Protein language models enable prediction of polyreactivity of monospecific, bispecific, and heavy-chain-only antibodies.

Background: Early assessment of antibody off-target binding is essential for mitigating developability risks such as fast clearance, reduced efficacy, toxicity, and immunogenicity. The baculovirus particle (BVP) binding assay has been widely utilized to evaluate polyreactivity of antibodies. As a complementary approach, computational prediction of polyreactivity is desirable for counter-screening antibodies from in silico discovery campaigns. However, there is a lack of such models. Methods: Herein, we present the development of an ensemble of three deep learning models based on two pan-protein foundational protein language models (ESM2 and ProtT5) and an antibody-specific protein language model (PLM) (Antiberty). These models were trained in a transfer learning network to predict the outcomes in the BVP assay and the bovine serum albumin binding assay, which was developed as a complement to the BVP assay. The training was conducted on a large dataset of antibody sequences augmented with experimental conditions, which were collected through a highly efficient application system. Results: The resulting models demonstrated robust performance on canonical mAbs (monospecific with heavy and light chain), bispecific Abs, and single-domain Fc (VHH-Fc). PLMs outperformed a model built using molecular descriptors calculated from AlphaFold 2 predicted structures. Embeddings from the antibody-specific and foundational PLMs resulted in similar performance. Conclusion: To our knowledge, this represents the first application of PLMs to predict assay data on bispecifics and VHH-Fcs.

Off-target binding of an anti-amyloid beta monoclonal antibody to platelet factor 4 causes acute and chronic toxicity in cynomolgus monkeys.

ABT-736 is a humanized monoclonal antibody generated to target a specific conformation of the amyloid-beta (Aß) protein oligomer. Development of ABT-736 for Alzheimer's disease was discontinued due to severe adverse effects (AEs) observed in cynomolgus monkey toxicity studies. The acute nature of AEs observed only at the highest doses suggested potential binding of ABT-736 to an abundant plasma protein. Follow-up investigations indicated polyspecificity of ABT-736, including unintended high-affinity binding to monkey and human plasma protein platelet factor 4 (PF-4), known to be involved in heparin-induced thrombocytopenia (HIT) in humans. The chronic AEs observed at the lower doses after repeat administration in monkeys were consistent with HIT pathology. Screening for a backup antibody revealed that ABT-736 possessed additional unintended binding characteristics to other, unknown factors. A subsequently implemented screening funnel focused on nonspecific binding led to the identification of h4D10, a high-affinity Aß oligomer binding antibody that did not bind PF-4 or other unintended targets and had no AEs in vivo. This strengthened the hypothesis that ABT-736 toxicity was not Aß target-related, but instead was the consequence of polyspecificity including PF-4 binding, which likely mediated the acute and chronic AEs and the HIT-like pathology. In conclusion, thorough screening of antibody candidates for nonspecific interactions with unrelated molecules at early stages of discovery can eliminate candidates with polyspecificity and reduce potential for toxicity caused by off-target binding.

Targeting Anti-TGF-ß Therapy to Fibrotic Kidneys with a Dual Specificity Antibody Approach.

Targeted delivery of a therapeutic agent to a site of pathology to ameliorate disease while limiting exposure at undesired tissues is an aspirational treatment scenario. Targeting diseased kidneys for pharmacologic treatment has had limited success. We designed an approach to target an extracellular matrix protein, the fibronectin extra domain A isoform (FnEDA), which is relatively restricted in distribution to sites of tissue injury. In a mouse unilateral ureteral obstruction (UUO) model of renal fibrosis, injury induced significant upregulation of FnEDA in the obstructed kidney. Using dual variable domain Ig (DVD-Ig) technology, we constructed a molecule with a moiety to target FnEDA and a second moiety to neutralize TGF-ß After systemic injection of the bispecific TGF-ß + FnEDA DVD-Ig or an FnEDA mAb, chemiluminescent detection and imaging with whole-body single-photon emission computed tomography (SPECT) revealed significantly higher levels of each molecule in the obstructed kidney than in the nonobstructed kidney, the ipsilateral kidney of sham animals, and other tissues. In comparison, a systemically administered TGF-ß mAb accumulated at lower concentrations in the obstructed kidney and exhibited a more diffuse whole-body distribution. Systemic administration of the bispecific DVD-Ig or the TGF-ß mAb (1-10 mg/kg) but not the FnEDA mAb attenuated the injury-induced collagen deposition detected by immunohistochemistry and elevation in Col1a1, FnEDA, and TIMP1 mRNA expression in the obstructed kidney. Overall, systemic delivery of a bispecific molecule targeting an extracellular matrix protein and delivering a TGF-ß mAb resulted in a relatively focal uptake in the fibrotic kidney and reduced renal fibrosis.