Prediction and Reduction of the Aggregation of Monoclonal Antibodies

Abstract

Protein aggregation remains a major area of focus in the production of monoclonal antibodies. Improving the intrinsic properties of antibodies can improve manufacturability, attrition rates, safety, formulation, titers, immunogenicity, and solubility. Here, we explore the potential of predicting and reducing the aggregation propensity of monoclonal antibodies, based on the identification of aggregation-prone regions and their contribution to the thermodynamic stability of the protein. Although aggregation-prone regions are thought to occur in the antigen binding region to drive hydrophobic binding with antigen, we were able to rationally design variants that display a marked decrease in aggregation propensity while retaining antigen binding through the introduction of artificial aggregation gatekeeper residues. The reduction in aggregation propensity was accompanied by an increase in expression titer, showing that reducing protein aggregation is beneficial throughout the development process. The data presented show that this approach can significantly reduce liabilities in novel therapeutic antibodies and proteins, leading to a more efficient path to clinical studies.

Using extensional flow to reveal diverse aggregation landscapes for three IgG1 molecules

“…Taken together, our experiments point toward a common mechanism of mAb aggregation induced by the flow fields present in our device (Figure 1c) centered on the formation of activated, aggregation‐prone species that readily self‐associate, forming soluble, and then insoluble aggregates (Dobson et al, 2017; Roberts, 2007; Wang, Nema, & Teagarden, 2010). The flux through the pathway is governed by the ability of extensional flow to activate each native mAb into a perturbed structural state (van der Kant et al, 2017), the affinity of the exposed APR, the rate of relaxation from the activated species (which may itself be modulated by force [Bustamante, Chemla, Forde, & Izhaky, 2004]) and the productive collisional frequency. Consequently, mechanically robust proteins such as BSA (seventeen intra‐molecular disulfide cross‐links in the native state) or, like STT, those with reduced aggregation propensity, require the application of high strain rates and/or pass number to induce appreciable aggregation.…”

“…Differences in accelerated and innate aggregation propensities (Goldberg et al, 2017) may arise because the acceleration method increases the relative flux through certain pathways which are distinct to those traversed during production or upon storage (Chakroun, Hilton, Ahmad, Platt, & Dalby, 2016; Luo et al, 2011; Phillips et al, 2017; van der Kant et al, 2017). There is thus a need to develop stress tests that more closely replicate the conformational ensemble generated during processing and transport.…”

“…The ability to assess aggregation propensity is essential for any biopharmaceutical, including mAb-based products, to progress successfully from molecule to market (Jain et al, 2017;Tiller & Tessier, 2015; van der Kant et al, 2017). In silico analyses, such as TANGO (Fernandez-Escamilla, Rousseau, Schymkowitz, & Serrano, 2004), Zyggregator (Tartaglia & Vendruscolo, 2008), Waltz (Oliveberg, 2010), Aggrescan (Conchillo-Solé et al, 2007), and PASTA (Trovato, Seno, & Tosatto, 2007) can be used to predict the presence of aggregation-prone regions (APRs) within proteins using protein primary sequence information alone.…”

“…In silico analyses, such as TANGO (Fernandez-Escamilla, Rousseau, Schymkowitz, & Serrano, 2004), Zyggregator (Tartaglia & Vendruscolo, 2008), Waltz (Oliveberg, 2010), Aggrescan (Conchillo-Solé et al, 2007), and PASTA (Trovato, Seno, & Tosatto, 2007) can be used to predict the presence of aggregation-prone regions (APRs) within proteins using protein primary sequence information alone. In addition, CamSol (Sormanni, Aprile, & Vendruscolo, 2015) and SAP can be used to predict aggregation-prone (poorly soluble) surfaceexposed regions (Chennamsetty, Voynov, Kayser, Helk, & Trout, 2009;Trainor, Broom, & Meiering, 2017) while Solubis (Van Durme et al, 2016) integrates thermodynamic data, allowing identification of buried APRs that may be transiently exposed (van der Kant et al, 2017). A caveat of these latter in silico methods is the necessity of high resolution structural data for the molecule under study.…”

Designing Feature-Controlled Humanoid Antibody Discovery Libraries Using Generative Adversarial Networks

“…While many papers have been published trying to develop a predictable connection between an antibody's sequence and/or computed molecular structure and the molecule's various physical characteristics, the connection is elusive as it involves complex nonlinear interactions between the constituent amino acid residues. 11,[22][23][24][25][26][27][28][29][30][31][32][33][34][35][36][37] Frequently, such work involves an exceptionally small number of molecules, frequently under 200 and often under 50, from a non-diverse set of sequences -a small number of parental sequences, several parents with a small number of highly-related sequence variants, or a single antibody with mutational scanning. Such approaches give information on an individual antibody or small group, but are highly unlikely to generalize the complexity of residue interactions to other antibodies.…”