Biochemical Engineering Journal 83 (2014) 7989

Contents lists available at ScienceDirect

Biochemical Engineering Journal
j our nal homepage: www. el sevi er . com/ l ocat e/ bej
Regular Article
Enzyme immobilization by fouling in ultraltration membranes:
Impact of membrane conguration and type on ux behavior and
biocatalytic conversion efcacy
Jianquan Luo

, Anne S. Meyer, Gunnar Jonsson, Manuel Pinelo

Department of Chemical and Biochemical Engineering, Center for BioProcess Engineering, Building 229, Technical University of Denmark, DK-2800 Kgs.
Lyngby, Denmark
a r t i c l e i n f o
Article history:
Received 19 August 2013
Received in revised form
28 November 2013
Accepted 14 December 2013
Available online 21 December 2013
Keywords:
Membrane fouling
Enzyme immobilization
Alcohol dehydrogenase
Filtration
Catalysis
a b s t r a c t
Enzyme-immobilization in membranes accomplished by fostering membrane fouling was evaluated.
Four different membrane congurations and ve membranes were compared for immobilization of
alcohol dehydrogenase (ADH) in terms of enzyme loading, permeate ux and nal biocatalytic conver-
sion. The membrane conguration impacted the efciency of the enzyme-immobilization as well as the
biocatalytic-membrane reaction, and the sandwich mode, with an extra polypropylene support above
the membrane skin layer, worked best due to its high ux and stable conversion. Among the membranes,
a GR51PP polysulphone membrane allowed for the highest ux during the reaction with the enzyme-
immobilized membrane. At the same time, the lowest enzyme loading and low reaction stability were
achieved for this membrane. Satisfactory enzyme loadings, stable conversions, but low ux rates were
obtained for the PLTK and PLGC regenerated cellulose membranes. With these two highly hydrophilic
membranes, the ADH enzyme activity was fully retained even after 24 h of storage of the membrane.
Filtration blocking and resistance models were used to analyze the fouling/immobilization mechanisms
and give explanations for the different results. The work conrms that fouling-induced enzyme immo-
bilization is a promising option for enhancing biocatalytic productivity, and highlights the signicance
of the membrane type and conguration for optimal performance.
2013 Elsevier B.V. All rights reserved.
1. Introduction
Biocatalytic membranes, with enzymes physically docked in
or on the membrane, combine several distinct functions of bio-
logical membranes: localized biochemical reaction, immobilized
enzyme, andphysical separationof thereactionreactants andprod-
ucts [1,2]. Recently, designed biocatalytic membranes deploying
enzymes immobilized in industrially manufactured membranes
have attracted growing attention in both biosensor and bioreactor
applications [3,4]. The porous membrane, functioning as a selective
barrier as well as a support for enzyme immobilization allowing
enzyme re-use, may also help stabilize the enzymes, eliminate
product inhibition, and allow for continuous processing [2,5,6].
Immobilization of enzymes in/on membranes can be achieved via
adsorption, covalent attachment, cross-linking and entrapment
[710]. Enzyme immobilization by physical adsorption is simple
andeasy, but enzyme desorptionmay easilyoccur. Covalent attach-
ment andcross-linkingwithmultifunctional reagents ensure stable

Corresponding authors. Tel.: +45 4525 2950; fax: +45 4588 2258.
E-mail addresses: Jluo@kt.dtu.dk (J. Luo), mp@kt.dtu.dk (M. Pinelo).
immobilization of the enzymes in/on the membrane, but the initial
enzyme activity loss is likely to be quite high due to the possible
alterations near the enzymes active site [3]. Entrapment therefore
seems to be an advantageous approach for immobilizing enzymes
in membranes as long as the enzyme leakage, mass transfer lim-
itations, and ux decline across the membrane can be kept at a
minimum.
For the immobilization the enzymes can be incorporated during
the membrane fabrication [11] or, when the membrane is already
available, the enzymes can be entrapped in the membrane pores
via pressure-driven ltration [4]. Enzymes have been immobilized
in the sponge layer of an asymmetric capillary membrane by cross-
owultraltration (UF) [12]. This immobilization did not cause any
signicant subsequent loss of enzyme activity [13]. However, the
enzyme entrapment induced a dramatic ux decline of the mem-
brane, and the membrane permeability decreased by 4384% for
an immobilized enzyme load of 0.0090.052mg cm
2
[12].
Membrane fouling can be dened as membrane permeability
loss due to adsorption or precipitation of the solute on or in the
membrane [1417]. It is well known that surface adsorption,
pore blockage, inorganic precipitation, gel or cake formation, and
biological fouling are the main mechanisms for membrane fouling
1369-703X/$ see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.bej.2013.12.007
80 J. Luo et al. / Biochemical Engineering Journal 83 (2014) 7989
Nomenclature
C
i
NADH concentrations in feed (M)
C
0
enzyme concentration in feed (gL
1
)
C
p
enzyme and NADH concentrations in permeate
(gL
1
or M)
CP concentration polarization
C
r
enzyme concentrations in the mixture of retentate
and rinsing (gL
1
)
C
w
enzyme concentration in pressure-driven washing
permeate (gL
1
)
K constant
m
i
immobilized enzyme amount before washing (mg)
m

