Journal of Chromatography B, 848 (2007) 28–39

Review

Downstream processing of monoclonal antibodies—Application

of platform approaches夽
Abhinav A. Shukla a,1 , Brian Hubbard a , Tim Tressel b , Sam Guhan b , Duncan Low b,∗
a
Process Development, Amgen Inc., 1201 Amgen Court W., Seattle, WA 98119, United States
b Process Development, Amgen Inc., One Amgen Center Drive, Thousand Oaks, CA 91320, United States
Received 31 May 2006; accepted 8 September 2006
Available online 13 October 2006

Abstract
This paper presents an overview of large-scale downstream processing of monoclonal antibodies and Fc fusion proteins (mAbs). This therapeutic
modality has become increasingly important with the recent approval of several drugs from this product class for a range of critical illnesses.
Taking advantage of the biochemical similarities in this product class, several templated purification schemes have emerged in the literature. In
our experience, significant biochemical differences and the variety of challenges to downstream purification make the use of a completely generic
downstream process impractical. Here, we describe the key elements of a flexible, generic downstream process platform for mAbs that we have
adopted at Amgen. This platform consists of a well-defined sequence of unit operations with most operating parameters being pre-defined and
a small subset of parameters requiring development effort. The platform hinges on the successful use of Protein A chromatography as a highly
selective capture step for the process. Key elements of each type of unit operation are discussed along with data from 14 mAbs that have undergone
process development. Aspects that can be readily templated as well as those that require focused development effort are identified for each unit
operation. A brief description of process characterization and validation activities for these molecules is also provided. Finally, future directions
in mAb processing are summarized.
© 2006 Elsevier B.V. All rights reserved.

Keywords: Monoclonal antibodies; Fc fusion proteins; Cell culture harvest; Protein A chromatography; Viral inactivation; Viral filtration; Ultrafiltration/diafiltration;
Process characterization; Process validation

Contents

1. Monoclonal antibodies and Fc fusion proteins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2. Purification of mAbs—literature review and templated purification schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3. A platform approach to process development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4. A flexible, generic platform for mAb downstream processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.1. Cell culture harvest operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2. Protein A chromatography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3. Low pH viral inactivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.4. Polishing chromatographic steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.5. Viral filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.6. Ultrafiltration/diafiltration (UF/DF) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.7. Absolute filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5. Process characterization and validation activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

夽 This paper is a part of a special issue entitled “Polyclonal and Monoclonal Antibody Production, Purification, Process and Product Analytics”, guest edited by
A.R. Newcombe and K. Watson.
∗ Corresponding author. Tel.: +1 805 313 1754; fax: +1 805 480 0613.

E-mail address: [email protected] (D. Low).

1 Bristol-Myers Squibb, 6000 Thompson Road, East Syracuse, NY 13057, United States.

1570-0232/$ – see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.jchromb.2006.09.026
 A.A. Shukla et al. / J. Chromatogr. B 848 (2007) 28–39 29

6. Future directions in mAb purification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

7. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

1. Monoclonal antibodies and Fc fusion proteins these molecules by Protein A chromatography and facilitates
the use of a common platform approach for both classes of
Monoclonal antibodies and Fc fusion proteins have emerged molecules.
as one of the most exciting therapeutic modalities in the biophar- The advent of monoclonal antibodies and Fc fusion proteins
maceutical industry. Nineteen monoclonal antibodies and 3 Fc has significantly increased the production scales for biophar-
fusion protein-based therapeutics have been approved for sale maceuticals. Most biotechnology products that were approved
in the U.S. and the European Union [1] with combined annual until the mid 1990s (including a variety of vaccines, hormones
sales already exceeding $9 billion [2]. Several of these molecules and growth factors) required very small quantities of purified
serve significant unmet medical needs (Table 1). Nearly a quar- product. In contrast, due to the high doses and the large patient
ter of biologics undergoing clinical trials belong to this class populations in the indications they have been approved for, mon-
of molecules (PhRMA 2004 survey, www.phrma.org) ensuring oclonal antibodies and Fc fusion proteins commonly require
that the importance of this product class will continue to increase annual production of several hundred kilograms of bulk drug
over the coming years. substance. Thus, the considerations for large-scale production
Some of the crucial properties of monoclonal antibodies for of pharmaceutical grade antibodies and Fc fusion proteins are
biological applications include their specificity for in vivo dis- quite different from their routine laboratory scale purification
ease targets as well as the near infinite range of targets for which that have been described elsewhere [5].
they can be generated. All therapeutic antibodies are IgGs with Since monoclonal antibodies and Fc fusion proteins are typ-
IgG1 and IgG2 being the most common subclasses [3]. IgGs ically large, glycosylated molecules, they are most often pro-
have a well-defined biochemical structure consisting of two duced commercially by deep tank mammalian cell culture [6].
heavy and two light chains held together by intra-molecular Recent increases in cell culture titers to >2 g/L [7] have enabled
disulfide bonds. Each chain consists of constant and variable this production technology to stave off immediate competition
regions (heavy chain: CH 1, CH 2, CH 3 and VH ; light chain: CL from transgenic sources of production although these produc-
and VL ). Fig. 1 shows a schematic structure of a monoclonal tion methods might still find broad applicability in the future
antibody. Fc fusion proteins, as the name implies, consist of the [8]. This review deals with the recovery and purification of anti-
fusion of the Fc region (CH 2, CH 3 and hinge) of an antibody with bodies and Fc fusion proteins (termed mAbs for the rest of this
a fusion partner (commonly a receptor or a cytokine). In con- paper) from mammalian cell culture sources.
trast to antibodies, Fc fusion proteins do not possess a common
biochemical structure beyond the dimeric Fc region [4]. This 2. Purification of mAbs—literature review and
construct takes advantage of the interaction of the Fc region templated purification schemes
with a receptor on endothelial cells called the FcRn receptor
that rescues these molecules from intracellular degradation and Efficient recovery and purification of mAbs from cell culture
increases their half-life quite significantly. As will be explained media is a critical part of the production process and can dic-
later, the presence of the Fc tag also enables purification of tate a significant proportion of the total manufacturing costs [9].
The primary consideration during downstream process develop-
ment is purity. Another important consideration is the speed of
process development given that process development needs to
occur prior to introduction of a therapeutic candidate into clin-
ical trials. Other key considerations include overall yield and
process throughput. In addition, the process must meet several
manufacturability criteria including robustness, reliability and
scalability.
Important product purity attributes include process related
contaminants (e.g. host cell protein levels, DNA, endotoxin,
leached Protein A and some cell culture media additives) and
product related impurities (e.g. high molecular weight aggre-
gate and clipped/low molecular weight species). In addition, the
process must be capable of clearing viruses to ensure product
safety in the event of an undetected contamination.
A variety of preparative modes of chromatography have
Fig. 1. Structure of a monoclonal antibody. VH , variable region, heavy chain; been employed for the process-scale purification of mAbs. Most
CH , constant domain, heavy chain; VL , variable region, light chain; CL , constant schemes have involved the use of Protein A affinity chromatog-
domain, light chain. raphy exploiting the specific interactions that take place between
30 A.A. Shukla et al. / J. Chromatogr. B 848 (2007) 28–39

