Classification model of amino acid sequences prone to aggregation of therapeutic proteins

Abstract

Background: Total body clearance of biological drugs is for the most part dependent on the receptor mechanisms (receptor mediated clearance) and the concentration of antibodies aimed at administered drug - anti-drug-antibodies (ADA). One of the significant factors that induces the increase of ADA level after drug administration could be the aggregates present in the finished product or formed in the organism. Numerous attempts have been made to identify the sequence fragments that could be responsible for forming the aggregates - aggregate prone regions (APR).

Purpose: The aim of this study was to find physiochemical parameters specific to APR that would differentiate APR from other sequences present in therapeutic proteins.

Methods: Two groups of amino acid sequences were used in the study. The first one was represented by the sequences separated from the therapeutic proteins (n = 84) able to form APR. A control set (CS) consisted of peptides that were chosen based on 22 tregitope sequences.

Results: Classification model and four classes (A, B, C, D) of sequences were finally presented. For model validation Cooper statistics was presented.

Conclusions: The study proposes a classification model of APR. This consists in a distinction of APR from sequences that do not form aggregates based on the differences in the value of physicochemical parameters. Significant share of electrostatic parameters in relation to classification model was indicated.

Keywords: ADA; Aggregates; Immunogenicity; In silico; Proteins.

Fig. 1

A relationship between hydrogen bond acceptors (HBA) and ionization potential (IP-[eV – electronovolts]). APR aggregation prone regions (○; n = 84); tregitope sequences (□; n = 22), CS control set extracted from tregitopes (●; n = 42)

Fig. 2

A separation of 4 classes (a, b, c, d) of dependencies and the relationship between aqueous solubility (QPlogS) and arithmetic expression value (AEx), Ln(AM − IP + AC × ROT) − (QPCaco − NON). APR aggregation prone regions (○; n = 84), CS control set extracted from tregitopes (●; n = 42), AC number of carboxylic acid groups, AM number of non-conjugated amine groups, IP ionization potential, ROT number of non-trivial (not CX3), non-hindered (not alkene, amide, small ring) rotatable bonds, QPCaco predicted apparent Caco-2 cell permeability, NON number of ring atoms not able to form conjugated aromatic systems