Biologicals 44 (2016) 291e305

Contents lists available at ScienceDirect

Biologicals
journal homepage: www.elsevier.com/locate/biologicals

Determination of critical quality attributes for monoclonal antibodies

using quality by design principles
Nadja Alt a, *, Taylor Y. Zhang d, Paul Motchnik e, Ron Taticek d, Valerie Quarmby f,
Tilman Schlothauer b, Hermann Beck c, Thomas Emrich b, Reed J. Harris d
a
Pharma Technical Development, Roche Diagnostics GmbH, Nonnenwald 2, 82377 Penzberg, Germany
b
Pharma Research and Early Development, Roche Innovation Center Munich, Roche Diagnostics GmbH, Nonnenwald 2, 82377 Penzberg, Germany
c
Pharma Technical Development Biotech Europe, F. Hoffmann-La Roche Ltd, 4070 Basel, Switzerland
d
Pharma Technical Development, Genentech, South San Francisco, CA 94080, USA
e
Biologics Quality Control, Genentech, South San Francisco, CA 94080, USA
f
Research and Early Development, Genentech, South San Francisco, CA 94080 USA

a r t i c l e i n f o a b s t r a c t

Article history: Quality by design (QbD) is a global regulatory initiative with the goal of enhancing pharmaceutical
Received 9 June 2016 development through the proactive design of pharmaceutical manufacturing process and controls to
Accepted 10 June 2016 consistently deliver the intended performance of the product. The principles of pharmaceutical devel-
Available online 25 July 2016
opment relevant to QbD are described in the ICH guidance documents (ICHQ8-11). An integrated set of
risk assessments and their related elements developed at Roche/Genentech were designed to provide an
Keywords:
overview of product and process knowledge for the production of a recombinant monoclonal antibody.
Quality by design
This chapter describes the identification of critical quality attributes (CQAs) as an important first step for
Critical quality attribute
Risk ranking and filtering
QbD development of biopharmaceuticals. A systematic scientific based risk ranking and filtering
approach allows a thorough understanding of quality attributes and an assignment of criticality for their
impact on drug safety and efficacy. To illustrate the application of the approach and tools, a few examples
from monoclonal antibodies are shown. The identification of CQAs is a continuous process and will
further drive the structure and function characterization of therapeutic proteins.
© 2016 International Alliance for Biological Standardization. Published by Elsevier Ltd. All rights reserved.

1. Introduction
The identification of critical quality attributes (CQAs) is an
important step in the development of biopharmaceuticals that
depends on a thorough understanding of the potential for quality
attributes to affect safety and efficacy. The assessment of CQAs is
dependent on the definition of the necessary product attributes for
the expected product performance as defined by a quality target
product profile (QTPP) and taking other sources of information into
consideration (Fig. 1). It is further related to process characteriza-
tion since all CQAs identified need to be assessed for their vari-
ability within the manufacturing process in order to define their
acceptance criteria and a sound control strategy.
This chapter describes the risk ranking and filtering approach
developed by Roche/Genentech to assess CQAs for monoclonal
* Corresponding author.
E-mail address: nadja.alt@roche.com (N. Alt).
antibody (mAb) products. This approach has been used in several
development and licensing stage submissions, and has been refined
since its incorporation in the A-Mab case study [1]. Our method-
ology considers the impacts of product attributes on bioactivity,
pharmacokinetics, immunogenicity risk and safety. An assessment
of uncertainty regarding the basis for the impact severity scoring is
also applied to determine overall attribute criticality. The CQA
assessment tool as developed sets a high bar for declaration of
quality attributes as non-CQAs since non-criticality of a QA needs to
be actively proven by experimental and clinical data of this or a
related molecule or can only be assigned upon well accepted sci-
entific literature. Cut-off criteria for impact ranges were established
conservatively as described in detail within this chapter.
This risk ranking and filtering approach can be applied to
identify potential CQAs early in product development, enabling
refinement during the development period to support a suitable
control strategy at the licensing stage. It requires focusing on po-
tential patient impacts, without regard for process capability or

http://dx.doi.org/10.1016/j.biologicals.2016.06.005
1045-1056/© 2016 International Alliance for Biological Standardization. Published by Elsevier Ltd. All rights reserved.
292 N. Alt et al. / Biologicals 44 (2016) 291e305

Fig. 1. Quality by design risk assessment tools assess product quality attributes for criticality.

product history, consistent with the definition provided in ICH function studies. In addition, general product-specific clinical and
Q8(R1) [2]. non-clinical information is gained during product development.
The criticality of pCQAs is assessed using a risk ranking and
2. Critical quality attributes for monoclonal antibodies filtering approach with respect to impact bioactivity, pharmacoki-
netics (PK)/pharmacodynamics (PD), immunogenicity and safety.
A critical quality attribute is defined in the Quality by Design ICH Product specific information gained from clinical or non-clinical
Q8(R2) guidance document [2] as a physical, chemical, biological or observations, analytical and biological characterization, studies
microbiological property or characteristic that should be within an with related molecules, general (platform) antibody knowledge
appropriate limit, range, or distribution to ensure the desired prod- and published literature are used to complete this assessment.
uct quality. This definition refers only to potential patient impacts, The identification of pCQAs begins early in development and
which allows us to remove process capability from the assessment. evolves as more product understanding is gained [3]. Early identi-
In traditional approaches to pharmaceutical development, fication of pCQAs helps to focus development efforts on those at-
quality attributes were assigned to be product-related or process- tributes where more understanding or control may be needed. The
related substances or impurities, and used to determine specifica- final CQAs are generally confirmed at the later stages of commercial
tions (i.e., identify, strength, and purity) that would maintain the process development in anticipation of the finalization of the
analytical profile of the clinical batches. The robustness of the commercial control strategy. This systematic approach for identifi-
process to deliver the specified product was not the central feature cation of CQAs involves a greater analytical effort, but pays off with a
of process design. The more scientific, systematic and compre- deeper understanding of the structure-function relationship of
hensive QbD approach seeks to identify product quality attributes quality attributes and their impacts on product efficacy and safety.
that are critical for the safety and efficacy of the product, thereby
building product quality control into the manufacturing process 2.1. Definition of a QTPP
and pharmaceutical dosage form. The QbD tools described below
provide design itself, making product testing a risk management The quality profile of biopharmaceutical products, including
activity. This is achieved by a comprehensive and systematic mAbs, is defined by their Quality Target Product Profiles (QTPP)
assessment of product variants, process-related impurities, product
composition and strength, appearance and microbiological attri-
butes, as well as excipient and raw materials for their criticality.
The pathway towards a final list of CQAs for a mAb is outlined in
Fig. 2 and involves several assessment steps starting with the
definition of a Quality Target Product Profile (QTPP) which lists the
necessary product attributes required to deliver the expected
product performance as defined by the target product profile (TPP).
The QTPP and additional sources of product quality attribute in-
formation are the basis for assembling a list of all potential critical
quality attributes (pCQAs).
Quality attributes can be classified in different categories. The list
of pCQAs generally comprises product variants and process-related
impurities that may potentially impact the molecule's mechanisms
of action/mechanisms of toxicity (MoAs/MoTs), or general safety or
pharmacokinetic properties. These two categories will be assessed
using the risk ranking and filtering approach for CQA identification.
While a relationship of each pCQA to clinical outcomes cannot be
directly studied, potential patient impact can be inferred from
analytical and biological characterization, including structure- Fig. 2. Outline for workflow for CQA identification.
 N. Alt et al. / Biologicals 44 (2016) 291e305 293

representing a “prospective summary of the quality characteristics
of a drug product that ideally will be achieved to deliver the desired
quality, ensuring safety and efficacy of the drug product” [2]. The
QTPP is derived from the expected product positioning and per-
formance as well as critical attributes thereof as described in the
Target Product Profile (TPP) for products under development. In
addition, intrinsic pharmaceutical properties, scientific knowledge
and legal requirements are considered. Table 1 displays an example
QTPP listing information on indication and MoA(s), dosage and
administration, drug product requirements and product quality
requirements. The definition of the desired MoA of the molecule
represents a key content in the QTPP since the MoA(s) pertain(s) to
CQA assessment. In case a relevant MoA or an undesired MoT is not
considered in the assessment, the derived CQA list may be
incomplete and eventually propagate into undetected critical pro-
cess parameters (CPPs) without suitable process characterization/
validation (PC/PV) studies, and an incomplete control strategy.
The QTPP forms the basis of design for the development of the
product. It should be established during early development of a
product and revisited at key stages of product development, i.e.,
upon changes implemented to the TPP and before starting a CQA
assessment, with any changes approved by the appropriate
governance.
2.2. Initial assessment of quality attributes
A list of all potential critical quality attributes (pCQAs) needs to
be compiled (Fig. 1). According to ICH Q8R2, pCQAs are derived
from the QTTP by linking the expected quality profile to specific
attributes of the product. For example, if Fc effector function is
involved in the MoA, certain Fc glycosylation variants may need to
be considered as pCQAs. To be included on the pCQA list, a variant
may be identified in forced degradation or specific modification
studies, during method characterization or isolated from manu-
factured materials. Additional sources of information on pCQAs
include platform knowledge and scientific literature. Intrinsic
molecule properties such as the complementary determining re-
gions (CDRs) that directly determine antigen binding properties
need to be considered.
Certain attributes are determined as obligatory CQAs per stan-
dard expectations for recombinant protein therapeutics or com-
pendial requirements due to their high criticality for product
safety and/or efficacy and regulatory requirements. While some
obligatory CQAs would certainly be identified by the current tools,
others reflect health authority expectations for certain attributes or
tests on release testing. For example, adventitious agents and
contaminants from bacteria or other microbes are critical with
respect to safety and would be ranked as CQAs by the RFF tool.
Others, such as appearance parameters may not be identified crit-
ical by the CQA assessment tool unless a safety concern may be
linked, but are requirements by regulators.
Product variants and process-related impurities represent two
categories that will be assessed on a product-specific basis to build
the pCQA list.
3. Categorization of quality attributes
The approach to identification of pCQAs for a mAb is dependent
on the category of the quality attribute (QA) being assessed. Attri-
butes are divided into the following assessment categories:

Product variants

Process-related impurities

Obligatory CQAs

Raw materials and leachable compounds
Dividing attributes into these categories enables identification
of QAs that are product- or process-specific from those that are
common to platform mAb processes.
3.1. Product variants and process-related impurities
Product variants and process-related impurities are assessed for
their criticality on a product-specific basis to account for the unique
modifications, MoAs indication and route of administration. Prod-
uct variants formed at amino acid residues in or close to antigen- or
receptor-binding sites, glycan structures, and variants at the poly-
peptide termini may be considered for bioactivity and PK assess-
ment. Variants formed on all antibody sites need to be considered
for the assessment of immunogenicity and safety impact.
Quality attributes may include size-related variants, charge-
related variants, oxidation-related variants, glycosylation, struc-
tural variants and process-related impurities. Table 2 shows an
exemplary list of molecular variant pCQAs for a monoclonal anti-
body. Additional molecule-specific quality attributes may need to
be considered. Oxidation variants are observed to predominantly

Table 1
Example QTPP for a monoclonal antibody.