Table 1
Approved monoclonal antibodies and Fc fusion proteins
Trade name Indication Company Year of approval

(a) Approved monoclonal antibodies

Orthoclone OKT3 Acute kidney transplant rejection Ortho Biotech 1986
ReoPro Prevention of blood clot Centocor 1994
Rituxan Non-Hodgkin’s lymphoma Genentech/Biogen-IDEC 1997
Panorex Colorectal cancer GlaxoSmithKline 1995
Zenapax Acute kidney transplant rejection Hoffman-LaRoche 1997
Simulect Prophylaxis of acute organ rejection in Novartis 1998
allogenic renal transplantation
Synagis Respiratory synctial virus Medimmune 1998
Remicade Rheumatoid arthritis Centocor 1998
Herceptin Metastatic breast cancer Genentech 1998
Mylotarg Acute myelogenous lymphoma Wyeth-Ayerst 2000
Campath-1H B-cell chronic lymphocytic leukemia Millenium/ILEX 2001
Zevalin Non-Hodgkin’s lymphoma Biogen IDEC 2002
Humira Rheumatoid arthritis Abbott 2002
Bexxar Non-Hodgkin’s lymphoma Corixa/GSK 2003
Xolair Allergy Genentech/Novartis 2003
Erbitux Colon cancer Imclone/BMS/Merck 2004
Avastin Metastatic colon cancer Genentech 2004
Raptiva Psoriasis Genentech/Xoma 2004
Tysabri Multiple sclerosis Biogen-Idec 2006
Vectibix Metastatic colorectal cancer Amgen 2006
(b) Approved Fc fusion proteins
Enbrel Rheumatoid arthritis, psoriasis, Amgen 1998
ankylosing spondylitis
Amevive Psoriasis Biogen-Idec 2004
Orencia Rheumatoid arthritis Bristol Myers Squibb 2005

the Fc region of mAbs and immobilized Protein A which is a adopted as a generic purification scheme [15] for a number of
cell wall component of Staphylococcus aureus [10,11]. Protein monoclonal antibody products.
A affinity chromatography has been shown to be highly selec- While Protein A chromatography is highly selective for
tive for mAbs, resulting in >95% purity in a single step starting mAbs, the use of an immobilized protein as a ligand also lends
from complex cell culture media [12]. its own share of challenges to this mode of chromatography.
Other modes of chromatography have been combined with The ligand is prone to proteolysis and the cleaved domains can
Protein A chromatography to achieve pharmaceutically accept- adhere to product molecules creating a separation challenge.
able purity levels. These steps are typically chosen to provide Conventional Protein A ligands cannot be exposed to alkaline
orthogonal modes of interaction with the product to enable conditions that are commonly employed to sanitize other col-
effective separation from host cell proteins and other contami- umn modes thus necessitating the use of high concentrations of
nants. Direct capture by Protein A chromatography followed by chaotropes such as urea for column regeneration and sanitiza-
anion-exchange chromatography and size exclusion chromatog- tion. The use of high concentrations of chaotropes creates a cost
raphy has been employed for purifying a monoclonal antibody issue as well as a disposal challenge. The need to elute the col-
expressed by hybridoma cell culture [13]. Anion-exchange was umn at a low pH can induce product aggregation for some mAbs.
selected as the second chromatographic step for DNA and endo- Most significantly, the cost of Protein A resins is nearly an order
toxin clearance, while size-exclusion was employed as the last of magnitude higher than conventional chromatographic resins.
step for removal of aggregates and degradation products. An Clearly, there is a significant driver for the development of small
ultrafiltration/diafiltration (UF/DF) buffer exchange step was molecule ligands that can match the selectivity of Protein A for
employed prior to size exclusion chromatography to concen- binding to mAbs.
trate the product. Other modes of chromatography have also Hydrophobic Charge Induction Chromatography (HCIC)
been employed successfully for mAb purification including employs a heterocyclic ligand such as 4-mercaptoethanol (MEP)
hydroxyapatite and immobilized metal affinity chromatogra- that takes on an inducible positive charge at low pHs. This
phy (IMAC) [12]. Genentech has adopted the use of Protein resin has been reported to be selective for antibody separa-
A chromatographic capture followed by cation-exchange chro- tions [16,17]. However, more recent investigations [18] have
matography (CEX) and anion-exchange chromatography (AEX) found this mode of chromatography to be based on non-specific
operated in the flowthrough mode [14]. The CEX step clears hydrophobic interactions with electrostatic repulsion at low pH
host cell proteins, aggregats leached Protein A while the AEX being responsible for product elution. In the capture mode, this
flowthrough step removes DNA and achieves further reduction resin was nearly an order of magnitude less selective for mAbs
in host cell protein impurities. This sequence of steps has been over host cell proteins as compared to Protein A chromatog-
 A.A. Shukla et al. / J. Chromatogr. B 848 (2007) 28–39 31