i
immobilized enzyme amount after washing (mg)
m
t
total enzyme amount in feed (mg)
n fouling coefcient (0, 1, 1.5, 2)
P transmembrane pressure (Pa)
PET polyethylene terephthalate
PP polypropylene
R
cp
resistance of CP layer (m
1
)
R
if
resistance of irreversible fouling (m
1
)
R
m
intrinsic membrane resistance (m
1
)
R
rf
resistance of reversible fouling (m
1
)
R
t
total ltration resistance (m
1
)
t ltration time (s)
V permeate volume (m
3
)
V
p
permeate volume (mL)
V
r
volumeof themixtureof retentateandrinsingresid-
ual (mL)
V
w
volume of washing permeate (mL)
Greek letter
solvent viscosity (Pa s)
[14]. As illustratedinFig. 1, there are many parallels betweenmem-
brane fouling mechanisms and enzyme immobilization strategies
for obtaining biocatalytic membranes. Forced membrane fouling
can thus be employed directly as an immobilization strategy for
adsorption and entrapment of enzymes in a membrane [18,19].
When another functional reagent has been added into the enzyme
solution to promote covalent cross-linking, the approach has been
called combined fouling [20,21]. Although some reports are
thus available proving the principal workability of fouling-induced
enzyme immobilization [12,13,22], there is a scarcity of knowledge
about the signicance of the membrane type and notably of the
membrane conguration (immobilization on/in the skin layer
versus in the support layer) for the efciency of the immobilization
Fig. 1. Schematic of connection between membrane fouling mechanisms and
enzyme immobilization strategies.
as well as for the subsequent performance of the biocatalytic
reaction.
The present work was undertaken to assess the inuence of dif-
ferent congurations, pore size and membrane materials on the
fouling-induced enzyme immobilization and biocatalytic perfor-
mance(conversions) inmembranes. Alcohol dehydrogenase(ADH),
being a sensitive enzyme, was used in the study to catalyze the
conversion of formaldehyde (HCOH) to methanol (CH
3
OH), dur-
ing oxidation of NADH to NAD
+
(the third step of multi-enzymatic
catalysis of CO
2
to methanol [23]). This systemwas chosen to min-
imize interference from substrate and/or product induced fouling
HCOHand CH
3
OHbeing small molecules. In order to provide a bet-
ter base for solving the possible problems of lowpermeate ux and
enzyme leakage/inactivation, focusing on the long-term enzyme
stability and catalytic efciency, the fouling and the enzyme
immobilization mechanisms were modeled using two different
membrane fouling models.
2. Theory
2.1. Filtration blocking models
For a dead-end ltration process at constant pressure, the laws
of ltration can be stated as [24]:
d
2
t
dV
2
= K

dt
dV
n
(1)
where t is ltration time (s), V is permeate volume (m
3
), K is
constant and n can have different values depending upon differ-
ent types of fouling: n=2 for the complete blocking model, n=1.5
for the standard blocking model, n=1 represents the intermediate
blocking model, and n=0 indicates the cake layer model. By inte-
grating Eq. (1), four linear equations can be obtained when xing
the value of n [25,26] (shown in Table S1, see supplementary data).
Although the value of n will change during ltration due to the
evolution of fouling with time [24] it is possible to identify the
most possible fouling mechanisms for different membrane con-
gurations and membranes by tting the experimental ux data
obtained in the initial ltration period using these linear models
and comparing their regression coefcients.
2.2. Filtration resistance model
Generally, permeate ux (J
p
) can be described by a resistance
model without considering the osmotic pressure, which includes
various possible fouling resistances, given by Eq. (2),
J
p
=
P
R
t
=
P
(R
m
+R
cp
+R
rf
+R
if
)
(2)
where P is transmembrane pressure (Pa), is solvent viscosity
(Pa s) and R
t
is total ltration resistance (m
1
), R
m
is the intrinsic
membrane resistance (m
1
), R
cp
is the resistance of the concen-
tration polarization (CP) layer, R
rf
is the resistance resulting from
reversible fouling (e.g. particle deposit) (m
1
), R
if
is the irreversible
fouling including pore blocking or cake formation (m
1
).
3. Materials and methods
3.1. Chemicals and membranes
Alcohol dehydrogenase (ADH, EC 1.1.1.1) from Saccharomyces
cerevisiae, -nicotinamide adenine dinucleotide reduced form
(NADH) and formaldehyde (37%, w/w) were purchased from
SigmaAldrich (St. Louis, MO, USA). All enzyme and substrate solu-
tions were prepared using 0.1M phosphate buffer (pH=7.0). ADH
is a protein tetramer containing four equal subunits. The molecular
J. Luo et al. / Biochemical Engineering Journal 83 (2014) 7989 81
Table 1
Main characteristics of the ultraltration membranes used in the study.
Membrane PLGC GR61PP PLTK Biomax 30 GR51PP
Manufacturer Millipore Alfa Laval Millipore Millipore Alfa Laval
MWCO (kDa) 10 20 30 30 50
Skin material Regenerated cellulose Polysulphone Regenerated cellulose Polyethersulfone Polysulphone
Support material Polypropylene Polypropylene Polypropylene Polyolen Polypropylene
Thickness (m) 230 350
a
230 280 300
a
Water permeability (L m
2
h
1
bar
1
) 108 38
a
362 544 49
a
MWCO: molecular weight cut-off.
a
Own measurements at roomtemperature of 23

