Biotechnology and gene expression profiling for mechanistic understanding of cellular aggregation in mammalian cell perfusio
1. ARTICLE
Gene Expression Profiling for Mechanistic
Understanding of Cellular Aggregation in
Mammalian Cell Perfusion Cultures
Meile Liu, Chetan T. Goudar
Cell Culture Development, Global Biological Development, Bayer HealthCare, 800
Dwight Way, Berkeley, California 94710; telephone: 805-410-2005; fax: 805-447-1010;
e-mail: cgoudar@amgen.com
ABSTRACT: Aggregation of baby hamster kidney (BHK)
cells cultivated in perfusion mode for manufacturing re-
combinant proteins was characterized. The potential impact
of cultivation time on cell aggregation for an aggregating
culture (cell line A) was studied by comparing expression
profiles of 84 genes in the extracellular adhesion molecules
(ECM) pathway by qRT-PCR from 9 and 25 day shake flask
samples and 80 and 94 day bioreactor samples. Significant
up-regulation of THBS2 (4.4- to 6.9-fold) was seen in both
the 25 day shake flask and 80 and 94 day bioreactor samples
compared to the 9 day shake flask while NCAM1 was down-
regulated 5.1- to 8.9-fold in the 80 and 94 day bioreactor
samples. Subsequent comparisons were made between cell
line A and a non-aggregating culture (cell line B). A 65 day
perfusion bioreactor sample from cell line B served as the
control for 80 and 94 day samples from four different
perfusion bioreactors for cell line A. Of the 84 genes in
the ECM pathway, four (COL1A1, COL4A1, THBS2, and
VCAN) were consistently up-regulated in cell line A while two
(NCAM1 and THBS1) were consistently down-regulated. The
magnitudes of differential gene expression were much
higher when cell lines were compared (4.1- to 44.6-fold)
than when early and late cell line B samples were compared
(4.4- to 6.9-fold) indicating greater variability between
aggregating and non-aggregating cell lines. Based on the
differential gene expression results, two mechanistic models
were proposed for aggregation of BHK cells in perfusion
cultures.
Biotechnol. Bioeng. 2013;110: 483–490.
ß 2012 Wiley Periodicals, Inc.
KEYWORDS: aggregation; cell culture; gene expression
profiling; mammalian cells; perfusion
Introduction
Mammalian cell culture continues to be the method of
choice for the production of biotherapeutics that require
complex post-translational modifications for in vivo
efficacy. Since monoclonal antibodies comprise the largest
segment of licensed biopharmaceuticals (Aggarwal, 2007;
Walsh, 2010), the fed-batch mode of bioreactor operation in
bioreactor sizes of up to 25,000 L is the method of choice due
to operational simplicity and also because of product
stability under cultivation conditions of elevated tempera-
ture ($378C). The perfusion cultivation method is typically
reserved for highly unstable molecules, and despite
operational complexities, it allows for high density and
high viability cell cultivation over time periods in excess of
3 months under quasi steady-state conditions. Perfusion
reactors tend to be smaller, often with working volumes of a
few 100 L and varying perfusion rates that are optimized
to preserve product quality and maximize volumetric
productivity.
Unlike a fed-batch culture where cell density follows the
typical exponential and declining phases, perfusion systems
are typically operated at constant cell density and perfusion
rate over the entire duration of cultivation. This ensures a
consistent nutritional environment in the bioreactor which
can ultimately result in a product with consistent product
quality attributes. Hence accurate cell density control is
critical for robust long-term operation of perfusion systems.
Closed loop control of cell density in a perfusion bioreactor
is typically performed using indirect measurements such as
optical density or oxygen uptake rate which are calibrated
with cell counts that are performed in a traditional way using
either a hemocytometer or a cell density estimator (Goudar
et al., 2007, 2011; Konstantinov et al., 1994). Accurate cell
density estimation is thus essential for robust calibration of
the feedback control scheme to ensure robust long-term cell
density control at set-points.
Cell aggregation during bioreactor operation can result in
gross misrepresentation of the true viable cell density in a
bioreactor. While aggregated samples can be subjected to
The authors declare no conflicts of interest.
Chetan T. Goudar’s present address is Amgen Inc., 1 Amgen Center Drive, Thousand
Oaks, CA 91320.