raphy but was found to be a potentially useful polishing step. A generic process assumes that a pre-defined purification
Ligands that can mimic the binding pocket of Protein A for process works for all mAbs. However, in our experience, sig-
the Fc region of mAbs have been found [19] and developed nificant physicochemical differences exist among mAbs making
into Protein mimetic resins marketed as MAbSorbent A1P and this approach either impractical or resulting in a non-robust pro-
A2P [20]. In internal investigations, these resins have also been cess. Fig. 2 shows analytical cation-exchange elution profiles for
found to possess lower selectivity than Protein A. Thus, at this eight monoclonal antibodies under identical chromatographic
point none of the small molecule ligands can universally match conditions. As can be seen from the figure, the eight molecules
the selectivity offered by Protein A chromatography for mAb differ quite significantly from each other in their affinity and
separations. However, they might be useful additions to the elution behavior. Our experience with over 20 mAb candidates
downstream process sequence due to their orthogonal selectivity indicates that an inflexible generic process, with fully templated
with conventional modes of chromatography. operating conditions, is not a desirable approach given these dra-
Similar to what is done for other proteins, it is conceptu- matic differences in properties. Even if a single set of operating
ally possible to design a downstream process for mAbs with- conditions that work were to be arrived at, the resulting processes
out a Protein A affinity step by employing combinations of would clearly not be optimal or robust for each of the molecules.
conventional chromatographic modes. Three-step combinations Accordingly, our approach has been to adopt a platform strat-
of cation-exchange, anion-exchange flowthrough, hydrophobic egy instead of a fully generic set of operating conditions for all
interaction chromatography and mixed mode cation-exchange mAbs. The platform serves as a guidance document that defines
chromatography were found to deliver adequate clearance of the overall scheme of downstream processes and brackets the
host cell protein contaminants for a CHO derived monoclonal operating conditions for individual unit operations, thus limit-
antibody [21]. However, such purification schemes by-and-large ing the scope of experimentation required to reach a solution for
have not caught on in commercial downstream operations due a given molecule.
to the need to design the purification sequence separately for A platform approach has several advantages from a busi-
each mAb. Given the almost universal applicability of Protein A ness standpoint. Speed to the clinic is often the key determinant
chromatography and the development of workarounds for most of business advantage for biotechnology companies since very
of its limitations (described in Section 4.2), it appears that this often several companies may try and target similar biologi-
ligand will continue to be employed for commercial scale mAb cal pathways. The reduction in time and resources required
purification at least in the foreseeable future. to carry out process development are usually the primary eco-
The Protein A chromatographic step is typically employed nomic driver for adopting a platform approach. However, once
for direct capture of the product from cell culture supernatant such a strategy is adopted several other advantages also become
after harvest operations designed to remove cells and cell debris. apparent. Other organizations in the company such as Quality
In a few cases, the Protein A step is the second step in the process and Manufacturing can better align with Process Development
following capture on a conventional mode of chromatography and integrate templated documents into their systems. Since raw
[22]. This was done to protect the expensive Protein A resin from materials are now selected from a significantly limited list, bet-
possible fouling through direct exposure to cell culture harvest ter deals can be negotiated with vendors. In addition, efforts can
media. However, the development of effective column regenera- be directed towards multi-sourcing of critical raw materials to
tion schemes commonly allow Protein A resins to be employed better manage operating risk. The platform also lays down a
for over 100 cycles with direct load of the cell culture super- common, aligned philosophy that can be adopted across multi-
natant. This also eliminates the need for concentration or buffer ple geographic process development sites of a company (such
exchange of the harvest prior to chromatography. For the vast as Amgen), resulting in a site-independent process that can be
majority of commercial mAb processes, Protein A chromatog- transferred to multiple manufacturing sites. The platform can
raphy appears to be firmly ensconced as the primary capture step also serve as a planning tool while planning across the entire
that also delivers a high purification factor. organization since it lays down a common set of expectations.
For instance, while designing manufacturing facilities for mul-
3. A platform approach to process development tiple future products, the platform can serve as a guide of what
a process will look like and thus be a key planning tool.
Process development can often be the rate-limiting step in
the introduction of biopharmaceuticals into clinical trials [23]. 4. A flexible, generic platform for mAb downstream
Given the explosion in the numbers of mAbs entering clinical processing
trials, there is a clear driver for employing a templated approach
to process development. Indeed, if it was possible to have a The previous section described the concept behind the plat-
generic process that could be employed for all mAb candidates form process. In this section, we describe the downstream plat-
it would greatly reduce the time and resources needed for process form for mAbs that we have developed at Amgen and applied
development. This can have a significant impact on the number for the production of over 20 molecules over a range of scales
of clinical candidates who can be introduced into clinical trials, ranging from clinical production to commercial launch. Fig. 3
and some kind of a generic approach is increasingly forming the shows a schematic for the platform downstream process for
cornerstone of the business strategy of most companies focusing mAbs. Each of the unit operations shown in the figure is fur-
on this therapeutic modality. ther described below.
32 A.A. Shukla et al. / J. Chromatogr. B 848 (2007) 28–39

Fig. 2. Analytical cation-exchange linear gradient elution profiles for eight monoclonal antibodies. x-axis is time (min), y-axis is absorbance units at 280 nm measured
by an in-line UV spectrometer.

