You are on page 1of 10

Protein Fouling During Microfiltration:

Comparative Behavior of Different


Model Proteins
Sean T. Kelly, Andrew L. Zydney

Department of Chemical Engineering, University of Delaware,


Newark, Delaware 19716; telephone: (302) 831-2399; fax: (302) 831-1048;
e-mail: zydney@che.udel.edu
Received 21 June 1996; accepted 15 November 1996

Abstract: Recent studies of protein fouling have provided


considerable insight into both the underlying fouling
mechanisms and the mathematical description of the flux
decline. However, most of the data have been obtained
with a single model protein, making it difficult to generalize the results to commercially relevant process streams.
Experiments were thus performed using a range of proteins with different physicochemical characteristics to determine the relationship between the protein structure
and fouling behavior. Fouling in these systems occurred
by two distinct mechanisms: deposition of large protein
aggregates and chemical attachment of native proteins to
the growing deposit. The chemical attachment generally
occurred via the formation of intermolecular disulfide
linkages involving a free sulfhydryl group in the native
protein. Proteins without a free sulfhydryl group were
typically unable to form these intermolecular linkages.
The quasi-steady flux for the different proteins was proportional to the square of the protein surface charge density, consistent with a model in which protein deposition
occurs when the drag force on the proteins associated
with the convective filtrate flow is sufficient to overcome
electrostatic repulsive interactions. These results clearly
demonstrate the importance of the protein structure,
charge, and reactivity in determining the rate and extent
of protein fouling during microfiltration. 1997 John Wiley & Sons, Inc. Biotechnol Bioeng 55: 91100, 1997.

Keywords: microfiltration; membrane; protein; fouling;


filtration

INTRODUCTION
Microfiltration is used extensively for the clarification
and sterilization of many biological solutions, including
a variety of beverages, fermentation broths, and therapeutic proteins. Protein fouling is a critical factor in
many of these processes, resulting in a substantial reduction in membrane performance and a significant loss of
valuable product.
Recent studies have provided considerable insights
into the underlying fouling mechanisms as well as the
proper mathematical description of the flux decline in
these systems. Kelly et al. (1993) have shown that the
Correspondence to: A. L. Zydney
Contract grant sponsor: National Science Foundation

1997 John Wiley & Sons, Inc.

initial flux decline during the stirred cell filtration of


bovine serum albumin (BSA) was associated with the
deposition of large protein aggregates on the membrane
surface. Similar results were obtained by Chandavarkar
(1990) and Meireles et al. (1991) in cross-flow microfiltration experiments employing BSA. Scanning electron
micrographs clearly confirm the presence of BSA aggregates on the surface of fouled microfiltration membranes (Belfort et al., 1994; Kelly and Zydney, 1994;
Kim et al., 1993). Mathematical descriptions of protein
fouling have generally employed one of the standard
fouling models based on pore blockage, pore constriction, and/or cake formation (Belfort et al., 1994; Hlavacek and Bouchet, 1993).
Several recent studies have provided evidence that
BSA fouling in microfiltration is best described using a
two-stage fouling model in which the initial fouling is
described using some type of pore blockage model with
the longer-term fouling due to external cake formation
model (Jonsson et al., 1996; Tracey and Davis, 1994).
Kelly and Zydney (1995) developed a mathematical for
BSA fouling accounting for two distinct fouling mechanisms: pore blockage associated with aggregate deposition and chemical attachment of native BSA to the
growing deposit. Chemical attachment of native BSA
was shown to occur via the formation of an intermolecular disulfide linkage. This model is discussed in more
detail subsequently.
Although these studies provide considerable insights
into protein fouling during microfiltration, they were all
performed with the same protein, bovine serum albumin. BSA has been used as a model protein in these
systems largely because of its ready availability and low
cost. However, the physicochemical properties of BSA
are quite unique (Foster, 1977), making it difficult to
extrapolate from these results to the behavior of other,
more commercially relevant, proteins. Experimental
studies of the fouling characteristics of other proteins
have generally been more qualitative in nature, or they
have focused primarily on the effects of protein size
during fouling of partially retentive membranes (Pradanos et al., 1996). There is currently very little funda-

CCC 0006-3592/97/010091-10

mental understanding of the relationship between the


actual protein structure and physicochemical properties
and the observed fouling behavior during protein microfiltration.
In this study, we examined fouling of 9 different
model proteins with a range of physicochemical characteristics. Data were obtained in a stirred cell apparatus
with nitrogen pressurization to avoid the additional
complexities associated with pumping-induced protein
denaturation (Meireles et al., 1991). The flux decline
data were analyzed using the dual-mode fouling model
developed by Kelly and Zydney (1995). These results
were then used to explore the role of the protein structure and chemistry in determining the overall rate of
flux decline and the magnitude of the quasi-steady flux
during protein microfiltration.