Correspondence to: C. T. Goudar
Additional supporting information may be found in the online version of this article.
Received 19 May 2012; Revision received 3 August 2012; Accepted 10 September 2012
Accepted manuscript online 24 September 2012;
Article first published online 18 October 2012 in Wiley Online Library
(http://onlinelibrary.wiley.com/doi/10.1002/bit.24730/abstract)
DOI 10.1002/bit.24730
ß 2012 Wiley Periodicals, Inc. Biotechnology and Bioengineering, Vol. 110, No. 2, February, 2013 483
2. approaches such as trypsin treatment to break up the
aggregates, information on cell viability can be lost after such
treatments. Moreover, it is extremely difficult to get
representative samples from the bioreactor because large
aggregates may not be uniformly distributed in the
bioreactor and can also be preferentially retained in the
sample bags or during other steps of sample preparation.
This combination of inaccuracies during actual sample
measurement and non-representative samples makes robust
estimation of cell density from aggregated cultures difficult.
Aggregated cultures also have an adverse impact on on-line
optical density measurement and also accelerate the fouling
of optical density and pH and dissolved oxygen (DO) probes
during long-term perfusion cultivation. Hence, robust
operation of a perfusion system is challenging with an
aggregating culture. A mechanistic understanding of cell
aggregation is essential if robust mitigation measures are to
be designed.
Previous studies on the aggregation of CHO (Dube et al.,
2001; Han et al., 2006; Renner et al., 1993; Yamamoto et al.,
2000; Zanghi et al., 2000) BHK (Moreira et al., 1995a,b; Reid
et al., 1993) and human cells (Chan et al., 1990) have
identified medium and culture conditions (Han et al., 2006;
Zanghi et al., 2000), DNA released from lysed cells (Renner
et al., 1993), and elevated calcium concentration in the cell
culture medium (Han et al., 2006; Zanghi et al., 2000) as
potential causes for aggregation. Extracellular cell adhesion
molecules (ECM) have been also suggested as a cause for
aggregation in CHO cells (Yamamoto et al., 2000) but this
has not been explored in detail in previous studies. Increased
agitation rates (Moreira et al., 1995a) and the addition of
sulfated polyions (Dee et al., 1997) have been suggested as
approaches to mitigate cell aggregation but both these
approaches have limited utility in a commercial perfusion
process. Higher agitation rates cannot break up tightly
bound aggregates and the associated higher shear rates can
damage cells. Sulfated polyions substantially increase the
cost of cell culture medium and can also adversely impact
downstream process because they are highly charged.
The impact of cellular DNA and medium calcium
concentration was ruled out in this study due to high
culture viability and low Ca2þ
concentrations in the cell
culture medium. The focus was on the analysis of gene
expression in the extracellular matrix and cell adhesion
pathway both in aggregating and non-aggregating cultures
and over varying stages in the cultivation process, an
approach not explored in previous aggregation studies.
Mechanistic models for cell aggregation were developed
based on the analysis of differential gene expression in
aggregated and non-aggregated cultures.
Materials and Methods
Cell Line, Medium, and Cell Culture
Baby hamster kidney (BHK) cells were cultivated in protein-
free medium containing glucose and glutamine as the main
carbon and energy sources. Cryopreserved cells were thawed
and expanded in shake flasks at 37.08C and 5% CO2 for
bioreactor inoculation. Bioreactors were cultivated at a
temperature of 35.58C and an agitation rate of 40 rpm. The
dissolved oxygen (DO) concentration in the bioreactor was
maintained at 50% by membrane aeration using an oxygen–
nitrogen mixture. Bioreactor pH was controlled at 6.8 by
addition of 6% Na2CO3. The cell density during steady state
operation was 10 Â 106
or 20 Â 106
cells/mL (depending on
bioreactor campaign), which was maintained by automatic
cell-discard from the bioreactor.
Analytical Methods
Samples from the bioreactors were taken for cell density and
viability analysis using the CEDEX system (Innovatis,
Bielefeld, Germany) and were also analyzed for nutrient and
metabolite concentrations. Highly aggregated cells were
trypsinized to obtain more representative cell counts.
