Predictive modeling of concentration-dependent viscosity behavior of monoclonal antibody solutions using artificial neural networks

Abstract

Solutions of monoclonal antibodies (mAbs) can show increased viscosity at high concentration, which can be a disadvantage during protein purification, filling, and administration. The viscosity is determined by protein-protein-interactions, which are influenced by the antibody’s sequence as well as solution conditions, like pH, buffer type, or the presence of salts and other excipients. To predict viscosity, experimental parameters, like the diffusion interaction parameter (kD), or computational tools harnessing information derived from primary sequence, are often used, but a reliable predictive tool is still missing. We present a modeling approach employing artificial neural networks (ANNs) using experimental factors combined with simulation-derived parameters plus viscosity data from 27 highly concentrated (180 mg/mL) mAbs. These ANNs can be used to predict if mAbs exhibit problematic viscosity at distinct concentrations or to model viscosity-concentration-curves.

Biophysical Determinants for the Viscosity of Concentrated Monoclonal Antibody Solutions

“…The second virial coefficient B 2 B 2 = 2 π ∫ 0 ∞ normald r nobreak0em0.25em⁡ r 2 [ 1 − e − u ( r ) / k normalB T ] is related to A 2 , which can be determined by SLS (see above) via B 2 = A 2 · M W 2 N normalA We employ the reduced second virial coefficient B 2 * , defined as the ratio B 2 / B 2 HS ( B 2 HS : second virial coefficient for hard spheres) and linked to the stickiness parameter τ by B 2 * = B 2 / B 2 H S = 1 − 1 4 τ where B 2 HS = 2πσ 3 /3, with σ being the diameter of the hard spheres . In the case of the mAb samples studied here, we employ the approach by Da Vela et al .…”

“…The second virial coefficient ) is related to A 2 , which can be determined by SLS (see above) 41 via 116…”

“…In this work, we denote these antibodies as mAb1, mAb9, mAb12, mAb16, and mAb24, consistent with previous work on the same molecules. 41 All mAbs were manufactured in-house at Lonza AG/Ltd. Double gene vectors containing the heavy and light chains were transfected into CHOK1SV GS-KO cells 60 from Lonza Biologics (Slough, UK) and cultured under selection conditions as stable pooled cultures.…”

“…38,39 Apgar and co-workers applied the same approach optimizing the charge distribution and hydrophobicity in the variable region (VH and VL) to successfully reduce the viscosity of a mAb solution while preserving its stability and its affinity for the antigen. 40 Schmitt et al recently presented a predictive modeling approach for viscosity employing artificial neural networks and based on both experimental and simulation-derived parameters and viscosity data from 27 highly concentrated mAbs, 41 showing that sequence-based optimization of mAb properties is a powerful method for rational mAb design. In addition, recent studies have successfully investigated the link between viscosity and cluster formation using small-angle X-ray scattering (SAXS), microrheology, and coarse-grained simulations and used cluster theory to predict structure factors of highly concentrated mAb solutions.…”

“…Second Virial Coefficients A 2 Determined by SLS and Diffusion-Interaction Parameters k D from DLS Obtained for the Five mAbs by Schmitt et al41 …”