Co-optimization of therapeutic antibody affinity and specificity using machine learning models that generalize to novel mutational space.

Therapeutic antibody development requires selection and engineering of molecules with high affinity and other drug-like biophysical properties. Co-optimization of multiple antibody properties remains a difficult and time-consuming process that impedes drug development. Here we evaluate the use of machine learning to simplify antibody co-optimization for a clinical-stage antibody (emibetuzumab) that displays high levels of both on-target (antigen) and off-target (non-specific) binding. We mutate sites in the antibody complementarity-determining regions, sort the antibody libraries for high and low levels of affinity and non-specific binding, and deep sequence the enriched libraries. Interestingly, machine learning models trained on datasets with binary labels enable predictions of continuous metrics that are strongly correlated with antibody affinity and non-specific binding. These models illustrate strong tradeoffs between these two properties, as increases in affinity along the co-optimal (Pareto) frontier require progressive reductions in specificity. Notably, models trained with deep learning features enable prediction of novel antibody mutations that co-optimize affinity and specificity beyond what is possible for the original antibody library. These findings demonstrate the power of machine learning models to greatly expand the exploration of novel antibody sequence space and accelerate the development of highly potent, drug-like antibodies.

Toward Drug-Like Multispecific Antibodies by Design.

The success of antibody therapeutics is strongly influenced by their multifunctional nature that couples antigen recognition mediated by their variable regions with effector functions and half-life extension mediated by a subset of their constant regions. Nevertheless, the monospecific IgG format is not optimal for many therapeutic applications, and this has led to the design of a vast number of unique multispecific antibody formats that enable targeting of multiple antigens or multiple epitopes on the same antigen. Despite the diversity of these formats, a common challenge in generating multispecific antibodies is that they display suboptimal physical and chemical properties relative to conventional IgGs and are more difficult to develop into therapeutics. Here we review advances in the design and engineering of multispecific antibodies with drug-like properties, including favorable stability, solubility, viscosity, specificity and pharmacokinetic properties. We also highlight emerging experimental and computational methods for improving the next generation of multispecific antibodies, as well as their constituent antibody fragments, with natural IgG-like properties. Finally, we identify several outstanding challenges that need to be addressed to increase the success of multispecific antibodies in the clinic.

Photoresponsive azo-combretastatin A-4 analogues.

Colchicine analogues in which an azo group is incorporated into a molecule containing the key pharmacophore of colchicine, have found particular utility as switchable tubulin binding chemotherapeutics. Combretastatin is a related compound containing a stilbene fragment that shows different bioactivity for the cis and trans isomers. We have performed cell assays on 17 new compounds structurally related to a previously reported azo-analogue of combretastatin. One of these compounds showed enhanced potency against HeLa (IC50 = 0.11 µM) and H157 cells (IC50 = 0.20 µM) for cell studies under 400 nm irradiation and the highest photoactivity (IC50 with irradiation/IC50 in dark = 550). We have performed docking and physicochemical studies of this new compound (7). Kinetic studies in water reveal a longer half-life for the cis isomer of 7 which may be one factor responsible for the better IC50 values in cell assays and the improved photoresponsive behavior.

Synthesis, Characterization, and Bioactivity of the Photoisomerizable Tubulin Polymerization Inhibitor azo-Combretastatin A4.

Combretastatin A4 is a stilbenoid tubulin binding mitotic inhibitor whose conformation greatly influences its potency, making it an excellent candidate for adaptation as a photoactivatable tool. Herein we report a novel synthesis, the facile isomerization with commercial grade equipment, and biological activity of azo-combretastatin A4 in vitro and in human cancer cells. Photoisomerized azo-combretestatin A4 is at least 200-fold more potent in cellular culture, making it a promising phototherapeutic and biomedical research tool.

Metal complex catalysis in living biological systems.