Glucose, lactate, glutamine, and glutamate concentrations
were determined using a YSI Model 2700 analyzer (Yellow
Springs Instruments, Yellow Springs, OH). The pH, pCO2,
and pO2 were analyzed in a Rapidlab1
248 blood gas
analyzer (Bayer HealthCare, Tarrytown, NY).
RNA Extraction and DNA Removal
Total RNA was extracted from fresh BHK cells using the
RNeasy Mini kit according to manufacturer’s instructions
(Qiagen, Austin, TX). The RNA yield and purity were
determined by measuring the absorbance at 260 and 280 nm
using a UV spectrophotometer. Isolated RNA was of high
purity, with OD260/OD280 ratios >1.8. Trace quantities of
DNA contamination were removed using RNAse-Free
DNAse Set (Qiagen).
RT2
Profiler PCR Arrays
The RT2
Profiler PCR Array System (extracellular matrix
and adhesion molecules pathway) was used for quantifica-
tion of aggregation related genes (SABiosciences, Frederick,
MD). This approach allows detection of 84-cell adhesion
and extracellular matrix genes in the mouse genome. The
array also includes controls for evaluating genomic DNA
contamination, RNA quality, and PCR performance. The
96-well plate format was used and each sample was run on a
separate plate to obtain expression data for the 84 genes in
the extracellular matrix and adhesions molecules pathway.
Comparison of gene expression data between samples was
performed using the DDCt method with normalization of
the raw data to housekeeping genes.
RNA samples were first converted to cDNA using the RT2
First Strand Kit (SABiosciences) following manufacturer’s
instructions. Each set of cDNA samples was diluted in
nuclease free H2O and 2Â SABiosciences RT2
qPCR Master
484 Biotechnology and Bioengineering, Vol. 110, No. 2, February, 2013
3. mix (SABiosciences). The experimental cocktail was loaded
into the PCR array and the loaded plates were analyzed using
a two-step cycling program on the Strategene Mx3000p
instrument (Strategene, Santa Clara, CA). The first cycle was
at 958C for 10 sec, followed by 40 cycles at 958C for 15 sec,
and 608C for 1 min. A melting curve program was
immediately run after the above cycling program for quality
control and evaluation of primer specificity.
Experimental Approach
Two BHK cell lines demonstrating different aggregation
characteristics were thawed and cultivated in shake flasks
(Fig. 1). The aggregation-prone cell line was designated as
cell line A and was expanded and used to inoculate four
laboratory scale perfusion bioreactors (BR1, BR2, BR3, and
BR4). The non-aggregating cell line was designated as cell
line B and was scaled-up in shake flasks and used to
inoculate a single laboratory scale perfusion bioreactor
(BR5). Previous experience with cell line A indicated a clear
increase in aggregation over the course of bioreactor
operation while this was not the case for cell line B. RNA
samples were collected both in early shake flask culture and
towards the end of the bioreactor cultures for cell line A
(Table I). A single time point sample was taken for cell line B
because no time related aggregation changes were observed.
This combination of samples coupled with replicate
bioreactors BR1–BR4 allowed evaluation of biological and
technical replicates in addition to comparison of gene
expression across cell age and cell lines.
Results
Reproducibility and Specificity of qPCR Arrays
The amount of genomic information on BHK cells is limited
in the GenBank database and as a result, PCR arrays specific
to BHK are not commercially available. Nevertheless, the
study of gene expression in BHK is possible through cross-
species hybridization to mouse, which in addition to being
closely related to BHK, has abundant genomic sequence
information (Bult et al., 2008; Eppig et al., 2005; Wlaschin et
al., 2006). Because the level of identity between the mouse
and BHK sequences is between 75% and 97% and the level of
homology varies with each gene (Wlaschin et al., 2006), the
reproducibility of the PCR arrays and the specificity to the
BHK genes of interest must be assessed. The reproducibility
and specificity of the PCR arrays were evaluated through
technical replicates. In the reproducibility studies, the same
RNA samples were processed on two PCR arrays and
representative data for the RNA sample from BR5 are shown
in Figure 2 where the Log10ð2ÀDCt Þ values from both arrays
for all 84 genes were within the Æ1 fold-change limit with a
majority along the y ¼ x line (R2
¼ 0.99). These results
suggest that cross-hybridization of the mouse primers to
BHK RNA is feasible for the genes evaluated in this study.
