Reduction of monoclonal antibody viscosity using interpretable machine learning.

Early identification of antibody candidates with drug-like properties is essential for simplifying the development of safe and effective antibody therapeutics. For subcutaneous administration, it is important to identify candidates with low self-association to enable their formulation at high concentration while maintaining low viscosity, opalescence, and aggregation. Here, we report an interpretable machine learning model for predicting antibody (IgG1) variants with low viscosity using only the sequences of their variable (Fv) regions. Our model was trained on antibody viscosity data (>100 mg/mL mAb concentration) obtained at a common formulation pH (pH 5.2), and it identifies three key Fv features of antibodies linked to viscosity, namely their isoelectric points, hydrophobic patch sizes, and numbers of negatively charged patches. Of the three features, most predicted antibodies at risk for high viscosity, including antibodies with diverse antibody germlines in our study (79 mAbs) as well as clinical-stage IgG1s (94 mAbs), are those with low Fv isoelectric points (Fv pIs < 6.3). Our model identifies viscous antibodies with relatively high accuracy not only in our training and test sets, but also for previously reported data. Importantly, we show that the interpretable nature of the model enables the design of mutations that significantly reduce antibody viscosity, which we confirmed experimentally. We expect that this approach can be readily integrated into the drug development process to reduce the need for experimental viscosity screening and improve the identification of antibody candidates with drug-like properties.

Recent advances in generative biology for biotherapeutic discovery.

Generative biology combines artificial intelligence (AI), advanced life sciences technologies, and automation to revolutionize the process of designing novel biomolecules with prescribed properties, giving drug discoverers the ability to escape the limitations of biology during the design of next-generation protein therapeutics. Significant hurdles remain, namely: (i) the inherently complex nature of drug discovery, (ii) the bewildering number of promising computational and experimental techniques that have emerged in the past several years, and (iii) the limited availability of relevant protein sequence-function data for drug-like molecules. There is a need to focus on computational methods that will be most practically effective for protein drug discovery and on building experimental platforms to generate the data most appropriate for these methods. Here, we discuss recent advances in computational and experimental life sciences that are most crucial for impacting the pace and success of protein drug discovery.

Utility of physiologically based pharmacokinetic modeling to predict inter-antibody variability in monoclonal antibody pharmacokinetics in mice.

In this investigation, we tested the hypothesis that a physiologically based pharmacokinetic (PBPK) model incorporating measured in vitro metrics of off-target binding can largely explain the inter-antibody variability in monoclonal antibody (mAb) pharmacokinetics (PK). A diverse panel of 83 mAbs was evaluated for PK in wild-type mice and subjected to 10 in vitro assays to measure major physiochemical attributes. After excluding for target-mediated elimination and immunogenicity, 56 of the remaining mAbs with an eight-fold variability in the area under the curve (AUC0-672h: 1.74 × 106 -1.38 × 107 ng∙h/mL) and 10-fold difference in clearance (2.55-26.4 mL/day/kg) formed the training set for this investigation. Using a PBPK framework, mAb-dependent coefficients F1 and F2 modulating pinocytosis rate and convective transport, respectively, were estimated for each mAb with mostly good precision (coefficient of variation (CV%) <30%). F1 was estimated to be the mean and standard deviation of 0.961 ± 0.593, and F2 was estimated to be 2.13 ± 2.62. Using principal component analysis to correlate the regressed values of F1/F2 versus the multidimensional dataset composed of our panel of in vitro assays, we found that heparin chromatography retention time emerged as the predictive covariate to the mAb-specific F1, whereas F2 variability cannot be well explained by these assays. A sigmoidal relationship between F1 and the identified covariate was incorporated within the PBPK framework. A sensitivity analysis suggested plasma concentrations to be most sensitive to F1 when F1 > 1. The predictive utility of the developed PBPK model was evaluated against a separate panel of 14 mAbs biased toward high clearance, among which area under the curve of PK data of 12 mAbs was predicted within 2.5-fold error, and the positive and negative predictive values for clearance prediction were 85% and 100%, respectively. MAb heparin chromatography assay output allowed a priori identification of mAb candidates with unfavorable PK.

Development of in silico models to predict viscosity and mouse clearance using a comprehensive analytical data set collected on 83 scaffold-consistent monoclonal antibodies.