MATERIALS AND METHODS


Proteins
Experiments were performed using bovine serum albumin (BSA), cysteinylated BSA, ovalbumin, lysozyme,
pepsin, myoglobin, a-lactalbumin, and b-lactoglobulin,
all obtained from Sigma Chemical Co. (St. Louis, MO).
Sigma Catalog numbers, source, and physical property
data for the different proteins are provided in Table I.
Protein solutions were prepared by carefully dissolving the lyophilized protein powder in a phosphate buffer
solution (pH 6.88) at room temperature. No stirring was
used during the dissolution of the protein. The phosphate buffer solutions were composed of 0.025 M
KH2PO4 , 0.025 M Na2HPO4?7H2O, and 0.001 M EDTA
(tetrasodium salt) and were prepared by dissolving preweighed quantities of the salts in the desired volume of
deionized distilled water (resistivity . 18 Mohm-cm).
Buffer solutions were prefiltered through 0.2-mm-poresize Supor-200 membranes (Gelman Sciences, Ann
Arbor, MI) prior to use. The protein solutions were
then prefiltered through 10-mm coarse filters (Gelman
Sciences) to remove large particulates and any undis-

solved protein. In order to avoid bacterial contamination, protein solutions were stored at 48C and used
within 48 h of preparation.
In some experiments, the protein solutions were prefiltered through 100,000 (100K) molecular weight cutoff (MWCO) Omega polyethersulfone ultrafiltration
membranes (Filtron Technology, Northborough, MA)
to remove protein aggregates and oligomers prior to
use in the microfiltration experiments. This prefiltration
was performed in a 64-mm-diameter stirred ultrafiltration cell (model 8200, Amicon, Beverly, MA) using nitrogen pressurization (36 psi) with the stirring speed
maintained between 90 and 120 rpm. The feed solution
reservoir and filtrate collection flasks were maintained
at 588C throughout the prefiltration to minimize the
possibility of protein denaturation. The prefiltration was
typically performed using protein solutions having
about twice the desired (final) concentration. The filtrate was then diluted with phosphate-buffered saline
(PBS) (prefiltered with a clean membrane of identical
MWCO) to obtain the desired bulk protein concentration.
Protein concentrations were determined by reacting
the proteins with copper to form a purple copper
protein complex. Two milliliters of Total Protein Reagent 541 (Sigma Chemical Co.) were added to a 4-mL
sample cuvette containing 800 mL of protein solution.
The solutions were mixed by gentle inversion and allowed to sit for 15 min. The absorbance was then measured at 540 nm using a Lambda 4B spectrophotometer,
with the actual protein concentrations determined using
appropriate calibration curves. Concentrations of BSA
and cys-BSA were determined by reaction with bromcresol green (Sigma Chemical Co.) with the absorbance
measured at 628 nm. The accuracy of the concentration
measurements was approximately 0.1 g/L.
Proteins were also analyzed using discontinuous pH
sodium dodecyl sulfatepolyacrylamide gel electrophoresis (SDSPAGE). Gradient gels of approximately
3.520% acrylamide were cast in a model SE250 minigel
apparatus (Hoefer Scientific Instruments, San Francisco, CA). The upper and lower buffer reservoirs were

Table I. Summary of protein physical properties.

Protein

Sigma
catalog
number

Source

MW
(kD)

SSa

SHb

pI

BSA
Cys-BSA
Ovalbumin
Lysozyme
b-Lactoglobulin
a-Lactalbumin
Pepsin
Myoglobin

A7906
A0161
A5503
L6876
L6879
L5385
P6887
M0630

Bovine serum
Bovine serum
Chicken egg
Chicken egg
Cow milk
Cow milk
Pig stomach
Horse muscle

67
67
40
14
18
15
36
17

17
18
1
4
2
4
3
0

1
0
4
0
1
0
0
0

4.7
4.7
4.6
11.0
5.3
5.1
1.0
7.0

a
b

92

Number of internal disulfide linkages.


Number of free thiol groups.

BIOTECHNOLOGY AND BIOENGINEERING, VOL. 55, NO. 1, JULY 5, 1997

set to between pH 8.3 and 8.4, while the pH of the


running and stacking gels was set to pH 8.8 and 6.8,
respectively. A constant current of 15 mA was maintained by gradually increasing the voltage from approximately 110 V to 210 V over the course of the electrophoresis (approximately 8090 min). Molecular weight
markers were obtained from Sigma Chemical Co.
(M4038). Gels were stained with Coomassie blue and
were quantified using a model SI personal densitometer
(Molecular Dynamics, Sunnyvale, CA). Additional details on the electrophoresis are provided by Bollag and
Edelstein (1991).
Filtration Experiments
Filtration experiments were performed using Durapore
0.22-mm-pore-size polyvinylidene fluoride (PVDF)
membranes (lot nos. H1AM92208A and H35M00899,
Millipore Corp., Bedford, MA). The Durapore PVDF
membranes have been rendered hydrophilic via surface
modification by the manufacturer.
All filtration experiments were conducted using a 25mm stirred ultrafiltration cell (model 8010, Amicon
Corp.) connected to a nitrogen-pressurized acrylic solution reservoir. The stirred cell and solution reservoir
were initially filled with PBS, with the saline flux measured as a function of time at a constant pressure until
steady state was attained (usually after 20 min). The
solution reservoir and stirred cell were then emptied of
PBS and rapidly refilled with a protein solution. The
system was repressurized (within 2 min of the end of
the saline filtration), and the stirring speed set to 600
rpm using a Strobotac type 1531-AB strobe light (General Radio Co., Concord, MA). The solution flow rate
was measured using timed collection. Protein concentrations in the filtrate and stirred cell (bulk) were evaluated
periodically over the course of the experiment. All
experiments were conducted at room temperature
(22 6 28C).
RESULTS
Experimental data for the filtration of 2 g/L solutions of
lysozyme, pepsin, and cysteinylated-BSA at a constant
pressure of 14 kPa (2 psi) are shown in Figure 1. The
filtrate flux data have been normalized by the initial flux
at the start of the filtration ( J0), with the values of J0
being within 5% of the steady-state saline flux evaluated
just prior to the protein filtration. The rate of flux decline
(lower panel) was evaluated directly from the filtrate
flux data as
K52