Since bioreactors BR1–BR3 originated from the same
cryovial and because bioreactor samples were collected at 80
culture days for analysis (Table I), these samples can be
considered as biological replicates (BR4 was not included
because it was collected on Day 94). A total of three
Figure 1. An overview of the experimental approach used in this study. Cell lines A (aggregating) and B (non-aggregating) were thawed and scaled up in shake flasks for
bioreactor inoculation. RNA samples were taken as described in Table I and used to analyze gene expression in the ECM/cell adhesion pathway.
Liu and Goudar: Cellular Aggregation During Perfusion Cultivation 485
Biotechnology and Bioengineering
4. comparisons were possible from these three cultures and in
all three cases, the maximum differences in gene expression
were <4-fold with a majority of the expression differences
much <4-fold (Supplementary Information). Based on
these observations for biological replicates, only gene
expression changes >4-fold were considered significant in
subsequent comparisons.
Impact of Culture Vessel and Cell Age
Previous experience with cell line A indicated a clear increase
in aggregation over the course of the cultivation. Lower
levels of aggregation were seen in shake flasks where
the culture duration was 10–30 days. Upon subsequent
bioreactor inoculation, aggregation increased and the
highest levels were seen towards the end of the culture
(80–100 days). In addition to the culture age differences, cell
densities in shake flask cultures were typically <2.0 Â
106
cells/mL while the steady-state bioreactor cell density
was in the 10–20 Â 106
cells/mL range. It is thus likely that
higher aggregation in bioreactor cultures was a combination
of these two factors.
In an attempt to decouple these two factors, two sets of
gene expression comparisons were made. In the first, shake
flask samples on Day 9 (SF9) and 25 (SF25) were compared
and using the 4-fold difference as the threshold for
significant change (based on the biological replicate data),
genes THBS2 and ICAM1 were up-regulated in SF25 relative
to SF9 with corresponding fold changes of 4.1 and 4.4,
respectively (Table II).
Subsequent analysis compared the SF9 shake flask sample
with bioreactor samples BR1–BR4 which corresponded to
80–94 culture days and a different cultivation mode (high
density perfusion). Results from these 4 comparisons are
shown in Table II where THBS2 was up-regulated in three of
four bioreactor samples (BR1, BR2, and BR4 with fold
changes of 4.9, 6.5, and 6.9, respectively). In addition,
NCAM1 was down-regulated in all four bioreactor samples
with 5.1- to 8.9-fold changes (Table II) while THBS1 was
significantly down-regulated (5.6-fold) only in the BR4
sample. Overall, the primary differentiators of the early
shake flask (SF9) and late bioreactor samples (BR1–BR4) for
cell line A were the expression levels of genes THBS2 and
NCAM1 (Table II).
Comparison of Cell Lines A and B
Cell lines A and B had widely varying aggregation properties
with aggregation in cell line A getting worse with culture
duration while cell line B was not aggregated even after
extended periods of cultivation. In an attempt to relate this
aggregation difference to gene expression changes related to
ECM/cell adhesion molecules, RNA from a 65 day bioreac-
tor sample from cell line B (BR5) was compared with an
early shake flask culture of cell line A (SF9). An early sample
from cell line A was first chosen for comparison since it was
representative of the least aggregated state for cell line A. A
comparison of gene expression data for these two samples is
shown in Table II where three genes COL4A1, THBS2, and
VCAN were up-regulated in cell line A while gene STY1 was
down-regulated. The gene COL4A1 that codes for basement
membrane constituents was up-regulated 40.1-fold in cell
line A while genes THBS2 and VCAN were up-regulated
11.6- and 12.9-fold, respectively (Table II). In addition, gene
STY1, which translates into a transmembrane molecule, was
9.3-fold down-regulated in cell line A cultured in shake
flasks.
In a subsequent analysis, gene expression in bioreactor
samples for cell lines A and B was compared. The bioreactor
sample for cell line B was collected on Day 65 while multiple
samples for cell line A were collected at either 80 or 94 days
in culture (Table I). There were thus similarities in culture
age and cultivation modes for these samples with cell
line being the dominant distinction. Comparisons for each
of the four bioreactor samples from cell line A with the cell
line B bioreactor sample are shown in Table II. In all four
comparisons, genes COL4A1, THBS2, and VCAN, were
Table I. Shake flask and bioreactor RNA samples from cell lines A and B
analyzed using the EMC/cell adhesion qPCR array.
