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.

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.

Development of BioRad NGC and GE ÄKTA pure systems for highly automated three column protein purification employing tandem affinity, buffer exchange and size exclusion chromatography.

Affinity purification, such as Protein A (ProA) followed by size exclusion chromatography (SEC) remains a popular method to obtain research scale proteins. With the need for higher throughput protein production increasing for discovery research, there is substantial interest in the automation of complex protein purification processes, which often start with a ProA step followed by SEC. However, the harsh elution conditions from ProA based chromatography can destabilize some proteins resulting in particulates, which in turn can cause column fouling and potential cross-contamination of subsequent purifications. We modified both Bio Rad NGC and ÄKTA Pure systems to run a three-column process (ProA to buffer exchange to SEC) enabling automated tandem affinity to SEC purification while minimizing the risk of SEC column fouling and subsequent cross-contamination. The intervening buffer exchange column, unlike the final SEC column, can be rapidly regenerated using harsh methods between runs, and these automated systems are capable of processing up to six samples per day without user intervention.

A High-Throughput Automated Protein Folding System.

In vitro protein folding can be employed to produce complex proteins expressed as insoluble inclusion bodies in E. coli from laboratory to commercial scale. Often the most challenging step is identification of renaturation conditions that will enable the denatured protein to form the native structure at an acceptable yield. Generally this requires screening a matrix of buffers and stabilizers to find an appropriate solution. Herein, we describe an automated and quantitative method to identify optimal in vitro protein folding parameters with a high rate of success.

Immunoaffinity capture coupled with capillary electrophoresis - mass spectrometry to study therapeutic protein stability in vivo.

Protein engineering is at an all-time high in biopharmaceutics. As a result, absorption, distribution, metabolism and excretion (ADME) of proteins has become more important to understand in the context of engineering strategies to optimize therapeutic properties of potential lead constructs. Immunoaffinity capture coupled with a newly developed capillary electrophoresis - mass spectrometry (CE-MS) system was used to characterize intact protein mass analysis of a wild type Fc-FGF21 construct and a sequence re-engineered Fc-FGF21 construct from an in vivo study. A number of truncated forms were observed and the time courses of the various proteolytic products were identified and compared between the two constructs. The abundances of the intact and truncated forms were used to provide the basis to semi-quantify ADME properties of the two protein forms. The use of this immunoaffinity capture followed by CE-MS based intact mass analysis workflow provided a qualitative and quantitative analysis of the pharmacokinetic profiles of the two proteins. The platform presented here holds great potential in characterization of the ADME properties of proteins.

A Novel Fc-FGF21 With Improved Resistance to Proteolysis, Increased Affinity Toward ß-Klotho, and Enhanced Efficacy in Mice and Cynomolgus Monkeys.

Fibroblast growth factor (FGF) 21 is a natural hormone that modulates glucose, lipid, and energy metabolism. Previously, we engineered an Fc fusion FGF21 variant with two mutations, Fc-FGF21(RG), to extend the half-life and reduce aggregation and in vivo degradation of FGF21. We now describe a new variant developed to reduce the extreme C-terminal degradation and improve the binding affinity to ß-Klotho. We demonstrate, by introducing one additional mutation located at the C terminus of FGF21 (A180E), that the new molecule, Fc-FGF21(RGE), has gained many improved attributes. Compared with Fc-FGF21(RG), Fc-FGF21(RGE) has similar in vitro potency, preserves ß-Klotho dependency, and maintains FGF receptor selectivity and cross-species reactivity. In vivo, Fc-FGF21(RGE) showed reduced susceptibility to extreme C-terminal degradation and increased plasma levels of the bioactive intact molecule. The circulating half-life of intact Fc-FGF21(RGE) increased twofold compared with that of Fc-FGF21(RG) in mice and cynomolgus monkeys. Additionally, Fc-FGF21(RGE) exhibited threefold to fivefold enhanced binding affinity to coreceptor ß-Klotho across mouse, cynomolgus monkey, and human species. In obese and diabetic mouse and cynomolgus monkey models, Fc-FGF21(RGE) demonstrated greater efficacies to Fc-FGF21(RG), resulting in larger and more sustained improvements in multiple metabolic parameters. No increased immunogenicity was observed with Fc-FGF21(RGE). The superior biophysical, pharmacokinetic, and pharmacodynamic properties, as well as the positive metabolic effects across species, suggest that further clinical development of Fc-FGF21(RGE) as a metabolic therapy for diabetic and/or obese patients may be warranted.

