AI-based antibody discovery platform identifies novel, diverse, and pharmacologically active therapeutic antibodies against multiple SARS-CoV-2 strains.

Background: We are entering a new era of antibody discovery and optimization where machine learning (ML) processes will become indispensable for the design and development of therapeutics. Methods: We have constructed a Humanoid Antibody Library for the discovery of therapeutics that is an initial step towards leveraging the utility of artificial intelligence and ML. We describe how we began our validation of the library for antibody discovery by isolating antibodies against a target of pandemic concern, SARS-CoV-2. The two main antibody quality aspects that we focused on were functional and biophysical characterization. Results: The applicability of our platform for effective therapeutic antibody discovery is demonstrated here with the identification of a panel of human monoclonal antibodies that are novel, diverse, and pharmacologically active. Conclusions: These first-generation antibodies, without the need for affinity maturation, exhibited neutralization of SARS-CoV-2 viral infectivity across multiple strains and indicated high developability potential.

A comprehensive engineering strategy improves potency and manufacturability of a near pan-neutralizing antibody against HIV.

Anti-HIV envelope broadly neutralizing antibodies (bnAbs) are alternatives to conventional antiretrovirals with the potential to prevent and treat infection, reduce latent reservoirs, and/or mediate a functional cure. Clinical trials with "first generation" bnAbs used alone or in combination show promising antiviral effects but also highlight that additional engineering of "enhanced" antibodies will be required for optimal clinical utility, while preserving or enhancing cGMP manufacturing capability. Here we report the engineering of an anti-CD4 binding-site (CD4bs) bnAb, N49P9.3, purified from the plasma of an HIV elite-neutralizer. Through a series of rational modifications we produced a variant that demonstrates: enhanced potency; superior antiviral activity in combination with other bnAbs; low polyreactivity; and longer circulating half-life. Additional engineering for manufacturing produced a final variant, eN49P9, with properties conducive to cGMP production. Overall, these efforts demonstrate the feasibility of developing enhanced anti-CD4bs bnAbs with greatly improved antiviral properties as well as potential translational value.

A candidate antibody drug for prevention of malaria.

Over 75% of malaria-attributable deaths occur in children under the age of 5 years. However, the first malaria vaccine recommended by the World Health Organization (WHO) for pediatric use, RTS,S/AS01 (Mosquirix), has modest efficacy. Complementary strategies, including monoclonal antibodies, will be important in efforts to eradicate malaria. Here we characterize the circulating B cell repertoires of 45 RTS,S/AS01 vaccinees and discover monoclonal antibodies for development as potential therapeutics. We generated >28,000 antibody sequences and tested 481 antibodies for binding activity and 125 antibodies for antimalaria activity in vivo. Through these analyses we identified correlations suggesting that sequences in Plasmodium falciparum circumsporozoite protein, the target antigen in RTS,S/AS01, may induce immunodominant antibody responses that limit more protective, but subdominant, responses. Using binding studies, mouse malaria models, biomanufacturing assessments and protein stability assays, we selected AB-000224 and AB-007088 for advancement as a clinical lead and backup. We engineered the variable domains (Fv) of both antibodies to enable low-cost manufacturing at scale for distribution to pediatric populations, in alignment with WHO's preferred product guidelines. The engineered clone with the optimal manufacturing and drug property profile, MAM01, was advanced into clinical development.

Framework Mutations of the 10-1074 bnAb Increase Conformational Stability, Manufacturability, and Stability While Preserving Full Neutralization Activity.