Cell line Culture vessel RNA sample Culture age (days)
A Shake flask SF9 9
SF25 25
Bioreactor BR1 80
BR2 80
BR3 80
BR4 94
B Bioreactor BR5 65
Figure 2. Log10ð2ÀDCt Þ values for BR-5 technical replicate RNA samples. Each
of the 84 genes was within Æ1 fold-change with the majority along the y ¼ x line.
486 Biotechnology and Bioengineering, Vol. 110, No. 2, February, 2013
5. up-regulated in cell line A while gene COL1A1 was up-
regulated in three of four instances. Gene NCAM1 was
down-regulated in cell line A in all four instances while
THBS1 was down-regulated in three instances and STY1 in a
single instance (Table II).
Overall, up-regulation of genes COL4A1, THBS2, and
VCAN for cell line A was seen both in shake flask and
bioreactor samples while COL1A1 was only up-regulated in
the cell line A bioreactor samples (Table II). While gene
STY1 was down-regulated in the cell line A shake flask
samples, this down-regulation was only seen in a single
bioreactor sample (BR3). In addition, down-regulation of
genes NCAM1 and THSB1 was only seen in the cell line A
bioreactor samples and not in the shake flask sample
(Table II).
Specificity of Mouse Primers to BHK RNA
The specificity of mouse primers to target the corresponding
genes in the BHK RNA samples was evaluated by
examining the dissociation curves generated using the
melting curve program immediately following qPCR.
Representative examples for genes COL4A1 and THBS2
which were differentially expressed in multiple comparisons
(Table II) are shown in Figure 3. In both cases, a single peak
was observed around 848C suggesting high specificity
between the mouse primers against the BHK RNA.
Similar results were observed for all the other differentially
expressed genes in Table II (data not shown).
Discussion
Overview of Differentially Expressed Genes
Results from the qPCR array analysis resulted in the
identification of 4 ECM/cell adhesion genes COL1A1
(collagen, type 1, and alpha 1), COL4A1 (collagen, type 4,
and alpha 1), THBS2 (thrombospondin-2), and VCAN
(versican) that were consistently up-regulated and two genes
NCAM1 (neural cell adhesion molecule 1) and THBS1
(thrombospondin-1) that were down-regulated in cell line A
(aggregating culture) when compared with cell line B
(non-aggregating). Gene THBS2 was also up-regulated in
three of four comparisons between the cell line A bioreactor
samples (more aggregated) and early shake flask samples
(less aggregated) while expression of NCAM1 was down-
regulated in all four comparisons (Table II). These
observations are consistent with those seen from a
comparison of cell lines A and B were up-regulation of
THBS2 and down-regulation of NCAM1 were consistently
associated with the more aggregated samples (Table II).
Furthermore, the number of genes that were differently
expressed and the magnitudes of differential expression were
higher across cell lines (A vs. B) than across different samples
from the same cell line (shake flask vs. bioreactor samples for
cell line A).
It must be recognized that changes in mRNA levels do not
necessarily correlate with protein levels. While not performed
in this study, the gene expression changes presented above can
be validated by analyzing the associated proteins using
antibody staining and detection through immunofluorescence
microscopy or western blot. Alternatively, gene silencing can
be used to verify the role of differentially expressed genes in cell
aggregation. For instance, aggregating BHK cells could be
transfected with shRNA plasmids targeting THBS2 and a
reduction in cell aggregation would be a verification of the
contribution of THBS2 to aggregation. If successful, such
approaches can be used in the creation of cell lines, which in
addition to the desirable growth and recombinant protein
production attributes, also possess favorable aggregation
characteristics.
Relationships Between Differentially Expressed Genes
The interactions and/or associations between the differen-
tially expressed genes were visualized using the Gene
Network Central online tool (http://gncpro.sabiosciences.
com/gncpro/gncpro.php; Fig. 4). The four genes (COL1A1,
Table II. Summary of genes differentially expressed between RNA samples in shake flasks and bioreactors from cell lines A and B.
