Vous êtes sur la page 1sur 10

Understanding and Modeling Alternating Tangential Flow Filtration for Perfusion

Cell Culture
William Kelly and Jennifer Scully
Dept. of Chemical Engineering, Villanova University, 318 White Hall, 800 Lancaster Ave, Villanova, PA 19085

Di Zhang and Gang Feng


Dept. of Mechanical Engineering, Villanova University, Tolentine Hall, 800 Lancaster Ave, Villanova, PA 19085

Mathew Lavengood, Jason Condon, John Knighton, and Ravinder Bhatia


API-LM Pharmaceutical Development and Manufacturing Sciences, Janssen R&D, 1400 McKean Rd, Spring House, PA
DOI 10.1002/btpr.1953
Published online August 6, 2014 in Wiley Online Library (wileyonlinelibrary.com)

Alternating tangential flow (ATF) filtration has been used with success in the Biopharmaceutical industry as a lower shear technology for cell retention with perfusion cultures. The
ATF system is different than tangential flow filtration; however, in that reverse flow is used
once per cycle as a means to minimize fouling. Few studies have been reported in the literature that evaluates ATF and how key system variables affect the rate at which ATF filters
foul. In this study, an experimental setup was devised that allowed for determination of the
time it took for fouling to occur for given mammalian (PER.C6) cell culture cell densities
and viabilities as permeate flow rate and antifoam concentration was varied. The experimental results indicate, in accordance with DArcys law, that the average resistance to permeate flow (across a cycle of operation) increases as biological material deposits on the
membrane. Scanning electron microscope images of the post-run filtration surface indicated
that both cells and antifoam micelles deposit on the membrane. A unique mathematical
model, based on the assumption that fouling was due to pore blockage from the cells and
micelles in combination, was devised that allowed for estimation of sticking factors for the
cells and the micelles on the membrane. This model was then used to accurately predict the
increase in transmembane pressure during constant flux operation for an ATF cartridge
C 2014 American Institute of Chemical Engineers Biotechused for perfusion cell culture. V
nol. Prog., 30:12911300, 2014
Keywords: fouling, tangential flow, filtration, mammalian cell, perfusion

Introduction
Perfusion is the most efficient mode of production for animal cell culture products. The high cell and product concentrations are achievable because of the feeding of fresh
nutrients and removal of metabolites.1 Alternating tangential
flow (ATF) filters have been used with success in the biopharmaceutical industry as a lower shear technology for
mammalian cell retention with perfusion cultures.2 The longer growth cycles and the higher cell densities achievable
using ATFs for perfusion cell culture can result in membrane
fouling, which is detected by an inability of the permeate
pump to maintain the desired permeate flow rate. A fouling
rate is then defined by how long the filter is used before
fouling is detected. The potential mechanisms for ATF fouling are presumably the same as those that cause fouling in
tangential flow filtration (TFF) systems: complete pore
blockage (due to cells or aggregates of molecules), pore constriction (due to molecular buildup within the pores), or cake
formation (due to the buildup of a biological matrix on the
Correspondence concerning this article should be addressed to
W. Kelly at william.j.kelly@villanova.edu.
C 2014 American Institute of Chemical Engineers
V

outer membrane surface composed of cells and/or molecules).3 Deposition of yeast cells and latex beads (312 mm)
on membrane surfaces has been observed using a noninvasive Direct Observation through Membrane (DOTM) technique.4 These results indicated that at one fixed flux rate, the
larger (6.4 mm) particles were observed rolling across the
membrane surface while the smaller (3 mm) particles were
observed forming a flowing cake layer. In another study, for
a spin filter and cross flow filtration system, it was determined that filter fouling was due to the deposition of nucleic
acids and dead cells.5,6 In other studies, membrane fouling
due to antifoam has been studied.7
Mathematical models have been developed to describe
how transmembrane pressure (DP) should increase during
constant flux dead-end filtration. This increase in pressure
can be attributed most generally to an increase in total resistance (Rtot) due to the increased amounts of biological material deposited on the membrane per
DP=Q=A5Rtot

(1)

where Q/A is the flux or liquid velocity passing through the


membrane. For constant flux operation, an increase in Rtot
1291

1292

Biotechnol. Prog., 2014, Vol. 30, No. 6

concentration and viability could change with time that could


affect the likelihood of fouling. It is not uncommon these
days to need to replace an ATF cartridge at least one time
during the course of a perfusion run. It has been observed
that an ATF cartridge is more prone to fouling as the cell
density and run duration increases, and viability starts to
drop. In this study, a unique experimental setup was utilized
to assess the impact of key operating parameters (i.e., permeate flow rate, cell and antifoam concentrations, and cell viability) on the fouling rate of a specific type of ATF cartridge
being used to filter suspended PER.C6 cells. The ultimate
goal was to develop a mathematical model that could be
used to predict when an ATF cartridge would foul during a
perfusion cell culture run with these same cells, and in general to provide a framework for modeling other similar ATF
systems for the purpose of minimizing the number of filter
exchanges per perfusion run.
Figure 1.

