High-Throughput Multi-Product Liquid Chromatography for Characterization of Monoclonal Antibodies

Published on: 
BioPharm International, BioPharm International-01-01-2011, Volume 24, Issue 1

An approach to biopharmaceutical development that combines Quality by Design with a suite of visual informatics tools to reduce scale-up risks.


Glycosylation can significantly influence the safety and efficacy profiles of biopharmaceuticals. Changes in product glycosylation during scale-up can lead to serious regulatory and commercial problems because of the risk of altered clinical performance. This article presents an approach to biopharmaceutical development and manufacturing that reduces these scale-up risks. It combines a Quality by Design (QbD) framework and a suite of visual informatics tools designed to aid the implementation of QbD for therapeutics with complex glycosylation.

Biopharmaceutical glycosylation is highly complex and a major source of within-batch heterogeneity and between-batch variability. Designers and manufacturers of biologics, therefore, must effectively measure and control glycosylation throughout the drug lifecycle, particularly during scale-up and after significant manufacturing changes.1

(Ludger Ltd.)


The possibility of nonuniform glycosylation during scale-up is inherent in the traditional approach to biopharmaceutical development, which typically is:

  • produce a drug product for early-stage development in a small-scale bioreactor (this becomes the reference material for subsequent comparability studies)

  • characterize the glycosylation of the reference material, invoking ICH Q6B and other relevant guidelines to direct the analyses

  • scale up in stages, optimizing cell -culture conditions and downstream processing to keep the scaled-up material similar to the reference material.

  • invoke ICH Q5E to compare the glycosylation material from large- and small-scale bioreactors. During this exercise, focus on the overall shape of oligosaccharide patterns by comparing the relative amounts of major glycan species.

There are three main risks of this approach. First, the cells of the expression system could behave differently in the scaled-up process, thus resulting in altered glycosylation—so the comparability exercise fails. The risk is the possible modification of safety or efficacy profiles. Second, the overall f pattern might look the same after scale-up but the safety or efficacy profiles do change significantly because of alterations in relative quantities of minor glycoforms with high bioactivity that were not considered significant in the glycan structure analyses (glycoforms are forms of the drug that share the same protein backbone but have different oligosaccharides). Third, the glycosylation could alter after scale-up, leading to modified clinical performance of the drug, but the structural changes may not have been picked up during glycoprofiling.

This article outlines a five-step approach to the realization of glycoprotein therapeutics that reduces scale-up risks. It combines a Quality by Design (QbD) framework for the development and manufacture of the drug and a suite of visual informatics tools designed to aid the implementation of QbD for drugs with complex glycosylation.


The first step in this approach is to establish a comparability system for glycosylation that takes into account the clinical performance of the drug's component glycoforms rather than just glycan primary structures. This is the heart of the QbD framework and is achieved by basing the measurement and control of glycosylation on glycosylation critical quality attributes (GCQAs), i.e., glycosylation parameters that affect patient safety and clinical efficacy. Table 1 shows potential GCQAs for a typical monoclonal antibody (MAb).


Table 1. Candidate glycosylation critical quality attributes (GCQAs) for a typical monoclonal antibody

Determining critical quality attributes (CQAs) is a crucial part of implementing QbD. ICH Q8 describes failure mode and effects analysis (FMEA) as a tool for determining the criticality of a drug's attributes. However, as discovered in the A-MAb project,2 there are two significant problems with applying FMEA to drug glycosylation. First, the traditional FMEA equation (risk priority number = severity × occurrence × detection) applies to the detection and occurrence of faults during the manufacturing process. However, the criticality of a therapeutic's structural attributes should depend on its effect on clinical performance, but neither on the influence of the manufacturing process on that attribute nor on the effectiveness of the analytical tools. Second, drug developers using FMEA generally recognize the problems with the standard RPN equation and will modify it, but there is no consensus on what factors should be in the new version and how those factors should be scored. In the case of A-MAb, a MAb for the treatment of non-Hodgkin's Lymphoma, the therapeutic effects of B cell killing occur primarily by antibody-dependent cell-mediated cytotoxicity (ADCC) and are influenced by the drug's Fc glycans. The A-MAb report details the use of two FMEA variants for assessing the criticality of various glycosylation parameters (galactose, afucosylation, sialylation, high mannose, and nonglycosylated heavy chain). The two tools scored criticality in different ways and gave roughly similar results for the risk ranking of the A-MAb glycosylation parameters but showed dissimilar risk levels. For example, tool #1 (where criticality = impact × uncertainty) categorized sialylation as high risk while tool #2 (criticality = severity × likelihood) scored it as moderate risk.