C in normal mode.
weights of ADH, NADHandformaldehyde are 141, 0.7and0.03kDa,
respectively. Fivecommercial UFmembranes weretested, andtheir
mainproperties andsuppliers aresummarizedinTable1. Two extra
membrane brous supports, made of polyethylene terephthalate
(PET) and polypropylene (PP), respectively, were used to congure
the sandwich membrane (Fig. S1). The PP support, with a higher
specic weight than the PET support (0.306 vs. 0.187gcm
3
), was
obtained fromthe PLTK membrane by removing its skin layer.
3.2. Experimental set-up and procedure
The dead-end ltrations were carried out in a stirred cell (Ami-
con 8050, Millipore, USA). The working volume of the cell was
50mL, the stirring speed was xed at 100rpm, and the effective
membrane surface area was 13.4cm
2
. Aconstant pressure was gen-
erated by lling nitrogen gas into the cell and the permeate was
collected in a beaker on an electronic scale (PJ300, Mettler, USA)
or in a precision cylinder (Duran, Germany) in order to monitor
the permeate ux (permeate density was dened as 1gmL
1
). All
experiments were performed at a controlled temperature of 23

C
and a new membrane was used for each experiment. As seen in
Fig. 2, the membranes were placed on the membrane holder in
three congurations: (a) normal mode (skin layer facing feed); (b)
reverse mode (support layer facing feed); or (c) sandwich mode
(own support layer facing feed with an extra support between the
skin layer and the membrane holder).
The virgin GR61PP, Biomax 30 and GR51PP membranes were
rst cleaned by 0.1% NaOH solution for 1h with a pressure of 1
or 0.1bar, whereas the new PLGC and PLTK membranes were rst
dipped in a 5% NaCl solution for 30min and then ltrated with
deionized water for another 30min at 0.1bar (procedures accord-
ing to the manufacturers instructions). Afterwards, the water
permeability of the membranes was measured at 2bar with phos-
phate buffer for 2030min. Then, 50mL of ADH enzyme solution
with an enzyme protein concentration of 0.1gL
1
was put into the
cell for the subsequent enzyme immobilization operations.
3.2.1. Enzyme immobilization
Enzyme immobilizationwas carriedout at a pressure of 2bar for
1h, in four series: normal mode (Fig. 2a), reverse mode (Fig. 2b),
PET sandwich mode, and PP sandwich mode (Fig. 2c), i.e. the
PET and PP support layers were used in the two sandwich modes,
respectively. For the normal mode, the permeate ux was so high
that another 40mL buffer (without enzyme) was added into the
cell during the immobilization in order to obtain the same ltra-
tion time as in the other modes. In each case, the permeate was
collected in precision cylinders for analysis and the cylinders were
replacedmanually for every 4mL. At the endof ltration, the fouled
membrane was rinsed 5 times by buffer without pressure (5mL
of buffer each time). The nal retentate and rinsing residual were
combined in order to calculate the amount of immobilized enzyme
by mass balance. Then, the fouled membrane was washed by buffer
at a pressure of 2bar, and the permeate was collected for analysis
until the enzyme concentration in the permeate was close to zero
(except for Biomax 30). The nal steady buffer ux was recorded
as permeability for the fouled membrane.
The various ltration resistances were calculated as follows:
rstly, the intrinsic membrane resistance (R
m
) was acquired from
the initial membrane permeability, secondly, the total ltration
resistance (R
t
) was calculated from the nal permeate ux dur-
ing immobilization; thirdly, the sumof R
m
, R
rf
and R
if
was obtained
fromthe membrane permeability at the beginning of the washing
step with pressure because the CP layer was removed by the mild
rinsing; fourthly, the sum of R
m
and R
if
could be gained from the
membrane permeability at the end of the washing step with pres-
sure as the reversible fouling layer was wiped off by the washing
with pressure and agitation.
3.2.2. Enzymatic reaction
As shown in Fig. 2, the enzyme catalyzed reaction was carried
out in four different membrane conguration modes: (d) normal
mode; (e) reverse mode; (f) sandwich mode; (g) switch mode.
The rst three modes had the same congurations as in the immo-
bilization step. The switch mode (g) was achieved by turning the
fouled membrane to normal mode manually after immobilizing
the enzyme in sandwich mode so the clean skin layer was facing
the reaction solution (Fig. 2g).
The NADH-mediated reduction of formaldehyde into methanol
can be catalyzed by ADH [27]:
CHOH+NADH+H
+
ADH
CH
3
OH+NAD
+
(3)
This reaction was used to evaluate the activity of immobilized
ADH enzyme. The procedure of reaction was the same in the
four different reaction modes. 25mL substrate mixture (CHOH
(0.1M) +NADH(100M)) was pouredintothestirredcell equipped
with the particular biocatalytic membrane and a pressure of 2bar
was set immediately. When 20mL permeate was obtained (col-
lectedinaliquots of 4mL), the ltrationwas suspendedandanother
20mL substrate mixture was added for the next cycle. At the end
of reaction, the retentate was collected for analysis to ensure that
no reaction occurred in bulk solution.
After having accomplished 46 reaction cycles, the reactor
equipped with the specic catalytic membrane was rinsed with
buffer andstoredinthefridge(4

C) with20mL buffer overnight.