1 dJ
J dt

(1)

All derivatives were evaluated numerically using a central difference representation that accounted for the

Figure 1. Normalized filtrate flux (top panel) and rate of flux decline
(bottom panel) for the filtration of 2 g/L solutions of lysozyme, pepsin,
and cysteinylated BSA through 0.22-mm PVDF membranes at 14 kPa
and 600 rpm. Solid curves are model calculations.

nonuniform time spacing of the flux data and was accurate to O(Dt 2). The solid curves are model correlations
which are discussed in more detail subsequently. The
most rapid flux decline was obtained with pepsin, while
the least rapid decline was obtained with lysozyme. The
fouling time, defined as the time required for the filtrate
flux to decline to 20% of its initial value, ranged from
t 20 5 12 min for pepsin to t 20 5 54 min for cysBSA
and t 20 . 160 min for lysozyme. Repeat measurements
performed under identical experimental conditions
yielded filtrate flux data that were within 15% of the
values shown in Figure 1 throughout the filtration, with
the calculated values of the fouling parameters (discussed subsequently) varying by less than 30%. However, the averaged flux data for multiple runs could not
be used to evaluate the rate of flux decline (K) due to the
large errors introduced by the numerical differentiation.
Thus, all of the flux decline calculations shown in this

KELLY AND ZYDNEY: PROTEIN FOULING DURING MICROFILTRATION

93

study were for filtrate flux data obtained in a single


experimental run.
The rate of flux decline data for lysozyme and cysBSA were qualitatively very similar. Both of these proteins show an essentially constant value of K over most
of the filtration, with only a slight decrease seen at very
long times. The rate of flux decline for pepsin decreases
continuously with time and appears to decay to zero at
long times as the flux approaches a quasi-steady value.
The flux decline seen with these proteins was directly
due to the deposition of large protein aggregates that
were present in the initial protein preparations. These
protein aggregates could be eliminated by prefiltration
of the protein solutions through 100,000 MWCO membranes (confirmed by SDSPAGE), and this prefiltration step essentially eliminated the flux decline. For
example, Figure 2 shows experimental data for microfiltration of 5 g/L 100K-prefiltered solutions of cys-BSA
and lysozyme at a constant pressure of 14 kPa. The cysBSA showed essentially no flux decline over the 2-h
filtration. A very small (,20%) flux decline was seen
with lysozyme. This was due almost entirely to a reduction in Jv right at the start of the run and might simply
reflect an overestimation of the initial flux ( J0) or the
presence of a slight membrane compaction upon repressurization.
Experimental data for the filtration of b-lactoglobulin, BSA, and ovalbumin, conducted under conditions
identical to those in Figure 1, are shown in Figure 3.
The data are again plotted as the normalized filtrate
flux (upper panel) and the rate of flux decline (lower
panel). The fouling times for these proteins ranged from
about t 20 P 8 min for b-lactoglobulin to t 20 5 9 min for
ovalbumin and t 20 5 25 min for BSA. The initial flux
decline for these proteins was again due to the deposi-

Figure 2. Normalized filtrate flux during the filtration of 100K-prefiltered solutions of lysozyme and cysteinylated BSA through 0.22-mm
PVDF membranes at 14 kPa and 600 rpm.

94

Figure 3. Normalized filtrate flux (top panel) and rate of flux decline
(bottom panel) for the filtration of 2 g/L solutions of b-lactoglobulin,
BSA, and ovalbumin through 0.22-mm PVDF membranes at 14 kPa
and 600 rpm. Solid curves are model calculations.

tion of large aggregates, and this fouling was almost


completely eliminated by prefiltering the protein solutions through a 100K membrane, as shown in Figure 4
(the flux decline seen at long times for the 100K-prefiltered BSA and ovalbumin is discussed in more detail
subsequently). In sharp contrast to the data in Figure
1, the rate of flux decline for these proteins increased
significantly at short times, passing through a very distinct maximum before declining at long times. The maximum value of the rate of flux decline (K max) was greatest
for b-lactoglobulin, although the ratio K max /K 0 was
greatest for BSA. The time at which the maximum occurred was also different for the different proteins, ranging from about 5 min for b-lactoglobulin to 25 min
for BSA.
The observed increase in the rate of flux decline at
the start of the microfiltration has been discussed in
detail by Kelly and Zydney (1995) with respect to BSA
and has been attributed to the presence of two distinct

BIOTECHNOLOGY AND BIOENGINEERING, VOL. 55, NO. 1, JULY 5, 1997

Figure 4. Normalized filtrate flux during the filtration of 100K-prefiltered solutions of BSA and ovalbumin through 0.22-mm PVDF membranes at 14 kPa and 600 rpm.

