Enhancing antibody affinity through experimental sampling of non-deleterious CDR mutations predicted by machine learning.

The application of machine learning (ML) models to optimize antibody affinity to an antigen is gaining prominence. Unfortunately, the small and biased nature of the publicly available antibody-antigen interaction datasets makes it challenging to build an ML model that can accurately predict binding affinity changes due to mutations (ΔΔG). Recognizing these inherent limitations, we reformulated the problem to ask whether an ML model capable of classifying deleterious vs non-deleterious mutations can guide antibody affinity maturation in a practical setting. To test this hypothesis, we developed a Random Forest classifier (Antibody Random Forest Classifier or AbRFC) with expert-guided features and integrated it into a computational-experimental workflow. AbRFC effectively predicted non-deleterious mutations on an in-house validation dataset that is free of biases seen in the publicly available training datasets. Furthermore, experimental screening of a limited number of predictions from the model (<10^2 designs) identified affinity-enhancing mutations in two unrelated SARS-CoV-2 antibodies, resulting in constructs with up to 1000-fold increased binding to the SARS-COV-2 RBD. Our findings indicate that accurate prediction and screening of non-deleterious mutations using machine learning offers a powerful approach to improving antibody affinity.

Generation of bispecific antibodies by structure-guided redesign of IgG constant regions.

Bispecific antibodies (BsAbs) form an exciting class of bio-therapeutics owing to their multispecificity. Although numerous formats have been developed, generation of hetero-tetrameric IgG1-like BsAbs having acceptable safety and pharmacokinetics profiles from a single cell culture system remains challenging due to the heterogeneous pairing between the four chains. Herein, we employed a structure-guided approach to engineer mutations in the constant domain interfaces (CH1-CL and CH3-CH3) of heavy and κ light chains to prevent heavy-light mispairing in the antigen binding fragment (Fab) region and heavy-heavy homodimerization in the Fc region. Transient co-transfection of mammalian cells with heavy and light chains of pre-existing antibodies carrying the engineered constant domains generates BsAbs with percentage purity ranging from 78% to 85%. The engineered BsAbs demonstrate simultaneous binding of both antigens, while retaining the thermal stability, Fc-mediated effector properties and FcRn binding properties of the parental antibodies. Importantly, since the variable domains were not modified, the mutations may enable BsAb formation from antibodies belonging to different germline origins and isotypes. The rationally designed mutations reported in this work could serve as a starting point for generating optimized solutions required for large scale production.

Recognition of heparan sulfate by clinical strains of dengue virus serotype 1 using recombinant subviral particles.

Dengue is the most important arthropod-borne viral disease in humans, with an estimated 3.6 billion people at risk for infection and more than 200 million infections per year. Identification of the cellular receptors for dengue virus (DV), the causative agent of dengue, is important toward understanding the pathogenesis of the disease. Heparan sulfate (HS) has been characterized as a DV receptor in multiple model systems, however the physiological relevance of these findings has been questioned by observations that flaviviruses, including DV, can undergo cell culture adaptation changes resulting in increased binding to HS. It thus remains unclear whether HS is utilized by clinical, non-cell culture-adapted strains of DV. To address this question, herein we describe a set of methodologies using recombinant subviral particles (RSPs) to determine the utilization of HS by clinical strains of DV serotype 1 (DV1). RSPs of clinically isolated strains with low cell culture passage histories were used to study HS interaction. Biochemically characterized RSPs showed dose-dependent binding to immobilized heparin, which could be competed by heparin and HS but not structurally related glycosaminoglycans chondroitin sulfate A and hyaluronic acid. The relevance of heparin and HS biochemical interactions was demonstrated by competition of RSP and DV binding to cells with soluble heparin and HS. Our results demonstrate that clinical strains of DV1 can specifically interact with heparin and HS. Together, these data support the possibility that HS on cell surfaces is utilized in the DV-human infection process.