After 24h, the buffer permeability of the membrane was then mea-
suredagainandanother 12reactioncycles werecarriedout. Inthis
series of experiments, the permeate ux and conversion in sand-
wich and switch modes (Fig. 2f and g) were compared after 24h
storage.
3.3. Analytical methods
The concentration of the ADHenzyme was measured as protein
concentration with a spectrophotometer (Perkin Elmer lambda20
UV/VIS, Germany) at 280nm [28] or using Bradford protein assay
during reaction. The NADH concentration was monitored by the
absorbance at 340nm[28]. The absorbance stability of the enzyme
82 J. Luo et al. / Biochemical Engineering Journal 83 (2014) 7989
Fig. 2. Schematic diagrams of membrane congurations for enzyme immobilization and subsequent catalytic reaction. (a) Enzyme immobilization in normal mode; (b)
enzyme immobilization in reverse mode; (c) enzyme immobilization in sandwich mode; (d) enzymatic reaction in normal mode; (e) enzymatic reaction in reverse mode;
(f) enzymatic reaction in sandwich mode; (g) enzymatic reaction in switch mode.
and the NADH during the ltration was conrmed in preliminary
experiments (data not shown).
3.4. Calculated parameters
The average apparent rejection of the enzyme was dened as:
R
app
(%) =

1
C
p
C
0

100 (4)
where C
p
and C
0
are the enzyme concentrations in the permeate
and in the feed (initially) during immobilization, respectively.
The amount of immobilized enzyme was calculated from the
mass balance equation in two approaches:
m
i
= m
t
C
p
V
P
C
r
V
r
(5)
m