At Ludger, rather than using FMEA, we use two informatics tools developed for reliable GCQA determination and ranking: safety and efficacy (SE) profiling and impact maps. These tools are used iteratively to build up and refine GCQAs throughout the product lifecycle. They address factors embedded in the A-Mab FMEA tools described above, but deal with them in a different way.

GCQA Tool #1: SE Profiles

SE profiling is a tool for discovering potential GCQAs. It involves classifying the component glycoforms of a drug into different safety and efficacy (SE) categories and analyzing the structural relationships between glycoforms in those categories. For a simple classification, there are three safety categories: S (safe), N (not safe), and U (unknown safety level) and four efficacy categories: L (low efficacy), M (medium efficacy), H (high efficacy), and V (unknown efficacy). There are 12 combined SE categories for glycoforms. These include SL, SM, SH, and so forth. The glycoforms that we put into a SE profile could be produced by our candidate expression system. The parameters considered include all those mentioned in relevant regulatory guidelines such as ICH Q6B and the latest EMA guideline for MAbs.3 Figure 1 shows an example of an SE profile for glycosylation of a therapeutic MAb.

Figure 1. The annotations above and to the right of the matrix indicate the key glycosylation features of drug glycoforms occupying the corresponding matrix squares. Each square represents a safety–efficacy category. Potential safety and efficacy glycosylation critical quality attributes (GCQAs) are shown in the rows and columns respectively. A drug with optimized glycosylation is composed of glycoforms from the SM and SH categories (indicated by the squares with thickened borders).

The SE profile is like a chessboard where the position of the pieces (the glycoforms) denotes their therapeutic power. The GCQAs relate to transformations of structural form and the power found when moving from one square to another. The greater the increase or decrease in power for a move, the greater the effect of that GCQA on the clinical performance of the drug.

An SE profile reveals two distinct classes of GCQA: Type I GCQAs connect glycoforms in different SE categories (i.e., information indicates that they alter a drug's clinical performance) and Type U GCQAs are glycosylation parameters that place glycoforms into the U and V categories (i.e. whether or not they alter a drug's clinical performance is uncertain).

GCQA Tool # 2: Impact Maps

The GCQAs identified in SE profiles are further analyzed and prioritized using impact maps. These are mathematical graphs showing (a) the effects (i.e., the degree and type of influence) of a drug's glycosylation attributes on its biological behaviors, (b) the effects of those behaviors on the clinical safety and efficacy profiles, and (c) evidence for those effects. The effects are represented by lines linking attributes to behaviors and then to safety or efficacy, the degree of impact being denoted by line thickness. Figure 2 shows an impact map. The maps are annotated with references to evidence for the effects. These references are weighted with knowledge strength. This is an index composed of a veracity index (v1 = idea with no experimental evidence, v2 = expert opinion, v3 = peer reviewed data, v4 = data on the drug) and a relevance index (r1 = data for another molecule, r2 = data for a related molecule, r3 = data for your drug). The 12 combinations of v and r can be classified into five groups of knowledge strength from very weak (v1–r1) through weak, medium, strong, to very strong (v4–r3). The knowledge strength can be denoted by the color of impact lines to allow focus on the most reliable data. Interactions between glycosylation attributes or behaviors also can be denoted on impact maps as can evidence for lack of effects.

