1 INTRODUCTION
Total sales of biologics in the United States reached
~$63.6 billion in 2012[1]. In all the class of biologics,
monoclonal antibody (mAbs) biopharmaceuticals still
maintained their ranking as the highest selling with an
18.3% growth over their 2011 sales. Glycosylation, a
process of post-translational modification on glycoprotein
of mAbs, involves the transfer of a core monosaccharide
from nucleotide sugar donors (NSDs) to the recipient
protein within the endoplasmic reticulum (ER) and Golgi
apparatus of the host cell. The terminal glycosylation of
N-linked glycan will affect the bioactivity and therapeutic
efficacy of the monoclonal antibody. For example, the
absence of core fucose can increase antibody-dependent
cell-mediated cytotoxicity (ADCC)[2,3]; terminal
sialylation can affect anti-inflammatory properties[4];
galactose terminating structures have a substantial effect
on complement -dependent cytotoxicity (CDC)[5]; high
mannose content and terminal n-acetylglucosamine
(GlcNAc) can have an effect on serum half-life[6,7].
FDA issued a Quality by Design (QbD) guidance for
pharmaceutical development in 2009[8], in which the
foundations are defining the critical quality attributes
(CQAs) and analyzing process characterizations. N-linked
glycosylation, a typical CQA, is investigated by
researchers to support the analysis and design of process
operation in silico. The statistical design of experiments
(DOE) methodology is used to analyze the effect of
metabolic factors or measurements on the cell-line specific
galactosylation of mAb N-glycans[9]. The high
dimensional process gain matrices are determined
This work is partially supported by National Natural Science
Foundation of China (61174128, 61403017, 61473025), Beijing Natural
Science Foundation (4132044) and Fundamental Research Funds for the
Central Universities of China (YS1404).
according to the DOE and the Analysis of Variance
(ANOVA) scheme to assess the controllability of various
glycan classes and of specific glycoforms[10]. Then the
appropriate manipulated variables are identified for
developing an effective and comprehensive strategy to
control glycosylation on-line [11].
In this study, a N-linked glycosylation process model
is modified from the traditional KB2005 model. KB2005
model[12] is presented by Kramberk and Betenbaugh to
obtain further detailed information on microheterogeneity.
This model increases to 7565 potential glycoforms arising
from 22,871 reaction networks with 11 enzymes. Then,
statistic process analysis is finished based on the modified
N-linked glycosylation model at the micro-scale. The
glycosylation process is an extremely high-dimensional,
very nonlinear and coupled system, so it is almost
impossible to analyze the process directly in the original
manipulation space. A high-dimensional process gain
matrix is obtained numerically via systematical statistical
analysis and design of experiment scheme. This process
gain can extract efficiently the main process
informationand quantitatively describe the functional
relationship between process input (factor) and process
output (response). Then the process gain decomposition
and the controllability analysis are performed, which help
the pharmaceutical manufacturer to investigate the
glycosylation process.
2 MATHEMATICAL MODEL OF
N-LINKED GLYCOSYLATION
The mathematical model can assist in identifying the
functional relationship between final production quality
and the material attributes and guiding control strategy
design. Complex reaction scheme and network structure
result that these mechanistic models must be exceedingly
high-dimensional, nonlinear and severe coupling. The
KB2005 model involves 11 Enzymes, 4 nucleotide sugar
Statistic Process Analysis of Monoclonal Antibodies N-linked Glycosylation
Jing Wang
1
, Daiwei Yang
1
, Jinglin Zhou
1
, Devesh Radhakrishnan
2
, Babatunde A Ogunnaike
2
1. Department of Automation, Beijing University of Chemical Technology, Beijing,100029, China
E-mail: jwang@mail.buct.edu.cn
2. Department of Chemical and Biomolecular Engineering, University of Delaware, Newark, Delaware,19716,USA
E-mail: {devesh & Ogunnaik}@udel.edu
Abstract: Glycosylation can affect the bioactivity and therapeutic efficacy of the monoclonal antibody. In this paper, a
simplified N-linked glycosylation model is given first, and a deep statistic process investigation is carried out at the
micro-scale. The enzyme concentrations in cell are considered as the manipulated variables. The output variable is the
glycan class distribution. The process gain matrix obtained based on statistically DOE and ANOVA strategy is used to
catch the data relationship between the enzyme concentrations and the glycan distribution. Singular value decomposition
of the process gain matrix can transform a high nonlinear system in the original space into a couple of SISO linear
systems in the new coordinate space. Then the controllability can be analyzed in the latent variable space easily.
Key Words: N-linked glycosylation; monoclonal antibodies; process gain matrix; statistic analysis.