i
= m
t
C
p
V
P
C
r
V
r
C
w
V
w
(6)
where m
i
(m

i
) and m
t
are immobilized and total enzyme amounts,
respectively; C
r
is the enzyme concentrations in the mixture of
retentate and rinsing residual, and C
w
is the enzyme concentra-
tion obtained in the pressure-driven washing permeate; V
p
, V
r
and
V
w
are the volume of the permeate for immobilization, the mix-
ture of retentate and rinsing residual, and the washing permeate,
respectively.
The immobilizationefciency was expressed as enzyme loading
per unit area of membrane (mg cm
2
) or loading efciency (%):
loadingefciency(%) =
m
i
C
0
V
p
100 (7)
The enzyme activity was evaluated from the conversion of
NADH:
conversion(%) =
C
i
C
p
C
i
100 (8)
where C
i
andC
p
are the NADHconcentrations infeedandpermeate,
respectively.
4. Results and discussion
4.1. Effect of membrane conguration
4.1.1. Membrane permeability before immobilization
The effect of membrane conguration on initial membrane per-
meability without immobilized enzyme was evaluated using the
GR51PPmembraneinabasic ltrationwithbuffer. This asymmetric
membranehadmuchhigher permeabilitywhenoperatedinnormal
mode than in reverse mode (Table 2). Since there was no enzyme
immobilized in the membrane this difference had to be a result
of differences in membrane physics during the ltration. Bohonak
and Zydney [29] thus reported that membrane compaction with
the skin-side down resulted in a 1020% reduction in permeability
for an asymmetric UF membrane (polyvinylidene uoride, PVDF).
Persson et al. [30] also found that a polysulphone skin layer was
more sensitive to compaction than other materials. Therefore, in
this case, the compaction of the polysulphone skin layer when fac-
ing the hard membrane holder is likely to be the main reason for
the lowpermeability obtained in the reverse mode.
The permeability of the membranes with an extra macroporous
support (made of either PET or PP, see Section3.2.1) placedbeneath
the skin layer, to congure a sandwich mode, was also examined.
Inthis sandwich operatingmode, as seeninTable2andFig. S2, the
PP support performed better than the PET support, and the mem-
brane permeability of the PP sandwich mode was much higher
and more stable with operating time (pressure =2bar) than that
of the PET sandwich mode, and moreover close to that obtained
for the normal mode. The superior performance of the PP support
as compared to the PET support could be due to the PP polymer
material being more exible, which might help cushion the com-
paction of the skin layer during the ltration. In addition, the PP
support material is believed to have a more even pore distribution
than the PET material, which might support a more leveled ow
across the membrane area and produce an overall better perme-
ability.
4.1.2. Enzyme immobilization
As shown in Fig. 3a, during the enzyme immobilization a dra-
matic ux decline occurred when the support layer was facing
the enzyme solution. This ux decline was most likely due to CP
and membrane fouling resulting from the desired adsorption and
immobilization of the ADHenzyme molecules in the support layer.
The two different support layer materials, PET and PP, used in
the sandwich mode, produced strikingly different ux responses;
hence, the initial permeate ux levels for the PP sandwich mode
were high, starting at the level recorded for the normal mode, but
thendecreasedrapidlywithincreasingaccumulativepermeatevol-
ume (Fig. 3a). In contrast, for the PET sandwich mode, as well as
for the reverse mode, the initial ux was reduced immediately, but
then decreased at a lower rate than that of the PP sandwich mode
as the immobilization operation progressed (Fig. 3a). At the same
timetheconcentrationof theenzymeinthepermeateof thenormal
mode conguration was lowand steady during the immobilization
(Fig. 3b), whereas the permeate concentration of the enzyme for
the other membrane congurations, i.e. the reverse mode and the
J. Luo et al. / Biochemical Engineering Journal 83 (2014) 7989 83
Table 2
Membrane performance and enzyme loading for different membrane congurations (GR51PP membrane).
Membrane conguration Initial membrane
permeability
(L m
2
h
1
bar
1
)
Initial apparent
rejection
a
(%)
Permeability loss
after enzyme
immobilization (%)
Enzyme loading
(mg cm
2
)
Loading efciency
(%)
Normal mode 49.2 3.0 86.6 1.1 40.1 3.0 0.022 0.009 6.4 2.5
Reverse mode 22.8 1.8 85.4 1.1 55.7 7.7 0.156 0.004 69.9 1.9
PET sandwich mode 28.6 2.8 88.0 1.4 53.7 3.0 0.180 0.017 67.4 2.3
PP sandwich mode 47.0 2.2 89.2 1.0 44.8 0.7 0.148 0.010 59.1 1.3
a
Measured after collecting 4mL permeate.
two sandwich mode congurations increasedafter the rst 12mL
of permeate were collected (Fig. 3b).
The increases in the concentration of enzyme in the permeate
for the reverse mode and the two sandwich modes indicated that
the enzyme passed through the membrane once a support layer
was present on top of the skin layer facing the feed side. This effect
is congruent with negligible shear-induced and Brownian motion
back-diffusions of the solute (i.e. the enzyme) in the brous sup-
port layer; in contrast, the low permeate concentration, coupled
with the very low enzyme loading achieved for the normal mode
(Table 2), indicated that signicant back-diffusion of the enzyme
took place during enzyme immobilization in the normal mode.
When comparing the results for the reverse mode, the PET sand-
wich mode, and the PP sandwich mode, it was evident that the
enzyme concentration in the permeate was higher in the PP sand-
wich mode, presumably due to less compaction of the skin layer,
which in turn produced a higher permeability (Table 2).
Regarding to the enzyme loading (Eq. (5)), intuitively, a higher
permeate ux would result in more enzymes entering into the
membrane during the same ltration time and in turn produce
a higher enzyme immobilization load; this would explain why
a higher enzyme loading was achieved with the PET sandwich
mode than for the reverse mode; however, this mechanism con-
trast the lower enzyme loading of the normal mode membrane and
does not fully explain that the enzyme loading for the PP sand-
wich mode was lower than that obtained for the PET sandwich
mode. The quite low enzyme loading in the normal mode was
thus caused by lowaccessibility of enzymes into membrane (small
pore size of skin layer and charge repulsion between membrane
and enzymes). Moreover, membrane compaction in sandwich
mode could increase the enzyme rejection by the skin layer and
thus induced a higher enzyme loading, which explained why the
enzyme loading in PET sandwich mode was higher than that
in PP sandwich mode (the latter had a higher permeate ux
but less compact skin layer), as shown in Fig. 3 and Table 2. It
is worth mentioning that when comparing normal mode with
PP sandwich mode, the former had similar permeability loss
but much lower enzyme loading, which might be caused by the
different fouling mechanisms in these two modes. It is speculated
that in normal mode, a few enzymes decreased the hydrophilicity
of skin layer and then produced a large permeability loss.
4.1.3. Fouling mechanism
The ltration blocking model was used to analyze the fouling
mechanisms during immobilization. The permeate ux data for
the rst 30min were tted using four empirical fouling linear
models (Table S1 and Fig. S3), and the regression coefcients
obtained for each model are shown in Table 3. For the GR51PP
membrane in normal mode, the modeling veried that cake layer
formation was the dominant fouling mechanism. This cake layer
apparently deposited on the skin layer, but could be wiped off
easily (during washing of the membrane), resulting in a net, low
enzyme loading (Table 2). In contrast, for the reverse and the PET
sandwich modes, complete and standard blocking were the main
fouling types that were built by the enzyme molecules during the
immobilization operation, indicating that a signicant fraction
of enzymes, i.e. about 70% of the available enzymes according
to Table 2, were successfully entrapped and adsorbed in/on the
support layer or the membrane pores of skin layer. However,
the PP sandwich mode apparently produced a different fouling
mechanism, since complete blocking was not its dominant fouling
mechanism. This result was in complete accordance with the
rst interpretation of the permeate ux data (Table 2, Fig. S2 and
Fig. 3), and could be explained by the skin layer in PP sandwich
mode being less compact and thus that the enzymes entered into
the pores, and to a certain extent passed through the membrane,
80 60 40 20 0
Accumulative permeate volume (mL)
Normal mode
0
20
40
60
80
100
Reverse mode
PP "sandwich" mode
P
e
r
m
e
a
t
e

f
l
u
x

(
L
m
-
2
h
-
1
)
Add buffer
PET "sandwich" mode
a
0.00
0.02
0.04
0.06
0.08
0.10
P
e
r
m
e
a
t
e