fouling mechanisms: aggregate deposition and chemical


attachment of native protein to the growing deposit.
Kelly and Zydney hypothesized that the increase in
K at short times was specifically due to the chemical
attachment of native protein to the deposit. At the start
of the filtration, the membrane is free of any protein
deposit; thus the rate of chemical addition of protein to
the existing deposit is necessarily zero. As deposition
continues, the rate of chemical attachment increases,
with the net result that the overall rate of flux decline
also increases. The solid curves in the upper and lower
panels of Fig. 3 represent the predictions of this dualmode fouling model and are discussed below.
In order to understand the differences in the fouling
characteristics of the different proteins, it is necessary
to examine the structural properties of these macromolecules. In particular, Table I shows the number of free
sulfhydryl (SH) groups. All of the proteins that have
at least one free sulfhydryl group (BSA, ovalbumin, and
b-lactoglobulin) showed a characteristic maximum in
the rate of flux decline (Fig. 3). In contrast, those proteins without any free sulfhydryl groups (cys-BSA, lysozyme, and pepsin from Fig. 1) showed either constant
or continually decaying values of K throughout the filtration. These results clearly indicate that the increase
in K at short times, and thus the rate of chemical addition of native protein to the growing deposit, is directly
related to the presence of a free sulfhydryl group in the
protein structure. The one exception to this behavior
was myoglobin, which appeared to add to the growing
deposit by a very different mechanism (not involving a
free sulfhydryl group). This is discussed in more detail subsequently.
The chemistry of the sulfhydryl group has been discussed in some detail by Cecil and McPhee (1959),

Torchinsky (1981), and Jocelyn (1972). Free sulfhydryls are known to participate in thiol oxidation and
thioldisulfide interchange reactions, both of which
result in the formation of intermolecular disulfide
linkages, which can in turn generate large aggregate
structures (Huggins et al., 1951; Liu et al., 1991). The
rate of these reactions is determined by the number,
accessibility, and reactivity of the free thiols and the
reacting disulfides.
Kelly and Zydney (1994) showed that the thiol
disulfide interchange reaction plays a critical role in the
fouling characteristics of BSA solutions; changes in pH
or the addition of chelating agents altered the rate of
these intermolecular reactions and significantly affected
the rate of flux decline. The results in Figures 1 and 3
for cys-BSA and BSA clearly confirm this behavior.
The cys-BSA is prepared directly from the normal BSA
product and is reported to differ from the normal BSA
only in the capping of the free sulfhydryl group on
the native protein. This cysteinylation eliminates the
reactivity of this sulfhydryl group, thereby eliminating
the increase in the rate of flux decline seen during protein microfiltration.
In order to further quantify the fouling behavior of
the different proteins, the filtrate flux data in Figures 1
and 3 were fit to the dual-mode fouling model developed
by Kelly and Zydney (1995). This model accounts for
two distinct mechanisms of protein fouling: pore
blockage associated with the deposition of large protein
aggregates and the chemical attachment of native (monomeric) protein to the growing deposit. The rate of
change in the number of open pores, Nopen , is expressed as
dNopen
5 2aC bAJ v,open 2 bC bAJv,open(N0 2 Nopen)
dt
(2)
where N0 is the total number of pores, A is the membrane area, C b is the bulk protein concentration, and
Jv,open is the filtrate flux through the open pores. The
parameter a is related to the rate of aggregate deposition as well as the efficiency of pore blockage. Here, a
is thus assumed to be proportional to the fraction of
the total protein present as aggregates in the bulk solution. The parameter b describes the rate of chemical
addition of the native protein to the growing deposit.
This rate is assumed to be proportional to both the
convective transport of native protein toward the membrane (Jv,openC b) and the number of aggregates already
on the membrane surface. Since the protein deposit
is permeable to the filtrate, Kelly and Zydney (1995)
evaluated the filtrate flow through the blocked pores
using a classical Darcys law model accounting for the
resistance of the membrane and protein layer in series.
The total flux through the open and blocked pores
was evaluated by appropriate integration of Eq. (2),
yielding (Kelly and Zydney, 1995)

KELLY AND ZYDNEY: PROTEIN FOULING DURING MICROFILTRATION

95

Jv
5
J0

G DG

Jss
bN0
bN0
aA
112
1 2 exp J0C b
11
t
a
J0
N0
a

GD

bN0
bN0
aA
11
t
1 exp J0C b
a
N0
a

(3)
where Jss is the steady-state filtrate flux obtained at long
times. The instantaneous rate of flux decline (K) can
be calculated by direct substitution of Eq. (3) into Eq.
(1). Equation (3) is clearly phenomenological, and it
provides no insights into the actual location of the deposited protein on (or in) the membrane. However, this
model does provide a quantitative description of the
relative importance of the two fouling steps as characterized by the values of the parameters a and b.
The best fit values of the model parameters (a/N0 ,
b, and Jss) for the different protein filtration experiments
were evaluated by minimizing the sum of the squared
residuals between the experimental data for the filtrate
flux and the model [Eq. (3)]. The results are summarized
in Table II, with the error estimates determined directly
from the least-squares optimization. The model calculations (solid curves in both the upper and lower panels
of Figs. 1 and 3) were in excellent agreement with the
data for both Jv /J0 and K. The model properly captures
the initial increase in K, the rise through a maximum,
and the subsequent decline in K seen in Figure 3 for
b-lactoglobulin, BSA, and ovalbumin, and it properly
describes the continual decline in K seen in Figure 1
for lysozyme, pepsin, and cys-BSA.
All of the proteins possessing at least one free sulfhydryl (b-lactoglobulin, ovalbumin, and BSA) had relatively large values of the parameter b, which characterizes the rate of chemical addition of native protein to the
growing deposit. As discussed previously, this reaction
apparently occurs via the thioldisulfide interchange reaction and is thus dependent on the presence of a free
thiol group in the native protein structure. The actual
value of b should be related to the number, accessibility,
and reactivity of the free thiols and reactive disulfides.
This is discussed in more detail below.
In contrast to the results for b-lactoglobulin, ovalbumin, and BSA, the best fit value of the parameter b
for lysozyme was b 5 0.002 6 0.04 3 1023 g21, which