Figure 2. A partial impact map for monoclonal glycosylation. The diagram shows part of the impact map for core Fucα1,6 and bisecting GIcNAc residues of an antibody relying on ADCC activity for its therapeutic effect. The map shows the degree of effects of glycosylation features on ADCC activity and key evidence for the effects. For this drug, core Fucα1,6 is an efficacy GCQA as there is a critical path (high effects) from Fuc to ADCC to efficacy, but bisecting GIcNAc is not. Note that biGIcNAc could still be a GCQA if it had an impact on another biological behavior of the drug that affected safety or efficacy.

GCQAs are identified in the maps by critical paths with high effects linking glycosylation parameters by biological behavior to safety or efficacy. GCQAs are prioritized by ranking all behaviors affecting safety in descending order of the degree of impact, doing the same for behaviors affecting efficacy, and then ranking the GCQAs alongside those behaviors.

Impact maps are simple but powerful visual tools to reveal the evidence for assigning criticality to any glycosylation parameter. They should be updated as knowledge is acquired throughout the drug's lifecycle. Note that in this system, the criticality of a glycosylation attribute is independent of both its measurement and control during biomanufacturing. A critical attribute remains critical even when it is well controlled and detected, which are the goals of the next steps.


A flexible glycoprofiling system should be developed that is tuned to the drug's GCQAs (so they are measured reliably) and tunable to different stages of your drug's lifecycle. This tuning can be aided by measurement maps. These are extensions of impact maps and are obtained by listing all the glycoprofiling methods available and drawing measurement lines from each method to each GCQA that it could measure. Figure 3 shows an example of a measurement map.

Figure 3. A measurement map for nonhuman Galα1,3Gal in a monoclonal antibody. Galα1,3Gal has been identified as a GCQA from the drug safety–efficacy profile and impact maps. This measurement map shows the relative merits of various glycoprofiling methods for Galα1,3Gal for this drug. The thickness of each measurement line is proportional to the suitability of the method. The stars indicate the key methods chosen to measure Galα1,3Gal at different stages of the drug lifecycle.

The glycoprofiling methods considered should include your inhouse glycoanalyses, methods commonly used in the industry, and those outlined in the new USP Chapter <1084> on glycoprotein and glycan analysis and improved variants. In general, one needs analyses that separate and identify glycan or glycoform species. This is typically achieved using a combination of high-performance liquid chromatography (HPLC) or capillary electrophoresis (CE) and mass spectrometry (MS) plus sequencing grade exoglycosidases to analyze released oligosaccharides, glycopeptides, and intact glycoproteins.1,4,5

The reliability of measurements is denoted on the measurement map by the thickness of the measurement lines, and the ease of measurement for different stages of the drug lifecycle can be denoted by coloring or annotating the lines. Measurement reliability should be assessed according ICH Q2(R1) or similar guidelines, for analyses performed on a particular drug. Measurement maps to aid the selection of a set of glycoprofiling methods that provide either relative or absolute quantitation of the full range of GCQAs.

It is usually not feasible to measure all GCQAs at every step in a product's lifecycle. However, they should all be measured before and after product scale-up. Also, for effective comparability studies, companies should aim to identify all the major peaks in HPLC, CE, and MS glycoprofiles early on in drug development.


The next step in the QbD framework is to define the drug's quality target product profile (QTPP). For novel biopharmaceuticals, this should include optimized glycosylation to maximize the drug's clinical performance.

In terms of SE profiles, a completely optimized drug will have only SH (safe, high activity) and SM (safe, medium activity) glycoforms. Compared to the unoptimized therapeutic, such a drug would tend to be:

  • less sensitive to changes in its clinical performance following glycosylation changes typically found during scale-up

  • more potent, so requiring a smaller level of scale-up

  • less heterogeneous, being composed of fewer glycoforms

  • lower in cost per therapeutic dose.