Experimental setup, showing location of flow and


pressure (P1, P2, and P3) sensors. [Color figure can
be viewed in the online issue, which is available at
wileyonlinelibrary.com.]

corresponds to an increase in DP across the membrane. With


pore blockage, the increase in Rtot is due to a decrease in
open pore area over time8; whereas with cake formation the
increase in Rtot is due to the increasing thickness of a biological layer across the entire membrane surface.
For TFF and ATF, there is a crossflow velocity that minimizes the tendency to foul a membrane, and transmembane
pressure (TMP; not DP) indicates the pressure drop driving
permeate flow across the membrane. Stressman and Moresoli9 suggested that the rate of formation of open pore area
is a result of the crossflow, and correlated mathematically to
a constant (that is a function of crossflow velocity) multiplied by the total amount of biomaterial currently on the
membrane. Belfort and Zydney10 suggest that biomaterial
can be peeled off the membrane by crossflow due to an
inertial lift force which increases as the square of the tangential shear rate and the cube of the particle size. Several
recent studies have reported the merits of mathematical models that consider more than one fouling mechanism.11 Ho
and Zydney12 developed a new mathematical model to
describe the change in TMP during constant flux microfiltration of protein solutions. Here, fouling was assumed to occur
first by pore blockage, and then with a cake forming over
the blocked areas of the membrane. In another study,13 yeast
cells and extra-polymeric-substance suspensions were filtered
at a controlled shear rate in a cone and plate viscometer
through a microfiltration membrane. The results indicated an
increased fouling rate with increasing feed concentration and
pore size and decreasing shear stress. The rise in TMP (or
Rtot) over filtration time (at constant permeate flux) was
greater with unwashed, as opposed to washed cells, presumably due to interactions between the cells and the media
components and the membrane surface.
The ATF system is different than TFF, in that reverse
flow is used once per cycle as a means to minimize fouling
by peeling off the biological layer. There are few published
studies on ATF. In recent studies, Clinke and Chotteau evaluated the performance of ATF but did not attempt to model
membrane fouling.14,15 Maximum cell densities were determined for ATF vs. TFF systems. Use of an ATF with perfusion culture is especially challenging in that the cell

Materials and Methods


Experimental methods
An experimental setup was devised with the intention of
running conditions that would promote fouling in < 8 h, so
that the effects of the following variables on fouling rate
could be studied: permeate flow rate, cell density, cell viability, and antifoam concentration. Mammalian (suspended
PER.C6) cells were grown in a preculture and then transferred to a Corning 500 mL spinner flask with impeller for
the ATF experiments. The preculture media and that added to
the spinner flask was a proprietary serum free and antifoam
(Thermo fisher, Hyclone antifoam, #SH30897.02) free media
(except in some experiments, where antifoam was added to a
specified concentration right before filtration began) from
Janssen. Preliminary spinner flask experiments at very high
cell densities, approximately 150 million cells/mL, confirmed
that the cell viability remained relatively constant over 67 h
in these flask with no nutrients being fed. Oxygen and carbon
dioxide exchange occurred via the head space being vented to
the room through air filters. This indicated that experiments
with ATF cartridges could be run for this duration without a
change in cell concentration or quality due to conditions in
the flask, provided there was no cell damage that occurred in
the ATF (due to shear effects etc.). In addition to this fouling
study, a single perfusion run was conducted.
All experiments used GE ATF2 hollow fiber cartridges
with polyethersulfone (PES) membranes (model # CFP-2-E4X2MA). The membranes had pores that were 0.2 mm in
size and the lumen were 66.7 cm long with an ID of 1 mm
and a membrane (i.e., lumen wall) thickness of approximately 200 mm. A C-24 controller from Refine Technology
drove flow in two directions, depending on the phase of the
ATF cycle: flow moved toward the bioreactor during the
pressure phase and away from the bioreactor during the
exhaust phase (See Figure 1). Every experiment utilized a
new filter cartridge that was autoclaved prior to use. The
ATF system was equipped with three PendoTECH single use
pressure sensors (see Figure 1 for locations of sensor P1, P2,
and P3) to allow for calculation of TMP, and a disposable
rotary flow sensor to measure the flow rate of the permeate
flow. The permeate flow was pulled from the hollow fiber
cartridge by a Masterflex Easy Load II peristaltic pump. The
pressure and flow readings were read by a PendoTECH
PMAT3P monitor and readings were collected into