4.1. Cell culture harvest operations into single-use cartridges that can be stacked in a housing and
pressurized to drive fluid flow through the system. While conven-
Since mAbs are secreted into the cell culture medium dur- tionally, depth filters have been regarded solely as a particulate
ing mammalian cell culture, the first step in the downstream removal operation, recent evidence suggests that the adsorptive
process is to remove cells and cell debris. At commercial scale properties of depth filters can be exploited to remove soluble
this is accomplished by centrifugation using a continuous disk- species as well [27]. In this work, host cell protein contaminants
stack centrifuge. Centrifugation [24] is preferred over other were shown to be effectively removed from cell culture harvest
harvesting technologies such as cross-flow microfiltration [25] supernatant of a mAb by appropriate selection of depth filter size
due to its scalability and economical operation for large vol- and flux. Removal of these contaminants was shown to prevent
umes (typically 2–15,000 L/batch). Large-scale centrifugation problems with turbidity during elution of the capture Protein A
acts as the primary harvesting step but cannot accomplish com- chromatographic column.
plete removal of cells and cell debris, which must be removed Depth filters typically do not come with an absolute pore size
prior to chromatography. rating unless they include a membrane layer at the end of the
For this reason, centrifugation is followed by depth filtration flow path. The depth filter is followed by a filter with an absolute
step(s) to remove residual cellular debris. Depth filtration refers pore size rating (typically 0.45 ␮m or 0.2 ␮m) that ensures the
to the use of a porous medium that is capable of retaining partic- removal of solid particulates (and bacteria in case of the 0.2 ␮m
ulates throughout its matrix rather than just on its surface [26]. filter) from the cell culture harvest supernatant.
Depth filters employed in bioprocessing typically consist of a Operating conditions for the harvest operations can be suc-
fibrous bed of cellulose or polypropylene fibers along with a filter cessfully templated for practically all mAbs. The combination
aid (diatomaceous earth) and binder. The flat sheets are packed of various harvest operations results in a process that is robust
 A.A. Shukla et al. / J. Chromatogr. B 848 (2007) 28–39 33

well as provide additional clearance steps that help assure viral

safety. The Protein A step also adds another impurity into the
process in the form of leached Protein A ligand that is typically
cleaved by proteases present in the cell culture supernatant.
Protein A chromatography does suffer from several limita-
tions. The primary disadvantage is the high cost of the resin,
which can be up to 10 times as expensive as conventional chro-
matographic supports. The high cost of the resin often leads
to an operating strategy in which a smaller Protein A column is
cycled several times while purifying a batch of cell culture super-
natant. Protein A column loading is usually the rate-limiting
step during this unit operation, since a large volume of cell cul-
ture supernatant is loaded on a relatively smaller column. This
in turn means that throughput during Protein A chromatogra-
phy becomes a key consideration. Differences in the dynamic
binding capacity at various flow rates [28] and in the pressure-
flow characteristics of various Protein A chromatographic media
can result in wide variations in throughput [29,30] and were an
important consideration during selection of a platform Protein
A chromatographic resin. Interestingly, as titers achievable in
cell culture increase, the focus will shift from how fast one can
load the column to how much one can load (i.e. dynamic binding
capacity).
Another key limitation of Protein A chromatography is the
need to carry out product elution at low pHs. Exposure to low
pH conditions can result in the formation of soluble high molec-
ular weight aggregates (as can be detected by analytical size
exclusion chromatography) and/or insoluble precipitate forma-
Fig. 3. Platform downstream process for mAbs. tion during product elution. Table 2 shows the occurrence and
qualitative severity of these phenomena with 14 mAbs that
enough to account for variations in cell density at harvest and underwent process development. As can be seen from the table,
final cell viability between molecules. these problems occur quite frequently during Protein A chro-
matography. High molecular weight aggregate formation can
4.2. Protein A chromatography lead to a reduction in product yield if a significant level of the
product species aggregate. This also places an added burden on
Protein A chromatography serves as the capture step in the the polishing steps to achieve clearance since aggregate species
platform process. The Protein A step also serves as the key vol-
ume reduction step in the process since the product stream is
concentrated from a relatively dilute cell culture supernatant Table 2
to the eluate, which is typically at a concentration of >10 g/L. Aggregation and precipitation during Protein A chromatography
This step has proved to be highly selective for mAbs and can Molecule Soluble high molecular weight Insoluble precipitate formation
in many cases yield >99% purity starting from the cell culture aggregate (measured by SEC) (measured byOD410)
supernatant. The overall scheme of operating a Protein A step 1 − −
also lends itself readily to a platform format. The cell culture 2 + +
supernatant can be directly loaded on the column (at a neutral 3 + −
pH) and the product is eluted from the column at low pHs. A 4 + ++
5 + +
wash step introduced between column load and elution is often 6 + +++
at an intermediate pH and removes host cell protein and other 7 + +++
contaminants. Finally, the column is stripped (an acidic solution 8 + −
of pH ∼2) and regenerated (high concentrations of chaotropes 9 + −
such as urea or guanidine hydrochloride are employed since con- 10 ++ −
11 ++ −
ventional Protein A ligands are not alkaline stable). The high 12 +++ −
purities that are achieved by Protein A chromatography make 13 +++ +++
the concept of a platform process possible for mAbs. The pol- 14 +++ −
ishing steps have to remove remaining levels of host cell protein Soluble HMW: (+++) >10%; (++) 4–10%; (+) 1–4%; (−) <1%. SEC refers to
contaminants, DNA and product-related species (high molecular size-exclusion chromatography. OD410 refers to an optical density measurement
weight aggregate and low molecular weight clipped species) as at 410 nm as an indication of turbidity.
34 A.A. Shukla et al. / J. Chromatogr. B 848 (2007) 28–39

Fig. 4. Strategies adopted for addressing aggregation/precipitation during Protein A chromatography.