Attribute Target

Indication Non-Hodgkin's Lymphoma (indolent & aggressive NHL)

Chronic Lymphocytic Leukemia (CLL)
Diffuse Large B-Cell Lymphoma (DLBCL)
Mechanism of Action B-cell depletion:
- antibody-dependent cellular cytotoxicity (ADCC)
- antibody-dependent cellular phagocytosis (ADCP)
- direct cell death induction (apoptosis-like)
Critical Features Impacting MoA Type II CD20 binding, ADCC activation
Dosage Form Sterile, preservative-free liquid for infusion
Dosage Strength 1000 mg per vial, 40 mL at 25 mg/mL
Mode of Administration Intravenous, diluted with isotonic saline, max. 1000 mg/h
Drug Product Primary Container 50 mL type 1 borosilicate glass vials, fluoro-resin laminated stopper
Drug Product Shelf-Life Minimal claim at submission 30 months (target) at 2e8  C)
Compatibility with Application Devices Compatibility with intravenous bags and application lines in concentrations
and Stability during Administration of 0.4e20 mg/mL and at infusion speed  4 mL/h without requirement of inline
filter. Stable solution for 24 h at room temperature
Drug Product Quality Requirements Meets pharmacopoeial requirements for parenteral dosage forms (PhEur, USP, JP)
Degradants and Impurities Acceptable patient risk due to process-related and product-related impurities in relation to the benefit
294 N. Alt et al. / Biologicals 44 (2016) 291e305

Table 2
List of molecular variant pCQAs for a monoclonal antibody.

Category Quality attributea

Size-related Variants High Molecular Weight Species (HMWS)

Low Molecular Weight Species (LMWS)
Charge-related Variants (Acidic) Deamidation in CDR
Deamidation in Non-CDR
Glycation in CDR
Glycation in Non-CDR
Charge-related Variants (Basic) Aspartic Acid Isomerization in CDR
Aspartic Acid Isomerization in Non-CDR
N-Terminal Leader Sequence (may be molecule specific)
N-Terminal Pyroglutamic Acid
C-Terminal Lysine
C-Terminal Proline (IgG1) or Leu (IgG4) Amidation
Oxidation-related Variants Oxidation in CDR (Met, Trp)
Oxidation in Non-CDR (Met, homo-variant)
Oxidation in Non-CDR (Met, hetero-variant)
Fc Glycosylation Afucosylation
Galactosylation
High-Mannose
Sialylation (NANA, NGNA)
Non-Glycosylated Heavy Chain
Structural Variants Cysteine Forms
Sequence Variants
Protein Structure
a
Certain low abundance variants may need to be added to the list of general known variants such as advanced glycation
end-products, hydroxylysine, or oxidative carbonylation.

occur on specific methionine and/or tryptophan residues, whereas 3.2.2. Adventitious agent controls
deamidation and isomerization usually occurs on asparagine or Adventitious agents and contaminants and impurities from
aspartate residues. For a glyco-engineered antibody additional bacteria or other microbes are also classified as obligatory CQAs.
glycosylation variants such as bisecting N-acetylglucosamine These include:
(bGlcNAc) or hybrid glycans may become relevant pCQAs to be
included in the pCQA list [4] For an antibody containing Fab
Viruses
glycosylation, the respective glycan structures need to be consid-
Microbiological impurities (Bacteria, Mycoplasma)
ered separately since their criticality may differ from Fc glycosyla-
Bacterial endotoxins
tion. Cysteine forms represents a quality attribute subgroup that
should be assessed for free sulfhydryl variants as well as unex-
pected disulfide linkages and disulfide bond modifications such as
3.2.3. Drug product-specific QAs
trisulfides, thioethers (lanthionine) and cysteinylation.
Obligatory CQAs for a liquid drug product in a vial may include:

Subvisible Particles
3.2. Obligatory CQAs
Visible Particles

Extractable Volume
Due to their high criticality to product efficacy and safety or
Sterility
regulatory requirements, certain DS and DP quality attributes such
as composition and strength, drug product appearance, and Other attributes may need to be considered for freeze-dried
freedom from adventitious agents are classified as obligatory CQAs products, for combination products with delivery devices, or for
by default. These CQAs are therefore not assessed with the risk non-parenteral delivery.
ranking and filtering approach described in this chapter. The
highest criticality impact score (20) is assigned.
3.3. Raw materials

Raw materials used in the drug substance manufacturing pro-

3.2.1. Composition and strength cess are evaluated for toxicity by considering a theoretical esti-
The following DS/DP composition and strength related QAs are mated daily intake (EDI), compared to a threshold of toxicological
classified as obligatory CQAs because of their impact on product concern (TTC), provided by a toxicologist. This EDI value is derived
efficacy, stability and safety: by summing up the complete amount of the respective raw mate-
rial that is introduced into the process, divided by the minimum

Protein Content yield. This approach is very conservative because it does not take

Osmolality into account any removal of a raw material in unit operations after

pH the ones in which they are introduced into the process. The ob-

Appearance (Color, Opalescence, Clarity) tained value is multiplied with the maximum dose for the product

Buffer Content to obtain the final EDI value.

Excipient Content Clearance is typically established by testing for raw materials