c
o
n
c
e
n
t
r
a
t
i
o
n

(
g
L
-
1
)
Reverse mode
80 60 40 20 0
Normal mode
PET "sandwich" mode
PP "sandwich" mode
Accumulative permeate volume (mL)
b
Add buffer
Fig. 3. Comparison of permeate ux (a) and enzyme concentration in permeate (b) during immobilization for the four membrane congurations (GR51PP membrane, feed
concentration=0.1gL
1
).
84 J. Luo et al. / Biochemical Engineering Journal 83 (2014) 7989
Table 3
Coefcient R
2
of linear regression of tting experimental data according to ltration blocking model for different membrane congurations (GR51PP membrane).
Membrane conguration/blocking model Complete blocking Standard blocking Intermediate blocking Cake layer
Normal mode 0.9051 0.9139 0.9416
a
0.9845
a
Reverse mode 0.9689
a
0.9492
a
0.9239 0.8616
PET sandwich mode 0.9857
a
0.9763
a
0.9602 0.9115
PP sandwich mode 0.9353 0.9658
a
0.9738
a
0.9443
a
Indicating the best two tting models for each group of experimental data.
rather than completely blocked them up. Therefore, entrapment
and adsorption (e.g. hydrophobic interaction) might be the main
enzyme immobilization mechanismin play for this membrane.
4.1.4. Enzymatic reaction
Fig. 4 shows the permeate ux and conversions obtained during
the enzymatic reactions in four different membrane congura-
tions using the GR51PP membrane. Although the fouling-induced
enzyme immobilizationinthe reverse mode hadhighenzyme load-
ing, it suffered from very low ux during the reaction step, as
compared to the normal mode. When using an extra support, the
permeate ux for the reaction increased by 76% in the PET sand-
wich mode and by 225% in PP sandwich mode. The conversion
was still very high for PP sandwich mode even if the contact time
between substrate and enzymes decreased by 56% compared with
that for reverse mode (Fig. 4b); the conversion data also indicated,
that although the net enzyme loading differed for the different
membrane congurations, the amount of enzyme loaded was suf-
cient to achieve a high conversion, and the enzyme immobilization
produced anextended, steady, and highconversionfor the congu-
rations with the support layer facing the feed (i.e. the reverse mode
and both of the sandwich congurations). In the normal mode,
however, the conversion dropped profoundly as the membrane
reactor reaction was progressing (Fig. 4b) presumably because the
immobilized enzymes were detached from the skin layer to the
bulk solution during the reaction, and then activity loss of the rel-
atively fragile enzyme ensued. This same mechanismmight be the
explanation for the gradual reduction of conversion in fourth and
fth reaction cycles (beyond 80mL of accumulative permeate vol-
ume) observed for the reverse and both of the sandwich modes
(Fig. 4b). On the other hand, it is also possible that the immobilized
enzymes lost their activity gradually as the enzyme performance
was not favored by the present immobilization mechanisms.
Because of the convective transport toward the membrane
support, the enzymes cannot readily diffuse back to the bulk,
and therefore more likely leak into the permeate. By selecting a
membrane with suitable pore size of the skin layer, this enzyme
leakage may be avoided in sandwich mode. Moreover, enzyme
activity loss caused by low chemical afnity between enzymes
and membrane materials may be another possible reason for
the observed gradual reduction in conversion [3]. Therefore, it is
important to investigate the effect of membrane pore size and
material on enzyme immobilization and reactive stability [31].
4.2. Effect of membrane pore size and material
4.2.1. Enzyme immobilization
Fiveasymmetric membranes withdifferent poresizes andmate-
rials (Table 1) were tested in the PP sandwich mode. As seen in
Tables 1 and 4, the order of membrane permeability and enzyme
rejection was not fully in accordance with pore size probably
because the different membrane materials and microstructures
also affected the enzyme immobilization. The Biomax 30 mem-
branehadthehighest permeabilitypossiblyduetoits highporosity,
although an effect of the membrane material (polyolen) cannot
be excluded. Biomax 30 membrane also had the highest appar-
ent rejection of enzyme because of its extremely high permeate
ux at the beginning of ltration, as seen in Fig. 5a. Compared
to their permeabilities in the normal mode (Table 1), the perme-
abilities achieved in the sandwich mode were generally slightly
decreased except for the GR61PP membrane. Although the initial
permeate uxes were quite different for the ve membranes, the
nal uxfor the fouledmembranes was almost the same, indicating
that the initial ux behavior was highly dependent on membrane
properties (enzymeclean-membrane interaction), while the nal
ux behavior became more independent of membrane properties
(pore size, materials, etc.) and was largely controlled by the foul-
ing layer during the enzyme immobilization (enzymedeposited
enzyme interaction [32,33]). Except for the GR51PP membrane, for
which a gradual increase of enzyme in the permeate was recorded
(Fig. 5b), the enzyme concentration in the permeate of the different
membranes generally stayed at around 0.01gL
1
with increasing
0
20
40
60
80
100
P
e
r
m
e
a
t
e

f
l
u
x

(
L
m
-
2
h
-
1
)
Normal mode
100 80 60 40 20 0
+225%
Reverse mode
Accumulative permeate volume (mL)
+76%
a
PET "sandwich" mode
PP "sandwich" mode
100 80 60 40 20 0
0
20
40
60
80
100
C
o
n
v
e
r
s
i
o
n

