Additionally, comparisons of the ranking of predicted viscosities per molecule from your high concentration at 180mg/mL did not align to the ultra-high concentration regime for any of the model equation fits (Table S3). higher protein aggregation risk, phase separation, and elevated answer viscosity.2Viscosity, a fluids resistance to circulation or rate of deformation, is mechanistically characterized by examining electroviscous effects, mAb molecular size, and surface potential distributions, with these factors determining the likelihood of proteinprotein interactions (PPIs).2,3 Developed from colloidal principles, the primary electroviscous effect explains the distortion of the electrical double-layer in the ultra-dilute regime with varying ionic strength. Changes in counterions and the hydration shell around mAbs impact their hydrodynamic volume and Brownian motion.4,5With the growing demand for ultra-high concentration mAbs, secondary electroviscous effects reduce mAb solubility while increasing their solution viscosity.2,6Increased crowding and decreased inter-particle distance increase the pair interaction potential.7The interaction potential is quantified by the second virial coefficient (B22) or diffusion interaction parameter (kD) that are correlated with solution-phase viscosity.8,9However, B22and kDdo not fully capture anisotropic interactions due to surface potential variations. Beyond pair-wise interactions, cluster formation from soluble mAb oligomerization drives elevated viscosity for high concentration mAb solutions. Small-angle X-ray scattering experiments and coarse-grained computational simulations have revealed mechanisms of mAb self-assembly and microstructure formation, directly correlating with increased viscosity.1012 The complexity of interactions contributing to high viscosity, along with manufacturability and injectability risks, has driven the development ofin silicosequence and structure-based models. Regression and clustering models1315have recognized correlations between molecular descriptors derived from three-dimensional homology constructs and high concentration mAb viscosity. To mitigate the risks of overfitting of small, non-diverse datasets, machine learning classification tools are used to categorize mAb viscosity risks.11,1618 Considering manufacturability, product quality, and injectability risks of highly viscous mAb formulations, a variety of mitigation strategies have been investigated to improve the likelihood of mAb translation to the clinic.19 Currently, there is a knowledge gap in selecting appropriate viscosity prediction models for high mAb solution viscosity, with no prior cross-comparisons in the ultra-high mAb concentration regime. This study comprehensively assesses viscosity fit and prediction models for nine anti-IL-8 mAbs (eight mutant variants and wild-type (WT)) previously manufactured, focusing on high- and ultra-high mAb concentration regimes. Using a combined computational and experimental approach, we compare the effectiveness of these models for triaging mAb developability. Our findings spotlight the necessity of measuring viscosity at dose relevant ultra-high concentrations and reveal the limitations of predictive models, low concentration hydrodynamic properties andin silicomolecular descriptors. == Materials and methods == In silicostructural modeling and generation of molecular descriptors was performed in Molecular Leukadherin 1 Operating Environment (MOE) software, version 2020.0901 (Chemical Computing Group, Montreal, Canada). == Homology constructs of anti-IL-8 Fv structures == Homology models19for nine anti-IL-8 antibody variable fragment (Fv) regions, including eight single-point mutants were constructed based on the anti-IL-8 IgG1 Fab domain name crystal structure (PDB: 505B). Using theAntibody modellerfeature (version 2020.0901) in MOE with default refinement and forcefield settings, we created homology constructs and introduced single-point mutationsviatheResidue Scanfeature. The same methodology was applied to construct homology models for four in-house mAbs. == In silicomolecular descriptors == We computed sequence and structure based physicochemical descriptors using theProtein Propertiestool andDescriptorsFeature inBioMOE(version 2021-11-18, Chemical Computing Group, Montreal, Canada). Viscosity-relevant parameters for input into predictive models are reported in Supplementary Table S1. == Aggregation propensity tools: Leukadherin 1 TANGO and WALTZ. == The TANGO20,21(http://tango.crg.es/tango.jsp.) and WALTZ22,23(https://waltz.switchlab.org/) sequence-based aggregation propensity tools were used to predict cross-beta-sheet formation in all anti-IL-8 IgGs examined. == DeepSCM == A convolutional neural network (https://github.com/Lailabcode/DeepSCM.) was used to assess charge distributions of the assumed Fv structure over molecular dynamic simulations.17,24All anti-IL-8 heavy and light chain variable sequences were inputted separately as FASTA files and the code was run in the terminal on a Linux system. == Viscosity prediction from Fv construct molecular descriptors == Three empirical models derived from the regression of viscosity data and molecular descriptors were used Rabbit Polyclonal to OR2T2 to directly Leukadherin 1 predict viscosity at either 150 mg/mL13or 180 mg/mL.9,14 Theviscosity modelby Liet al. uses the structure-based isoelectric point and WALTZ aggregation propensity Leukadherin 1 score, normalized by the number of amino acid residues to generate relative viscosity predictions.13 whereis the relative viscosity,the number.