Surfactant Content whose estimated daily intake without clearance is above the TTC.
 N. Alt et al. / Biologicals 44 (2016) 291e305
Process platform knowledge regarding clearance may be leveraged
for common raw materials used in a platform mAb process,.
Excipients are not assessed using the EDI/TTC approach since
safety information may be available through experience with
marketed biopharmaceutical products or gained through the clin-
ical trials with a new product.
3.4. Leachable compounds
The approach for identification of leachables as CQAs is
dependent on whether a specific compound can be detected in the
final DS or DP. If a specific leachable is shown to exceed acceptable
and safe levels, e.g. as determined based on ICH M7, that compound
is designated as a CQA.
4. Assessing criticality of quality attributes
The criticality assessment for QAs is a team effort involving
various technical, clinical and non-clinical functions with expertise
in bioactivity, PK and safety/immunogenicity assessments as well
as a QA representative. Documentation of the thought process of
building rationales using product experience and literature that
leads to criticality risk scoring is required by Health Authorities.
pCQAs are assessed for their criticality using a risk ranking and
filtering tool (RRF). This allows the assessment of the impact of each
attribute on patient safety and product efficacy.
4.1. Rationale for a risk assessment approach
Risk assessment is used to identify CQAs because it systemati-
cally guides the identification of hazards and the evaluation of risks
to patient safety and product quality associated with exposure to
those hazards. Several risk assessment tools described in ICH
guidance Q9 were evaluated as potential approaches to identifying
CQAs. Our initial QbD efforts used FMEA tools to try to assign CQAs,
but these assessments were confounded by process capability,
stability and control considerations.
A key aspect of our strategy is that CQAs are product attributes
with the potential to affect quality, regardless of the capability of
the manufacturing process to control them. The criticality of a QA
on bioactivity and PK may be assessed independently of actual
levels present in the reference standard or clinical trial materials for
a broader spectrum of attribute levels using non-clinical models
and analytical tools.
On the other hand, the criticality assessment of a QA with
respect to safety and immunogenicity cannot be performed inde-
pendently of actual levels present. Where the impact relies on
clinical data we are limited to using material within the capability
of the process for part of the assessment. However, this data is
considered in a broader context considering further knowledge and
general concerns about a variant or impurity taken from literature
and experience with other antibody products to conclude whether
there are concerns with a variant or impurity at an elevated level.
Clinical experience with quality attributes such as aggregates or
variants is highly relevant for assessing immunogenicity and safety
risks, but this should not be confused with maintaining separation
between attribute criticality and process capability, which refers to
the ability of the process to maintain certain quality attributes at
suitable levels.
Because of limitation in available knowledge, the CQA-ACs for
potentially safety or immunogenicityeimpacting attributes are
generally set closer to clinical experience than for bioactivity or
PKeimpacting attributes.
Converting the ICH Q8 R2 definition of CQAs to a practical tool
that could apply to a range of proteins required defining impact
295
categories and risk scoring. The central challenge in designing
this RRF was to reasonably divide practical situations into
different categories that best reflect the levels of risk and con-
cerns that product safety or efficacy would be impacted. Several
examples of known high- and low-risk quality attributes were
assessed using the tool in order to test the scoring and outcomes
for logic of the categorization. An overall cut-off level of the risk
for categorization of quality attributes as critical or non-critical
was defined at a conservative level assuring that moderate to
very high impact assessments will result in ranking as a CQA,
regardless of the level of uncertainty. The uncertainty risk
scoring was set to ensure that an attribute that initially appears
to have a low probability of impact will be further evaluated as a
pCQA if the data to support that initial conclusion has high
uncertainty.
The approach presented in this chapter, including the four
impact parameters (bioactivity, PK/PD, immunogenicity and safety)
and an uncertainty factor, was reviewed at several conferences
during 2008e2009, and was later incorporated into the A-Mab case
study as Quality Attribute Assessment Tool #1 [1]. With some re-
finements, this approach was used to define CQAs for marketing
applications of two monoclonal antibodies.
4.2. Use of platform knowledge and/or published literature
The impact of the QA on bioactivity, PK or safety and immu-
nogenicity derived from literature is usually assigned with a high
uncertainty score since no data with the specific molecule is
available. Literature may be leveraged for the assessment of
biological and PK impact when it is clear and well established
that no impact comes from a certain variant or impurity. The
assessment then can result in a non-CQA. For the assessment of
safety and immunogenicity impacts the use of literature from
molecules with other indication, mechanism of action, patient
population or administration/dosage may be possible but needs
justification.
For mAbs, experience has been gained throughout the last three
decades on the formation and relevance of certain variants of these
molecules, particularly in the conserved regions of mAbs. For
example, the CQA ranking of C-terminal lysine heterogeneity can be
performed based on external literature only since evidence has
been gathered by several sources that this Lys residue is rapidly
removed in vivo by endogenous carboxypeptidase activity in pa-
tients when administered intravenously and, therefore, it does not
impact efficacy and safety as described below in part 6.3. For these
reasons, the classification of C-terminal lysine as non-critical
attribute for mAbs administered intravenously can be made with
sufficient confidence. Structure-function studies reported in liter-
ature for a specific molecule may in some cases be applied for the
assessment of related molecules and can reduce the uncertainty
reflected in the RRF.
Broader experience has also been gained for mAb products
regarding safety of some impurities, since biopharmaceuticals
usually contain residual impurity levels derived from the host cell
or process. Whether information and clinical experience can be
transferred from previous biopharmaceutical products, however,
has to be carefully justified considering the production process,
exposure levels, mechanism of action, indication, route of admin-
istration, and patient population. Endogenous antibodies charac-
terization literature would also be considered to justify the
potential low risk of immunogenicity and safety impacts for attri-
butes (such as Fc glycation and deamidation) present both in
endogenous antibodies and therapeutic antibodies.
296 N. Alt et al. / Biologicals 44 (2016) 291e305
4.3. Mechanisms of action: selection of relevant bioassays
Detailed understanding of the possible contributions of
different mechanisms to the in vivo bioactivity of each molecule is
essential for selecting the appropriate analytical assay strategy for
testing the impact of QAs on bioactivity. Besides the primary MoA
(e.g., signaling disruption by receptor binding) additional mecha-
nisms can potentially contribute to the overall bioactivity. For
cytotoxic mAbs, such mechanisms may include but are not
restricted to Fc effector functions. The extent to which the different
mechanisms contribute to bioactivity is specific for each molecule.
The in vivo relevance of different mechanisms should be assessed
based on pre-clinical profiling and, if available, by clinical data to
decide which mechanisms need to be considered within a CQA
assessment for bioactivity impact. In-house as well as published
data should be taken into account. Direct information on the clin-
ical relevancy of potential mechanism of actions is usually not
obtainable, so any plausible mechanism that cannot be ruled out
must be evaluated. The resulting assessment of the molecule's
bioactivity profile and the functional domains involved should be
scientifically sound and robust. Consistency with pre-clinical
pharmacology sections of submission regulatory documents is
essential.
The primary MoA is usually reflected by the potency assay used
for release and stability testing. This assay should be part of the
bioactivity assay panel for CQA assessment. If additional mecha-
nisms are deemed to be relevant, the assay strategy for CQA RRF
depends on the functional domains involved and on the compari-
son of product variants' behavior in the assays reflecting these
mechanisms.
If an additional relevant mechanism relies on the same func-
tional domain as the primary MoA, and activity of relevant product
variants is affected similarly in both the primary and the additional
mechanism, the potency assay can be used as a surrogate for both
mechanisms. Such an approach should be justified by correlation
data as appropriate. If comparison of relevant product variants in
primary and additional mechanism exhibit differential behavior, an
assay reflecting the additional mechanism should be included in
the CQA assessment. If an additional mechanism involves func-
tional domains other than those involved in the primary MoA, an
appropriate additional assay should be included in the CQA
assessment. For assays covering Fc function the suitability for their
intended use needs to be demonstrated. A formal qualification or
validation of those assays is not considered as required.
Potency measured by the bioassay is a quality characteristic of
the molecule under consideration, but not considered a CQA itself.
Potency can be defined as a quantitative measure of relevant bio-
logic function based on the integrated presence and impact of
specific quality attributes that are linked to relevant biologic
properties. Therefore, measurement of potency may be used to
determine the criticality of an attribute with respect to impact on
bioactivity.
4.4. Analytical and biological characterization of pCQAs
For certain attributes on the pCQA list, analytical and biological
characterization studies may be performed to assess the structure-
function relationship between product variants and potential
impact on bioactivity and PK (Fig. 1). For these studies, product
variants need to be available either by isolation from product
comprising a mixture of variants or generated by applying specific
stress conditions or enzymatic treatment to create enriched mix-
tures. The product-specific data allow assessment of criticality with
a higher confidence as compared to assessments based on infor-
mation derived from other molecules or literature.
4.4.1. Generation and isolation of variants for CQA assessment
Product variants may be generated by applying specific stress
conditions [5]. Alternatively, variants that may be enriched under
stress conditions may be isolated from representative materials by
chromatographic separation. Glycosylation variants may be ob-
tained by manipulating cell culture conditions or by in vitro
modification [6,7] with mock controls available for bioactivity
testing. In some cases the impact may also be derived from data of
development batches, i.e. in the case of afucosylation correlating
with ADCC bioactivity.
For experimental structure-function studies of product variants,
the highest possible level of a variant in a sample to be studied
should be generated without changing levels of other variants. The
criticality ranking of a CQA based on bioanalytical or PK data re-
quires a minimum level of 50% of an enriched variant. Assuming a
high impact CQA with a bioactivity of <59% or >140%, the variant
has to be present in a mixture with at least 50% relative abundance
to shift the overall activity <80% or >120% (50% þ 50% x
0.59 ¼ 79.5%) enough to result in a confident ranking of the variant
as a non-CQA. The specific levels for ranking will be explained later
in this chapter (refer to Section 4.5).
In some cases a variant cannot be isolated or generated at levels
above 50%. In those instances, multiple measurements of samples
containing a range of values may allow a correlation of the pCQA
level and function. An extrapolation of the impact of lower levels of
a variant to higher levels has to be thoroughly justified, and the
assay precision and statistical confidence of the values obtained
have to be considered.
4.4.2. Homo- and hetero-modified variants
Some proteins have multiple identical subunits, and the modi-
fication of one or more polypeptide chains may impact criticality.
Standard mAbs consist of two heavy chains and two light chains,
both identical with respect to their amino acid sequence. For bis-
pecific monoclonal antibodies the two Fab arms usually differ with
respect to amino acid sequences and their complementarity
determining regions (CDRs). Depending on the mechanism of ac-
tion for standard antibodies, binding of only one HC/LC arm may be
required (monovalent binding); in other cases binding of both arms
(bivalent binding) is required. Also with respect to interactions of
the heavy chain Fc region, the homo-modified Fc variant (one heavy
chain modified e for example by a specific glycoform) may differ in
the impact on function compared to the hetero-modified variant
(both HCs modified).
Within CQA assessment studies the potential mono- vs. bivalent
interactions with regard to receptor binding should be considered.
For instance it has been shown that only Met252 homo-oxidized
variants (the Met252 residues on both HCs are oxidized) have
reduced PK as evidenced by reduced area under the curve (AUC)
and FcRn binding [8,9]. Therefore, only the homo-oxidized variant
is considered as a CQA whereas the hetero-oxidized variant is
categorized a non-CQA. In contrast, it might be the case that for
effective target binding and function, bivalent target binding needs
to occur, such as for trastuzumab's antiproliferative activity, where
a CDR isoaspartyl modification of one heavy chain is sufficient to
reduce antiproliferation [10]. Afucosylated glycans may only need
to be present on one heavy chain to enhance ADCC activity [11], but
other studies demonstrate that Fcg effector function is clearly
reduced also for a hetero-glycosylated antibody [12]. Therefore, it
has to be considered whether only the content of the hetero-
modified variant may need to be considered or the homo-
modified variant in case bivalent binding is crucial when
deducing the impact on activity or PK.
If each of the individual homodimer or heterodimer species
cannot be measured directly, the variants are often identified based
 N. Alt et al. / Biologicals 44 (2016) 291e305 297