Biotechnol. Prog., 2014, Vol. 30, No. 6

1293

Table 1. Run Conditions for the ATF2 Experiments


Initial Cell
Conc. (3106
cells/mL)

Initial Cell
Viability (%)

Permeate
Rate (mL/min)

Antifoam
Level (ppm)

80.6
80.1
69.3
91.1
91.3
87.2
50.0
76.6
45.2

83.4
81.4

77.0

56.9
92.5
91.9

70
120
170
170
170
120
120
70
120
120
120
170
120
120
120
20
30
30

0
0
0
0
0
0
0
0
0
6,000
0
0
3,000
3,000
4,500
0
3,000
0

46.8
42.1
27.9
37.4
28.9
39.7
24.9
30.9
45.9
0
29.6
32.6
0
40.0
0
34.6
42.7
67.1

Figure 2.

Total resistance (Rtot) with just media (no cells or


antifoam).

PendoTECHs data acquisition software. The data acquisition


system captured data every 2 s, making it possible to capture
approximately 12 data points (during the 24 s long cycle) at
an ATF rate of 0.5 lpm (which was used for all of the runs).

Analytical methods
Throughout the duration of each experiment, 10 mL samples
were taken from the bioreactor. The sample was analyzed using
Siemens RAPIDLab 248 Systems blood gas analyzer to obtain
values for pH, pO2, and pCO2. In addition, a Cedex HiRes
Analyzer was used to measure average cell diameter (mm), viability, viable cell concentration, dead cell concentration, total
cell concentration, and cell aggregate rate. The sample was
centrifuged at 1,500 rpm for 5 min, with the supernate of the
sample then being extracted and filtered through a 0.2 mm filter. Metabolites, such as glucose, glutamine, glutamate ammonium, pH, potassium, lactate, sodium, and ionized calcium
were measured using BioProfile FLEX Analyzer. Finally, a
Cedex BioHT Analyzer measured Lactate Dehydrogenase
(LDH) levels in the supernate.
After some of the runs, the used filter cartridge was
drained by gravity and then the two lumen ends and permeate port were sterilely wrapped. The filters were stored in the

freezer, until they were prepared for scanning electron


microscope (SEM) analysis. The preparation consisted of
cutting the cartridge near both ends, and then removing one
representative fiber. A one inch section was cut near the
axial center of the fiber. This section was frozen for a minute
or so in liquid nitrogen and then a new razor blade was used
to make a cut through the fiber cross section. This procedure
was used to minimize fiber compression during the cutting
process. The topographies of fibers were characterized using
a Hitachi S-4800 SEM. The fiber samples were glued on the
SEM sample stage using carbon tape. To avoid the SEM
charging effect, the acceleration voltage was fixed at 3 kV.
Experimental conditions
The range of conditions explored in the study is shown in
Table 1. These values represent those at the start of an ATF
experiment, and (as seen in the Results and Discussion section)
did not change much during the course of the run. Three
experiments were conducted with no cells present, at different
antifoam concentrations. A sufficient number of levels were
evaluated for each of the key variables: permeate flow rates
(170, 120, 70, and 2030 mL/min), cell concentrations (higher :
>45 3 106 cells/mL, medium: 3545 3 106 cells/mL, lower:
<35 3 106 cells/mL and no cells), cell viabilities (high: 90
100%, medium-high: 8090%, medium: 7080% and low:
<70%), and antifoam levels (0, 3,000, 4,500, and 6,000
ppm).The range of experimental cell concentrations (24.967.1
3 106 cells/mL) and cell viabilities (50.092.5%) is consistent
with what can be encountered near the end of typical perfusion
cultures. For the experiments that called for cell cultures with
lower viabilities, this was obtained naturally by extending the
culture duration until the concentration of essential nutrients
(glucose and glutamine) in the culture were depleted. Most of
the permeate flow rates used were at levels (120 and 170 mL/
min) considerably higher than those for a typical perfusion run
with an ATF2 cartridge to promote faster fouling rates; however a few of the runs used permeate flow rates (2030 mL/
min) more typical of actual perfusion operation. The ATF rate
was fixed at 0.5 L/min for this study. The normal range of
operation for the ratio of the ATF to Permeate flow rates is
20200 (per Refine technologies). With an ATF rate of 0.5 L/
min, this ratio of ranged in this study then between 25 (at
20 mL/min permeate flow rate) and 3 (at 170 mL/min permeate flow rate). The higher permeate flow rates were of interest
in this study, to promote faster fouling rates and because of the
potential cost benefits associated with increasing and maximizing the capacity of the filters. The duration of the experiments
was between 5 min and 8 h, with most being less than 3 h.

