Structure guided homology model based design and engineering of mouse antibodies for humanization

Accurate prediction of CDR-H3 loop structures of antibodies with deep learning

Accurate prediction of the structurally diverse complementarity determining region heavy chain 3 (CDR-H3) loop structure remains a primary and long-standing challenge for antibody modeling. Here, we present the H3-OPT toolkit for predicting the 3D structures of monoclonal antibodies and nanobodies. H3-OPT combines the strengths of AlphaFold2 with a pre-trained protein language model and provides a 2.24 Å average RMSDCα between predicted and experimentally determined CDR-H3 loops, thus outperforming other current computational methods in our non-redundant high-quality dataset. The model was validated by experimentally solving three structures of anti-VEGF nanobodies predicted by H3-OPT. We examined the potential applications of H3-OPT through analyzing antibody surface properties and antibody-antigen interactions. This structural prediction tool can be used to optimize antibody-antigen binding and engineer therapeutic antibodies with biophysical properties for specialized drug administration route.

GNUV201, a novel human/mouse cross-reactive and low pH-selective anti-PD-1 monoclonal antibody for cancer immunotherapy

BACKGROUND

Several PD-1 antibodies approved as anti-cancer therapies work by blocking the interaction of PD-1 with its ligand PD-L1, thus restoring anti-cancer T cell activities. These PD-1 antibodies lack inter-species cross-reactivity, necessitating surrogate antibodies for preclinical studies, which may limit the predictability and translatability of the studies.

RESULTS

To overcome this limitation, we have developed an inter-species cross-reactive PD-1 antibody, GNUV201, by utilizing an enhanced diversity mouse platform (SHINE MOUSE™). GNUV201 equally binds to human PD-1 and mouse PD-1, equally inhibits the binding of human PD-1/PD-L1 and mouse PD-1/PD-L1, and effectively suppresses tumor growth in syngeneic mouse models. The epitope of GNUV201 mapped to the "FG loop" of hPD-1, distinct from those of Keytruda® ("C'D loop") and Opdivo® (N-term). Notably, the structural feature where the protruding epitope loop fits into GNUV201's binding pocket supports the enhanced binding affinity due to slower dissociation (8.7 times slower than Keytruda®). Furthermore, GNUV201 shows a stronger binding affinity at pH 6.0 (5.6 times strong than at pH 7.4), which mimics the hypoxic and acidic tumor microenvironment (TME). This phenomenon is not observed with marketed antibodies (Keytruda®, Opdivo®), implying that GNUV201 achieves more selective binding to and better occupancy on PD-1 in the TME.

CONCLUSIONS

In summary, GNUV201 exhibited enhanced affinity for PD-1 with slow dissociation and preferential binding in TME-mimicking low pH. Human/monkey/mouse inter-species cross-reactivity of GNUV201 could enable more predictable and translatable efficacy and toxicity preclinical studies. These results suggest that GNUV201 could be an ideal antibody candidate for anti-cancer drug development.

The Potential of Gut Microbiota Metabolic Capability to Detect Drug Response in Rheumatoid Arthritis Patients

Gut microbiota plays an essential role in the development of rheumatoid arthritis (RA) and affects drug responses. However, the underlying mechanism remains elusive and urgent to elucidate to explore the pathology and clinical treatment of RA. Therefore, we selected methotrexate (MTX) as an example of RA drugs to explore the interactions between the gut microbiota and drug responses and obtain an in-depth understanding of their correlation from the perspective of the metabolic capability of gut microbiota on drug metabolism. We identified 2,654 proteins and the corresponding genes involved in MTX metabolism and then profiled their abundances in the gut microbiome datasets of four cohorts. We found that the gut microbiota harbored various genes involved in MTX metabolism in healthy individuals and RA patients. Interestingly, the number of genes involved in MTX metabolism was not significantly different between response (R) and non-response (NR) groups to MTX, but the gene composition in the microbial communities significantly differed between these two groups. Particularly, several models were built based on clinical information, as well as data on the gene, taxonomical, and functional biomarkers by using the random forest algorithm and then validated. Our findings provide bases for clinical management not only of RA but also other gut microbiome–related diseases. First, it suggests that the potential metabolic capability of gut microbiota on drug metabolism is important because they affect drug efficiency; as such, clinical treatment strategies should incorporate the gene compositions of gut microbial communities, in particular genes involved in drug metabolism. Second, a suitable model can be developed to determine hosts’ responses to drugs before clinical treatment.

A universal in silico V(D)J recombination strategy for developing humanized monoclonal antibodies

BACKGROUND

Humanization of mouse monoclonal antibodies (mAbs) is crucial for reducing their immunogenicity in humans. However, humanized mAbs often lose their binding affinities. Therefore, an in silico humanization method that can prevent the loss of the binding affinity of mAbs is needed.