Biologic drug discovery pipelines are designed to deliver protein therapeutics that have exquisite functional potency and selectivity while also manifesting biophysical characteristics suitable for manufacturing, storage, and convenient administration to patients. The ability to use computational methods to predict biophysical properties from protein sequence, potentially in combination with high throughput assays, could decrease timelines and increase the success rates for therapeutic developability engineering by eliminating lengthy and expensive cycles of recombinant protein production and testing. To support development of high-quality predictive models for antibody developability, we designed a sequence-diverse panel of 83 effector functionless IgG1 antibodies displaying a range of biophysical properties, produced and formulated each protein under standard platform conditions, and collected a comprehensive package of analytical data, including in vitro assays and in vivo mouse pharmacokinetics. We used this robust training data set to build machine learning classifier models that can predict complex protein behavior from these data and features derived from predicted and/or experimental structures. Our models predict with 87% accuracy whether viscosity at 150 mg/mL is above or below a threshold of 15 centipoise (cP) and with 75% accuracy whether the area under the plasma drug concentration-time curve (AUC0-672 h) in normal mouse is above or below a threshold of 3.9 × 106 h x ng/mL.

Microfluidic viscometer by acoustic streaming transducers.

Measurement of fluid viscosity represents a huge need for many biomedical and materials processing applications. Sample fluids containing DNA, antibodies, protein-based drugs, and even cells have become important therapeutic options. The physical properties, including viscosity, of these biologics are critical factors in the optimization of the biomanufacturing processes and delivery of therapeutics to patients. Here we demonstrate an acoustic microstreaming platform termed as microfluidic viscometer by acoustic streaming transducers (µVAST) that induces fluid transport from second-order microstreaming to measure viscosity. Validation of our platform is achieved with different glycerol content mixtures to reflect different viscosities and shows that viscosity can be estimated based on the maximum speed of the second-order acoustic microstreaming. The µVAST platform requires only a small volume of fluid sample (∼1.2 µL), which is 16-30 times smaller than that of commercial viscometers. In addition, µVAST can be scaled up for ultra-high throughput measurements of viscosity. Here we demonstrate 16 samples within 3 seconds, which is an attractive feature for automating the process flows in drug development and materials manufacturing and production.

VERITAS: Harnessing the power of nomenclature in biologic discovery.

We are entering an era in which therapeutic proteins are assembled using building block-like strategies, with no standardized schema to discuss these formats. Existing nomenclatures, like AbML, sacrifice human readability for precision. Therefore, considering even a dozen such formats, in combination with hundreds of possible targets, can create confusion and increase the complexity of drug discovery. To address this challenge, we introduce Verified Taxonomy for Antibodies (VERITAS). This classification and nomenclature scheme is extensible to multispecific therapeutic formats and beyond. VERITAS names are easy to understand while drawing direct connections to the structure of a given format, with or without specific target information, making these names useful to adopt in scientific discourse and as inputs to machine learning algorithms for drug development.

Expression liabilities in a four-chain bispecific molecule.

Multispecific antibodies, often composed of three to five polypeptide chains, have become increasingly relevant in the development of biotherapeutics. These molecules have mechanisms of action that include redirecting T cells to tumors and blocking multiple pathogenic mediators simultaneously. One of the major challenges for asymmetric multispecific antibodies is generating a high proportion of the correctly paired antibody during production. To understand the causes and effects of chain mispairing impurities in a difficult to express multispecific hetero-IgG, we investigated consequences of individual and pairwise chain expression in mammalian transient expression hosts. We found that one of the two light chains (LC) was not secretion competent when transfected individually or cotransfected with the noncognate heavy chain (HC). Overexpression of this secretion impaired LC reduced cell growth while inducing endoplasmic reticulum stress and CCAAT/enhancer-binding protein homologous protein (CHOP) expression. The majority of this LC was observed as monomer with incomplete intrachain disulfide bonds when expressed individually. Russell bodies (RB) were induced when this LC was co-expressed with the cognate HC. Moreover, one HC paired promiscuously with noncognate LC. These results identify the causes for the low product quality observed from stable cell lines expressing this heteroIgG and suggest mitigation strategies to improve overall process productivity of the correctly paired multispecific antibody. The approach described here provides a general strategy for identifying the molecular and cellular liabilities associated with difficult to express multispecific antibodies.

Nanoscale integration of single cell biologics discovery processes using optofluidic manipulation and monitoring.