Results and Discussion


Media-only runs
An experiment with only media (no cells or antifoam) at
a single ATF rate (0.5 L/min) and a single and typical (and
high, for this study) permeate flow rate of 120 mL/min
the new Figure 2 shows that there was little increase in
TMP or Rtot over an extended filtration time (4.5 h). A
TFF experiment at the same permeate flow rate provided a
similar result, indicating that the Rtot value for this ATF2
cartridge and this media combination can be approximated
as being relatively constant in the absence of cells and antifoam (over a several hour period, even at a high permeate
flow rate).

1294

Biotechnol. Prog., 2014, Vol. 30, No. 6

Figure 3. (a) Total cell concentration vs. time during the ATF experiments and (b) cell viability vs. time during the ATF experiments.

Cell viability during the experiments


To consider the effects of cell concentration and viability
on ATF fouling, the experimental system was devised such
that the cell viability would not change during the experiment as a result of nutrient depletion or excessive shear
rates. The results shown in Figures 3a,b indicate that this
was achieved. Each line represents a different experiment.
After an initial drop in total cell concentration (Figure 3a),
due to the cell solution from the spinner flask mixing with
the media that was in the ATF cartridge, the cell concentration remained relatively constant (as did the cell viability,
shown in Figure 3b) for the remainder of the runseven for
the few runs that lasted up to 8 h. The variability, within an
experiment, in total cell concentration and cell viability was
within the variability range for the assay (which is approximately 30%). Furthermore, there was little (<10%) change
in cell size and aggregation rates during the duration of the
experiments (data not shown), and LDH data (not shown)
confirmed the cell viability to be relatively constant during
the experiments. These results make sense in light of the relatively constant nutrient and metabolite (pO2, pCO2, glutamine etc.) levels that were measured during the experiments.
Cycle-averaged TMP results
For each of the experiments, cycle-averaged TMP and
Rtot were plotted vs. run time. These plots show the

increase in average TMP and Rtot values with time, as a


result of the deposition of biological material on the membrane, but do not show the fluctuations in TMP and Rtot
within a given cycle. From Figures 4a,b, it is apparent that
the permeate flow rate cannot be maintained by the permeate pump at the desired flow rate when the TMP exceeds
approximately 14 psi. This TMP value was observed to be
indicative of filter fouling for all of the runs in the
study, and hence deemed a critical fouling pressure. For
the flow rate used in this experiment (170 mL/min), this
critical TMP corresponds to a Rtot value of approximately
3 3 1012 m21. Interestingly, this Rtot value is quite comparable to a limiting resistance of a fouling layer that
resulted from the filtration of extracellular polysaccharides, as reported by Harcoat et al.16 The two runs represented by Figures 4a,b were both conducted at: ATF rate
of 0.5 lpm, permeate flow rate of 170 mL/min, and a cell
viability of 91% for the solution. The total cell densities
were different, however, in that the data from Figure 3a
was obtained with a cell density of 37 3 106 cells/mL vs.
28 3 106 for the data in Figure 4b. Interestingly, this difference did seem to effect the fouling rate, in that it took
approximately 320 min for the higher cell density solution
to drop to a permeate flow rate of 140 mL/min; whereas it
took significantly (and proportionately) longer (approximately 430 min) for the lower cell density to drop to a
permeate flow rate of 140 mL/min. This result suggests

Biotechnol. Prog., 2014, Vol. 30, No. 6

1295

Assessing a mechanism for membrane fouling (i.e.,


cartridge fouling)
In summary, the SEM images indicate that both cells and
micelles deposit on the inner membrane surface of the ATF
cartridge. Furthermore, most of the ATF runs with cells and/
or antifoam fouled within 0.53 h; whereas the TFF and
ATF runs with only media (no antifoam or cells) showed no
signs of fouling (see Figure 2). These results suggest that the
ATF membranes are fouling due to the cells and/or antifoam
either blocking pores or forming a cake layer. The comparable r-squared values in Figure 6 indicate that both models
predict reasonably well the change in TMP vs. filtration
time. The SEM results, however, from filters that have
fouled, show a relatively thin (<10 mm) layer that: is not
likely to be made up of multiple cells (or micelles on top of
each other) and does not cover the entire surface of the
membrane. Furthermore, (unlike with deposited protein or
protein aggregates), permeate flow is unlikely to pass
through these attached cells and micelles as the path of least
resistance for this flow would be through open membrane
area. Therefore, additional deposition (of cells, micelles and
other biological material) would be more likely to occur on
open membrane area, as opposed to on top of already
attached particles (cell, micelles etc). Zydney and Ho12 notes
self-leveling of layers comprised of protein aggregates,
that occurs presumably due to this effect but also because
deposited material that is on top of existing attached particles (i.e., further out into the tangential flow field) is more
accessible to the flow and so more readily peeled-off (than
material being deposited on or closer to the membrane surface) by shear forces. For these reasons then, fouling in this
study was modeled by assuming pore blockage alone as the
mechanism.