The broadly neutralizing anti-HIV antibody, 10-1074, is a highly somatically hypermutated IgG1 being developed for prophylaxis in sub-Saharan Africa. A series of algorithms were applied to identify potentially destabilizing residues in the framework of the Fv region. Of 17 residues defined, a variant was identified encompassing 1 light and 3 heavy chain residues, with significantly increased conformational stability while maintaining full neutralization activity. Central to the stabilization was the replacement of the heavy chain residue T108 with R108 at the base of the CDR3 loop which allowed for the formation of a nascent salt bridge with heavy chain residue D137. Three additional mutations were necessary to confer increased conformational stability as evidenced by differential scanning fluorimetry and isothermal chemical unfolding. In addition, we observed increased stability during low pH incubation in which 40% of the parental monomer aggregated while the combinatorial variant showed no increase in aggregation. Incubation of the variant at 100 mg/mL for 6 weeks at 40°C showed a 9-fold decrease in subvisible particles ≥2 µm relative to the parental molecule. Stability-based designs have also translated to improved pharmacokinetics. Together, these data show that increasing conformational stability of the Fab can have profound effects on the manufacturability and long-term stability of a monoclonal antibody.

Intermolecular interactions involving an acidic patch on immunoglobulin variable domain and the γ2 constant region mediate crystalline inclusion body formation in the endoplasmic reticulum.

Full-length immunoglobulins (Igs) are widely considered difficult to crystallize because of their large size, N-linked glycosylation, and flexible hinge region. However, numerous cases of intracellular Ig crystallization are reported in plasma cell dyscrasias. What makes some Ig clones more prone to crystallize during biosynthesis as well as the biochemical and cell biological requirements for this cryptic event are poorly understood. To investigate the underlying process of intracellular Ig crystallization we searched for model IgGs that can induce crystalline inclusions during recombinant overexpression. By testing various subunit combinations through mixing and matching of individual subunit chains derived from a panel of human IgG clones, we identified one secretion competent IgG2λ that induced needle-like crystalline inclusions in transfected HEK293 cells. Ig crystallization rarely occurred at steady-state cell growth conditions but was easily induced when ER-to-Golgi transport was pharmacologically blocked. Homology modeling revealed the presence of a prominent negatively-charged patch on the variable domain surface. The patch was composed of eight aspartic acids, of which five were in the heavy chain variable region and three were in the light chain. Crystallization occurred only when the two subunits were co-transfected and the intracellular crystals co-localized with ER resident proteins. Furthermore, subtype switching from IgG2 to IgG1 and stepwise neutralization of the acidic patch independently abrogated Ig crystallization events. The evidence supported that the formation of needle-like crystalline inclusions in the ER was underscored by multivalent intermolecular interactions between the acidic patch and undefined determinants present on the γ2 subunit constant region.

A rare IL33 loss-of-function mutation reduces blood eosinophil counts and protects from asthma.

IL-33 is a tissue-derived cytokine that induces and amplifies eosinophilic inflammation and has emerged as a promising new drug target for asthma and allergic disease. Common variants at IL33 and IL1RL1, encoding the IL-33 receptor ST2, associate with eosinophil counts and asthma. Through whole-genome sequencing and imputation into the Icelandic population, we found a rare variant in IL33 (NM_001199640:exon7:c.487-1G>C (rs146597587-C), allele frequency = 0.65%) that disrupts a canonical splice acceptor site before the last coding exon. It is also found at low frequency in European populations. rs146597587-C associates with lower eosinophil counts (ß = -0.21 SD, P = 2.5×10-16, N = 103,104), and reduced risk of asthma in Europeans (OR = 0.47; 95%CI: 0.32, 0.70, P = 1.8×10-4, N cases = 6,465, N controls = 302,977). Heterozygotes have about 40% lower total IL33 mRNA expression than non-carriers and allele-specific analysis based on RNA sequencing and phased genotypes shows that only 20% of the total expression is from the mutated chromosome. In half of those transcripts the mutation causes retention of the last intron, predicted to result in a premature stop codon that leads to truncation of 66 amino acids. The truncated IL-33 has normal intracellular localization but neither binds IL-33R/ST2 nor activates ST2-expressing cells. Together these data demonstrate that rs146597587-C is a loss of function mutation and support the hypothesis that IL-33 haploinsufficiency protects against asthma.

Anti-PCSK9 antibody pharmacokinetics and low-density lipoprotein-cholesterol pharmacodynamics in nonhuman primates are antigen affinity-dependent and exhibit limited sensitivity to neonatal Fc receptor-binding enhancement.