on peptide map analysis and the binomial distribution can be used As a practical and relatively robust method, IE-HPLC has some
to estimate the levels of each provided that independence can be clear advantages avoiding the deamidation and isomerization ar-
justified. The RRF separately evaluates the potential impact if the tifacts caused by sample preparation in peptide mapping methods.
variant is present on one site (hetero-modified) or both sites The acidic region of IE-HPLC has been used as an effective surrogate
(homo-modified) of the symmetrical antibody. to assess process impact on deamidation in PC/PV studies. A full
characterization of the acidic region and understanding of the
4.4.3. Variants present at low abundance presence of only common acidic variants or additional novel spe-
Improvements in analytical techniques such as mass spec- cies is a prerequisite for consideration to use IE-HPLC as a regional
trometry have been shown to allow for the detection and analysis assay for acidic variants. The lack of specificity of IE-HPLC method
of molecular variants and impurities at increasingly low levels: can be mitigated by LC-MS analysis by spot-checking selected
down to the ppm level and below. Recently, this has resulted in samples for acidic region to confirm the general trend and extreme
identification of new quality attributes in biopharmaceuticals, such cases.
as oxidative carbonylation [13] and advanced glycation end-
products [14,15] usually at very low levels. In many cases it may
4.5. CQA identification for product variants and process-related
not be possible for these low level variants to be either isolated or
impurities
generated at sufficient levels to perform a CQA assessment. In
addition, many of these variants at low levels naturally occur on a
Following analytical and biological characterization the
patient's endogenous proteins. The assessment of these low
product-specific information on structure-function relationships is
abundance variants may, therefore, be aligned to the analysis and
subjected to the RRF for categorizing CQAs (Fig. 1). In addition,
levels regarded as relevant for naturally occurring sequence vari-
clinical and non-clinical observations as well as external informa-
ants. Mistranslation frequencies resulting in sequence variants are
tion from related molecules and literature will be used for assessing
reported in the range from 105 up to 103 [16e19]. Considering a
the criticality of pCQAs.
translational error rate of 105 for a monoclonal antibody
The risk ranking approach incorporates two factors: impact and
comprising e.g. 663 amino acids, at least one amino acid substitu-
the uncertainty of that impact (Fig. 3). Impact is the known or
tion will be found at levels in a detectable range of 0.1e1%. A gen-
potential effect an attribute may have on patient safety and product
eral threshold as considered for reliable quantification of sequence
efficacy (the effects on safety and efficacy together constitute
variants, therefore, may be reasonable which determines whether a 00 harm00 ). Uncertainty is related to the relative degree of confidence
quality attribute has to be included in a pCQA list. However, the
that the impact is correctly assigned for the QA of interest. The
threshold may be lower for sequence variants where structural or
impact and uncertainty rankings have different scales (2e20 for
in silico analysis, or precedent data indicate potential general safety
impact and 1e7 for uncertainty) to reflect the relative importance
concerns or concerns related to mechanism of action (i.e., potential
of the two factors, with impact outweighing uncertainty. Numerical
low level antagonist activity). As for sequence variants, selected
values are assigned to impact and uncertainty and multiplied to
sample testing should be performed to confirm the variants are
generate a relative risk score, which is used in ranking. Filters, in
present consistently only at very low levels.
the form of cut-offs for risk scores, are then used to identify attri-
butes that are high risk (classified as CQAs) and low risk (classified
4.4.4. Assays for quantification of pCQAs
as non-CQAs).
Different functional regions within a molecule require a site-
specific assessment of CQAs in order to determine whether a
pCQA is located in or close enough to affect a functional site. After 4.5.1. CQA impact score and categories
site-specific identification of a variant is completed often using We needed to define general terms such as “low impact” or
LCMS peptide mapping, further structure-function studies may “high impact” in a way that would allow our expert teams to assess
involve analytical methods that sum up multiple potential modi- patient risks for each quality attribute by adapting internal or in-
fication sites of the same type into “regions of interest” in the dustry standards, or by developing a graduated patient impact
protein (e.g., glycation or methionine oxidation in the Fc including score based on expert-led discussions within the QbD strategy
Met252 and Met428 residues). For such variants affecting multiple team. These definitions are provided in Table 3. Impact is assigned
sites there is context-specific information needed to support the on a 2e20 point scale that considers the effect that an attribute has
criticality assessment. A measurement of the total level of a on biological activity, PK/PD, immunogenicity and safety. The
modification (and not information on the levels of specific sites of bioactivity and PK assessments are associated with demonstrating
concern based on impact to potency or PK) may complicate criti-
cality assessments. Establishing the relative reactivity of framework
vs. CDR or Fc residues could be helpful in interpreting the total
levels. “Regional assays” are of further advantage since they are
usually less complex and laborious, allowing high throughput
analysis benefitting process characterization/process validation
studies.
Ion exchange HPLC (IE-HPLC) is a common mAb characterization
and quality control method. This method purportedly measures the
charge distribution in the protein, although other interactions of
the protein with the stationary phase may lead to separation based
by non-charge mechanisms. Unless there is a unique selectivity for
a particular variant, it is often not possible to achieve the baseline
separation required for isolation of individual variants to enable
specific CQA assessment. Frequently, the regions in the chromato-
gram are referred to as “main peak”, “acidic region”, and “basic
region”, with each comprising multiple species. Fig. 3. Risk score defined by impact and uncertainty.
298 N. Alt et al. / Biologicals 44 (2016) 291e305

Table 3
Impact score for the risk ranking and filtering tool.

Impact and rating Biological Activitya PKb Immunogenicityc,d Safetyd

Very High (20) >100% change >40% change on PK ATAs detected that may be life threatening Irreversible or life-threatening AEs
Highe (16) 40%e100% change 20%e40% change with impact on PD ATAs detected that may be associated with Reversible AEs that are not life threatening
non-life-threatening loss of efficacy
Moderate (12) 20%e40% change 20%e40% change with no impact on PD ATA detected with effect that AEs that can be managed by clinical
can be managed by treatment (e.g., dose titration, medication)
clinical treatment (i.e., dose
titration, medication, etc.)
Low (4) <20% change <20% change with no impact on PD ATAs detected with effect on PK or PD, Safety effect with minimal
but no effect on safety or efficacy clinical significance
None (2) No change No impact on PK or PD ATAs not detected or ATAs detected with No effect on safety
no effect on PK, PD, safety, or efficacy

AE ¼ adverse event; ATA ¼ anti-therapeutic antibody ¼ ADA (anti-drug antibody); PD ¼ pharmacodynamics; PK ¼ pharmacokinetics.
a
Based on clinically relevant potency assay results of isolated variants relative to reference standard.
b
Serum exposure (area under the curve) or relevant in vitro clearance receptor binding (FcRn and mannose receptor). PD considered if information is available.
c
Effect observed in clinical studies. Prior to clinical experience, product variants known to be present on plasma-derived antibodies are given a score of 2.
d
The potential link to immunogenicity and safety are also considered based on literature and the general concerns associated with the attributes, such as HMWS.
e
Attributes for which no relevant information is available must be ranked as High (16).