can be potentially antigenic. Insoluble aggregate formation can

be the result of either the product species or impurities such as
host cell proteins precipitating. In either event, there exists the
risk of reduction in column lifetime if precipitation occurs dur-
ing elution. If the product precipitates there might be a threat to
product activity.
A variety of strategies have evolved to address the issue
of aggregation/precipitation during Protein A elution and are
reviewed elsewhere [31]. Fig. 4 shows a schematic of the strate-
gies that have been taken in such situations. The modification
of the Protein A elution buffer to make the buffer conditions
more conducive to product stability is often the simplest solu-
tion. Stabilizers such as arginine have been added to the Protein Fig. 5. Elution buffer pH during Protein A chromatography for 14 mAbs.
A elution buffer to reduce aggregation of eluting antibodies [32].
If host cell protein impurities precipitate and prove to be remov- A media with this alternative ligand can lend themselves more
able during harvest depth filtration, appropriate selection of the readily to templated elution buffer pH as shown in Fig. 6.
depth filter chemistry, flux and loading can form a viable solu- Binding capacity on Protein A media also vary quite signifi-
tion [27]. Low temperature operation of the Protein A step can cantly between molecules. Fig. 7 shows the operational loading
in some cases reduce product aggregation. Yet another strat- capacity varying between 10 and 40 g/L resin for 14 molecules.
egy can be to influence the slope of the pH transition from Clearly, loading capacity also requires experimental determi-
wash to elution buffers. Even though Protein A chromatogra- nation for each molecule. Fig. 7 plots data for the same set of
phy is operated under step gradient conditions, the mixing of molecules as in Fig. 5 but employs a different numbering scheme
the two buffers during the transition creates a pH transition that to avoid an erroneous correlation of the data in the two figures.
can be steep or gradual depending on the choice of wash and The numbering scheme for each of the figures in this text is
elution buffers and their strengths. Clearly, molecule specific unique to that particular plot although the same set of molecules
solutions have to be developed and a templated set of condi- are employed through the text.
tions will not apply in the event of aggregation/precipitation Column regeneration for Protein A columns is typically
issues. carried out with high concentrations of chaotropes although
Given the conventional wisdom that mAbs interact with Pro- there are recent reports suggesting the use of weakly alka-
tein A through their Fc regions and the fact that >95% sequence line conditions [35]. High concentrations of chaotropes (e.g.
homology exists even between the four Fc sub-types, one might
expect Protein A elution pH to be readily templated. Fig. 5 plots
the elution buffer pH employed in the downstream process for
14 mAbs. As can be seen from the figure, elution pH varies
quite substantially from close to pH 3.0 to 4.1, even though
the molecules belonged either to the IgG1 or IgG2 subclasses.
A recent explanation for this has been provided through the
demonstration of monoclonal antibody interactions with Pro-
tein A through their variable regions [33]. These interactions
were shown to be eliminated on SuRe Protein A media in which
the ligand consists solely of the B domain of Protein A which is Fig. 6. Comparison of elution pH on MAbSelect® (conventional ligand) and
not implicated in variable region interactions [34]. Thus, Protein SuRe (engineered ligand) resins for 14 mAbs.
 A.A. Shukla et al. / J. Chromatogr. B 848 (2007) 28–39 35

atively easy to include a low pH incubation step to inactivate

viruses. Low pH treatment has been shown to successfully inac-
tivate retroviruses for a variety of biotechnology products [38].
The pH of the Protein A elution pool is adjusted to pH ≤3.8 by
addition of an acid solution (e.g. 0.5 M phosphoric acid). The
use of strong acids such as HCl is avoided despite the advantage
of low volume addition due to the risk of product denaturation
in the localized region where the solution is added. In addition,
storage of high concentrations of halide containing solutions can
cause corrosion concerns for stainless steel vessels, especially
when these solutions are at a low pH.
Retroviral inactivation kinetics in the specific solution dictate
Fig. 7. Maximum operational load capacity during Protein A chromatography.
the duration of the incubation step. For molecules that are stable
under low pH conditions (true for most monoclonal antibodies),
urea, guanidine hydrochloride) are costly and require special a generic low pH condition that assures complete inactivation
handling during disposal. The MAbSelect SuRe® Protein A of retroviruses can be selected [39]. Following acid inactivation,
resin with the engineered ligand is designed to be alkaline the solution is neutralized to move the product into a more stable
stable and permits the use of sodium hydroxide for column pH range. Once again, the use of strong bases (such as sodium
regeneration [36]. hydroxide solutions) is avoided and the use of higher concentra-
Table 3 summarizes the operating parameters for Protein A tions of weaker bases (e.g. Tris base solution) is preferred. An
chromatography and highlights the ones that can be templated important consideration during neutralization is the presence of
and those that require development effort specific to a molecule. a buffering species that can help maintain pH around the final
The operational load capacity, wash and elution buffers require target pH. This is important to help prevent pH overshoot. For
development although the platform for this step does define a example, if an acetate buffer is employed for Protein A elution
set of possible buffers. and the target pH following neutralization is pH 7.0 where this
species cannot buffer, it might be required to add a buffering salt
4.3. Low pH viral inactivation such as phosphate along with the high pH solution.
Some mAb solutions might exhibit a turbid appearance fol-
The FDA Q5A guidance document [37] requires the use of lowing neutralization. If the product species is not involved in
two dedicated orthogonal steps for viral reduction in addition the precipitation, this might not be as much of a concern as
to the clearance achieved on chromatographic steps to assure precipitation during Protein A column elution. Filtration of the
safety of products produced by mammalian cell culture. Since solution with 0.45 ␮m/0.2 ␮m absolute filtration or in the case of
the Protein A column eluate is at low pH and given that most excessive turbidity, use of depth filtration followed by absolute
mAbs are stable in solution under low pH conditions, it is rel- filtration might be an expedient process solution.

Table 3 4.4. Polishing chromatographic steps

Operational parameters for Protein A chromatography

The subsequent chromatographic steps are aimed at reducing

host cell protein impurities, high molecular weight aggregates,
low molecular weight clipped species, DNA and leached Pro-
tein A that remain after the Protein A chromatographic step to
acceptably low levels that assure safety of the product. At least
two subsequent chromatographic steps are typically employed
in mAb downstream processes, with a sufficient level of redun-
dancy between them that assures robust operation of the entire
process. These steps are typically referred to as the polishing
steps in the downstream process.
In our platform, we typically select the polishing steps from
cation-exchange chromatography (CEX), anion-exchange chro-
matography (AEX), hydrophobic interaction chromatography
(HIC) and hydroxyapatite. Typically, one of the two polish-
ing steps is operated in the flowthrough mode (in which the
product does not bind to the column whereas impurity species
are retained). AEX and HIC steps are often operated in the
flowthrough mode for monoclonal antibodies since in general
Parameters in gray are templated across molecules, others these molecules possess high pIs (isoelectric points). Higher
require molecule specific development. column loadings are usually possible in the flowthrough mode
36 A.A. Shukla et al. / J. Chromatogr. B 848 (2007) 28–39