Figure 4.

(a) At higher cell density (37 3 106 cells/mL): TMP,


Rtot, and permeate flow vs. time and (b) at lower cell
density (28 3 106 cells/mL): TMP, Rtot, and permeate flow vs. time.

that the fouling rate increases with the cell density of the
solution, when all other variables are the same.
SEM analysis of filtration membranes
The fibrous structure of pristine or unused membranes is
evident under SEM (Figure 5a). It was evident from looking at membranes that had been exposed to cells, that cells
had been deposited on the surface. In general, most of
these membranes had fouled, and the samples indicated
that much but not all of the surface was covered with a
(cell containing) cake layer. Figure 5b shows a section of
membrane surface that has a cake layer with several
hills that correspond to the size (approximately 10 mm)
of the cells used in the experiments. Two of the filtration
experiments were run with only media and antifoam, but
no cells were present. According to the vendor, the antifoam should form micelles in the media that are 2540 mm
in diameter. Figure 5c shows a large section of the inner
lumen membrane surface from an experiment with media
and antifoam (no cells present), on which particles are
deposited. At closer magnification (Figure 5d), these particles are seen to be approximately 3040 mm and are
assumed to be micelles.

Developing a model that considers pore blockage during


ATF cell filtration
Based on the previous sections, it was assumed that either
cells or antifoam micelles were blocking pores in the ATF
system studied here. The pore blockage model suggests that
the slope of the 1/DP vs. V plot is proportional to a fouling
coefficient (r), and the inverse of; the fluid viscosity (l), the
permeate flow rate (Q) and the resistance of a new membrane (Rm)
1=DP51=DPo 2rV=Rm lQ

(2)

17

Hermia suggested that, for dead-end filtration, r is equal


to the product of the number concentration (C) of the particles blocking the pores multiplied by the cross-sectional
area (Ap) of one of these particles. Because of the cross-flow
velocity in the ATF system, 1/DP in Eq. 2 is better represented by 1/TMP and it is proposed here that the relationship
from Hermia17 is modified to include a sticking factor (a)
such that
r5aCAp

(3)

This sticking factor could presumably depend on factors


including: the type of particle (i.e., cell, antifoam micelle,
etc.) being deposited on the membrane surface, the magnitude of the cross-flow velocity, flow features associated with
the changing of flow direction in an ATF system, as well as
cell viability. In this study, the ATF rate was kept constant,

1296

Biotechnol. Prog., 2014, Vol. 30, No. 6

Figure 5. (a) SEM image: inner surface of a new fiber, (b) SEM image: inner surface, covered with cells after a perfusion run, (c)
SEM image: inner lumen surface, covered with antifoam micelles after ATF run, and (d) SEM image: inner lumen surface,
close up of micelles on membrane surface.

decrease in TMP, at a constant permeate flow rate, during


this phase is due to an decrease in Rtot resulting from a net
removal of biological material (i.e., cells and/or antifoam)
from the membrane surface back into solution (also per Eq.
1). Figure 7b illustrates this effect for one cycle, by plotting
in the form (1/TMP vs. V) consistent with the pore blockage
model. The slope for the entire run is an average of the
slopes across the many cycles within an experimental run. If
it is assumed that there are only cells (i.e., no antifoam
micelles in solution), then it can be shown from Figure 7b
that for a given cycle the overall slope (Slopenet-cycle) can be
related to the slopes during the two phases by
Slopenet-cycle 5 slopeexhaust2phase 1 slopepressure
Figure 6. Evaluation of fouling mechanisms: pore blockage
and cake formation.