is not statistically different than zero. The best fit values


of b for cys-BSA and pepsin (and also for the filtration
of a-lactalbumin) were actually slightly negative (with
absolute values less than 0.3 g21), although in each case
these values were not statistically different than zero.
The b values for these four proteins were thus set equal
to zero and the model was refit to the data using only
two adjustable parameters, a/N0 and Jss . The values of
these recalculated parameters (shown in Table II) were
all within 10% of the values originally determined from
the three-parameter fit. Thus, those proteins having no
free thiols (lysozyme, cys-BSA, a-lactalbumin, and pepsin) all had b 5 0. The absence of a free thiol thus
completely eliminated the chemical addition of native
protein to the deposit, leading to the absence of any
initial increase in the rate of flux decline during the
filtration of these protein solutions.
The one exception to the behavior discussed above
was myoglobin (from horse skeletal muscle). The rate
of flux decline for myoglobin was very similar to that
of BSA and ovalbumin, with K showing a significant
increase at short times, passing through a maximum,
and then decaying at long times. Thus, the flux decline
for myoglobin displayed the same characteristic behavior as was seen for the proteins that were able to chemically add to the growing deposit via the thioldisulfide
interchange reaction. In fact, the value of b for myoglobin was the largest of all of the proteins examined in
this study, despite the complete absence of any free
sulfhydryls. Myoglobin does have a large hydrophobic
region surrounding its iron binding site, and it is possible
that these regions allow the native myoglobin to attach
to the growing deposit via strong hydrophobic interactions. The presence of EDTA, a metal-chelating agent,
in the phosphate buffer solution used in these experiments may have facilitated these intermolecular interactions. Additional studies with different myoglobin preparations would clearly be required to fully identify the
fouling mechanism for this particular protein.
To obtain additional insights into the fouling behavior
of BSA, ovalbumin, and lysozyme, a series of sequential filtration experiments were performed. The 0.22mm Durapore membranes were first used to filter a
standard (unfiltered) solution of the protein until the
flux decreased by about 10%. This corresponded to ap-

Table II. Best fit values of model parameters (a/N0), b, and Jss.
Protein
BSA
Cys-BSA
Ovalbumin
Lysozyme
b-Lactoglobulin
a-Lactalbumin
Pepsin
Myoglobin

96

a/N0
(g21)

b
(g21)

Jss
(mm/s)

q
(pH 7)

Surface area
(m2 3 1016)

6
6
6
6
6
6
6
6

2.6 6 0.02
0
3.9 6 0.6
0
5.0 6 0.9
0
0
8.5 6 0.8

11.6
11.2
11.7
15.2
10.8
15.9
7.3
7.0

218
218
210
19
211
28
214
0

1.63
1.63
0.97
0.53
0.94
0.53
0.74
0.47

0.4
0.7
2.3
0.3
4.7
0.7
5.8
1.0

0.02
0.03
0.1
0.03
0.4
0.01
0.2
0.03

BIOTECHNOLOGY AND BIOENGINEERING, VOL. 55, NO. 1, JULY 5, 1997

proximately 30 mL for ovalbumin (with C b 5 1 g/L),


50 mL for BSA (C b 5 1 g/L), and 60 mL for lysozyme
(C b 5 7 g/L). The different volumes (and thus filtration
times) were consistent with the different rates of fouling
seen in Figure 1. The membranes and stirred cell were
then gently rinsed with PBS before performing a second
filtration using the 100K-prefiltered protein solutions.
These sequential filtration experiments thus provide a
direct measure of the rate of chemical attachment of
native protein to an existing deposit without the complications associated with the continued deposition of protein aggregates as occurs during the standard filtration
runs (the prefiltered protein solutions are, at least initially, completely free of any aggregates).
The filtrate flux data for the second filtration step are
plotted in Fig. 5, with the flux normalized by the initial
flux through the membrane at the start of this second
filtration. These initial fluxes were similar, ranging from
2.7 3 1024 m/s for lysozyme to 3.0 3 1024 m/s for ovalbumin. The filtrate flux for lysozyme remained essentially constant over the 160 min filtration, declining by
less than 10% over this time period. In contrast, the
BSA and ovalbumin flux declined by more than an order
of magnitude after 4 h of filtration. The rate of flux
decline for ovalbumin was considerably higher than that
for BSA, with a t 20 value of about 45 min compared to
150 min for BSA. This behavior is completely consistent
with the rate of flux decline data shown in Figures 1
and 3. Lysozyme is unable to chemically add to the
existing deposit and thus shows no flux decline in this
type of sequential filtration experiment. Ovalbumin
chemically adds to the existing deposit more rapidly
than BSA, which is likely due to the presence/reactivity

