Toward generalizable prediction of antibody thermostability using machine learning on sequence and structure features.

Over the last three decades, the appeal for monoclonal antibodies (mAbs) as therapeutics has been steadily increasing as evident with FDA's recent landmark approval of the 100th mAb. Unlike mAbs that bind to single targets, multispecific biologics (msAbs) have garnered particular interest owing to the advantage of engaging distinct targets. One important modular component of msAbs is the single-chain variable fragment (scFv). Despite the exquisite specificity and affinity of these scFv modules, their relatively poor thermostability often hampers their development as a potential therapeutic drug. In recent years, engineering antibody sequences to enhance their stability by mutations has gained considerable momentum. As experimental methods for antibody engineering are time-intensive, laborious and expensive, computational methods serve as a fast and inexpensive alternative to conventional routes. In this work, we show two machine learning approaches - one with pre-trained language models (PTLM) capturing functional effects of sequence variation, and second, a supervised convolutional neural network (CNN) trained with Rosetta energetic features - to better classify thermostable scFv variants from sequence. Both of these models are trained over temperature-specific data (TS50 measurements) derived from multiple libraries of scFv sequences. On out-of-distribution (refers to the fact that the out-of-distribution sequnes are blind to the algorithm) sequences, we show that a sufficiently simple CNN model performs better than general pre-trained language models trained on diverse protein sequences (average Spearman correlation coefficient, ρ, of 0.4 as opposed to 0.15). On the other hand, an antibody-specific language model performs comparatively better than the CNN model on the same task (ρ= 0.52). Further, we demonstrate that for an independent mAb with available thermal melting temperatures for 20 experimentally characterized thermostable mutations, these models trained on TS50 data could identify 18 residue positions and 5 identical amino-acid mutations showing remarkable generalizability. Our results suggest that such models can be broadly applicable for improving the biological characteristics of antibodies. Further, transferring such models for alternative physicochemical properties of scFvs can have potential applications in optimizing large-scale production and delivery of mAbs or bsAbs.

TorsinA folding and N-linked glycosylation are sensitive to redox homeostasis.

The Endoplasmic Reticulum (ER) is responsible for the folding and post-translational modification of secretory proteins, as well as for triaging misfolded proteins. During folding, there is a complex yet only partially understood interplay between disulfide bond formation, which is an enzyme catalyzed event in the oxidizing environment of the ER, along with other post-translational modifications (PTMs) and chaperone-supported protein folding. Here, we used the glycoprotein torsinA as a model substrate to explore the impact of ER redox homeostasis on PTMs and protein biogenesis. TorsinA is a AAA+ ATPase with unusual oligomeric properties and controversial functions. The deletion of a C-terminal glutamic acid residue (∆E) is associated with the development of Early-Onset Torsion Dystonia, a severe movement disorder. TorsinA differs from other AAA+ ATPases since it is an ER resident, and as a result of its entry into the ER torsinA contains two N-linked glycans and at least one disulfide bond. The role of these PTMs on torsinA biogenesis and function and the identity of the enzymes that catalyze them are poorly defined. Using a yeast torsinA expression system, we demonstrate that a specific protein disulfide isomerase, Pdi1, affects the folding and N-linked glycosylation of torsinA and torsinA∆E in a redox-dependent manner, suggesting that the acquisition of early torsinA folding intermediates is sensitive to perturbed interactions between Cys residues and the quality control machinery. We also highlight the role of specific Cys residues during torsinA biogenesis and demonstrate that torsinA∆E is more sensitive than torsinA when these Cys residues are mutated.

Inhibition of Tumor Cell Growth and Cancer Stem Cell Expansion by a Bispecific Antibody Targeting EGFR and HER3.

The frequent activation of HER3 signaling as a resistance mechanism to EGFR-targeted therapy has motivated the development of combination therapies that block more than one receptor tyrosine kinase. Here, we have developed a novel tetravalent, bispecific single-chain diabody-Fc fusion protein targeting EGFR and HER3 (also known as ErbB3) that integrates the antigen-binding sites of a humanized version of cetuximab as well as a recently developed anti-HER3 antibody, IgG 3-43. This bispecific antibody combines the binding and neutralizing properties of the parental antibodies, as observed in biochemical and in vitro two-dimensional and three-dimensional cell culture assays, and gave rise to long-lasting growth suppression in a subcutaneous xenograft head and neck tumor model. In triple-negative breast cancer (TNBC) cell lines, treatment with the bispecific antibody inhibited the proliferation and oncosphere formation efficiency driven by HER3 signaling. In an orthotopic MDA-MB-468 tumor model, this translated into antitumor effects superior to those obtained by the parental antibodies alone or in combination and was associated with a reduced number of cells with stem-like properties. These findings demonstrate that the bispecific antibody efficiently blocks not only TNBC proliferation, but also the survival and expansion of the cancer stem cell population, holding promise for further preclinical development.

Inhibition of HER3 activation and tumor growth with a human antibody binding to a conserved epitope formed by domain III and IV.

Human epidermal growth factor receptor 3 (HER3, also known as ErbB3) has emerged as relevant target for antibody-mediated tumor therapy. Here, we describe a novel human antibody, IgG 3-43, recognizing a unique epitope formed by domain III and parts of domain IV of the extracellular region of HER3, conserved between HER3 and mouse ErbB3. An affinity of 11 nM was determined for the monovalent interaction. In the IgG format, the antibody bound recombinant bivalent HER3 with subnanomolar affinity (KD = 220 pM) and HER3-expressing tumor cells with EC50 values in the low picomolar range (27 - 83 pM). The antibody competed with binding of heregulin to HER3-expressing cells, efficiently inhibited phosphorylation of HER3 as well as downstream signaling, and induced receptor internalization and degradation. Furthermore, IgG 3-43 inhibited heregulin-dependent proliferation of several HER3-positive cancer cell lines and heregulin-independent colony formation of HER2-overexpressing tumor cell lines. Importantly, inhibition of tumor growth and prolonged survival was demonstrated in a FaDu xenograft tumor model in SCID mice. These findings demonstrate that by binding to the membrane-proximal domains III and IV involved in ligand binding and receptor dimerization, IgG 3-43 efficiently inhibits activation of HER3, thereby blocking tumor cell growth both in vitro and in vivo.