Predictive glycoengineering of biosimilars using a Markov chain glycosylation model

Abstract

Biosimilar drugs must closely resemble the pharmacological attributes of innovator products to ensure safety and efficacy to obtain regulatory approval. Glycosylation is one critical quality attribute that must be matched, but it is inherently difficult to control due to the complexity of its biogenesis. This usually implies that costly and time-consuming experimentation is required for clone identification and optimization of biosimilar glycosylation. Here, a computational method that utilizes a Markov model of glycosylation to predict optimal glycoengineering strategies to obtain a specific glycosylation profile with desired properties is described. The approach uses a genetic algorithm to find the required quantities to perturb glycosylation reaction rates that lead to the best possible match with a given glycosylation profile. Furthermore, the approach can be used to identify cell lines and clones that will require minimal intervention while achieving a glycoprofile that is most similar to the desired profile. Thus, this approach can facilitate biosimilar design by providing computational glycoengineering guidelines that can be generated with a minimal time and cost.

Keywords: Biosimilars; CHO cells; Erythropoietin; Glycoengineering; Markov model.

Figure 1

(A): Glycoprofiles on Rituximab, expressed in CHO-S cell line before any glycoengineering (“Initial”). The target profile shows the glycoprofile of the innovator drug as reported previously (Visser et al., 2013). (B)–(D): Results from the Markov-model based glycoengineering simulation (two representative runs shown) considering only FucT (B), FucT+GalT (C) or FucT+GalT+GnTI (D) reaction sets for perturbation. Profile mismatch evolution plots (left) show the worst, best and mean profile mismatches in each generation of the genetic algorithm. Perturbation plots (right) show the quantity of predicted perturbation yielding the best match in each generation. Glycoprofiles (far right) are calculated with the averages of perturbations throughout all independent simulation runs. Error bars are standard deviation. (E): Reaction network for the glycoprofiles on Rituximab and partial reaction network reconstructed for the glycoprofile on erythropoietin. Reactions are color-coded according to the glycosylation enzyme involved.

Figure 2

(A): Glycoprofiles on EPO, expressed in CHO-GS (Wildtype and knock-out data from (Yang et al., 2015)). The simulation infers the required perturbations in SiaT, GnTIV and GnTV that transform the wildtype profile into the observed knock-out profiles (one representative run shown). Error bars are standard deviation. (B): Glycoengineering of two EPO-producing CHO lines with varying degrees of branching and sialylation. The reaction sets dependent on the branching enzymes GnTIV & GnTV and sialyltransferase (SiaT) are subject to perturbation in simulations (one representative run shown). Glycoprofiles are calculated with the averages of perturbations throughout all independent simulation runs. Error bars are standard deviation. Brackets indicate the range of the predicted averages when optimal predicted perturbations are varied by 10% (Methods).