Control Test
Differentially expressed genes
COL1A1 COL4A1 THBS2 VCAN ICAM1 NCAM1 THBS1 STY1
SF9 SF25 4.1 4.4
SF9 BR1 4.9 À8.9
SF9 BR2 6.5 À8.0
SF9 BR3 À7.0
SF9 BR4 6.9 À5.1 À5.6
BR5 SF9 40.1 11.6 12.9 À9.3
BR5 BR1 4.1 36.2 27.7 22.7 À21.0
BR5 BR2 6.8 38.6 37.0 17.7 À18.8 À7.7
BR5 BR3 31.3 21.5 10.4 À16.5 À7.5 À9.4
BR5 BR4 4.6 44.6 39.6 17.9 11.9 À10.7
The cell ages for control and test samples are shown in Table I.
Liu and Goudar: Cellular Aggregation During Perfusion Cultivation 487
Biotechnology and Bioengineering
6. Figure 3. Representative dissociation curves for COL4A1 (A) and THBS2 (B) from 3 ECM/cell adhesion PCR arrays.
Figure 4. The interactions or associations between FN1, COL4A1, COL1A1, VCAN, and THBS2.
488 Biotechnology and Bioengineering, Vol. 110, No. 2, February, 2013
7. COL4A1, THBS2, and VCAN) that were up-regulated in the
aggregating cell line are known to be co-expressed in kidney
cells, along with FN1 (fibronectin-1) and MMP2 (matrix
metalloproteinase-2). Although the expression of FN1 was
not significantly higher (<4-fold) in cell line A, it is directly
involved in the interaction of genes COL1A1, COL4A1,
VCAN, and THBS2 and has also been found to associate
with these molecules directly and can promote cell adhesion
(Lawler, 1986; Mumby et al., 1984; Nagata et al., 1985; Wu
et al., 2005). Fibronectin binds to a variety of collagen types
including types I and IV through specific sequence of amino
acids on the collagen chain to form a fibronectin–collagen
complex. Cells can bind to this complex in the presence of
Ca2þ
and Mg2þ
(Kleinman et al., 1981; Peshwa et al., 1993;
Zanghi et al., 2000) both of which were present in the cell
culture medium used in this study. Fibronectin also has
affinity to thrombospondin, which has specific binding sites
for collagen and other adhesive glycoproteins (Frazier,
1987). Thrombospondin is an adhesive glycoprotein that
can form aggregates and agglutinate platelets and nucleated
cells (Frazier, 1987). Fibronectin has also been shown to
interact with versican, which is a chondroitin sulfate
proteoglycan, and collagen type 1 to reduce cell adhesion
in melanoma cells (Wu et al., 2005). The increased
expression of versican in cell line A could be a negative
feedback reaction to reduce aggregation.
Theoretical Model for Aggregation
From the collective data, we propose two theoretical models
for aggregation in cell line A. The first involves thrombos-
pondin as a key element in promoting cell aggregation.
Thrombospondin was first identified as an adhesive
protein secreted by platelets as a result of exposure to
thrombin to promote hemmaglutination (Frazier, 1987;
Mumby et al., 1984). It is also synthesized and secreted by
several cells in culture, including endothelial cells and
fibroblasts (McKeown-Longo et al., 1984). Other studies
have shown that thrombospondin can be involved in cell–
cell adhesion through the terminal globular domain which
has high affinity for platelets (Mosher, 1990). In addition,
thrombospondin can self-aggregate when intramolecular
thiol-disulfide exchange occurs between unprotected cyto-
sine in the chains of the thrombospondin molecule (Mosher,
1990). As illustrated in Figure 5A, THBS2 secreted by cells in
culture can initially form aggregates and the aggregated
THBS2 molecules can recruit cells to bind to their amino
terminal globular domain, forming an aggregation complex
of THBS2 molecules and cells.
In addition to thrombospondin, collagen and fibronectin
can also promote aggregation of cells in culture. Cells in
culture, including fibroblasts and endothelial cells, can
produce fibronectin (Jaffe and Mosher, 1978; Yamamoto
et al., 2000) which can bind to collagen substrate or tissue
culture plastic surface and promote cell attachment (Reid
et al., 1993). Collagen is also synthesized by most cells in
culture and can form cross-linked collagen fibers (Chan
et al., 1990; Levene et al., 1972) which have specific binding
sites to fibronectin for cells to bind (Reid et al., 1993). Many
of these cell–fibronectin–collagen interactions can lead to a
large aggregation complex, as shown in Figure 5B. This
complex can also bind to aggregated thrombospondin which
are bound to cells since thrombospondin also has an affinity
(albeit low) to type I and type IV collagen.