(4)

As [(CAp)/(RmlQ)] is not expected to change in the pressure vs. exhaust phases, Eqs. 2, 3, and 4 can be combined to
obtain
acells

and so only the relationship between particle type and cell


viability on the sticking factor was investigated.
Finally, it is important to assess the relative contributions
of the pressure and exhaust phases of the ATF cycle. The
data presented in Figure 7a is typical of that seen in all of
the ATF runs in this study. The TMP increases during the
exhaust phase of the cycle, as the cell solution is pulled
from the bioreactor into the ATF cartridge. This increase in
TMP, at a constant permeate flow rate, during this phase is
due to an increase in Rtot resulting from a net deposition of
biological material (i.e., cells and/or antifoam etc.) on to the
membrane surface per Eq. 1. The TMP decreases during the
pressure phase of the cycle, as the cell solution is pushed
from the ATF cartridge back into the bioreactor. This

phase =2

net-cycle

5acells

exhaust-phase 1acells pressure-phase =2

(5)

In Eq. 5, the term (acells_pressure-phase) is expected to be


negative (i.e., an anti-sticking factor) to account for the net
removal of biological deposits or decrease in TMP and Rtot
during this phase, and the magnitude of (acells_exhaust-phase)
must be greater than the magnitude of (acells_pressure-phase) to
have a net accumulation (or sticking) of biological material on to the membrane across a cycle.
Evaluating the mathematical model with cycle-averaged
experimental ATF data
Three experiments were run with only media and antifoam
(i.e., no cells present). Using the cycle-averaged data and the
mathematical model described in the preceeding section, the

Biotechnol. Prog., 2014, Vol. 30, No. 6

1297

Figure 7. (a) P2 sensor (see Figure 1 for sensor location) reading (psi-gauge) and TMP (psi) within three ATF cycles (Note: data
points are every 2 s) and (b) change in 1/TMP during the pressure and exhaust phases of the ATF cycle for a typical cycle.
[Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Figure 8. Cell membrane sticking factor (acells_net-cycle) vs.


cell.[Color figure can be viewed in the online issue,
which is available at wileyonlinelibrary.com.]

sticking factor for the antifoam micelles (aantifoam_net) was


determined to be approximately 2 3 1028 for all three experiments. For two of these runs, all variables were kept the same
(i.e., permeate flow rate of 120 mL/min, ATF rate of 0.5 lpm)
except antifoam concentration (4,500 and 6,000 ppm). Interestingly, the run at 4,500 ppm fouled after approximately 12
min and run at 6,000 ppm fouled after approximately 8 min
which is consistent with that predicted by Eqs. 2 and 3. From
the remaining runs, where cells and media were used (i.e., no

antifoam present), the overall cycle-averaged sticking factor


for the cells (acells_net) was determined to increase sharply as
the viability of the cell solution decreased per Figure 8. This
might be explained by the fact that cell lysis can result in the
release DNA that could promote cells sticking to the membrane surface. The cycle-average (net) sticking factors for the
cells ranged from a low of 3.6 3 1025 for high viability
(92.5%) cells to a high of 6.5 3 1023 for low viability (50%)
cells. Interestingly, all of the sticking factors for the cells are
at least three orders of magnitude more than for the antifoam
micelles, which makes sense given that the (PES) membrane
is hydrophillic.
To test the ability of the model to predict the rise in TMP
(and Rtot) vs. runtime during an ATF run (and ultimately
predict when the critical fouling TMP value of approximately 14 psi will occur), two ATF experiments were conducted in which the solution fed to the ATF contained both
cells and antifoam. In the first experiment, the solution consisted of a cell concentration of 40 3 106 cells/mL and an
antifoam level of 3,000 ppm and the permeate flow rate was
set at 120 mL/min. As a result of the lower viability (77%)
of the cells in this experiment, along with the presence of
antifoam, the filter fouled very quicklyafter only 5 min
(Figure 9a). The model, using a cell sticking factor of 1.7 3
1023 from Figure 8, predicted the rise in TMP over the 5
min quite wellespecially when the model incorporated the
effect of the cells as well as the antifoam. The model predicted (Figure 9b) for this run that most of the biological
deposition on the membrane surface was cells and not antifoam, due to the high cell sticking factors associated with a
77% viable cell solution. In the second experiment, the solution again consisted of a cell concentration of 40 3 106

1298

Figure 9. (a) Model vs. expt (40 3 106 cells/mL, 3,000 ppm AF,
120 mL/min perm flow, 77% viability) and (b) model
prediction for membrane coverage (same conditions as
Figure 9a). [Color figure can be viewed in the online
issue, which is available at wileyonlinelibrary.com.]

cells/mL and an antifoam level of 3,000 ppm; however, the


permeate flow rate was set to a low value of 30 mL/min. As
a result of the high viability (92.5%) of the cells in this
experiment, this filter ran over 3 h without fouling (Figure
10a). The model, using a cell sticking factor of 3.5 3 1025
from Figure 8, predicted the rise in TMP over the 3 h quite
well. Contrary to the first experiment, the model predicted
for this run (Figure 10b) that most of the biological deposition on the membrane surface was antifoam and not cells,
due to the low cell sticking factors associated with a 92.5%
viable cell solution. The minor differences between the
model prediction and the experimental results, in Figures 9a
and10a, might indicate some degree of interaction between
the cells and the antifoam that is not accounted for by the
model.

