Humanization of antibodies using a machine learning approach on large-scale repertoire data.

MOTIVATION: Monoclonal antibody (mAb) therapeutics are often produced from non-human sources (typically murine), and can therefore generate immunogenic responses in humans. Humanization procedures aim to produce antibody therapeutics that do not elicit an immune response and are safe for human use, without impacting efficacy. Humanization is normally carried out in a largely trial-and-error experimental process. We have built machine learning classifiers that can discriminate between human and non-human antibody variable domain sequences using the large amount of repertoire data now available. RESULTS: Our classifiers consistently outperform the current best-in-class model for distinguishing human from murine sequences, and our output scores exhibit a negative relationship with the experimental immunogenicity of existing antibody therapeutics. We used our classifiers to develop a novel, computational humanization tool, Hu-mAb, that suggests mutations to an input sequence to reduce its immunogenicity. For a set of therapeutic antibodies with known precursor sequences, the mutations suggested by Hu-mAb show substantial overlap with those deduced experimentally. Hu-mAb is therefore an effective replacement for trial-and-error humanization experiments, producing similar results in a fraction of the time. AVAILABILITY AND IMPLEMENTATION: Hu-mAb (humanness scoring and humanization) is freely available to use at opig.stats.ox.ac.uk/webapps/humab. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Persistent Polyfunctional Chimeric Antigen Receptor T Cells That Target Glypican 3 Eliminate Orthotopic Hepatocellular Carcinomas in Mice.

BACKGROUND AND AIMS: Glypican 3 (GPC3) is an oncofetal antigen involved in Wnt-dependent cell proliferation that is highly expressed in hepatocellular carcinoma (HCC). We investigated whether the functions of chimeric antigen receptors (CARs) that target GPC3 are affected by their antibody-binding properties. METHODS: We collected peripheral blood mononuclear cells from healthy donors and patients with HCC and used them to create CAR T cells, based on the humanized YP7 (hYP7) and HN3 antibodies, which have high affinities for the C-lobe and N-lobe of GPC3, respectively. NOD/SCID/IL-2Rgcnull (NSG) mice were given intraperitoneal injections of luciferase-expressing (Luc) Hep3B or HepG2 cells and after xenograft tumors formed, mice were given injections of saline or untransduced T cells (mock control), or CAR (HN3) T cells or CAR (hYP7) T cells. In other NOD/SCID/IL-2Rgcnull (NSG) mice, HepG2-Luc or Hep3B-Luc cells were injected into liver, and after orthotopic tumors formed, mice were given 1 injection of CAR (hYP7) T cells or CD19 CAR T cells (control). We developed droplet digital polymerase chain reaction and genome sequencing methods to analyze persistent CAR T cells in mice. RESULTS: Injections of CAR (hYP7) T cells eliminated tumors in 66% of mice by week 3, whereas CAR (HN3) T cells did not reduce tumor burden. Mice given CAR (hYP7) T cells remained tumor free after re-challenge with additional Hep3B cells. The CAR T cells induced perforin- and granzyme-mediated apoptosis and reduced levels of active ß-catenin in HCC cells. Mice injected with CAR (hYP7) T cells had persistent expansion of T cells and subsets of polyfunctional CAR T cells via antigen-induced selection. These T cells were observed in the tumor microenvironment and spleen for up to 7 weeks after CAR T-cell administration. Integration sites in pre-infusion CAR (HN3) and CAR (hYP7) T cells were randomly distributed, whereas integration into NUPL1 was detected in 3.9% of CAR (hYP7) T cells 5 weeks after injection into tumor-bearing mice and 18.1% of CAR (hYP7) T cells at week 7. There was no common site of integration in CAR (HN3) or CD19 CAR T cells from tumor-bearing mice. CONCLUSIONS: In mice with xenograft or orthoptic liver tumors, CAR (hYP7) T cells eliminate GPC3-positive HCC cells, possibly by inducing perforin- and granzyme-mediated apoptosis or reducing Wnt signaling in tumor cells. GPC3-targeted CAR T cells might be developed for treatment of patients with HCC.

A comparative study of the developability of full-length antibodies, fragments, and bispecific formats reveals higher stability risks for engineered constructs.

Engineered antibody formats, such as antibody fragments and bispecifics, have the potential to offer improved therapeutic efficacy compared to traditional full-length monoclonal antibodies (mAbs). However, the translation of these non-natural molecules into successful therapeutics can be hampered by developability challenges. Here, we systematically analyzed 64 different antibody constructs targeting Tumor Necrosis Factor (TNF) which cover 8 distinct molecular format families, encompassing full-length antibodies, various types of single chain variable fragments, and bispecifics. We measured 15 biophysical properties related to activity, manufacturing, and stability, scoring variants with a flag-based risk approach and a recent in silico developability profiler. Our comparative assessment revealed that overall developability is higher for the natural full-length antibody format. Bispecific antibodies, antibodies with scFv fragments at the C-terminus of the light chain, and single-chain Fv antibody fragments (scFvs) have intermediate developability properties, while more complicated formats, such as scFv- scFv, bispecific mAbs with one Fab exchanged with a scFv, and diabody formats are collectively more challenging. In particular, our study highlights the propensity for fragmentation and aggregation, both in bulk and at interfaces, for many current engineered formats.

Advances in computational structure-based antibody design.

Antibodies are currently the most important class of biotherapeutics and are used to treat numerous diseases. Recent advances in computational methods are ushering in a new era of antibody design, driven in part by accurate structure prediction. Previously, structure-based antibody design has been limited to a relatively small number of cases where accurate structures or models of both the target antigen and antibody were available. As we move towards a time where it is possible to accurately model most antibodies and antigens, and to reliably predict their binding site, there is vast potential for true computational antibody design. In this review, we describe the latest methods that promise to launch a paradigm shift towards entirely in silico structure-based antibody design.