Table 4
Modes of chromatography employed as polishing steps in mAb processes for
clearing specific kinds of contaminants
Impurity Mode of chromatography

High molecular weight aggregate HIC, CEX

Host cell protein impurities AEX, HIC, CEX
Leached Protein A Hydroxyapatite, HIC, CEX
Viral clearance AEX, CEX, HIC, HA

for removing these impurities as compared with leached Pro-

Fig. 8. Host cell protein contaminant levels after Protein A chromatography. tein A or clipped low molecular weight species. Table 4 lists
the modes of chromatography that have typically been success-
than in the bind and elute mode. In addition, operating in the ful for removing a particular kind of impurity in our experience.
flowthrough mode can make for a more robust operation. Suc- Note that this discussion does not include the requirement for the
cessful operation of the step depends primarily on accuracy in removal of product variant species that are sometimes required.
achieving the correct column load conditions whereas bind and Since variant removal (typically misfolded or inactive product
elute operation has a greater number of degrees of freedom in species) is not a typical occurrence, strategies for their removal
which errors can potentially occur (load, wash and elution con- fall outside the definition of a platform process. An additional
ditions). factor in the selection of polishing chromatography steps is the
The choice of which modes of chromatography are employed number of logs of viral clearance they can offer. In general, oper-
depend on the nature of impurities that require clearance and how ation in the flowthrough mode can decrease the viral clearance
difficult they are to remove. Fig. 8 plots the host cell impurity capability except in the case of anion-exchange chromatogra-
levels after the Protein A chromatographic step. As can be seen phy. AEX flowthrough steps have been demonstrated to achieve
from the figure, host cell protein levels can vary quite widely >4 logs of retroviral and parvoviral clearance for monoclonal
between molecules. Separate experiments have indicated that antibodies [40].
these variations are linked to differences in surface chemistry as
well as cell culture and harvest conditions between molecules. 4.5. Viral filtration
The percent high molecular weight aggregate levels after Pro-
tein A chromatography are shown in Fig. 9. In this case, the data Viral filtration is employed in the platform process to com-
was obtained from a generic set of operating conditions on Pro- plement the low pH viral inactivation step. Viral filters can be
tein A and subsequent optimization along the lines discussed classified on the basis of their pore sizes into retroviral (<50 nm)
earlier successfully reduced percent aggregate to <10% in all and parvoviral (<20 nm) grade filters [41]. Viral filters are placed
cases. Furthermore, it is quite clear that aggregate levels can in the platform process following either one of the polishing
also vary between molecules. A large proportion of these aggre- chromatographic steps based on the volume of the intermediate
gates is produced during cell culture (as demonstrated by the product stream to be filtered or the volume of solution that can
use of non-Protein A and neutral pH capture steps). The extent be filtered per unit surface area of membrane. Viral filters are
of host cell proteins and percent aggregate after Protein A chro- typically operated at constant pressure. Due to their relatively
matography is usually indicative of the degree of difficulty in small pore sizes, viral filters (especially parvoviral grade filters)
developing appropriate polishing steps for a given molecule. can clog relatively quickly in the presence of particulate aggre-
Most molecules require a larger time and resource investment gates. As a result, pressurized tanks are employed to drive the
fluid through the filters in preference to the use of pumps which
can generate particulates due to shear through their moving parts.
While the type of viral filter employed can be standardized across
molecules, the placement and size of filters employed typically
vary from product to product.

4.6. Ultrafiltration/diafiltration (UF/DF)

Following the completion of downstream purification,

the product is buffer exchanged into the formulation buffer
[42]. This is best accomplished by the use of an ultrafil-
tration/diafiltration setup. The type of membrane used, the
transmembrane pressure employed, the cross-flow rate and
the concentration at which diafiltration is carried out can be
Fig. 9. Percent high molecular weight aggregate levels after Protein A chro- templated for all mAbs. With several formulations requiring
matography. the use of very high protein concentrations, issues of viscosity
 A.A. Shukla et al. / J. Chromatogr. B 848 (2007) 28–39 37