Conclusions
Aggregation of BHK cells in perfusion cultures was
characterized by analyzing the expression profiles of 84
genes associated with intracellular adhesion molecules. Both
aggregating and non-aggregating cell lines were studied in
laboratory-scale bioreactors and by comparing gene
expression across these two systems, differentially expressed
(>4-fold difference) genes were identified. Specifically,
four genes (COL1A1, COL4A1, THBS2, and VCAN) were
consistently up-regulated in the aggregating cells while two
genes were consistently down-regulated (NCAM1 and
THBS1). This information was used to propose two
mechanistic models of cell aggregation. The first suggested
cell aggregation was mediated by secreted thrombospondin,
while collagen and fibronectin proteins which have affinity
for one another as well as binding sites for cells in culture
were implicated in the second mechanistic model. Since
Figure 5. Theoretical models for aggregation of cell line A. A: Unprotected
cytosine and intrachain disulfide bonds can lead to intermolecular thiol-disulfide
exchange and thrombospondin aggregation formation. Cells in culture are then bound
to the aggregated thrombospondin complex mediated by the amino-terminal globular
domains. B: Collagen fibers bind to fibronectin through specific binding sites. The
attached fibronectin also has binding sites for cells, resulting in a large aggregation
complex.
Liu and Goudar: Cellular Aggregation During Perfusion Cultivation 489
Biotechnology and Bioengineering
8. aggregation directly impacts cell density estimation in
perfusion systems, its mitigation is essential for robust
bioreactor operation. Approaches such as those presented in
this study enhance our understanding of the underlying
mechanisms leading up to cell aggregation and can lead to
the development of non-aggregating cell lines which also
have desirable productivity and product quality attributes.
References
Aggarwal S. 2007. What’s fueling the biotech engine? Nat Biotech 25:1097–
1104.
Bult CJ, Eppig JT, Kadin JA, Richardson JE, Blake JA. 2008. The Mouse
Genome Database (MGD): mouse biology and model systems. Nucleic
Acids Res 36:D724–D728.
Chan D, Lamande SR, Cole WG, Bateman JF. 1990. Regulation of procolla-
gen synthesis and processing during ascorbate-induced extracellular
matrix accumulation in vitro. Biochem J 269:175–181.
Dee KU, Shuler ML, Wood HA. 1997. Inducing single-cell suspension of
BTI-TN5B 1–4 in sect cells: I. The use of sulfated polyanions to prevent
cell aggregation and enhance recombinant protein production. Bio-
technol Bioeng 54:191–205.
Dube B, Luke HJ, Aumailley M, Prehm P. 2001. Hyaluronan reduces
migration and proliferation in CHO cells. Biochim Biophys Acta
1538:283–289.
Eppig JT, Bult CJ, Kadin JA, Richardson JE, Blake JA, Anagnostopoulos A,
Baldarelli RM, Baya M, Beal JS, Bello SM, Boddy WJ, Bradt DW,
Burkart DL, Butler NE, Campbell J, Cassell MA, Corbani LE, Cousins
SL, Dahmen DJ, Dene H, Diehl AD, Drabkin HJ, Frazer KS, Frost P,
Glass LH, Goldsmith CW, Grant PL, Lennon-Pierce M, Lewis J, Lu I,
Maltais LJ, McAndrews-Hill M, McClellan L, Miers DB, Miller LA, Ni
L, Ormsby JE, Qi D, Reddy TBK, Reed DJ, Richards-Smith B, Shaw DR,
Sinclair R, Smith CL, Szauter P, Walker MB, Walton DO, Washburn
LL, Witham IT, Zhu Y. 2005. The Mouse Genome Database (MGD):
from genes to mice—A community resource for mouse biology.
Nucleic Acids Res 33:D471–D475.
Frazier WA. 1987. Thrombospondin: A modular adhesive glycoprotein of
platelets and nucleated cells. J Cell Biol 105:625–632.
Goudar CT, Matanguihan R, Long E, Cruz C, Zhang C, Piret JM,
Konstantinov KB. 2007. Decreased pCO2 accumulation by eliminating
bicarbonate addition to high cell-density cultures. Biotechnol Bioeng
96:1107–1117.