Using the model to understand fouling within a single ATF


cycle
Analysis of the variation in TMP across both the pressure
and exhaust phases of the ATF cycles revealed interesting
results. First, the overall slope of the 1/TMP vs. V plot is
steeper for low viability solutions presumably because of
their higher overall or net sticking factors (acells_net-cycle)
per Figure 8. On analysis of multiple ATF cycles from

Biotechnol. Prog., 2014, Vol. 30, No. 6

Figure 10. (a) Model vs. expt (40 3 106 cells/mL, 3,000 ppm
AF, 30 mL/min perm flow, 92.5% viability) and (b)
model prediction for membrane coverage (same
conditions as Figure 10a).

experimental runs with lower and higher viability cell solutions; however, it became evident that the sticking factors
for the exhaust phase of a cycle (acells_exhaust-phase) are greater
for lower viability cell solutions, while the (anti)sticking factors for the pressure phase are greater for the higher viability
cells. This is evident from Figure 11, which shows one cycle
of data for a low (50%) cell viability solution (Figure 11a)
and a high (91%) cell viability solution (Figure 11b). The
slope during the exhaust phase (which is proportional to the
sticking factor) is steeper for the low viability cells
(20.0031) as opposed to the higher viability cells
(20.0027); whereas the slope during the pressure phase is
steeper for the high viability cells (0.0025) as opposed to the
lower viability cells (0.0016). This result suggests that there
is variation between low and high viability cells in the
degree to which they attach (during the exhaust phase) as
well as detach (during the pressure phase) from the
membrane.
It was also observed that, for most of the experiments, the
sticking factor for the exhaust cycle (acells_exhaust-phase)
increased as the cycle-averaged TMP increased during the
course of a given experiment. This corresponded to a greater
decrease in TMP during the exhaust phase near the end of
the experiment as compared to the beginning. Interestingly,
the (anti)sticking factor for the pressure phase (acells_pressurephase) increased by the same degree, resulting in the same net
sticking factor (acells_net-cycle, from Figure 8) at the beginning
of an experiment (i.e., where cycle-averaged TMP is low)
beginning) as at the end of an experiment (i.e., where cycle-

Biotechnol. Prog., 2014, Vol. 30, No. 6

1299

result of existing biological deposits on the membrane surface facilitating deposition of additional deposits.
Using the model to predict fouling rate in a perfusion cell
culture
A perfusion cell culture experiment was run to test the ability of the model to predict the rise in TMP vs. runtime. As
per Figure 12, the permeate flow rate was increased exponentially over the course of the 10 day run, per a predefined program. Samples were taken once per day for cell concentration
and viability, and those results indicate an exponential rise in
cell concentration (Figure 12). For a 1 h period every day, at
the time when samples were taken, TMP vs. time was
recorded and an average TMP for that period was obtained.
Antifoam is added when needed, and for this run was added
on day one and day six. A stepwise representation of the permeate flow rate was used (Figure 12) such that the model,
based on a constant permeate flow rate assumption, could be
used across 24 h segments of time. The model accounted for
the concentrations of cells and antifoam in solution on a given
day. The model also used the appropriate values of acells_netcycle from Figure 8, as the viability changed during the experiment (ranging from 8292%). In this experiment, the experiment concluded before the ATF cartridge fouled; however,
Figure 13 shows that the model predicted the rise in TMP
between days 710 quite well, as the system began to
approach the critical fouling TMP (of approximately 14 psi).
Figure 11. (a) Experimental TMP data within an ATF cycle:
low (50% viable cells) and (b) experimental TMP
data within an ATF cycle: low (91% viable cells).
[Color figure can be viewed in the online issue,
which is available at wileyonlinelibrary.com.]

Figure 12.