To illustrate the importance of glycan optimization, consider a drug composed of two types of glycoforms, H and L (e.g., afucosyl and fucosylated MAb with ADCC as mode of action or highly sialylated and low sialylated gonadotropin), where the H form has 10-fold greater activity than the L. In an unoptimized drug with 88% of L form, 12% of H forms would contribute over 57% of the total drug activity. If the specification for the H forms was 12% ± 3% the drug activity could vary from 1.81 to 2.35 (~30% variation in total activity from low to high limits). In an optimized drug with 88% of H forms, those would contribute over 98% of the total drug activity. If the specification for the H forms was 88% ± 3%, the total drug activity could vary from 8.65 to 9.19 (~6% variation in total activity from low to high limits).

Glycan optimization starts with picking an expression system with the appropriate glycosylation machinery.6 For example, MAbs expressed in NS0 cells can contain glycosylation with potentially immunogenic nonhuman epitopes (Gala1,3Gal and NeuGc) as well as high levels of core fucosylation (so the dominant glycoforms fall into the undesirable N and L categories). Switching to Chinese hamster ovary (CHO) cells reduces the N and L category glycoforms at a stroke, thus resulting in a safer, more potent product. Further improvements could be made by using one of the growing number of alternative expression systems with improved or tunable glycosylation.


The glycosylation of biopharmaceuticals produced in mammalian expression systems is profoundly affected by the physicochemical, nutritional, and mechanical environment in the cell culture. This environment changes as bioreactors of different size and configuration are used during drug scale-up. A major cause of altered glycosylation in traditional scale-up schemes is that the large-scale tanks cannot be forced to behave like the small ones used to produce reference material. Scale-down systems overcome this problem by reversing the direction of mimicry and forcing small reactors to behave like the large bioreactors.7

In this step, scale-down bioreactors are used within the QbD framework to select a suitable clone, define the design and control spaces for drug manufacturing, and define and reproduce the cell culture conditions that will be used in the full-scale system and all intermediate-scale bioreactors.

There are some important things to note:

1. Clone selection should be performed with the glycoform pattern of the QTPP in mind. In particular, the selection criteria should include both drug titer and the quality of the GCQAs.

2. Sensitive glycoprofiling methods that measure the most highly ranked GCQAs are needed to support clone selection and experiments to determine the design space (DS).

3. There are many cell culture parameters that could affect glycosylation. These include dissolved oxygen, pH, temperature, ammonia, cell age, the extent of the harvesting period, nutrients, and the use of serum or serum-free media.8,9

4. The A-MAb project shows one approach for determining a DS that includes glycosylation parameters.2

5. Full factorial design of experiments (DoE) for optimizing the DS can be very expensive and time-consuming. This work could be reduced considerably by moving from conventional (e.g., 1 or 3 L) scale-down bioreactors to high-throughput micro-bioreactors that allow multifactorial experiments on hundreds of cultures in parallel.7

6. The results of the A-MAb project suggest that the early promises of expanded QbD DS, and therefore greater leeway for alterations in the product after manufacturing changes, may not be realized. What QbD does deliver, however, is a well-characterized manufacturing process for a well-characterized drug.

The prize for all this work should be a thorough understanding of how to manufacture your drug with consistent and optimized glycosylation at all scales.


Controlled scale-up, with maintenance of uniform drug glycosylation, should now be straightforward. From the QbD perspective, comparability after scale-up is demonstrated with evidence that the control space of your scaled-up process is in your predefined DS.

Scale-up to the full-scale bioreactor would typically be done in stages. At each stage, personnel would invoke ICH Q5E and Q8 (plus annex) to demonstrate comparability and focus on showing consistency with respect to the full set of GCQAs identified for the drug. If the drug has optimized glycosylation and is composed mainly of SM and SH glycoforms, then the analyses are generally straightforward as pesonnel will be searching for the high-abundance glycan species to show consistent efficacy. However, scientists still need glycoprofiling tools powerful enough to show the presence of low abundance nonsafe (N) glycosylation. Analyses of nonoptimized low-activity (L) glycoform dominated drugs need to be more rigorous because even small changes in the proportions of M and H type glycoforms can significantly alter the product's in vivo efficacy.