(
%
)
Accumulative permeate volume (mL)
Normal mode
Reverse mode
PET "sandwich" mode
b
PP "sandwich" mode
Fig. 4. Comparison of permeate ux (a) and conversion (b) during reaction for four membrane congurations (GR51PP membrane).
J. Luo et al. / Biochemical Engineering Journal 83 (2014) 7989 85
Table 4
Membrane performance and enzyme loading for different membrane types in PP sandwich mode.
Membrane type Initial membrane
permeability
(L m
2
h
1
bar
1
)
Initial apparent
rejection (%)
Permeability loss
after enzyme
immobilization (%)
Enzyme loading
a
(mgcm
2
)
Loading efciency
a
(%)
PLGC 70.9 1.9 88.4 2.6 90.4 0.1 0.142 0.030 72.3 9.8
GR61PP 62.7 15.8 90.1 1.0 88.9 2.7 0.207 0.012 81.9 0.2
PLTK 320 8.9 86.0 0.6 96.5 1.0 0.208 0.016 68.1 3.4
Biomax 30 420 12.9 91.7 0.4 98.4 0.1 0.196 0.010 76.0 2.2
GR51PP 47.0 2.2 89.2 1.0 44.8 0.7 0.088 0.010 35.1 2.3
a
Deducting the enzyme amount in the permeate during the washing step with pressure.
40 30 20 10 0
0
180
360
540
720
900
P
e
r
m
e
a
t
e

f
l
u
x

(
L
m
-
2
h
-
1
)
Accumulative permeate volume (mL)
PLGC
GR61PP
PLTK
Biomax 30
GR51PP
a
PLGC
GR61PP
40 30 20 10 0
0.00
0.02
0.04
0.06
0.08
0.10
PLTK
P
e
r
m
e
a
t
e

c
o
n
c
e
n
t
r
a
t
i
o
n

(
g
L
-
1
)
Accumulative permeate volume (mL)
Biomax 30
b
GR51PP
Fig. 5. Comparison of permeate ux (a) and enzyme concentration in permeate (b) during immobilization for ve membranes using PP sandwich mode (feed concentra-
tion =0.1gL
1
).
accumulative permeate volume (Fig. 5b). The low level of enzyme
leakage through the membrane implied that all the other four
membranes (except the GR51PP membrane with largest pore size)
couldwell reject the enzyme evenwhenoperatedinsuchanabnor-
mal operating mode.
In order to evaluate the enzyme loading more accurately, the
enzyme concentration obtained in the permeate during the wash-
ing step was recorded: large amount of enzyme was found to be
washed away from the GR51PP membrane, while for the PLGC,
GR61PP and PLTK membranes, after collecting the rst 4mL the
enzyme concentration in the washing permeate (containing some
enzyme residue in the ow channels), was close to zero, although
the enzyme leakage for the Biomax 30 membrane was around
0.01gL
1
(Fig. 6). This latter slightly elevated leakage from the
Biomax 30 membrane might be caused by the special structure and
properties of this super highly permeable membrane (Table 1).
4.2.2. Permeability loss and enzyme loading
As shown in Table 4, the membrane permeability generally
decreased by more than 90% upon loading in the enzyme, except
for the GR51PP, for which the enzyme loading was reduced signi-
cantly after the washing step with pressure; as already mentioned
above, the lowloading of the GR51PP might be due to enzyme leak-
age via the skin layer (this membrane having the largest pore size,
50kDa, among the membranes tested (Table 1). The PLGC mem-
brane had a lower enzyme loading than the GR61PP, PLTK and
Biomax 30 membranes, which was likely caused by the small pore
size (less accessibility of enzymes), and PLGC membrane also gave
the lowest permeate volume during immobilization for a constant
ltrationtime (lowest permeate ux). Loadingefciencywas calcu-
lated by the ratio of the actual loading and the theoretical one (Eq.
(7)), and the GR61PP membrane had the largest loading efciency.
This high loading efciency was presumably resulted fromits high
membrane thickness (Table 1) inessence giving more loading space
for the enzyme immobilization.
4.2.3. Fouling mechanism
Table 5 shows the tting results for ve membranes inPP sand-
wich modeusingltrationblockingmodels (alsoseeFig. S3). It was
interesting to nd that the fouling mechanism was dependent of
the membrane material andpore size. For the regeneratedcellulose
membranes, i.e. PLGC and PLTK, cake layer formation and inter-
mediate blocking were the dominant fouling mechanisms. Since
the regenerated cellulose was highly hydrophilic, the enzymes that
accumulated behind the skin layer were likely to forma cake layer
40 30 20 10 0
0.00
0.02
0.04
0.06
0.08
0.10
P
e
r
m
e
a
t
e

c
o
n
c
e
n
t
r
a
t
i
o
n

(
g
L
-
1
)
Accumulative washing permeate (mL)
PLGC
GR61PP
PLTK
Biomax 30
GR51PP
Fig. 6. Enzyme leakage to the permeate after the immobilization for the different
membranes during washing step with pressure (2bar).
86 J. Luo et al. / Biochemical Engineering Journal 83 (2014) 7989
Table 5
Coefcient R
2
of linear regression of tting experimental data according to ltration blocking model for different membranes (PP sandwich mode).
Membrane type/blocking model Complete blocking Standard blocking Intermediate blocking Cake layer
PLGC 0.5943 0.7990 0.9278
a
0.9977
a
GR61PP 0.9741
a
0.9950
a
0.9739 0.8730
PLTK 0.6343 0.5151 0.8453
a
0.9933
a
Biomax 30 0.9093 0.9869
a
0.9775
a
0.9181
GR51PP 0.9353 0.9658
a
0.9738
a
0.9443
a
Indicating the best two tting models for each group of experimental data.
-
2
-
1
20 0
0
20
40
60
80
100
P
e
r
m
e
a
t
e

f
l
u
x

(
L
m
-
2
h
-
1
)
Accumulative permeate volume (mL)
60 40
PLGC
GR61PP
PLTK
Biomax 30
GR51PP
100 80 120
a
b
0
20
40
60
80
100
C
o
n
v
e
r
s
i
o
n