Proprotein convertase subtilisin/kexin type 9 (PCSK9) has emerged as an attractive therapeutic target for cardiovascular disease. Monoclonal antibodies (mAbs) that bind PCSK9 and prevent PCSK9:low-density lipoprotein receptor complex formation reduce serum low-density lipoprotein-cholesterol (LDL-C) in vivo. PCSK9-mediated lysosomal degradation of bound mAb, however, dramatically reduces mAb exposure and limits duration of effect. Administration of high-affinity mAb1:PCSK9 complex (1:2) to mice resulted in significantly lower mAb1 exposure compared with mAb1 dosed alone in normal mice or in PCSK9 knockout mice lacking antigen. To identify mAb-binding characteristics that minimize lysosomal disposition, the pharmacokinetic behavior of four mAbs representing a diverse range of PCSK9-binding affinities at neutral (serum) and acidic (endosomal) pH was evaluated in cynomolgus monkeys. Results revealed an inverse correlation between affinity and both mAb exposure and duration of LDL-C lowering. High-affinity mAb1 exhibited the lowest exposure and shortest duration of action (6 days), whereas mAb2 displayed prolonged exposure and LDL-C reduction (51 days) as a consequence of lower affinity and pH-sensitive PCSK9 binding. mAbs with shorter endosomal PCSK9:mAb complex dissociation half-lives (<20 seconds) produced optimal exposure-response profiles. Interestingly, incorporation of previously reported Fc-region amino acid substitutions or novel loop-insertion peptides that enhance in vitro neonatal Fc receptor binding, led to only modest pharmacokinetic improvements for mAbs with pH-dependent PCSK9 binding, with only limited augmentation of pharmacodynamic activity relative to native mAbs. A pivotal role for PCSK9 in mAb clearance was demonstrated, more broadly suggesting that therapeutic mAb-binding characteristics require optimization based on target pharmacology.

ADAM8 as a drug target in pancreatic cancer.

Pancreatic ductal adenocarcinoma (PDAC) has a grim prognosis with <5% survivors after 5 years. High expression levels of ADAM8, a metalloprotease disintegrin, are correlated with poor clinical outcome. We show that ADAM8 expression is associated with increased migration and invasiveness of PDAC cells caused by activation of ERK1/2 and higher MMP activities. For biological function, ADAM8 requires multimerization and associates with ß1 integrin on the cell surface. A peptidomimetic ADAM8 inhibitor, BK-1361, designed by structural modelling of the disintegrin domain, prevents ADAM8 multimerization. In PDAC cells, BK-1361 affects ADAM8 function leading to reduced invasiveness, and less ERK1/2 and MMP activation. BK-1361 application in mice decreased tumour burden and metastasis of implanted pancreatic tumour cells and provides improved metrics of clinical symptoms and survival in a Kras(G12D)-driven mouse model of PDAC. Thus, our data integrate ADAM8 in pancreatic cancer signalling and validate ADAM8 as a target for PDAC therapy.

Remediating agitation-induced antibody aggregation by eradicating exposed hydrophobic motifs.

Therapeutic antibodies must encompass drug product suitable attributes to be commercially marketed. An undesirable antibody characteristic is the propensity to aggregate. Although there are computational algorithms that predict the propensity of a protein to aggregate from sequence information alone, few consider the relevance of the native structure. The Spatial Aggregation Propensity (SAP) algorithm developed by Chennamsetty et. al. incorporates structural and sequence information to identify motifs that contribute to protein aggregation. We have utilized the algorithm to design variants of a highly aggregation prone IgG2. All variants were tested in a variety of high-throughput, small-scale assays to assess the utility of the method described herein. Many variants exhibited improved aggregation stability whether induced by agitation or thermal stress while still retaining bioactivity.

Entanglement model of antibody viscosity.