This feature article discusses synthetic metal complexes that are capable of catalyzing chemical transformations in living organisms. Photodynamic therapy exemplifies what is probably the most established artificial catalytic process exploited in medicine, namely the photosensitized catalytic generation of cell-damaging singlet oxygen. Different redox catalysts have been designed over the last two decades to target a variety of redox alterations in cancer and other diseases. For example, pentaazamacrocyclic manganese(ii) complexes catalyze the dismutation of superoxide to O(2) and H(2)O(2)in vivo and thus reduce oxidative stress in analogy to the native enzyme superoxide dismutase. Recently, piano-stool ruthenium and iridium complexes were reported to influence cellular redox homeostasis indirectly by catalytic glutathione oxidation and catalytic transfer hydrogenation using the coenzyme NADH, respectively. Over the last few years, significant progress has been made towards the application of non-biological reactions in living systems, ranging from the organoruthenium-catalyzed cleavage of allylcarbamates and a gold-catalyzed intramolecular hydroarylation to palladium-catalyzed Suzuki-Miyaura and Sonogashira cross-couplings within the cytoplasm or on the surface of living cells. The design of bioorthogonal catalyst/substrate pairs, which can passively diffuse into cells, combines the advantages of small molecules with catalysis and promises to provide exciting new tools for future chemical biology studies.

Catalytic azide reduction in biological environments.

In the quest for the identification of catalytic transformations to be used in chemical biology and medicinal chemistry, we identified iron(III) meso-tetraarylporphines as efficient catalysts for the reduction of aromatic azides to their amines. The reaction uses thiols as reducing agents and tolerates water, air, and other biological components. A caged fluorophore was employed to demonstrate that the reduction can be performed even in living mammalian cells. However, in vivo experiments in nematodes (Caenorhabditis elegans) and zebrafish (Danio rerio) revealed a limitation to this method: the metabolic reduction of aromatic azides.

Computational design of a ß-peptide that targets transmembrane helices.

The design of ß-peptide foldamers targeting the transmembrane (TM) domains of complex natural membrane proteins has been a formidable challenge. A series of ß-peptides was designed to stably insert in TM orientations in phospholipid bilayers. Their secondary structures and orientation in the phospholipid bilayer was characterized using biophysical methods. Computational methods were then devised to design a ß-peptide that targeted a TM helix of the integrin α(IIb)ß(3). The designed peptide (ß-CHAMP) interacts with the isolated target TM domain of the protein and activates the intact integrin in vitro.

Specificity for homooligomer versus heterooligomer formation in integrin transmembrane helices.

Transmembrane (TM) helices engage in homomeric and heteromeric interactions that play essential roles in the folding and assembly of TM proteins. However, features that explain their propensity to interact homomerically or heteromerically and determine the strength of these interactions are poorly understood. Integrins provide an ideal model system for addressing these questions because the TM helices of full-length integrins interact heteromerically when integrins are inactive, but isolated TM helices are also able to form homodimers or homooligomers in micelles and bacterial membranes. We sought to determine the features defining specificity for homointeractions versus heterointeractions by conducting a comprehensive comparison of the homomeric and heteromeric interactions of integrin alphaIIbbeta3 TM helices in biological membranes. Using the TOXCAT assay, we found that residues V700, M701, A703, I704, L705, G708, L709, L712, and L713, which are located on the same face of the beta3 helix, mediate homodimer formation. We then characterized the beta3 heterodimer by measuring the ability of beta3 helix mutations to cause ligand binding to alphaIIbbeta3. We found that mutating V696, L697, V700, M701, A703. I704, L705, G708, L712, and L713, but not the small residue-X(3)-small residue motif S699-X(3)-A703, caused constitutive alphaIIbbeta3 activation, as well as persistent focal adhesion kinase phosphorylation dependent on alphaIIbbeta3 activation. Because alphaIIb and beta3 use the same face of their respective TM helices for homomeric and heteromeric interactions, the interacting surface on each has an intrinsic "stickiness" predisposing towards helix-helix interactions in membranes. The residues responsible for heterodimer formation comprise a network of interdigitated side chains with considerable geometric complementarity; mutations along this interface invariably destabilize heterodimer formation. By contrast, residues responsible for homomeric interactions are dispersed over a wider surface. While most mutations of these residues are destabilizing, some stabilized homooligomer formation. We conclude that the alphaIIbbeta3 TM heterodimer shows the hallmark of finely tuned heterodimeric interaction, while homomeric interaction is less specific.