efficacy, and the safety and immunogenicity assessments address
safety concerns.
The assessment of individual variants and impurities within a
product is limited only to those that can be assessed specifically.
Variants that were enriched by specific generation or isolation from
the product are used for the impact assessment of biological ac-
tivity and PK. With respect to immunogenicity and safety the
assessment can only refer to the mixture of variants and impurities
within the drug. In case a variant occurs at decent levels in the
product, clinical experience may be justified as basis for the
assessment. In addition, literature and knowledge from other
products and in vivo occurrence of variants is taken into account for
judging the risk of a variant to contribute to safety and immuno-
genicity events in the product. High impact and uncertainty scores
are assigned when the variant or impurity is present in low
abundance and it cannot be excluded that it does not carry a safety
or immunogenicity risk.
4.5.1.1. Impact on biological activity. Biological activity may be
assessed based on a direct correlation between the level of a QA and
its potency in the most clinically relevant assay(s). The bioactivity
impact scale includes a 20% bioactivity change as the boundary
between Low and Medium impact, which was based on an internal
convention that commercial potency bioassays will usually have
acceptance criteria at least ±20% wide. The Moderate to High
impact transition was an assumed doubling (from ±20% to ±40%) of
the first threshold. Greater bioactivity changes including materials
with no bioactivity would have a High impact, and the Very High
impact category of >100% change was added as changes of this
magnitude require special consideration due to their potential for
undesired toxicity. The None category is used when the structural
difference could not affect bioactivity, such as the effect of C-ter-
minal Lys variability on antigen binding [20]; such variants repre-
sent a lower risk to patients than variation (such as modifications in
a complementarity-determining region) that could conceivably
affect bioactivity but whose measured effect is less than 20%.
Values determined in the relevant potency assay are normalized
to mock-treatment control generated along with stress treatment
or enrichment/isolation of variants. The unfractionated material or
recombined fractions considering their relative abundance may
serve as 100% reference for biological activity. However, consid-
ering the molecule without any modifications or degradation var-
iants as the most relevant reference, normalization to the main
peak may be considered best choice, if the main peak only com-
prises fully potent species.
Bioanalytical methods usually have an underlying greater vari-
ability compared to non-cell-based methods. Therefore, an indi-
vidual result or set of bioassay results may not have high statistical
confidence. To minimize bias and improve result 95% confidence
intervals (CIs) are calculated from individual values determined in
one analysis. More rigorous testing is done in case a statistically
meaningful result cannot be obtained based on the initial CIs. The
assignment of impact score based on the 95% CI is done according
to Table 3. A worst-case ranking to the higher impact level is done in
cases where the CI crosses a limit between two CQA rankings.
4.5.1.2. Impact on PK and PD. The PK/PD impact considers clear-
ance of the therapeutic protein as measured by the change in Area
Under the Curve (AUC) for a product concentration-time plot. For
the PK/PD scale, the Low to Medium threshold for the PK impact
scale was set at a >20% difference, which was based on the estab-
lished 2001 FDA guidance that allows the 90% confidence intervals
for the AUC for a test article to be within 80%e125% of the AUC of
the reference product to demonstrate bioequivalence [21]. The
impact is considered Moderate if the AUC change is within the
20%e40% range with no change in a pharmacodynamics (PD)
marker concentration change such as the B cell depletion during
rituximab therapy, where a 20% reduction in AUC is unlikely to
affect the nearly complete B cell depletion seen with rituximab
treatment due to CD20 binding saturation. The impact rating is
increased to a High impact if the PK change is associated with a PD
change such as with the IgE-binding omalizumab antibody, where a
20% decrease in serum PK could allow an increase in free IgE [22],
which could affect efficacy. The >40% change for the Very High
category is double the Moderate (20%) threshold. As with bioac-
tivity, the None and Low categories are differentiated by the po-
tential for an effect.
Clinical or non-clinical studies measuring AUC effects for quality
attributes are rare. For many quality attributes, including Fc mod-
ifications, FcRn binding can be used as a surrogate for clearance.
SPR FcRn assays were shown to be sensitive to biophysical pa-
rameters such as charge and size variants and oxidized methio-
nines in particular, reflecting species known to alter Fc/FcRn-
binding and PK [8,23,24]. An alternate model using immobilized
FcRn to assess relative Fc/FcRn binding has also been established
[25] that can be applied when significant differences are detected
in the SPR assay or for modifications known to impact FcRn binding
(e.g. oxidation, deamidation) of the antibody. In contrast to the SPR
measurement, the FcRn affinity chromatography method reflects
the avidity of the IgG-FcRn interaction due to the high amount of
 N. Alt et al. / Biologicals 44 (2016) 291e305
immobilized recombinant human FcRn. The applied pH gradient
from pH 5.5e8.8 displays Fc/FcRn-interaction in dependency of a
gradual pH-shift. Such pH-dependent Fc/FcRn-interaction is also
relevant in the physiological situation during endosomal recycling
of Fc/FcRn-complexes. We consider the interplay of both FcRn af-
finity methods well suitable to assess Fc/FcRn interactions and to be
complementary in their abilities to address an in vivo PK impact of a
variant.
Other PK factors to consider include administration route,
charge and glycosylation [26]. Oligomannose Fc glycans (such as
Man5) appear to accelerate clearance [27]. Chemical modifications
at many sites that make an antibody more basic have been shown
to increase clearance due to higher catabolism when those modi-
fications introduced >1 pI unit changes [26], which are far greater
than the 0.1e0.2 pI differences between the acidic, main and basic
peak forms resolved by IEC or IEF. The 2010 study by Khawli et al.
[23] showed that the acidic and basic forms of a recombinant
antibody are cleared at the same rate as the main peak, indicating
that the type of charge variants normally found on mAbs can be
considered to have low PK impact, although Fc modifications in the
FcRn binding regions may need to be assessed.
4.5.1.3. Impact on immunogenicity. Immunogenicity is defined as
the potential of a therapeutic to drive an unwanted immune
response. This is typically assessed by detection of anti-therapeutic
antibodies (ATAs, equals anti-drug antibodies or ADAs) in the cir-
culation. Immunogenicity risk is challenging to assess; currently
available tools (e.g. in silico algorithms) cannot predict clinical
immunogenicity, i.e., the incidence of ATA formation and the clin-
ical consequences of ATAs.
In-silico models and in vitro assays to predict human immuno-
genicity to therapeutic proteins have been developed [28,29]. Most
in silico models focus on identifying linear T-cell epitopes within a
protein and determining whether these epitopes are likely to be
presented by major histocompatibility complex (HLA) molecules.
Any potential T-cell epitopes that are identified can be confirmed
by their ability to activate T-cells in-vitro. Models for predicting B-
cell epitopes are also being developed. However, their value for
assessing the immunogenicity of lead molecules is limited since
most B-cell epitopes are conformational and are therefore difficult
to predict [28].
At present the predictive power of such in-silico models and
in vitro assays for human immunogenicity of protein therapeutics is
low. The relationship between the presence of a particular T-cell or
B-cell epitope and the development of specific ATAs directed
against a certain epitope is unknown and has not been clinically
validated. Moreover, to date, ATA incidence and the clinical con-
sequences of ATA, cannot be predicted.
Immunogenicity may have no consequences on efficacy or
safety, may change PK and/or PD and/or result in severe clinical
adverse events that may even be life-threatening. Prior to clinical
studies, clinical results from related therapeutic mAbs containing a
specific variant may be considered when assessing immunoge-
nicity, provided that the patient population is relevant. Clinical
experience with a particular variant is accrued as product devel-
opment proceeds. The final assessment of the immunogenicity
impact for any product variant will be assigned based on molecule-
specific clinical data with respect to ATA incidence and inci-
dence of immunogenicity-associated adverse events (AEs), e.g.
hypersensitivity.
The immunogenicity impact scale was developed by a Roche/
Genentech expert working group that considered the patient
impacts of ADAs; this ADA risk scale was directly transferred into
the CQA impact chart shown in Table 3. The immunogenicity
impacts scale considered the risk factors identified by an industry
299
group [30]. Each quality attribute is considered for its potential to
elicit an immune response. Attributes that are present on human
IgG are usually considered No Risk or Low Risk due to immune
tolerance. Some variants may elicit an immune response without
affecting safety or efficacy in a manner similar to the allotypic
variants, where Fc amino acid sequence differences can generate
anti-allotype responses that do not compromise clinical utility
[31]; these single site variants may be considered Low Risk. The
High and Very High risk categories are reserved for attributes
whose immunogenicity could lead to a loss of efficacy or the
development of debilitating conditions, such as when antibodies
to recombinant EPO neutralized both the administered product
and the patient's innate EPO, causing pure red cell aplasia [32].
Similar High or Very High risk attributes may exist for impurities
that cause an ATA response to insulin or the coagulation proteins
such as Factor VIII.
For variants or impurities present in clinical materials with no
established connection to immunogenicity in human subjects, the
immunogenicity impact scores can be based on the product's
overall clinical experience, in alignment with the ICH Q6B (speci-
fications) guidance that “If a consistent pattern of product hetero-
geneity is demonstrated, an evaluation of the activity, efficacy and
safety (including immunogenicity) of individual forms may not be
necessary”. It should be noted that this approach restricts this type
of classification to variant or impurity levels close to those present
in clinical study materials.
4.5.1.4. Impact on safety. The safety impact scale was also devel-
oped by an expert working group who reviewed the infrequent but
important situations where a critical quality attribute could cause a
safety problem. The safety assessment considers the potential
in vivo effects of structural variants and impurities. Attributes
whose safety issues stem from an anti-drug antibody (ADA)
response are assessed using the immunogenicity impact ranking.
Safety issues due to the innate immune responses and non-ADA
responses (including anti-HCP responses) belong in this category.
Attributes present on clinical materials are generally rated Low risk
if the product has a good clinical safety profile. Assessments based
on clinical experience are more relevant when the future levels are
expected to be close to clinical exposures.
The obligatory CQA list contains many of the CQAs with a safety-
based risk, such as virus or other adventitious agent contamina-
tions, and these are generally not reviewed on a product-specific
basis. Rare examples of CQAs linked to safety issues include
Gal(alpha1-3)Gal-epitope on N-linked Fab oligosaccharides that
triggered anaphylaxis in some cetuximab patients [33], and the
NGNA form of sialic acids that may be linked to serum sickness [34].
QAs affecting cytotoxic bioactivity, e.g., ADCC, associated with