of the greater number of free thiol groups in the ovalbumin structure compared to the single free sulfhydryl
in BSA.
The increased rate of flux decline seen at t . 60 min
in the experiments employing 100K-prefiltered solutions of BSA and ovalbumin (data from Fig. 4) was
apparently associated with the formation of new protein
aggregates during the actual filtration run. The formation of additional aggregates and oligomers was confirmed by sodium dodecyl sulfate polyacrylamide gel
electrophoresis which provides a measure of size via the
rate of protein migration though a polyacrylamide gel.
SDSPAGE showed a small increase in the number of
protein dimers and higher molecular weight oligomers
in the bulk protein solutions in the stirred cell over the
course of the ovalbumin and BSA filtration. The slow
onset of fouling seen for BSA and ovalbumin in Figure
4 is likely due to the slow kinetics of the intermolecular
thioldisulfide interchange reactions that are likely the
cause of aggregate formation for these proteins. Note
that cys-BSA and lysozyme, both of which lack any free
sulfhydryls, showed essentially no flux decline after the
removal of the protein aggregates (Fig. 2).
The data presented in Figure 3 suggest that the filtrate
flux approaches a quasi-steady value at long times. This
behavior is similar to the critical flux concept presented
by Bacchin et al. (1995), with Jss representing the filtrate
flux below which there is no longer any significant fouling. Palecek and Zydney (1994) hypothesized that this
quasi-steady flux is attained when the hydrodynamic
drag force on the proteins associated with the convective
filtrate flow toward the membrane is no longer sufficient
to overcome the repulsive (primarily electrostatic) interactions that exist between the native (bulk) protein and
the protein deposit on the membrane surface. As long as
the convective drag force is greater than the electrostatic
repulsion, proteins continue to add to the deposit and
the flux continues to decline. This model assumes that
fouling is governed primarily by the formation of a protein layer on the upstream surface of the membrane.
Internal (pore) adsorption is implicitly assumed to occur
quite rapidly upon exposure of the membrane to the
protein solution and thus has little direct effect on the
magnitude of the quasi-steady flux.
Palecek and Zydney (1994) estimated the electrostatic repulsive interaction using an expression for the
force between two charged spheres, while the hydrodynamic force was evaluated using the Stokes drag law.
The quasi-steady flux was then determined by equating
these forces, yielding (Palecek and Zydney, 1994)
Jv 5 JpI 1 vs 2 exp(2kh)

Figure 5. Normalized filtrate flux during the sequential filtration of


100K-prefiltered solutions of lysozyme, BSA, and ovalbumin at 14
kPa and 600 rpm through 0.22-mm PVDF membranes that had first
been used to filter a small volume of the unfiltered protein. Data are
only shown for the second filtration step.

(4)

where s is the protein surface charge density, k21 is the


Debye length, and h is the distance between the proteins. The term v is a proportionality constant related
to the ion valence and overall ionic strength, and JpI is

KELLY AND ZYDNEY: PROTEIN FOULING DURING MICROFILTRATION

97

the flux at the protein isoelectric point, i.e., at the pH


where the protein has no net charge.
The quasi-steady fluxes obtained during microfiltration of the different protein solutions [evaluated from
the best fit of the flux data to Eq. (3)] are plotted in
Figure 6 as a function of the square of the protein surface
charge density, as suggested by Eq. (4). The net protein
charge at pH 6.88 was obtained from pH titration data
(Gordon, 1971; Palecek and Zydney, 1994; Robbins et
al., 1967; Tanford, 1961; Tanford and Roxby, 1972) with
the values given in Table II. Also shown in Figure 6 are
additional data for BSA, hemoglobin, and ribonuclease
obtained by Palecek and Zydney (1994). The data for
BSA and ribonuclease actually include filtrations performed at several different pH from 4.7 to 7.45. The
protein surface areas were calculated as S 5 4pR 2 using
the effective hydrodynamic radii obtained from protein
diffusivity measurements (Tanford, 1961).
The quasi-steady fluxes for the different proteins are
very well-correlated by Eq. (4) with JpI P 6 mm/s independent of the protein molecular weight or physical
characteristics. This includes data obtained with proteins that are positively charged at pH 6.88 (lysozyme
and ribonuclease) as well as those that are negatively
charged at this pH (ovalbumin, a-lactalbumin, b-lactoglobulin, and BSA). The data for ribonuclease actually
cover both positive and negative charges since they include data at pH both above and below the pI. Note
that the quasi-steady flux for BSA and ovalbumin at
pH 6.88 are fairly similar, even though these proteins
have very different surface charges (218 and 210 for
BSA and ovalbumin, respectively), providing further
support for the dependence of the flux on the protein
surface charge density. Although Eq. (4) is clearly phenomenological in nature, it does provide a very effective

Figure 6. Quasi-steady filtrate flux obtained during protein microfiltration over a range of pH values as a function of the square of the
protein surface charge density. Calculations are described in the text.

