing formats, such as intensified fed-batch and continuous processing. Our JP3® scientists utilise high-throughput robotic solutions for process and product development and leverage data for learning and prediction. Process and long-term storage, which are traditionally determined through formulation activities, are defined in J.DESIGN during molecule optimisation activities, pushing the evaluation and lead selection of biotherapeutics earlier in the development cycle. Finally, the biotherapeutic is manufactured using the defined process in a J.POD®, a small footprint facility using disposable technologies and intensified processes resulting in flexible, deployable, and relatively low cost biomanufacturing. Our first J.POD® facility for biologics development and commercial manufacturing will be fully operational in Redmond, Washington by end of 2021.