systemic cytokine release may also have an impact on safety.
AEs include infusion related reactions (IRR), injection site re-
actions, hypersensitivity, anaphylaxis, rash, neutropenia and
thrombocytopenia. An example of a quality attribute that could be
rated as Moderate impact would be something that can lead to AEs
such as infusion-related reactions if not controlled by pre-dosing
with other medicines such as anti-inflammatory or steroid-based
medicines, or that can be minimized by changing the dosing
regimen such as by dividing the initial or loading dose into two or
more partial doses. Attributes that could lead to reversible AEs that
are not life-threatening are considered a High impact, such as
bacterial endotoxins. The Very High impact category includes those
attributes that could cause life-threatening AEs such as the
Gal(alpha1-3)Gal-glycans mentioned above, and those that cause
irreversible effects.
300 N. Alt et al. / Biologicals 44 (2016) 291e305
4.5.2. Uncertainty score
Uncertainty is based on the confidence we have in the relevance
of the information used in the impact assessment. These are
assigned on a 1e7 point scale; lower scores reflect more confidence.
The uncertainty rankings (1, 2, 3, 5 and 7) were chosen to reflect the
degree of confidence gained from the knowledge obtained from the
various sources of information listed in Table 4. A new attribute for
which no information is available results in the highest uncertainty
(7). Published data for an attribute in a related product is also
regarded as highly uncertain when the data are generated by an
external source and no detailed first-hand knowledge of the data
are available (5), such as the published studies on the PK impacts of
Fc oligomannose structures [35]. Non-clinical and in vitro or in silico
data on the specific molecule being evaluated, or in-house clinical
data on a similar class of molecule, provide moderate uncertainty
(3); and example of this was the study with pharmacokinetic study
of pertuzumab charge variants [23]. Clinical studies examining the
impact of a specific attribute of the molecule in question provide
the lowest uncertainty (1 or 2). Uncertainty in assigning immu-
nogenicity and safety impact is low if the variant has been present
in materials used in clinical trials including phase I to phase III
experience with the current product, with the caveat that the
clinical level of the variant restricts the quantitative range in which
high confidence can be claimed.
Certainty regarding safety and immunogenicity from clinical
studies can only extend to the levels of materials actually tested in
the clinics. For attributes of potential safety and immunogenicity
concern, all pertinent information, not just clinical observation, will
be considered in the impact assessment. For some CQAs, e.g., high
molecular weight species, there may be general higher concerns for
an immunogenicity risk based on literature [36,37] and this may
require an increased CQA impact score based on this external
information.
For PK, very low uncertainty only applies if the clearance of the
QA has been determined in the clinic by specific studies on clear-
ance of individual or classes of species after administration. Low (2)
and very low (1) uncertainty typically do not apply to biological
activity, because clinical evaluation of this parameter is difficult to
assess for specific variants. In general, high uncertainty scores
identify attributes that require further study, and the impact as-
sessments and can be lowered when more specific and relevant
information becomes available.
4.5.3. Risk ranking and filtering assessment
Finally, the overall criticality ranking of each pCQA is assessed by
calculating its risk to harm with respect to safety and efficacy. The
relative risk score for each pCQA (product variant or process-
related impurity) is obtained by multiplying the impact and un-
certainty scores. The highest risk score for an attribute for each of
the categories (bioactivity, PK, immunogenicity, safety) is used to
categorize the QA as CQA or non-CQA. Attributes having risk scores
greater than 13 (excluding the combination of impact ¼ 12,
uncertainty ¼ 1) in any single impact category are considered CQAs
(Table 5). Thirteen was selected as the cutoff based on applying the
Table 4
Uncertainty scale for the risk ranking and filtering tool.
Rank Uncertainty Description (product variants & host-cell-derived impurities)
7 Very High No information (new variant)
5 High Published external literature for variant in related molecule
3 Moderate Non-clinical or in vitro data with this molecule. Data (nonclinical, in vitro or clinical) from a similar class of molecule
2 Low Variant has been present in material used in clinical trials.a
1 Very Low Impact of specific form established in clinical studies
a
Applies to assigning impact for immunogenicity and safety; typically does not apply to PK or biological activity.
CQA RRF tool based on several examples of known high- and low-
risk QAs for licensed mAbs, and establishing that the risk scores of
established high-risk attributes were above 13. For example, for i.v.
administration it is well established in literature that C-terminal
lysine does not bear a risk for patient safety or efficacy. Therefore,
the combination of no impact (2) with high uncertainty for
applying literature (5) should still result in a risk score below the
cut-off limit. Variants, specifically tested in in vitro assays should
not result as CQA when no change above 20% on biological or PK is
observed. On the other hand, QAs with a moderate, high and very
high impact will remain CQAs; the impact strongly drives the table,
reduced uncertainty will not move such attributes into the low-risk
attribute category without corresponding reduction to Impact
Scores. The combination of impact 12 with an uncertainty of 1
represents a conservative exception to the cut-off filter but is
justified since there is certainty that an impact occurs.
A second important constraint of this tool is that any attribute
for which no relevant information is available are scored by default
as high impact (impact score of 16) and very high uncertainty
(uncertainty score of 7) until sufficient information becomes
available to reduce the scorings. Application of this approach dur-
ing early development enables identification of high-risk product
variants and impurities that should be studied further to lower the
uncertainty and/or the impact based on the new information ob-
tained. Ideally, at the time of filing additional information will
enable refinement of the assessment and uncertainty scores.
4.6. Attribute interactions and stability considerations
The RRF for CQA identification was designed to give conserva-
tive assessments in the context of the specific molecule and its
clinical indication. Depending upon the number of mechanisms of
action the mAb exhibits, the number of attributes categorized as
CQAs will vary. Additionally, attributes such as bacterial endotoxins
and sterility, which could impact safety and efficacy, have already
been classified as obligatory CQAs.
The RRF approach for CQAs assesses attributes individually.
Although the majority of mAb quality attributes may have already
been classified as critical, potential interactions and stability im-
pacts of quality attributes could affect impact scores. This will be an
important consideration when the attribute testing requirements
are eventually decided. Therefore, an additional assessment is
conducted to capture these potential attribute interactions and
stability impacts.
A potential interaction of CQAs may result in higher impact
levels. For example, stability can be impacted by residual proteases
(that may be measured using host cell protein impurity assays),
and/or metal ion co-factors for neutral proteases. A CQA in-
teractions assessment based on systematic multivariate analysis is
not feasible, as the possible combinations are too numerous. A gap
assessment considering potential interactions can be performed
that may lead to an increased CQA impact ranking.
The interactions gap assessment should consider all QAs,
including product variants, process-related impurities and
 N. Alt et al. / Biologicals 44 (2016) 291e305
Table 5
Risk scoring for product variants and process-related impurities.
obligatory CQAs, based on preclinical and clinical information,
product stability information, relevant literature and platform
knowledge. Unexpected changes observed for materials subjected
to stress conditions may indicate quality attribute interactions.
This additional assessment of CQA interactions is grouped into
two categories:
1. Those that synergistically affect bioactivity, PK, immunogenicity,
or safety (such as host cell DNA/lipopolysaccharide). The syn-
ergistic effects arise from interaction of two or more QAs that
produces an effect greater than the sum of their individual ef-
fects. These need to reflected in the CQA assessment.
2. Those that affect stability (e.g., sequential or simultaneous
change resulting in further degradation or causing stability is-
sues in other attributes, such as a host cell protein protease/
LMWS). These need to be considered in the control system.
5. CQAs in the product lifecycle
Presumptive CQAs are identified during product development
and their criticality assessment is updated based on increasing
product and platform knowledge. Final CQAs are defined during the
late Phase III clinical studies by the systematic and risk-based
criticality assessments as described in this chapter, and are then
presented in the registration dossier for Health Authority review.
CQAs are revised and updated in post-approval submissions, if
necessary. For example, a reassessment could be triggered by Phase
IV clinical findings, new indications/patient population consider-
ations, or identification of new variants throughout improved
analytical methods (better resolution of sensitivity) or from
external publications. C-terminal lysine represents an example for a
CQA where its criticality assessment depends on the administration
route and may have to be reassessed if the administration route is
changed. A general review of the CQAs is also guided by the post
approval lifecycle plan (PALM) plan [38].
CQA impact and uncertainty scores are expected to change
throughout development and, potentially, after approval as product
301
understanding increases and new information becomes available.
Attributes with high uncertainty and/or high impact may be tar-
geted for additional studies to increase understanding. In some
cases, a default assumption of high impact (16) could potentially be
reduced to low impact (4) based on new data.
6. Examples of CQA identification
In the following, examples from various molecules are pre-
sented to which we have applied the CQA RRF tool. For an overview
of scorings please refer to Table 6. The CQA assessments are
considered highly molecule specific and outcomes cannot be
broadly applied to other molecules. However, some experimental
data from structure-function studies may be transferred for the
assessments of related molecules, if justified.
6.1. Afucosylation (CQA)
The FcgRIIIa binding affinity and subsequent effector functions
such as antibody dependent cellular cytotoxicity (ADCC) elicited by
human IgG1 antibodies strongly depend on the carbohydrate
moiety linked to the Fc region of the protein. A large body of
experimental evidence has demonstrated the key role that the core
fucose plays in ADCC activity [39]. The absence of the core fucose
residue (i.e., the a1,6-linked fucose on the GlcNAc residue at the
reducing terminus of the oligosaccharide) imparts higher ADCC
activity to the antibody [40,41]. ADCC is a key mechanism of action
for many cytotoxic antibodies. Glycoengineering based on the
overexpression of recombinant 1,4-N-acetylglucosaminyltransfer-
ase III (GnTIII) and a-mannosidase II (ManII) leads to higher afu-
cosylation, which results in strongly increased ADCC bioactivity
compared to the unmodified antibody [42].
The impact of afucosylation levels on ADCC bioactivity can be
assessed by correlating data on attribute levels in development
batches with potency results. Since afucosylation is the main
parameter impacting ADCC bioactivity (as compared to other
glycan characteristics), a clear correlation could be established for
302 N. Alt et al. / Biologicals 44 (2016) 291e305
Table 6