The new and rapid advancement in the complexity of biologics drug discovery has been driven by a deeper understanding of biological systems combined with innovative new therapeutic modalities, paving the way to breakthrough therapies for previously intractable diseases. These exciting times in biomedical innovation require the development of novel technologies to facilitate the sophisticated, multifaceted, high-paced workflows necessary to support modern large molecule drug discovery. A high-level aspiration is a true integration of "lab-on-a-chip" methods that vastly miniaturize cellulmical experiments could transform the speed, cost, and success of multiple workstreams in biologics development. Several microscale bioprocess technologies have been established that incrementally address these needs, yet each is inflexibly designed for a very specific process thus limiting an integrated holistic application. A more fully integrated nanoscale approach that incorporates manipulation, culture, analytics, and traceable digital record keeping of thousands of single cells in a relevant nanoenvironment would be a transformative technology capable of keeping pace with today's rapid and complex drug discovery demands. The recent advent of optical manipulation of cells using light-induced electrokinetics with micro- and nanoscale cell culture is poised to revolutionize both fundamental and applied biological research. In this review, we summarize the current state of the art for optical manipulation techniques and discuss emerging biological applications of this technology. In particular, we focus on promising prospects for drug discovery workflows, including antibody discovery, bioassay development, antibody engineering, and cell line development, which are enabled by the automation and industrialization of an integrated optoelectronic single-cell manipulation and culture platform. Continued development of such platforms will be well positioned to overcome many of the challenges currently associated with fragmented, low-throughput bioprocess workflows in biopharma and life science research.

Distinct domains of tRNA synthetase recognize the same base pair.

Synthesis of proteins containing errors (mistranslation) is prevented by aminoacyl transfer RNA synthetases through their accurate aminoacylation of cognate tRNAs and their ability to correct occasional errors of aminoacylation by editing reactions. A principal source of mistranslation comes from mistaking glycine or serine for alanine, which can lead to serious cell and animal pathologies, including neurodegeneration. A single specific G.U base pair (G3.U70) marks a tRNA for aminoacylation by alanyl-tRNA synthetase. Mistranslation occurs when glycine or serine is joined to the G3.U70-containing tRNAs, and is prevented by the editing activity that clears the mischarged amino acid. Previously it was assumed that the specificity for recognition of tRNA(Ala) for editing was provided by the same structural determinants as used for aminoacylation. Here we show that the editing site of alanyl-tRNA synthetase, as an artificial recombinant fragment, targets mischarged tRNA(Ala) using a structural motif unrelated to that for aminoacylation so that, remarkably, two motifs (one for aminoacylation and one for editing) in the same enzyme independently can provide determinants for tRNA(Ala) recognition. The structural motif for editing is also found naturally in genome-encoded protein fragments that are widely distributed in evolution. These also recognize mischarged tRNA(Ala). Thus, through evolution, three different complexes with the same tRNA can guard against mistaking glycine or serine for alanine.

Stereoselective incorporation of an unsaturated isoleucine analogue into a protein expressed in E. coli.

The unsaturated amino acid 2-amino-3-methyl-4-pentenoic acid (E-Ile) was prepared in the form of its (2S,3S),(2R,3R) and (2S,3R),(2R,3S) stereoisomeric pairs. The translational activities of SS-E-Ile and SR-E-Ile were assessed in an E. coli strain rendered auxotrophic for isoleucine. SS-E-Ile was incorporated into the test protein mouse dihydrofolate reductase (mDHFR) in place of isoleucine at a rate of up to 72 %; SR-E-Ile yielded no conclusive evidence for incorporation. ATP/PPi exchange assays indicated that SS-E-Ile was activated by the isoleucyl-tRNA synthetase at a rate comparable to that characteristic of isoleucine; SR-E-Ile was activated approximately 100-times more slowly than SS-E-Ile.

Non-canonical amino acids in protein engineering.

Methods for engineering proteins that contain non-canonical amino acids have advanced rapidly in the past few years. Novel amino acids can be introduced into recombinant proteins in either a residue-specific or site-specific fashion. The methods are complementary: residue-specific incorporation allows engineering of the overall physical and chemical behavior of proteins and protein-like macromolecules, whereas site-specific methods allow mechanistic questions to be probed in atomistic detail. Challenges remain in the engineering of the translational apparatus and in the design of schemes that can be used to encode both canonical and non-canonical amino acids.