Antibody solutions are typically much more viscous than solutions of globular proteins at equivalent volume fraction. Here we propose that this is due to molecular entanglements that are caused by the elongated shape and intrinsic flexibility of antibody molecules. We present a simple theory in which the antibodies are modeled as linear polymers that can grow via reversible bonds between the antigen binding domains. This mechanism explains the observation that relatively subtle changes to the interparticle interaction can lead to large changes in the viscosity. The theory explains the presence of distinct power law regimes in the concentration dependence of the viscosity as well as the correlation between the viscosity and the charge on the variable domain in our antistreptavidin IgG1 model system.

Small-angle neutron scattering study of a monoclonal antibody using free-energy constraints.

Monoclonal antibodies (mAbs) contain hinge-like regions that enable structural flexibility of globular domains that have a direct effect on biological function. A subclass of mAbs, IgG2, have several interchain disulfide bonds in the hinge region that could potentially limit structural flexibility of the globular domains and affect the overall configuration space available to the mAb. We have characterized human IgG2 mAb in solution via small-angle neutron scattering (SANS) and interpreted the scattering data using atomistic models. Molecular Monte Carlo combined with molecular dynamics simulations of a model mAb indicate that a wide range of structural configurations are plausible, spanning radius of gyration values from ∼39 to ∼55 Å. Structural ensembles and representative single structure solutions were derived by comparison of theoretical SANS profiles of mAb models to experimental SANS data. Additionally, molecular mechanical and solvation free-energy calculations were carried out on the ensemble of best-fitting mAb structures. The results of this study indicate that low-resolution techniques like small-angle scattering combined with atomistic molecular simulations with free-energy analysis may be helpful to determine the types of intramolecular interactions that influence function and could lead to deleterious changes to mAb structure. This methodology will be useful to analyze small-angle scattering data of many macromolecular systems.

Native-state solubility and transfer free energy as predictive tools for selecting excipients to include in protein formulation development studies.

In the present report, two formulation strategies, based on different aggregation models, were compared for their ability to quickly predict which excipients (cosolutes) would minimize the aggregation rate of an immunoglobulin G1 monoclonal antibody (mAb-1) stored for long term at refrigerated and room temperatures. The first formulation strategy assumed that a conformational change to an aggregation-prone intermediate state was necessary to initiate the association process and the second formulation strategy assumed that protein self-association was instead controlled by the solubility of the native state. The results of these studies indicate that the stabilizing effect of excipients formulated at isotonic concentrations is derived from their ability to solubilize the native state, not by the increase of protein conformational stability induced by their presence. The degree the excipients solvate the native state was determined from the apparent transfer free energy of the native state from water into each of the excipients. These values for mAb-1 and two additional therapeutic antibodies correlated well to their long-term 4°C and room temperature aggregation data and were calculated using only the literature values for the apparent transfer free energies of the amino acids into the various excipients and the three-dimensional models of the antibodies.

In vivo crystallization of human IgG in the endoplasmic reticulum of engineered Chinese hamster ovary (CHO) cells.

Protein synthesis and secretion are essential to cellular life. Although secretory activities may vary in different cell types, what determines the maximum secretory capacity is inherently difficult to study. Increasing protein synthesis until reaching the limit of secretory capacity is one strategy to address this key issue. Under highly optimized growth conditions, recombinant CHO cells engineered to produce a model human IgG clone started housing rod-shaped crystals in the endoplasmic reticulum (ER) lumen. The intra-ER crystal growth was accompanied by cell enlargement and multinucleation and continued until crystals outgrew cell size to breach membrane integrity. The intra-ER crystals were composed of correctly folded, endoglycosidase H-sensitive IgG. Crystallizing propensity was due to the intrinsic physicochemical properties of the model IgG, and the crystallization was reproduced in vitro by exposing a high concentration of IgG to a near neutral pH. The striking cellular phenotype implicated the efficiency of IgG protein synthesis and oxidative folding exceeded the capacity of ER export machinery. As a result, export-ready IgG accumulated progressively in the ER lumen until a threshold concentration was reached to nucleate crystals. Using an in vivo system that reports accumulation of correctly folded IgG, we showed that the ER-to-Golgi transport steps became rate-limiting in cells with high secretory activity.

Chemical modifications in therapeutic protein aggregates generated under different stress conditions.