and product aggregation can occur for certain molecules and Table 5
have to be dealt with on a case-by-case basis. Definition of non-key, key and critical performance and operational parameters
Critical performance parameter: parameter that is a direct measure of the func-
4.7. Absolute filtration tionality of a step
Key performance parameter: parameter that measures process consistency and
performance
Filtration through an absolute filter with a microfiltration Critical operational parameter: an operational parameter that when varied within
membrane is often employed during the downstream process the CR will cause process to fail acceptance criteria in one or more critical
to ensure bioburden control or to remove small amounts of par- performance parameters
ticulates. Such filters are often employed as in-line filters to a Key operational parameter: a parameter that if varied within the CR will signif-
chromatographic column to help prevent fouling due to particu- icantly impact the performance of the process but not cause it to fail in terms
of acceptance criteria around critical performance parameters
lates. Following buffer exchange, the UF/DF retentate is filtered Non-key operational parameter: a parameter that if varied within CR will have
to generate the bulk drug substance which is often stored prior no significant impact on product or process
to the fill and finish operations. The type of the absolute filter
can be standardized for the downstream process and validated to
demonstrate bioburden clearance. Filter sizing can vary depend- classified as non-key. As shown in Fig. 10, a subset of the key
ing on the volume being filtered and the extent of particulates in operational parameters will be critical, in that they can cause
the feed stream. failure of the process. Clearly, this sub-classification can only
be determined from separate experiments called worst case runs
5. Process characterization and validation activities in which operating parameters for a step are combined to result in
a worst-case scenario with respect to a particular critical perfor-
Process characterization is a set of activities conducted to mance attribute. The subsequent steps are then operated at their
demonstrate robustness of a commercial manufacturing process center-points to determine if that particular impurity is cleared to
through studies conducted at a small-scale [43]. The clinical acceptable levels. If not, that operating parameter is deemed to
entry (first-in-human) manufacturing process is usually not char- be critical and should be carefully controlled during large-scale
acterized in detail. However for a biologic entering late-stage manufacturing operation. For such a parameter, the acceptable
clinical trials, process characterization is an essential component range might be as narrow as the operating range defined during
of the regulatory filing package. Operational and performance process development and not wider.
parameters for each unit operation are categorized as non-key, Process validation is defined in the Q7A guidance docu-
key or critical based on this exercise. This also helps to guide ment from the FDA as “. . . providing documented evidence
the setting of alert and action limits for these parameters and the that the process, operated within established parameters, can
setting of acceptance criteria for process validation studies. perform effectively and reproducibly to produce an intermedi-
Designing successful process characterization studies ate or API meeting its predetermined specifications and quality
requires a significant amount of planning, determining which attributes”. Consequently, process validation is typically carried
performance and operational parameters need to be studied on out at large-scale during a series of runs termed the conformance
the basis of process history and understanding gained during or validation campaign. These runs are carried out at large-scale
development or formalized risk quantitative risk analysis such following the engineering/practice runs and prior to start of the
as FMEA analysis [44]. Since process characterization is car- actual manufacturing campaign to produce and stockpile bulk
ried out at small-scale, qualification of a scale-down model is drug substance in anticipation of commercial launch. Data from
essential [45]. these runs are aimed at demonstrating control over the entire
For the characterization studies themselves, the characteri- process. A written validation master plan is created prior to
zation range for operational parameters are set to be at least as carrying out process validation activities. Some process qualifi-
wide as the normal operating ranges and narrower than the zone cation activities are carried out at small-scale using a qualified
of failure. A variety of fractional factorial experimental designs scale-down model of the process. These are employed to com-
are employed to study the impact of varying various operating
parameters within the predetermined range used for the char-
acterization study. A typical study is a two-level, resolution IV,
single replicate design with center points to assist in an indepen-
dent estimation of error. An important outcome of the process
characterization is the identification of parameters as non-key,
key and critical. Prior to carrying out the characterization study,
performance parameters (which include a set of in-process prod-
uct quality analytical assays as well as parameters which monitor
consistency of the run) are classified as shown in Table 5. If
an operational parameter when varied within the characteriza-
tion range causes significant variation in a critical performance
parameter, it is classified as key. If on the other hand, it does Fig. 10. Classification of performance and operational parameters resulting after
not impact a critical performance parameter for that step, it is process characterization.
38 A.A. Shukla et al. / J. Chromatogr. B 848 (2007) 28–39