98

means of correlating, and interpreting, quasi-steady flux


data in protein microfiltration.
DISCUSSION
The data obtained in this study clearly demonstrate
that the rate and extent of flux decline during protein
microfiltration is directly linked to the structural characteristics of the protein molecule. In particular, the presence of a free thiol group causes an initial increase in
the rate of flux decline due to the chemical attachment of
native protein to the growing deposit via intermolecular
thioldisulfide interchange reactions. Blocking this free
sulfhydryl (e.g., by cysteinylation of BSA) completely
eliminates the chemical addition of native protein to the
growing deposit. Those proteins without free sulfhydryls
(with the exception of myoglobin) were only able to foul
the membrane by physical deposition of large protein
aggregates. The quasi-steady flux obtained during protein microfiltration was unaffected by the presence (or
absence) of the free sulfhydryls but was instead directly
related to the surface charge density of the protein. In
this case, protein deposition continues until the convective drag on the proteins associated with the filtrate flow
is unable to overcome the repulsive (primarily electrostatic) interactions between the bulk proteins and the
existing deposit.
As discussed above, the initial flux decline in this
system was apparently due to the deposition of large
protein aggregates on the membrane surface. Removal
of these protein aggregates by prefiltration essentially
eliminated protein fouling of the PVDF microfiltration
membranes. Detailed studies of the mechanisms and
kinetics of protein aggregation are relatively scarce.
However, it is likely that BSA, ovalbumin, and b-lactoglobulin aggregate via the same type of intermolecular
thioldisulfide interchange reactions that were responsible for the chemical attachment of native protein to the
growing deposit (Kelly and Zydney, 1994; Liu et al.,
1991). The cys-BSA (Sigma A0161) was prepared from
a standard BSA preparation (Sigma A4503); thus, the
protein aggregates present in this product may have
been formed by thioldisulfide interchange reactions
prior to cysteinylation.
The origin of the protein aggregates for those molecules without free sulfhydryls is less clear. The aggregates could arise from strong coulombic and/or hydrophobic interactions, possibly involving partially
denatured proteins formed during the commercial preparation (e.g., during precipitation, crystallization, and/
or lyophilization). For example, the large-scale production of a-lactalbumin typically involves ammonium sulfate precipitation from acidified skim milk at pH values
as low as 2 (Gordon, 1971). a-Lactalbumin has been
shown to form large aggregates at pH , 5.2, which is
coincident with a number of changes in the protein
structure whereby certain groups become available for

BIOTECHNOLOGY AND BIOENGINEERING, VOL. 55, NO. 1, JULY 5, 1997

intermolecular interaction (Gordon, 1971). Pepsin is unstable at pH values above 6 (Bovey and Yanan, 1960),
and lysozyme is known to form dimers at pH above 5
(Sophianopoulos and Van Holde, 1964).
Protein aggregation can also occur due to pumping
in cross-flow filtration systems (Chandavarkar, 1990;
Meireles et al., 1991), although this type of aggregation
was unlikely to be significant in the nitrogen-pressurized
stirred cell employed in this study. It is also likely that
the chemistry and morphology of the membrane, as well
as the hydrodynamic characteristics of the membrane
module, can affect the rate and extent of aggregate
deposition. Additional research is needed to determine
the actual mechanisms of aggregate formation for different proteins and to develop appropriate strategies for
eliminating the formation and deposition of these large
aggregate structures during protein filtration.
As discussed in the text, the greater rate of chemical
addition of ovalbumin to the growing deposit relative
to BSA may have been due to the presence of four free
sulfhydryls in ovalbumin compared to the single free
sulfhydryl in BSA. However, it was not possible to actually correlate the value of b with the number of free
sulfhydryls. For example, the largest value of b was
obtained with b-lactoglobulin, which only has a single
free sulfhydryl group. The actual rate of the intermolecular thiol-disulfide interchange reaction is dependent
on the reactivity (and not just the number) of the free
sulfhydryls, as well as the nature of the existing disulfide
bonds and the steric accessibility of the reacting moieties. A much more complete analysis of the detailed
protein structure in the vicinity of the free sulfhydryls
would be required to actually predict the value of b for
a given protein. In addition, the reactivity of the free
sulfhydryls can be strongly affected by the solution
chemistry. Kelly and Zydney (1994) have shown that
the presence of trace amounts of copper significantly
increased the rate of flux decline for BSA through the
catalytic activity of this divalent metal cation for these
sulfhydryl-mediated reactions. Metal chelators, which
effectively remove metal cations from solution, have
been successfully used to reduce the rate of fouling
(Kelly and Zydney, 1994).
Although this study provides the clearest demonstration of the importance of the free sulfhydryl group in
the fouling characteristics of different proteins, Lee and
Merson (1976) previously demonstrated that the rate
and extent of flux decline during the ultrafiltration of
cottage cheese whey could be significantly reduced by
addition of very small amounts of N-ethylmaleimide.
This chemical agent is known to block the free thiol
groups of both b-lactoglobulin and BSA (Lee and Merson, 1976), which together comprise over 50% of the
whey proteins. The ultrafiltration flux increased by
nearly 50% upon addition of 0.0015 M N-ethylmaleimide and by more than 75% upon addition of 0.0015 M
N-ethylmaleimide and 0.2 M CaCl2 . The dramatic effect

of the N-ethylmaleimide on the flux is completely consistent with the data obtained in this study for cys-BSA;
the cysteinylation eliminated the reactivity of the free
sulfhydryl, thereby dramatically reducing the rate of flux
decline. A more detailed understanding of the chemistry
of these intermolecular interactions, and their relationship to the underlying protein structure and properties,
will hopefully provide new strategies for the effective
control of protein fouling in both ultrafiltration and
microfiltration.