Goudar CT, Piret JM, Konstantinov KB. 2011. Estimating cell specific
oxygen uptake and carbon dioxide production rates for mammalian
cells in perfusion culture. Biotechnol Prog 27:1347–1357.
Han Y, Liu XM, Liu H, Li SC, Wu BC, Ye LL, Wang QW, Chen ZL.
2006. Cultivation of recombinant Chinese hamster ovary cells grown
as suspended aggregates in stirred vessels. J Biosci Bioeng 102:430–
435.
Jaffe EA, Mosher DF. 1978. Synthesis of fibronectin by cultured human
endothelial cells. J Exp Med 147:1779–1791.
Kleinman HK, Klebe RJ, Martin GR. 1981. Role of collagenous matrices in
the adhesion and growth of cells. J Cell Biol 88:473–485.
Konstantinov K, Chuppa S, Sajan E, Tsai Y, Yoon S, Golini F. 1994. Real-
time biomass-concentration monitoring in animal-cell cultures. Trends
Biotechnol 12:324–333.
Lawler J. 1986. The structural and functional properties of thrombospon-
din. Blood 67:1197–1209.
Levene CI, Bates CJ, Bailey AJ. 1972. Biosynthesis of collagen cross-links in
cultured 3 T6 fibroblasts; effect of lathyrogens and ascorbic acid.
Biochim Biophys Acta 263:574–584.
McKeown-Longo PJ, Hanning R, Mosher DF. 1984. Binding and degrada-
tion of platelet thrombospondin by cultured fibroblasts. J Cell Biol
98:22–28.
Moreira JL, Alves PM, Aunins JG, Carrondo MJ. 1995a. Hydrodynamic
effects on BHK cells grown as suspended natural aggregates. Biotechnol
Bioeng 46:351–360.
Moreira JL, Alves PM, Feliciano AS, Aunins JG, Carrondo MJT. 1995b.
Serum-free and serum-containing media for growth of suspended BHK
aggregates in stirred vessels. Enzyme Microb Technol 17:437–444.
Mosher DF. 1990. Physiology of thrombospondin. Annu Rev Med 41:
85–97.
Mumby SM, Abbott-Brown D, Raugi GJ, Bornstein P. 1984. Regulation of
thrombospondin secretion by cells in culture. J Cell Physiol 120:280–
288.
Nagata K, Humphries MJ, Olden K, Yamada KM. 1985. Collagen can
modulate cell interactions with fibronectin. J Cell Biol 101:386–394.
Peshwa MV, Kyung YS, McClure DB, Hu WS. 1993. Cultivation of
mammalian cells as aggregates in bioreactors: Effect of calcium con-
centration of spatial distribution of viability. Biotechnol Bioeng
41:179–187.
Reid GG, Gorham SD, Lackie JM. 1993. The attachment, spreading and
growth of baby hamster kidney cells on collagen, chemically modified
collagen and collagen-composite substrata. J Mater Sci Mater Med
4:201–209.
Renner WA, Jordan M, Eppenberger HM, Leist C. 1993. Cell-cell adhesion
and aggregation: Influence on the growth behavior of CHO cells.
Biotechnol Bioeng 41:188–193.
Walsh G. 2010. Biopharmaceutical benchmarks 2010. Nat Biotech 28:917–
924.
Wlaschin KF, Seth G, Hu WS. 2006. Toward genomic cell culture engineer-
ing. Cytotechnology 50:121–140.
Wu YJ, La Pierre DP, Wu J, Yee AJ, Yang BB. 2005. The interaction of
versican with its binding partners. Cell Res 15:483–494.
Yamamoto S, Matsuda H, Takahashi T, Xing XH, Tanji Y, Unno H. 2000.
Aggregate formation of rCHO cells and its maintenance in repeated
batch culture in the absence of cell adhesion materials. J Biosci Bioeng
89:534–538.
Zanghi JA, Renner WA, Bailey JE, Fussenegger M. 2000. The growth factor
inhibitor suramin reduces apoptosis and cell aggregation in protein-
free CHO cell batch cultures. Biotechnol Prog 16:319–325.
490 Biotechnology and Bioengineering, Vol. 110, No. 2, February, 2013