In this study, we characterized the chemical modifications in the monoclonal antibody (IgG(2)) aggregates generated under various conditions, including mechanical, chemical, and thermal stress treatment, to provide insight into the mechanism of protein aggregation and the types of aggregate produced by the different stresses. In a separate study, additional biophysical characterization was performed to arrange these aggregates into a classification system (Joubert, M. K., Luo, Q., Nashed-Samuel, Y., Wypych, J., and Narhi, L. O. (2011) J. Biol. Chem. 286, 25118-25133). Here, we report that different aggregates possessed different types and levels of chemical modification. For chemically treated samples, metal-catalyzed oxidation using copper showed site-specific oxidation of Met(246), His(304), and His(427) in the Fc portion of the antibody, which might be attributed to a putative copper-binding site. For the hydrogen peroxide-treated sample, in contrast, four solvent-exposed Met residues in the Fc portion were completely oxidized. Met and/or Trp oxidation was observed in the mechanically stressed samples, which is in agreement with the proposed model of protein interaction at the air-liquid interface. Heat treatment resulted in significant deamidation but almost no oxidation, which is consistent with thermally induced aggregates being generated by a different pathway, primarily by perturbing conformational stability. These results demonstrate that chemical modifications are present in protein aggregates; furthermore, the type, locations, and severity of the modifications depend on the specific conditions that generated the aggregates.

Diverse functions of reactive cysteines facilitate unique biosynthetic processes of aggregate-prone interleukin-31.

Interleukin-31 (IL-31) is a member of the four helical-bundle gp130/IL-6 cytokine family. Despite its implicated roles in inflammatory diseases, the biosynthetic processes of IL-31 have been poorly investigated. A detailed understanding of IL-31 biosynthesis and the nature of ligand-receptor interactions can provide insights into effective strategies for the design of therapeutic approaches. By using various heterologous protein expression systems, we demonstrated that murine IL-31 was secreted as inter-molecularly disulfide-bonded covalent aggregates. Covalently aggregated IL-31 appeared while trafficking in the secretory pathway, but was not actively retained in the ER. The aggregate formation was not caused by a dysfunctional ER quality control mechanism or an intrinsic limitation in protein folding capacity. Furthermore, secreted IL-31 aggregates were part of a large complex composed of various pleiotropic secretory factors and immune-stimulators. The extent and the heterogeneous nature of aggregates may imply that IL-31 was erroneously folded, but it was capable of signaling through cognate receptors. Mutagenesis revealed the promiscuity of all five cysteines in inter-molecular disulfide formation with components of the hetero-aggregates, but no cysteine was required for IL-31 secretion itself. Our present study not only illustrated various functions that cysteines perform during IL-31 biosynthesis and secretion, but also highlighted their potential roles in cytokine effector functions.

A diversity of antibody epitopes can induce signaling through the erythropoietin receptor.

Stimulation of red cell production through agonism of the erythropoietin receptor (EpoR) has historically been accomplished through administration of erythropoietin (EPO), the native ligand. The short half-life of EPO has led to the development of a variety of other agonists, including antibodies. It is of considerable interest to understand how these agents might activate the EpoR and whether or not it is important to bind in a manner similar to the native ligand. The binding epitopes of a panel of eight agonistic, single-chain antibody (scFv-Fc) constructs were determined through scanning alanine mutagenesis as well as more limited arginine mutagenesis of the receptor. It was found that while some of these constructs bound to receptor epitopes shared by the ligand, others bound in completely unique ways. The use of a panel of agonists and scanning mutagenesis can define the critical binding regions for signaling; in the case of the EpoR, these regions were remarkably broad.

Histidine residue mediates radical-induced hinge cleavage of human IgG1.