As part of continuous improvement and lifecycle management, the details of tracked GCQAs should continually be reviewed as new information emerges. This will involve updating SE profiles and impact maps. In particular, companies should aim to gain information to eliminate Type U GCQAs, re-assigning them to either Type I GCQAs or non-GCQA status.

Scientists also should update their set of glycoprofiling tools as new analytical instruments and glycoanalysis methods emerge. These measures will help mitigate the clinical and commercial risks of inconsistent drug glycosylation.


This article outlines a QbD-based strategy for maintaining comparability of glycosylation during biopharmaceutical scale-up. It also describes visual informatics tools that aid the implementation of QbD and simplify the analysis of complex information on drug glycosylation. This strategy is not suitable for all drug developers—it requires a serious consideration of glycosylation early on in biopharmaceutical development, an adoption of new models for drug realization, and the consequent commitment of scientific resources. However, the reward should be a therapy that has better clinical performance, is readily scalable, and that can be brought to market in a timely way.


I thank the following for interesting discussions on QbD, the A-MAb project, and drug glycosylation: Ron Taticek, PhD, Lynne Krummen, PhD, and Paul Motchnik, PhD, of Genentech, and Michael DeFelippis, PhD, and Bruce Meiklejohn, PhD, of Eli Lilly.

Daryl Fernandes is the chief executive officer of Ludger Ltd., Cuham Science Centre, Oxfordshire, UK, +44 1865 408 554, daryl.fernandes@ludger.com


1. Fernandes, DL. A QbD approach to biopharmaceutical glycosylation. In: Lyscom N, editor. Quality for biologics—critical quality attributes, process and change control, product variation, characterisation, and regulatory concerns. Hampshire, UK: Biopharm Knowledge Publishing; 2008 Dec.

2. International Society for Pharmaceutical Engineering. CMC Biotech Working Group. A-Mab: A case study in bioprocess development. An e-publication; 2009. Available from: www.ispe.org.

3. European Commission. EMEA guideline on development, production, characterisation and specifications for monoclonal antibodies and related products EMEA/CHMP/BWP/157653/2007. London, UK: 2008 Dec.

4. Domann PJ, Pardos-Pardos AC, Fernandes DL, Spencer DI, Radcliffe CM, Royle L, Dwek RA, Rudd PM. Separation-based glycoprofiling approaches using fluorescent labels. Proteomics. 2007;Suppl 1:70–76.

5. Fernandes DL, Spencer, DI. Glycoprotein analysis. In: Fraser-Reid BO, Tatsuta K, Thiem J, editors. Glycoscience—chemistry and chemical biology. New York, NY: Springer Berlin Heidelberg; 2008.

6. Van Berkel PH, Gerritsen J, Perdok G, Valbjørn J, Vink T, van de Winkel JG, Parren PW. N-linked glycosylation is an important parameter for optimal selection of cell lines producing biopharmaceutical human IgG. Biotechnol Prog. 2009;25:244–51.

7. Legmann R, Schreyer HB, Combs RG, McCormick EL, Russo AP, Rodgers ST. A predictive high-throughput scale-down model of monoclonal antibody production in CHO cells. Biotechnol Bioeng. 2009 Dec 15;104(6):1107–20.

8. Butler M. Animal cell cultures: Recent achievements and perspectives in the production of biopharmaceuticals. Appl Microbiol Biotechnol 2005;68:283–91.

9. Hossler P, Khattak SF, Li ZJ. Optimal and consistent protein glycosylation in mammalian cell culture. Glycobiol. 2009;19:936–49.