Perfusion experiment: Model vs. experiment al prediction of TMP increase with time. [Color figure
can be viewed in the online issue, which is available
at wileyonlinelibrary.com.]

averaged TMP is higher). Larger pressure-phase (anti)sticking factors, seen near the end of the experiments, correspond
to an enhanced rate of removal of biological deposits from
the membrane which is consistent with Stressmans observation9 that the removal rate of deposited biological solids
from a membrane surface as a result of crossflow (during
TFF) increases when there is more material deposited on the
membrane. The increase in deposition rate during the
exhaust cycle (i.e., acells_exhaust-phase) near the end of the
experiments is more difficult to explain, and might be a

Conclusions
In this study, an experimental setup was devised that
allowed for determination of how long it took for fouling to
occur for given mammalian cell culture cell densities and viabilities as permeate flow rate and antifoam concentration was
varied. The experimental results indicate, in accordance with
DArcys law, that the average resistance to permeate flow
(across a cycle of operation) increases as biological material
deposits on the membrane. SEM images of the post-run filtration surface indicated that both cells and antifoam micelles
were depositing on the membrane. A unique mathematical
model, based on the assumption that fouling was due to pore
blockage from the cells and micelles in combination, was
devised that allowed for estimation of sticking factors for the
cells and the micelles that were used to accurately predict the
fouling rate of ATF cartridges as a function of cell and antifoam concentration, cell viability, and permeate flow rate for
a perfusion cell culture run. Future work might include evaluating the effects of changing the ATF rate and evaluating the
fluid dynamic mechanisms responsible for the removal of
some biological material from the membrane surface during
the pressure phase of the ATF cycles, along with determining
how ATF behavior (i.e., sticking factor values) varies with
different cell and membrane types.

Acknowledgment
We want to thank Dr. Ronald Balsamo, Director of the
Villanova University microscope facility, for assistance with
cutting the fibers using liquid nitrogen.

Literature Cited
1. Tang Y, Hamel J. Perfusion culture of hybridoma cells with a
wave bioreactor system. Biotechnol Prog. 2007;23:255264.

1300
2. Furey J. Scaleup of a cell culture perfusion processa low
shear filtration system that inhibits filter membrane fouling.
Genet Eng News. 2002;22:6262.
3. Stressman, M. Membrane fouling in constant permeate flux
cross-flow microfiltration of biological solutions. PhD Thesis.
University of Waterloo, Canada; 2008.
4. Li H, Fane A, Coster H, Yigneswaran S. Direct observation of
particle deposition on the membrane surface during crossflow
microfiltration. J Membr Sci. 1998;149:8397.
5. Escalade L, Carrel S, Peringer P. Influence of the screen material on the fouling of spin filters. Biotechnol Bioeng. 1991;38:
159168.
6. Mercille S, Johnson M, Lemieux R, Masse B. Filtration based
perfusion of hybridoma cultures in protein free medium. Biotechnol Bioeng. 1994;43:833840.
7. Yamagiwa K, Kobayashi H, Onodera M, Ohkawa, A. Antifoam
fouling and its reduction by surfactant precoat treatment of polysulphone ultrafilter. Biotechnol Tech. 1994;8:267270.
8. Hlavacek M, Bouchet F. Constant flowrate blocking laws and an
example of their applications to dead-end microfiltration of protein solutions. J Membr Sci. 1993;82:285295.
9. Stressman M, Moresoli C. Effect of pore size, shear rate and
harvest time during the constant permeate flux microfiltration of
CHO cell culture supernatant. Biotechnol Prog. 2008;24:890
897.
10. Belfort G, Davis R, Zydney A. The behavior of suspensions and
macromolecular solutions in crossflow microfiltration. J Membr
Sci. 1994;96:158.

Biotechnol. Prog., 2014, Vol. 30, No. 6


11. Bolton G, LaCasse D, Kuriyel R. Combined models of membrane fouling: development and application to microfiltration
and ultrafiltration of biological fluids. J Membr Sci. 2006;277:
7584.
12. Ho C, Zydney A. Transmembrane pressure profiles during constant flux microfiltration of bovine serum albumin. J Membr Sci.
2002;209:363377.
13. Hughes D, Field R. Crossflow filtration of washed and
unwashed yeast suspensions at constant shear under nominally
sub-critical conditions. J Membr Sci. 2006;280:8998.
14. Clinke M, Molleryd C, Zhang M. Very high density of CHO
cells in perfusion by ATF or TFF in WAVE Bioreactor. Part I.
Effect of the cell density on the process. Biotechnol Prog. 2013;
29:754767.
15. Clinke M, Molleryd C, Samani P. Very high density of CHO
cells in perfusion by ATF or TFF in WAVE Bioreactor. Part II:
applications for Antibody production and cryopreservation. Biotechnol Prog. 2013;29:768777.
16. Harcoat C, Jaffrin M, Bouzerar R, Courtois J. Influence of Fermentation conditions and microfiltration processes onmembrane
fouling during recovery of glucuronane polysaccharides from
fermentation broths. Biotechnol Bioeng. 1999;65:500511.
17. Hermia, J. Constant pressure blocking filtration lawsapplications to power law non-newtonian fluids. Trans Inst Chem Eng.
1982;60:183187.
Manuscript received May 12, 2014, and revision received Jul. 22,
2014.