Risk ranking and filtering of selected monoclonal antibody product variants.
Quality Attribute Category
Afucosylation Biological Activity
PK
Immunogenicity
Safety
Glycation in CDR Biological Activity
PK
Immunogenicity
Safety
C-Terminal Lysine (for IV administration) Biological Activity
PK
Immunogenicity
Safety
Free Thiols at Cys22/Cys99 Biological Activity
PK
Immunogenicity
Safety
calculation of the change in ADCC bioactivity upon change of afu-
cosylation levels. Some residual variance is observed and may
result from other minor factors also impacting ADCC bioactivity and
measurement variation.
The level of afucosylated antibody exhibits a major impact on
bioactivity and for antibodies comprising high afucosylation levels
changes in ADCC bioactivity are expected up to 100% change.
Therefore, a high impact ranking of 16 and an uncertainty of 3 are
assigned based on in vitro data.
The kinetics of elimination of completely fucosylated and afu-
cosylated complex type glycans produced by an alpha-1,6-
fucosyltransferase knock-out cell line in the presence of a glycosi-
dase inhibitor was similar in a mice PK study applying material [43]
suggesting no impact of core fucose on PK. Moreover, enzymatic
deglycosylation of the antibody resulted in a 20% lower FcRn on-
rate binding as determined by SPR analysis (relative to glycosy-
lated controls). However, applying a deglycosylated sample to an
FcRn affinity column no pronounced shift in pH gradient elution
retention time Therefore, the minor differences in the FcRn SPR
analysis were not regarded as relevant. Deglycosylation is seen to
represent a maximal possible change in glycans to impact PK.
Therefore, a low impact of 4 was assigned for PK and an uncertainty
of 3 based on these in vitro data.
Since species with a varying degree of core fucose have been
present in clinical trials for the intended indication reflecting the
administration route and dose claimed for market application, low
uncertainty scores were given for immunogenicity and safety.
Impact on immunogenicity was rated low because of the occur-
rence of afucosylated glycans on endogenous antibodies. The
impact on safety is considered moderate (score 12), as there are
indications that the afucose content has an impact on particular
AEs, such as first infusion reactions (IRRs).
Afucosylation is a CQA since it impacts biological activity and
safety (Table 6).
6.2. Glycation in CDR (CQA)
Non-enzymatic protein glycation is a chemical process that oc-
curs during cell culture production of therapeutic proteins [44,45]
and in biological fluids and tissues [46]. Reaction of reducing
sugars such as glucose and galactose with amino groups of lysine
residues and amino termini leads to glycation, which is reversible
until the Amadori rearrangement occurs.
The impact of CDR glycation on biological activity of an antibody
is potentially high given that a lysine is present at one light chain
residue and two heavy chain residues, which are located in the
Impact Uncertainty Risk score
16 3 48
4 3 12
4 2 8
12 2 24
16 3 48
2 5 10
2 2 4
4 2 8
2 3 6
4 3 12
2 2 4
2 2 4
4 3 12
2 5 10
4 2 8
4 2 8
LCDR and HCDR, respectively. CDR lysines highly glycated (>50%)
under forced conditions (1M glucose, 10 days, 37  C), as demon-
strated by LCMS peptide map analysis, showed a significant impact
of glycation on potency measured by ADCC bioassay (54 ± 17%
relative to control). Consequently, a high impact ranking of 16 was
assigned for bioactivity. Uncertainty is low since the conclusion is
based on in vitro results. Glycation in the CDR is unlikely to impact
PK since PK is mediated by the antibody Fc part [47].
Glycation of IgGs has been identified in healthy subjects [46],
without indication for autoantibody formation. As glycation occurs
on endogenous antibodies there are no safety and immunogenicity
concerns with the given dosage of the antibody administered
intravenously for the intended indication. Therefore, glycation of
IgGs is not expected to impact immunogenicity or safety. Uncer-
tainty is low as glycated material has been present in material used
in clinic.
CDR glycation is considered a CQA due to its impact to biological
activity (Table 6).
6.3. C-Terminal lysine (non-CQA)
Heavy chain C-terminal Lys residues whose presence is expected
based on the heavy chain gene sequence [48] are often absent on
monoclonal antibodies isolated from mammalian cell culture [49].
This absence is believed to be due to the activity of one or more
basic carboxypeptidases. In the example discussed below, the
criticality of C-terminal lysine variants is evaluated for an antibody
administered intravenously for an acute indication. In case of s.c.
administration there is less experience for the impact and fate of C-
terminal lysine in vivo which has to be considered for the CQA
assessment.
Charge-related variants resulting from removal of the heavy
chain C-terminal lysine are not expected to impact the biological
activity [20] or PK of an IgG antibody when administered intrave-
nously. The C-terminus of the heavy chain is not in the proximity of
the antigen binding Fab region. Crystal structures of Fc fragments in
complexes with Fc receptors (FcgR and FcRn) show that the heavy
chain C-terminus is not involved in the binding interface between
Fc and the Fc receptors [50]. Potency data on an isolated IE-HPLC
fraction containing mostly C-terminal lysine variants is fully
potent (101 ± 5% relative to controls without C-terminal Lys).
Analysis of the IE-HPLC fraction with the enriched C-terminal
Lys level by FcRn SPR procedure resulted in a lower FcRn on-rate
binding (~10% lower relative to RS) indicating a low impact on PK
based on in vitro data. In addition, applying the IE-HPLC fraction to
an FcRn column showed almost no shift in the pH dependent
 N. Alt et al. / Biologicals 44 (2016) 291e305
retention time. Taken together, these studies indicate that heavy
chain C-terminal forms (Lys variability and Pro-amidation) will not
affect the ability of the antibody to bind for target, FcgR or FcRn
receptors. Low bioactivity and PK impact scores were assigned
(2e4) with moderate uncertainty (3) based on literature and in vitro
data with this antibody.
Since forms lacking C-terminal lysine were present in material
of the antibody used in clinical studies, the impact and uncertainty
scores for immunogenicity and safety are relatively low. In addition,
C-terminal lysine variants are not expected to affect immunoge-
nicity and safety as they are most likely rapidly removed in vivo:
while the heavy chain gene of endogenous antibodies codes for C-
terminal lysine [48], it was not detected on circulating antibodies
[51]. In accordance with that, the C-terminal lysine of therapeutic
monoclonal antibodies is quickly removed in serum [52,53].
C-terminal lysine forms were assigned a low impact (2e4) in all
categories and are not a CQA (Table 6).
6.4. Unpaired cysteines (Non-CQA)
A pair of free thiols at Cys 22 and Cys 96 on the heavy chain of an
antibody could be resolved by reversed-phase high-performance
liquid chromatography (RP-HPLC). The identity of the free thiols
was confirmed by peptide mapping [54].
Analysis of the IE-HPLC fraction containing the free thiol variant
at Cys 22 and Cys 96 indicated that these variants have no impact
on ADCC and CDC activities. Furthermore, the free thiol containing
Fab was purified by IE-HPLC. The binding of free-thiol Fab and intact
Fab to the CD20 antigen was measured in a cell-based binding assay
and an ELISA. The unpaired cysteines in the Fab had no significant
impact on the binding to its target [54]. These free thiols were
found to be buried in the hydrophobic domain and could not be
detected by Ellman's assay under native conditions. Therefore, a
low impact score (4) and moderate uncertainty (3) score were
assigned to bioactivity.
The pair of free thiols at Cys 22 and Cys 96 are located in the
variable domain in HC and are located away from the FcRn binding
sites. Therefore, the variant should not impact the FcRn binding. A
score of no impact (2) and a high uncertainty score (5) were
assigned to PK.
The pair of free thiols has been present in clinical materials at
considerable levels (~18%). The assessment was preliminarily based
on phase I and II clinical data and was reconfirmed with pivotal
studies. Roche has another licensed mAb product with approxi-
mately 25% unpaired Cys in the VH region that also has very low
immunogenicity. So, while unpaired Cys could theoretically present
an elevated immunogenicity risk, this has not been our clinical
experience. Therefore, a low impact score (4) and low uncertainty
score (2) were assigned to immunogenicity and safety.
Free thiols at Cys22 and Cys96 were assigned with a no impact
score (2) for PK and a low impact score (4) for bioactivity, safety and
immunogenicity and are considered as a non-CQA (Table 6).
7. Lessons learned and future opportunities
The purpose of the CQA assessment is to identify quality attri-
butes that require control. Discipline is required to separate CQA
assessments, which are based on patient impacts, from control
strategies, which are also based on process capability and stability.
The application of our CQA RRF tool to the development of
several monoclonal antibodies supported this well-structured
procedure for identification of CQAs. We continue to refine the
approach for assessing impacts based on the availability of better
analytical tools and health authority feedback. The conservative
impact categorization in our CQA assessment tool results in only a
303
few non-CQAs, since proof of non-criticality is a high bar to clear.
For example, glycation has been conservatively ranked as a CQA in
regulatory submissions so far because the variant cannot be
generated at a high level to assess its impact on function. However,
there is evidence that overall high antibody glycation results in
comparable binding to Fc receptors compared to the non-glycated
control [55]. Since plasma proteins including antibodies are natu-
rally glycated, glycation of monoclonal antibodies is not expected to
impact safety and immunogenicity. Thus glycation has been ruled
as non-CQA for some products [55,56]. We may need to reconsider
whether some variants in conserved regions of the molecule that
are not a safety concern and which hardly form may be considered
as non-critical given that under harsh stress conditions lower levels
still do not translate into an in vitro effect on function. In general, for
such low abundance variants we will continue to learn more about
their readiness to form under realistic and relevant conditions in
relation to their relevance for molecule efficacy and safety.
The development of suitable assays to model mechanisms of
action and toxicity is critical for the assessment. This assessment
tool also highlights the need for improved in vitro methods to
model mechanisms of clearance.
Analytical technologies have been evolving for their ability to
detect new variants and known variants at continuously lower
levels. In the light of this increasing analytical sensitivity we need
to focus on the understanding of the patient impact risks for such
variants at trace levels and allow filtering that enables a reasonable
decision to include QAs in a criticality risk assessment. A 0.5% cut-
off level (at peptide level) for a variant formed in a stress experi-
ment may be considered reasonable to exclude low-level variants
from the pCQA list, and may be additionally justified when stress
conditions clearly exceed conditions in the production process and
normal shelf-life. This threshold may not apply for any quality
attribute with a potential risk to impact patient safety or
immunogenicity.
For recent projects, mostly in vitro structure-function studies
were performed to support QA criticality ranking. Only for few
molecule variants is their further fate well understood once
administered to the patient. C-terminal lysine is readily cleaved
from antibodies when injected intravenously into patients [53], but
there is uncertainty regarding the PK impacts of C-terminal Lys
variability on products administered by the subcutaneous route.
But there is still limited understanding of in vivo processing for the
majority of molecule variants and the site-dependency thereof.
Recovering samples from patient sera could provide further infor-
mation of the criticality of variants.
Overall, we will continuously improve our analytical techniques