Automated high-throughput dense matrix protein folding screen using a liquid handling robot combined with microfluidic capillary electrophoresis.

Modern molecular genetics technology has made it possible to swiftly sequence, clone and mass-produce recombinant DNA for the purpose of expressing heterologous genes of interest; however, recombinant protein production systems have struggled to keep pace. Mammalian expression systems are typically favored for their ability to produce and secrete proteins in their native state, but bacterial systems benefit from rapid cell line development and robust growth. The primary drawback to prokaryotic expression systems are that recombinant proteins are generally not secreted at high levels or correctly folded, and are often insoluble, necessitating post-expression protein folding to obtain the active product. In order to harness the advantages of prokaryotic expression, high-throughput methods for executing protein folding screens and the subsequent analytics to identify lead conditions are required. Both of these tasks can be accomplished using a Biomek 3000 liquid handling robot to prepare the folding screen and to subsequently prepare the reactions for assessment using Caliper microfluidic capillary electrophoresis. By augmenting a protein folding screen with automation, the primary disadvantage of Escherichia coli expression has been mitigated, namely the labor intensive identification of the required protein folding conditions. Furthermore, a rigorous, quantitative method for identifying optimal protein folding buffer aids in the rapid development of an optimal production process.

Automated two-step chromatography using an ÄKTA equipped with in-line dilution capability.

There has been a great emphasis on developing higher-throughput protein purification techniques to screen potential human therapeutics faster and more efficiently. Not only is it desirable to have high-throughput purification for initial screens but it is also desirable to efficiently purify selected protein therapeutics in the amounts and purity required for definitive assays. Current automated tandem technologies involve size exclusion as a second step that often fails to generate the required purity, is not robust and can only be operated at a limited scale. We have modified an ÄKTA to enable in-line dilution, assuring that the automated loading of a second column from a first column elution can be modified to a pH and ionic strength which is suitable for binding to the second column. For example, Protein A can be employed as a first step followed by direct loading on to a cation exchange column by conditioning the Protein A elution using the in-line diluter. Using this method as described, up to six samples of 1L each can be purified through two columns without human intervention per day per machine, and the system produces good yields of purified protein over a wide range of loading levels (12-300mg). In addition, the system employs guanidine HCl regeneration, followed by a sodium hydroxide wash between purification runs, minimizing the possibility of carryover contamination. The system is described at the 5mL and the 10mL column sizes; however, it could readily be programed for 100mL columns to enable larger-scale purifications. Using this system to automate two-column purifications minimizes human intervention, increases efficiency and minimizes the risk of human error.

Rationale-Based Engineering of a Potent Long-Acting FGF21 Analog for the Treatment of Type 2 Diabetes.

Fibroblast growth factor 21 (FGF21) is a promising drug candidate for the treatment of type 2 diabetes. However, the use of wild type native FGF21 is challenging due to several limitations. Among these are its short half-life, its susceptibility to in vivo proteolytic degradation and its propensity to in vitro aggregation. We here describe a rationale-based protein engineering approach to generate a potent long-acting FGF21 analog with improved resistance to proteolysis and aggregation. A recombinant Fc-FGF21 fusion protein was constructed by fusing the Fc domain of human IgG1 to the N-terminus of human mature FGF21 via a linker peptide. The Fc positioned at the N-terminus was determined to be superior to the C-terminus as the N-terminal Fc fusion retained the ßKlotho binding affinity and the in vitro and in vivo potency similar to native FGF21. Two specific point mutations were introduced into FGF21. The leucine to arginine substitution at position 98 (L98R) suppressed FGF21 aggregation at high concentrations and elevated temperatures. The proline to glycine replacement at position 171 (P171G) eliminated a site-specific proteolytic cleavage of FGF21 identified in mice and cynomolgus monkeys. The derived Fc-FGF21(RG) molecule demonstrated a significantly improved circulating half-life while maintaining the in vitro activity similar to that of wild type protein. The half-life of Fc-FGF21(RG) was 11 h in mice and 30 h in monkeys as compared to 1-2 h for native FGF21 or Fc-FGF21 wild type. A single administration of Fc-FGF21(RG) in diabetic mice resulted in a sustained reduction in blood glucose levels and body weight gains up to 5-7 days, whereas the efficacy of FGF21 or Fc-FGF21 lasted only for 1 day. In summary, we engineered a potent and efficacious long-acting FGF21 analog with a favorable pharmaceutical property for potential clinical development.