plement data from large-scale operation (in the case of extended [3] M.A. van Dijk, J.G.J. van de Winkel, Curr. Opin. Chem. Biol. 5 (4) (2001)
time/reuse studies) or in situations in which particular kinds of 368.
validation data cannot be obtained at large-scale (e.g. contami- [4] A. Aruffo, in: S. Chamow, A. Ashkenazi (Eds.), Antibody Fusion Proteins,
Wiley-Liss, New York, 1999, p. 221.
nant spiking studies including viral clearance validation studies [5] R.W. Scott, S.A. Duffy, B.J. Moellering, C. Prior, Biotechnol. Prog. 3
and clearance of certain types of process chemicals). Further (1987) 49.
details on process validation can be had from several references [6] H.E. Chadd, S.M. Chamow, Curr. Opin. Biotechnol. 12 (2001) 188.
[46,47]. [7] F. Wurm, Nat. Biotechnol. 22 (2004) 1393.
[8] T.J. Menkhaus, Y. Bai, C. Zhang, Z.L. Nikolov, C.E. Glatz, Biotechnol.
Prog. 20 (2004) 1001.
6. Future directions in mAb purification [9] K. Keller, T. Friedmann, A. Boxman, Trends Biotechnol. 19 (11) (2001)
438.
[10] H. Hjelm, K. Hjelm, J. Sjöquist, FEBS Lett. 28 (1972) 73.
Future challenges to the current paradigm in mAb purification [11] G. Kronvall, et al., J. Immunol. 111 (1973) 141.
will be provided by the scale of production for many of this class [12] P. Gagnon, Purification Tools for Monoclonal Antibodies, Validated
of products. Successful increases in cell culture titer are antici- Biosystems, Tucson, AZ, 1995.
pated to continue for the next several years, making the down- [13] L.S. Hanna, P. Pine, G. Reuzinsky, S. Nigam, D.R. Omstead, Biopharm.
stream process rate limiting. A large part of the limitations stem Int. (October) (1991) 33.
[14] R.L. Fahrner, H.L. Knudsen, C.D. Basey, W. Galan, D. Feuerhelm, M.
from limited tankage for buffers and process intermediates and Vanderlaan, G.S. Blank, Biotechnol. Gen. Eng. Rev. 18 (2001) 301.
the inability to increase large-scale column diameter to beyond [15] G. Blank, Recovery of Biological Products X Conference, Cancun, Mexico,
2 m without encountering significant issues with flow distribu- June, 2001.
tion and packing. Chromatographic operations thus become lim- [16] L. Guerrier, P. Girot, W. Schwartz, E. Boschetti, Bioseparation 9 (2000)
ited in terms of the throughput they can provide and necessitate 211.
[17] W. Schwartz, D. Judd, M. Wysocki, L. Guerrier, E. Birck-Wilson, E.
extensive cycling to process a single cell culture batch. Innova- Boschetti, J. Chromatogr. A 908 (2001) 251.
tive facility designs are likely to emerge as the first line of solu- [18] S. Ghose, B. Hubbard, S. Cramer, Biotechnol. Prog. 21 (2005) 498.
tions to this challenge. Further in future, non-chromatographic [19] R. Li, V. Dowd, D. Stewart, S. Burton, C. Lowe, Nat. Biotechnol. 16 (1998)
purification techniques such as selective precipitation or 190.
liquid–liquid separations employing highly selective ligands are [20] J. Curling, Genet. Eng. News 21 (20) (2001) 1.
[21] D. Follman, R.L. Fahrner, J. Chromatogr. A 1024 (2004) 79.
likely to emerge. Alternative ligands to Protein A are also likely [22] G. Follena-Wassermann, Presented at Recovery of Biological Products X,
to continue to be introduced in the market and their selectivity Cancun, Mexico, June, 2001.
will improve with time. All of these will result in a gradual evo- [23] G. Pisano, The Development Factory, HBS Press, Boston, MA, 1997.
lution of the downstream platform for mAb purification over the [24] J.P. Maybury, M. Hoare, P. Dunnill, Biotechnol. Bioeng. 67 (2000) 265.
next decade. For further information on this area, please consult [25] C.S. Parnham, R.H. Davis, Biotechnol. Bioeng. 47 (1995) 155.
[26] J.V. Fiore, W.P. Olson, S.L. Holst, in: J. Curling (Ed.), Methods of Plasma
the review by Low et al. [48] also included in this volume. Protein Fractionation, Academic Press, New York, 1980, p. 239.
[27] Y. Yigzaw, R. Piper, M. Tran, A.A. Shukla, Biotechnol. Prog. 22 (2006)
288.
7. Conclusions [28] R. Hahn, R. Schlegel, A. Jungbauer, J. Chromatogr. B 790 (2003) 35.
[29] R.L. Fahrner, D. Whitney, M. Vanderlaan, G. Blank, Biotechnol. Appl.
Platform processes have emerged as a direct result of the Biochem. 30 (1999) 121.
need to develop clinical manufacturing processes for a vast [30] R.L. Fahrner, H. Iyer, G. Blank, Bioproc. Eng. 21 (1999) 287.
pipeline of monoclonal antibody and Fc fusion proteins in the [31] A.A. Shukla, P. Hinckley, P. Gupta, Y. Yigzaw, B. Hubbard, Bioproc. Int.
3 (5) (2005) 36.
biopharmaceutical industry. This business strategy has resulted [32] T. Arakawa, J. Philo, K. Tsumoto, R. Yumioka, D. Ejima, Protein Express.
in significant savings in time and resources and harmonization Purif. 36 (2004) 244.
of practices and information flow across the process devel- [33] S. Ghose, M. Allen, B. Hubbard, C. Brooks, S.M. Cramer, Biotechnol.
opment, operations and quality organizations at Amgen. This Bioeng. 92 (2005) 665.
review described some of the essential elements of a platform [34] M. Starovasnik, M. O’Connell, W. Fairbrother, R. Kelley, Protein Sci. 8
(1999) 1423.
downstream process for large-scale production of mAbs. Com- [35] H. Johansson, A. Bergenstable, G. Rodrigo, K. Oberg, The Use of NaOH for
mon elements in the mAb platform are described for each unit CIP of rProtein A Media: a 300 Cycle Study Waterside Conference, Savan-
operation as well as aspects that pose a challenge to templated nah, May 2002. Available at http://www5.amershambiosciences.com/
conditions across molecules. A key focus of continued technol- applic/upp00738.nsf/vLookupDoc/250173076-P516/$file/18117764AA.
ogy development efforts is to template as many parameters as pdf.
[36] R. Hahn, K. Shimahara, F. Steindl, A. Jungbauer, J. Chromatogr. A 1102
possible for each unit operation. Improved understanding of the (2006) 224.
fundamentals of each step is essential in making this a reality in [37] FDA Q5A Guidance Document: Viral Safety Evaluation of Biotech-
the years to come. nology Products Derived from Cell Lines of Human or Animal Ori-
gin, Federal Register, vol. 63(185), 1998, p. 51074. Available at
http://www.fda.gov/cder/guidance/Q5A-fnl.PDF.
References [38] G. Sofer, Biopharm. Int. 16 (1) (2003) 50.
[39] K. Brorson, S. Krejci, K. Lee, E. Hamilton, K. Stein, Y. Xu, Biotechnol.
[1] G. Walsh, Biopharm. Intnl. 17 (12) (2004) 18. Bioeng. 82 (2003) 321.
[2] D. Milroy, C. Auchincloss, Monoclonal Antibodies–on the Crest of a Wave. [40] S. Curtis, K. Lee, G. Blank, Y. Xu, Biotechnol. Bioeng. 84 (2) (2003) 179.
Horizons, Wood Mackenzie, Boston, MA, 2003, p. 6. [41] T. Ireland, H. Lutz, M. Siwak, G. Bolton, Biopharm. Int. 11 (2004) 38.
 A.A. Shukla et al. / J. Chromatogr. B 848 (2007) 28–39 39

[42] R. van Reis, E. Goodrich, C. Yson, L. Frautschy, S. Dzengeleski, H. Lutz, [46] L. Conley, J. McPherson, J. Thommes, in: A. Rathore, G. Sofer (Eds.),
Biotechnol. Bioeng. 55 (1997) 737. Process Validation in Manufacturing of Biopharmaceuticals, Taylor and
[43] J. Seely, R. Seely, Biopharm. Int. (August) (2003) 24. Francis, Boca Raton, FL, 2005, p. 469.
[44] R. Seely, J. Haury, in: A. Rathore, G. Sofer (Eds.), Process Validation in [47] C. Bussineau, R. Seely, G. Lovitt, J. Fernandez, G. Blank, E. Brandreth, D.
Manufacturing of Biopharmaceuticals, Taylor and Francis, Boca Raton, Conrad, R. Devine, R. Ferris, K. de Heyder, R. Juffras, P. Levy, W. List, M.
FL, 2005, p. 13. Munk, B. Neely, S. Notarnicola, R. O’Leary, H. van Deinse, PDA Journal
[45] R. Godavarti, J. Petrone, J. Robinson, R. Wright, B. Kelley, in: A. Rathore, Technical Report #42 59 (2005) 1.
G. Sofer (Eds.), Process Validation in Manufacturing of Biopharmaceuti- [48] D. Low, R. O’Leary, H. Pujar, J. Chromatogr. (2006) (JCB-06-583R1),
cals, Taylor and Francis, Boca Raton, FL, 2005, p. 69. submitted for publication.