as well as our understanding of variants and impurities to impact
product quality. A vigilant survey of literature and comparison of
our own data to new publications relevant for a CQA assessment is a
key to finding the right evaluation of quality attributes and a
reasonable categorization of their relevance in the long term.
8. Conclusions
We describe tools that can make it possible to translate the ICH
Q8(R2) CQA definition into practical application for therapeutic
antibodies. Determination of CQAs is an important step for QbD
development of biopharmaceuticals. Using a systematic scientific
and risk-based approach provided a disciplined stepwise process
for assessing and documenting criticality of quality attributes. As
shown by the examples for quality attributes of two licensed mAbs,
the process of determination depends on comprehensive knowl-
edge through characterization of quality attributes.
The determination of CQAs using a scientific risk based sys-
tematic approach is an iterative process and evolves during the
304 N. Alt et al. / Biologicals 44 (2016) 291e305
development as more molecule knowledge becomes available.
Further mAb QbD development subsequent to CQA determination
involves the derivation of a control strategy and a design space,
which are described in later chapters. These subsequent activities
largely depend on the solid understanding of CQAs.
Acknowledgments
The authors would like to acknowledge Jochen Felix Kepert,
Brian Kelley, Alfred Schnüriger and Heidi Zhang for their contri-
butions as well as their constructive document review.
References
[1] CMC Biotech Working Group. A-MAb: a case study in bioprocess develop-
ment. Emeryville, CA: CASSS; 2009.
[2] International Conference on Harmonization of Technical Requirements for
Registration of Pharmaceuticals for Human Use. ICH harmonized tripartite
guideline, pharmaceutical development. Q8 R2. August 2009.
[3] Lynne K, Christof F. Introduction to the application of QbD principles for the
development of monoclonal antibodies. Biologicals 2016;44(5):282e90.
[4] Ferrara C, Brünker P, Suter T, Moser S, Püntener U, Umana P. Modulation of
therapeutic antibody effector functions by glycosylation engineering: influ-
ence of Golgi enzyme localization domain and co-expression of heterologous
beta1, 4-N-acetylglucosaminyltransferase III and Golgi alphamannosidase II.
Biotechnol Bioeng 2006;93(5):851e61.
[5] Haberger M, Bomans K, Diepold K, Hook M, Gassner J, Schlothauer T, et al.
Assessment of chemical modifications of sites in the CDRs of recombinant
antibodies: susceptibility vs. functionality of critical quality attributes. MAbs
2014;6:327e39.
[6] Raju TS, Briggs JB, Chamow SM, Winkler ME, Jones AJS. Glycoengineering of
therapeutic glycoproteins: in vitro galactosylation and sialylation of glyco-
proteins with terminal N-acetylglucosamine and galactose residues.
Biochemistry 2001;40:8868e76.
[7] Thomann M, Schlothauer T, Dashivets T, Malik S, Avenal C, Bulau P, et al.
In vitro glycoengineering of IgG1 and its effect on Fc receptor binding and
ADCC activity. PLoS One 2015;10:e0134949.
[8] Stracke JO, Emrich T, Rüger P, Schlothauer T, Kling L, Knaupp A, et al. A novel
approach to investigate the impact of methionine oxidation on pharmacoki-
netic properties of therapeutic antibodies. MAbs 2014;6(5):1229e42. http://
dx.doi.org/10.4161/mabs.29601.
[9] Haberger M, Heidenreich AK, Schlothauer T, Hook M, Gassner J, Bomans K,
et al. Functional assessment of antibody oxidation by native mass spectrom-
etry. MAbs 2015;7:891e900.
[10] Harris RJ, Kabakoff B, Macchi FD, Shen FJ, Kwong M, Andya J, et al. Identifi-
cation of multiple sources of charge heterogeneity in a recombinant antibody.
J Chromatogr B 2001;752:233e45.
[11] Ferrara C, Stuart F, Sondermann P, Brünker P, Umana P. The carbohydrate at
FcgRIIIa Asn-162. An element required for high affinity binding to non-
fucosylated IgG glycoforms. J Biol Chem 2006;281:5032e6.
[12] Ha S, Ou Y, Vlasak J, Li Y, Wang S, Vo K, et al. Isolation and characterization of
IgG1 with asymmetrical Fc glycosylation. Glycobiology 2011;21(8):1087e96.
[13] Yang Y, Stella C, Wang W, Schoneich C, Gennaro L. Characterization of
oxidative carbonylation on recombinant monoclonal antibodies. Anal Chem
2014;86(10):4799e806.
[14] Butko M, Pallat H, Cordoba A, Yu XC. Recombinant antibody color resulting
from advanced glycation end product modifications. Anal Chem 2014;86(19):
9816e23.
[15] Bozhinov A, Handzhiyski Y, Genov K, Daskalovska V, Niwa T, Ivanov I, et al.
Advanced glycation end products contribute to the immunogenicity of IFN-b
pharmaceuticals. J Allergy Clin Immunol 2012;129(3):855e8.
[16] Rakwalska M, Rospert S. The ribosome-bound chaperones RAC and Ssb1/2p
are required for accurate translation in Saccharomyces cerevisiae. Mol Cell
Biol 2004;24:9186e97.
[17] Salas-Marco J, Bedwell DM. Discrimination between defection in elongation
fidelity and termination efficiency provides mechanistic insights into trans-
lational readthrough. J Mol Biol 2005;248:801e15.
[18] Drummond DA, Wilke CO. Mistranslation-induced protein misfolding as a
dominant constraint on coding-sequence evolution. Cell 2008;134:341e52.
[19] Plant EW, Nguyen P, Russ JR, Pittman YR, Nguyen T, Quesinberry JT, et al.
Differentiating between near- and non-cognate codons in saccharomyces
cerevisiae. PLoS One 2007;2:e517.
[20] Antes B, Amon S, Rizzi A, Wiederkum S, Kainer M, Szolar O, et al. Analysis of
lysine clipping of a humanized Lewis-Y specific IgG antibody and its relation
to Fc mediated effector function. J Chromatogr B Anal Technol Biomed Life Sci
2007;852:250e6.
[21] US Department of Health and Human Services Food and Drug Administration.
Guidance for industry: statistical approaches to establishing bioequivalence.
Rockville, MD: FDA; 2001.
[22] Hochhaus G, Brookman L, Fox H, Johnson C, Matthews J, Ren S, et al. Phar-
macodynamics of omalizumab: implications for optimised dosing strategies
and clinical efficacy in the treatment of allergic asthma. Curr Med Res Opin
2003;19(6):491e8.
[23] Khawli LA, Goswami S, Hutchinson R, Kwong ZW, Yang J, Wang X, et al.
Charge variants in IgG1: isolation, characterization, in vitro binding properties
and pharmacokinetics in rats. mAbs 2010;2:613e24.
[24] Ferrant JL, Benjamin CD, Cutler AH, Kalled SL, Hsu Y, Garber EA, et al. The
contribution of Fc effector mechansims in the efficacy of anti-DC154 immu-
notherapy depends on the nature of the immune challenge. Int Immunol
2004;16(11):1583e94.
[25] Schlothauer T, Rueger P, Stracke JO, Hertenberger H, Fingas F, Kling L, et al.
Analytical FcRn affinity chromatography for functional characterization of
monoclonal antibodies. MAbs 2013;5(4):576e86.
[26] Bumbaca D, Boswell AC, Fielder PJ, Khawli LA. Physicochemical and
biochemical factors influencing the pharmacokinetics of antibody therapeu-
tics. AAPS 2012;14:554e8.
[27] Go€tze AM, Liu YD, Zhang Z, Shah B, Lee E, Bondarenko PV, et al. High-mannose
glycans on the Fc region of therapeutic IgG antibodies increase serum clear-
ance in humans. Glycobiology 2011;21:949e59.
[28] Bryson CJ, Jones TD, Baker MP. Prediction of immunogenicity of therapeutic
proteins. Biodrugs 2010;24(1):1e8.
[29] De Groot AS, Scott DW. Immunogenicity of protein therapeutics. Trends
Immunol 2007;28(11):482e90.
[30] Koren E, Smith HW, Shores E, Shankar G, Finco-Kent D, Rup B, et al. Recom-
mendations on risk-based strategies for detection and characterization of
antibodies against biotechnology products. J Immunol Methods 2008;333:
1e9.
[31] Van Schie KA, Wolbink G-J, Rispens T. Cross-reactive and pre-existing anti-
bodies to therapeutic antibodies e effects on treatment and immunogenicity.
mAbs 2015. http://dx.doi.org/10.1080/19420862.2015.1048411. accepted.
[32] Casadevall N, Nataf J, Viron B, Kolta A, Kiladjian JJ, Martin-Dupont P, et al. Pure
red-cell aplasia after treatment with recombinant erythropoietin. N Engl J
Med 2002;346:469e75.
[33] Chung CH, Mirakhur B, Chan E, Le QT, Berlin J, Morse M, et al. Cetuximab-
induced anaphylaxis and IgE specific for galactose-alpha-1-3-galactose.
N. Engl J Med 2008;358:1109e17.
[34] Higashi H, Naiki M. Antigen of “serum sickness” type of heterophile antibodies
in human sera: identification as gangliosides with N-glycoylneuraminic acid.
Biochem Biophys Res Commun 1977;79:388e95.
[35] Chen X, Lin DY, Flynn GC. The effect of Fc glycan forms on human IgG2
antibody clearance in humans. Glycobiology 2009;19:240e9.
[36] Rosenberg AS. Effects of protein aggregates: an immunologic perspective.
AAPS J 2006;8(3):E501e7.
[37] Wang W, Singh SK, Li N, Toler MR, King KR, Neema S. Immunogenicity
of protein aggregates e concerns and realities. Int J Pharm 2012;431:1e11.
[38] Kelley B, Cromwell M, Jerkins J. Integration of QbD risk assessment tools and
overall risk management. Biologicals 2016;44(5):341e51.
[39] Reusch D, Tejada ML. Fc glycans of therapeutic antibodies as critical quality
attributes. Glycobiology 2015;25(12):1325e34.
[40] Shields RL, Lai J, Keck R, O'Connell LY, Hong K, Meng YG, et al. Lack of fucose
on human IgG1 N-linked oligosaccharide improves binding to human
Fcgamma RIII and antibody-dependent cellular toxicity. J Biol Chem
2002;277(30):26733e40.
[41] Shinkawa T, Nakamura K, Yamane N, Shoji-Hosaka E, Kanda Y, Sakurada M,
et al. The absence of fucose but not the presence of galactose or bisecting N-
acetylglucosamine of human IgG1 complex-type oligosaccharides shows the
critical role of enhancing antibody-dependent cellular cytotoxicity. J Biol
Chem 2003;278:3466e73.
[42] Umana P, Jean-Mairet J, Moudry R, Amstutz H, Bailey JE. Engineered glyco-
forms of an antineuroblastoma IgG1 with optimized antibody-dependent
cellular cytotoxic activity. Nat Biotechnol 1999;17:176e80.
[43] Kanda Y, Yamada T, Mori K, Okazaki A, Inoue M, Kitajima-Miyama K, et al.
Comparison of biological activity among nonfucosylated therapeutic IgG1
antibodies with three different N-linked Fc oligosaccharides: the high-
mannose, hybrid, and complex types. Glycobiology 2007;17:104e18.
[44] Quan C, Alcala E, Petkovska I, Matthews D, Canova-Davis E, Taticek R, et al.
A study in glycation of a therapeutic recombinant humanized monoclonal
antibody: where it is, how it got there, and how it affects charge-based
behavior. Anal Biochem 2008;373:179e91.
[45] Zhang B, Yang Y, Yuk I, Pai R, McKay P, Eigenbrot C, et al. Unveiling a glycation
hot spot in a recombinant humanized monoclonal antibody. Anal Chem
2008;80:2379e90.
[46] Lapolla A, Fedele D, Garbeglio M, Martano L, Tonani R, Seraglia R, et al. Matrix-
assisted laser desorption/ionization mass spectrometry, enzymatic digestion,
and molecular modeling in the study of nonenzymatic glycation of IgG. J Am
Soc Mass Spectrom 2000;11:153e9.
[47] Jefferis R. Antibody therapeutics: isotype and glycoform selection. Expert Opin
Biol Ther 2007;7:1401e13.
[48] Ellison JW, Berson BF, Hood LE. The nucleotide sequence of a human immu-
noglobulin Cg1 gene. Nucleic Acids Res 1982;10:4071e9.
[49] Harris RJ. Processing of C-terminal lysine and arginine residues of proteins
isolated from mammalian cell culture. J Chrom A 1995;705:129e34.
[50] Sondermann P, Oosthuizen V. Mediation and modulation of antibody func-
tion. Biochem Soc Trans 2002;30:481e6.
 N. Alt et al. / Biologicals 44 (2016) 291e305
[51] Edelman GM, Cunningham BA, Gall WE, Gottlieb PD, Rutishauser U,
Waxdal MJ. The covalent structure of an entire gG immunoglobulin molecule.
PNAS 1969;63:78e85.
[52] Shyong B. A study to link the abundance of C-terminal lysine in an antibody
drug in blood to changes in bioavailability. Genentech Process Sciences In-
ternal Report. July 29, 2005.
[53] Cai B, Pan H, Flynn GC. C-terminal lysine processing of human immuno-
globulin G2 heavy chain in vivo. Biotechnol Bioeng 2011;108:404e12.
[54] Zhang T, Zhang J, Hewitt D, Tran B, Gao X, Qiu ZJ, et al. Identification and
characterization of buried unpaired cysteines in a recombinant monoclonal
305
IgG1 antibody. Anal Chem 2012;84(16):7112e23. http://dx.doi.org/10.1021/
ac301426h.
[55] Go€tze AM, Liu YD, Arroll T, Chu L, Flynn GC. Rates and impact of human
antibody glycation in vivo. Glycobiology 2012;22:221e34.
[56] Flynn GC, Nyberg GB. Using quality by design principles in setting a control
strategy for product quality attributes in: state-of-the-art and emerging
technologies for therapeutic monoclonal antibody characterization volume 1.
monoclonal antibody therapeutics: structure, function, and regulatory space
[Chapter 5] ACS Symp Ser 2014;1176:117e50.