References
Bacchin, P., Aimar, P., Sanchez, V. 1995. Model for colloidal fouling
of membranes. AIChE J. 41: 368376.
Belfort, G., Davis, R H., Zydney, A. L. 1994. The behavior of suspensions and macromolecular solutions in crossflow microfiltration. J.
Membrane Sci. 96: 158.
Bollag, D. M., Edelstein, S. J. 1991. Protein methods. Wiley-Liss,
New York.
Bovey, F. A., Yanari, S. S. 1960. Pepsin, pp. 6392. In: P. D. Boyer,
H. Lardy, and K. Myrback (eds.), The enzymes. Academic, New
York.
Cecil, R., McPhee, J. R. 1959. The sulfur chemistry of proteins, pp.
255389. In: C. B. Anfinsen, Jr., M. L. Anson, K. Bailey, and J. T.
Edsall (eds.), Advances in protein chemistry, vol. 14. Academic,
New York.
Chandavarkar, A. S. 1990. Dynamics of fouling of microporous membranes by proteins. Ph.D. Thesis, Massachusetts Institute of Technology, Cambridge, MA.
Foster, J. F. 1977. Some aspects of the structure and conformational
properties of serum albumin, pp. 5384. In: V. M. Rosenoer,
M. Oratz, and M. A. Rothschild (eds.), Albumin structure, function,
and uses. Pergamon, Oxford.
Gordon, W. G. 1971. In: H. A. McKenzie (ed.), Milk proteins: Chemistry and molecular biology, vol. 2. Academic, New York.
Hlavacek, M., Bouchet, F. 1993. Constant flowrate blocking laws and
an example of their application to dead-end microfiltration of protein solutions. J. Membrane Sci. 82: 285295.
Huggins, C., Tapley, D. F., Jensen, E. V. 1951. Sulfhydryl-disulphide
relationships in the induction of gels in proteins by urea. Nature
167: 592593.
Jocelyn, P. C. 1972. Biochemistry of the SH group. Academic, London.
Jonsson, G., Pradanos, P., Hernandez, A. 1996. Fouling phenomena
in microporous membranes. Flux decline kinetics and structural
modifications. J. Membrane Sci. 112: 171184.
Kelly, S. T., Zydney, A. L. 1994. Effect of thiol-disulfide interchange
reactions on albumin fouling during membrane microfiltration. Biotechnol. Bioeng. 44: 972982.
Kelly, S. T., Zydney, A. L. 1995. Fouling mechanisms during BSA
microfiltration. J. Membrane Sci. 107: 115127.
Kelly, S. T., Opong, W. S., Zydney, A. L. 1993. The influence of
protein aggregates on the fouling of microfiltration membranes
during stirred cell filtration. J. Membrane Sci. 80: 175187.
Kim, K. J., Chen, V., Fane, A. G. 1993. Some factors determining
protein aggregation during ultrafiltration. Biotechnol. Bioeng. 42:
260265.
Lee, D. N., Merson, R. L. 1976. Chemical treatments of cottage cheese
whey to reduce fouling of ultrafiltration membranes. J. Food Sci.
41: 778786.
Liu, W. R., Langer, R., Klibanov, A. M. 1991. Moisture-induced aggregation of lyophilized proteins in the solid state. Biotechnol. Bioeng.
37: 177184.
Meireles, M., Aimar, P., Sanchez, V. 1991. Albumin denaturation
during ultrafiltration: Effects of operating conditions and conse-

KELLY AND ZYDNEY: PROTEIN FOULING DURING MICROFILTRATION

99

quences on membrane fouling. Biotechnol. Bioeng. 38: 528


534.
Palecek, S. P., Zydney, A. L. 1994. Intermolecular electrostatic interactions and their effect on flux and protein deposition during protein
filtration. Biotechnol. Prog. 10: 207213.
Pradanos, P., Hernandez, A., Calvo, J. I., Tejerina, F. 1996. Mechanisms of protein fouling in cross-flow UF through an asymmetric
inorganic membrane. J. Membrane Sci. 114: 115126.
Robbins, F. M., Andreotti, R. E., Holmes, L. G., Kronman, M. J.
1967. Inter- and intramolecular interactions of a-lactalbumin. VII.
The hydrogen ion titration curve of a-lactalbumin. Biochim. Biophys. Acta 133: 3345.

100

Sophianopoulos, A. J., Van Holde, K. E. 1964. Physical studies of


muramidase. II. pH-dependent dimerization. J. Biol. Chem. 239:
25162524.
Tanford, C. 1961. Physical chemistry of macromolecules. Wiley,
New York.
Tanford, C., Roxby, R. 1972. Interpretation of protein titration curves.
Application to lysozyme. Biochemistry 11: 21922198.
Torchinsky, Y. M. 1981. Sulfur in proteins. Pergamon, Oxford.
Tracey, E. M., Davis, R. H. 1994. BSA fouling of track-etched polycarbonate microfiltration membranes. J. Colloid Interf. Sci. 167:
104116.

BIOTECHNOLOGY AND BIOENGINEERING, VOL. 55, NO. 1, JULY 5, 1997

You might also like