Hydroxyl radicals induce hinge cleavage in a human IgG1 molecule via initial radical formation at the first hinge Cys(231) followed by electron transfer to the upper hinge residues. To enable engineering of a stable monoclonal antibody hinge, we investigated the role of the hinge His(229) residue using structure modeling and site-directed mutagenesis. Direct involvement of His(229) in the reaction mechanism is suggested by a 75-85% reduction of the hinge cleavage for variants in which His(229) was substituted with either Gln, Ser, or Ala. In contrast, mutation of Lys(227) to Gln, Ser, or Ala increased hinge cleavage. However, the H229S/K227S double mutant shows hinge cleavage levels similar to that of the single H229S variant, further revealing the importance of His(229). Examination of the hinge structure shows that His(229) is capable of forming hydrogen bonds with surrounding residues. These observations led us to hypothesize that the imidazole ring of His(229) may function to facilitate the cleavage by forming a transient radical center that is capable of extracting a proton from neighboring residues. The work presented here suggests the feasibility of engineering a new generation of monoclonal antibodies capable of resisting hinge cleavage to improve product stability and efficacy.

Glutamine-linked and non-consensus asparagine-linked oligosaccharides present in human recombinant antibodies define novel protein glycosylation motifs.

We report the presence of oligosaccharide structures on a glutamine residue present in the V(L) domain sequence of a recombinant human IgG2 molecule. Residue Gln-106, present in the QGT sequence following the rule of an asparagine-linked consensus motif, was modified with biantennary fucosylated oligosaccharide structures. In addition to the glycosylated glutamine, analysis of a lectin-enriched antibody population showed that 4 asparagine residues: heavy chain Asn-162, Asn-360, and light chain Asn-164, both of which are present in the IgG1 and IgG2 constant domain sequences, and Asn-35, which was present in CDR(L)1, were also modified with oligosaccharide structures at low levels. The primary sequences around these modified residues do not adhere to the N-linked consensus sequon, NX(S/T). Modeling of these residues from known antibody crystal structures and sequence homology comparison indicates that non-consensus glycosylation occurs on Asn residues in the context of a reverse consensus motif (S/T)XN located on highly flexile turns within 3 residues of a conformational change. Taken together our results indicate that protein glycosylation is governed by more diversified requirements than previously appreciated.

A central nervous system-restricted isoform of the interleukin-1 receptor accessory protein modulates neuronal responses to interleukin-1.

Interleukin-1 (IL-1) has multiple functions in both the periphery and the central nervous system (CNS) and is regulated at many levels. We identified an isoform of the IL-1 receptor (IL-1R) accessory protein (termed AcPb) that is expressed exclusively in the CNS. AcPb interacted with IL-1 and the IL-1R but was unable to mediate canonical IL-1 responses. AcPb expression, however, modulated neuronal gene expression in response to IL-1 treatment in vitro. Animals lacking AcPb demonstrated an intact peripheral IL-1 response and developed experimental autoimmune encephalomyelitis (EAE) similarly to wild-type mice. AcPb-deficient mice were instead more vulnerable to local inflammatory challenge in the CNS and suffered enhanced neuronal degeneration as compared to AcP-deficient or wild-type mice. These findings implicate AcPb as an additional component of the highly regulated IL-1 system and suggest that it may play a role in modulating CNS responses to IL-1 and the interplay between inflammation and neuronal survival.

A role for the MHC class I-like Mill molecules in nutrient metabolism and wound healing.

MHC class I family members serve multiple functions beyond antigen presentation. We provide insight into the structure, expression and function of the Mill subfamily. This family includes two surface glycoproteins, Mill1 and Mill2. Protein sequences for Mill1 and Mill2 are most highly related to the NKG2D ligands, MICA and MICB, but neither of them bound to NKG2D. Computer-based protein modelling indicated that hereditary haemochromatosis protein (HFE), a molecule involved in iron uptake, was most similar. Mill1 and Mill2 were observed on cycling thymocytes, proliferating smooth muscle cells and fibroblasts. Using soluble Mill proteins, we found evidence for a soluble ligand in serum. Like HFE, the Mill family may be involved in nutrient metabolism. Skin was one of the only three organs found to express transcripts for both Mill1 and Mill2. Addition of antibodies specific for Mill2 to wounded skin enhanced healing. Our results suggest a role for the Mill proteins in cellular metabolism, with possible therapeutic significance.