METHODS

We developed an in silico V(D)J recombination platform in which we used V(D)J human germline gene sequences to design five humanized candidates of anti-tumor necrosis factor (TNF)-α mAbs (C1-C5) by using different human germline templates. The candidates were subjected to molecular dynamics simulation. In addition, the structural similarities of their complementarity-determining regions (CDRs) to those of original mouse mAbs were estimated to derive the weighted interatomic root mean squared deviation (wRMSDi) value. Subsequently, the correlation of the derived wRMSDi value with the half maximal effective concentration (EC50) and the binding affinity (KD) of the humanized anti-TNF-α candidates was examined. To confirm whether our in silico estimation method can be used for other humanized mAbs, we tested our method using the anti-epidermal growth factor receptor (EGFR) a4.6.1, anti-glypican-3 (GPC3) YP9.1 and anti-α4β1 integrin HP1/2L mAbs.

RESULTS

The R2 value for the correlation between the wRMSDi and log(EC50) of the recombinant Remicade and those of the humanized anti-TNF-α candidates was 0.901, and the R2 value for the correlation between wRMSDi and log(KD) was 0.9921. The results indicated that our in silico V(D)J recombination platform could predict the binding affinity of humanized candidates and successfully identify the high-affinity humanized anti-TNF-α antibody (Ab) C1 with a binding affinity similar to that of the parental chimeric mAb (5.13 × 10-10). For the anti-EGFR a4.6.1, anti-GPC3 YP9.1, and anti-α4β1 integrin HP1/2L mAbs, the wRMSDi and log(EC50) exhibited strong correlations (R2 = 0.9908, 0.9999, and 0.8907, respectively).

CONCLUSIONS

Our in silico V(D)J recombination platform can facilitate the development of humanized mAbs with low immunogenicity and high binding affinities. This platform can directly transform numerous mAbs with therapeutic potential to humanized or even human therapeutic Abs for clinical use.

Antibody Design and Humanization via In Silico Modeling

Antibody humanization process converts any nonhuman antibody sequence into humanized antibodies. This can be achieved using different methods of antibody design and engineering. This chapter will primarily focus on antibody design using a homology model followed by framework shuffling of murine to human germline sequence for humanization. Historically, mouse antibodies have been humanized using sequence-based approaches, in which all the murine frameworks are replaced with most homologous human germline sequence or related scaffold. Most often this humanized antibody design, when tested, has a significantly reduced binding or no binding to the cognate antigen. This is due to noncompatibility of mouse CDRs being supported by non-native human framework scaffold. This mismatch between mouse, human structural fold, and elimination of key conformational residues often leads to antibody humanization failures. Recently, there has been advent of homology modelor structure-guided antibody humanization. Instead of humanization based on linear sequence, this approach takes into account the tertiary structure and fold of the mouse antibody. A mouse homology model of the fragment variable is created, and based on sequence alignment with human germline, residues that are different in mouse are replaced with humanized sequence in the model. Energy minimization is applied to this humanized model that also delineates residues which might have steric clashes due to change in the overall tertiary conformation of the humanized antibody. Therefore, a homology model-guided with rational mutations, and reintroduction of key conformational residues from mouse antibody not only eliminates steric clashes but might also restore function in relation to binding affinity to its antigen.

VH-VL orientation prediction for antibody humanization candidate selection: A case study

Antibody humanization describes the procedure of grafting a non-human antibody's complementarity-determining regions, i.e., the variable loop regions that mediate specific interactions with the antigen, onto a β-sheet framework that is representative of the human variable region germline repertoire, thus reducing the number of potentially antigenic epitopes that might trigger an anti-antibody response. The selection criterion for the so-called acceptor frameworks (one for the heavy and one for the light chain variable region) is traditionally based on sequence similarity. Here, we propose a novel approach that selects acceptor frameworks such that the relative orientation of the 2 variable domains in 3D space, and thereby the geometry of the antigen-binding site, is conserved throughout the process of humanization. The methodology relies on a machine learning-based predictor of antibody variable domain orientation that has recently been shown to improve the quality of antibody homology models. Using data from 3 humanization campaigns, we demonstrate that preselecting humanization variants based on the predicted difference in variable domain orientation with regard to the original antibody leads to subsets of variants with a significant improvement in binding affinity.

Global analysis of VHHs framework regions with a structural alphabet

The VHHs are antigen-binding region/domain of camelid heavy chain antibodies (HCAb). They have many interesting biotechnological and biomedical properties due to their small size, high solubility and stability, and high affinity and specificity for their antigens. HCAb and classical IgGs are evolutionary related and share a common fold. VHHs are composed of regions considered as constant, called the frameworks (FRs) connected by Complementarity Determining Regions (CDRs), a highly variable region that provide interaction with the epitope. Actually, no systematic structural analyses had been performed on VHH structures despite a significant number of structures. This work is the first study to analyse the structural diversity of FRs of VHHs. Using a structural alphabet that allows approximating the local conformation, we show that each of the four FRs do not have a unique structure but exhibit many structural variant patterns. Moreover, no direct simple link between the local conformational change and amino acid composition can be detected. These results indicate that long-range interactions affect the local conformation of FRs and impact the building of structural models.