A novel anion-exchange resin suitable for both discovery research and clinical manufacturing purposes.

Strong ion-exchange protein chromatography is one of the most powerful and most common steps for protein purification in both discovery research and manufacturing. However, the demands on protein purification of early drug discovery and later stage manufacturing are quite different. In order to shorten the time of developing a purification process for new protein drug candidates, there is a need for a strong ion-exchange resin that will be optimum for both stages. This article details a novel anion-exchange resin suitable for research, as well as for clinical manufacturing. In this study, a novel Q resin anion-exchange prototype was evaluated and compared to the GE Healthcare Q Sepharose® Fast Flow (QFF) and Q Sepharose® High Performance (QHP) resins. This study specifically focused on the following: resolution, dynamic binding capacity, flow rate, back pressure, and scale up. The evaluation was performed in both small- and large-scale experiments. From all the comparable data, the prototype resin is adaptable for both discovery research and manufacturing. Its wide-range operation suitability could potentially shorten the time required to develop conventional purification protocols for clinical manufacturing.

Targeted codon optimization improves translational fidelity for an Fc fusion protein.

High levels of translational errors, both truncation and misincorporation in an Fc-fusion protein were observed. Here, we demonstrate the impact of several commercially available codon optimization services, and compare to a targeted strategy. Using the targeted strategy, only codons known to have translational errors are modified. For an Fc-fusion protein expressed in Escherichia coli, the targeted strategy, in combination with appropriate fermentation conditions, virtually eliminated misincorporation (proteins produced with a wrong amino acid sequence), and reduced the level of truncation. The use of full optimization using commercially available strategies reduced the initial errors, but introduced different misincorporations. However, truncation was higher using the targeted strategy than for most of the full optimization strategies. This targeted approach, along with monitoring of translation fidelity and careful attention to fermentation conditions is key to minimizing translational error and ensuring high-quality expression. These findings should be useful for other biopharmaceutical products, as well as any other transgenic constructs where protein quality is important.

FGF21 N- and C-termini play different roles in receptor interaction and activation.

Fibroblast growth factor-21 (FGF21) signaling requires the presence of beta-Klotho, a co-receptor with a very short cytoplasmic domain. Here we show that FGF21 binds directly to beta-Klotho through its C-terminus. Serial C-terminal truncations of FGF21 weakened or even abrogated its interaction with beta-Klotho in a Biacore assay, and led to gradual loss of potency in a luciferase reporter assay but with little effect on maximal response. In contrast, serial N-terminal truncations of FGF21 had no impact on beta-Klotho binding. Interestingly, several of them exhibited characteristics of partial agonists with minimal effects on potency. These data demonstrate that the C-terminus of FGF21 is critical for binding to beta-Klotho and the N-terminus is critical for fibroblast growth factor receptor (FGFR) activation.

Structural determinants of the ADAM inhibition by TIMP-3: crystal structure of the TACE-N-TIMP-3 complex.

TIMP-3 (tissue inhibitor of metalloproteinases 3) is unique among the TIMP inhibitors, in that it effectively inhibits the TNF-alpha converting enzyme (TACE). In order to understand this selective capability of inhibition, we crystallized the complex formed by the catalytic domain of recombinant human TACE and the N-terminal domain of TIMP-3 (N-TIMP-3), and determined its molecular structure with X-ray data to 2.3 A resolution. The structure reveals that TIMP-3 exhibits a fold similar to those of TIMP-1 and TIMP-2, and interacts through its functional binding edge, which consists of the N-terminal segment and other loops, with the active-site cleft of TACE in a manner similar to that of matrix metalloproteinases (MMPs). Therefore, the mechanism of TIMP-3 binding toward TACE is not fundamentally different from that previously elucidated for the MMPs. The Phe34 phenyl side chain situated at the tip of the relatively short sA-sB loop of TIMP-3 extends into a unique hydrophobic groove of the TACE surface, and two Leu residues in the adjacent sC-connector and sE-sF loops are tightly packed in the interface allowing favourable interactions, in agreement with predictions obtained by systematic mutations by Gillian Murphy's group. The combination of favourable functional epitopes together with a considerable flexibility renders TIMP-3 an efficient TACE inhibitor. This structure might provide means to design more efficient TIMP inhibitors of TACE.