(
%
)
0 40 20
Accumulative permeate volume (mL)
PLGC
GR61
PLTK
Biomax 30
GR51PP
80 60
C
PP
K
5
120 100
5 h
2 h
6 h
Fig. 7. Comparison of permeate ux (a) and conversion (b) during reaction for ve membranes using PP sandwich mode.
by hydrogen bonding. For the polysulphone and polyethersulfone
membranes, standard blocking was the main fouling mechanism,
presumably because of hydrophobic adsorption. However, for the
GR61PP membrane, due to its smaller pore size, complete blocking
was probably also taking place. Therefore, it could be speculated
that adsorption and entrapment occurred at the same time in
this fouling-inducedenzymeimmobilization. Furthermore, accord-
ing to the ltration resistance model, the various resistances for
different membranes were calculated, and the results are shown
in Table 6. Only irreversible fouling could be regarded as immo-
bilization, and high enzyme loading together with low ltration
Table 6
Analysis of membranefoulingbyenzymeafter immobilizationaccordingtoltration
resistance model for different membranes (PP sandwich mode).
Membrane type Filtration resistance (10
13
m
1
) m

i
/R
if
(10
13
)
Rm Rcp R
rf
R
if
PLGC 0.54 0.43 0.22 5.11 0.37
GR61PP 0.61 0.48 0.19 4.91 0.56
PLTK 0.12 0.52 0.41 3.31 0.84
Biomax 30 0.09 0.65 0.18 5.63 0.47
GR51PP 0.82 1.83 1.90 0.66 1.79
m

i
/R
if
: the ratio of enzyme loading to irreversible fouling resistance.
Fig. 8. Hypothesis regarding effect of fouling mechanism on substrate behavior
when passing through fouled membrane.
resistance was preferable. Thus, the ratio of enzyme loading to
irreversible fouling resistance (m

i
/R
if
) was calculated. In this light,
the GR51PP membrane gave the highest relative performance
of enzyme immobilization and membrane permeability, and the
PLTK membrane also outperformed the other three membranes
(Table 6).
4.2.4. Enzymatic reaction
When the permeate ux and conversion for the ve biocatalytic
membranes were compared under the same operating conditions
(sandwich mode) it became evident that the permeate uxduring
reaction was governed by the ltration resistances of the mem-
brane and irreversible fouling, and the GR51PP membrane thus had
highest ux (Fig. 7 and Table 6). Hence, the conversions achieved
with the PLGC and PLTK membranes, with regenerated cellulose
skinlayer (Table 1), were very highandstable for ve or six reaction
cycles, while the conversions for GR61PP and GR51PP membranes,
with polysulphone skin layer (Table 1), decreased after two or
three cycles (Fig. 7b). Enzyme leakage was certainly not the only
reason of the reduction in conversion for these latter two mem-
branes, because the GR61PP could well reject the enzyme and its
enzyme loading was high (Fig. 6 and Table 4). Thus, the enzyme
activity loss might be due to less chemical afnity between the
ADH enzyme and the membrane materials [3]. It is worth men-
tioning that the GR61PP membrane had much longer reaction time
for each cycle than the GR51PP due to their ux difference (i.e.
75 vs. 20min), inducing a relatively larger reduction in the conver-
sionfor the GR61PP membrane at the same accumulative permeate
volume (Fig. 7b). For the PLGC and the PLTK membranes, since
regenerated cellulose was hydrophilic and highly biocompatible
[34], the enzyme activity was almost fully retained. The Biomax
30 membrane, with a hydrophilic-modied polyethersulfone skin
layer and the highest water permeability (Table 1), showed a sta-
ble conversioninrst four cycles (240min) but the conversionthen
droppedgradually. Apparently, the fouling layer inthe Biomax 30
membrane was not stable and some leakage of enzyme occurred
J. Luo et al. / Biochemical Engineering Journal 83 (2014) 7989 87
GR51PP Biomax 30 PLTK GR61PP PLGC
0
10
20
30
40
50
60
P
e
r
m
e
a
t
e

f
l
u
x

(
L
m
-
2
h
-
1
)
Membrane type
0 h
24 h
a
GR51PP Biomax 30 PLTK GR61PP PLGC
0
20
40
60
80
100
C
o
n
v
e
r
s
i
o
n

(
%
)
Membrane type
0 h 24 h b
Fig. 9. Comparisonof permeate ux (a) andconversion(b) during biocatalytic membrane reactioninsandwichmode (Fig. 2f) initially (0h) andafter 24hfor ve membranes.
during the reaction (Fig. 6). The gradual activity drop could also
indicate that the hydrophilic modication of the Biomax 30 mem-
brane might not be sufcient for enhancingits afnitywiththe ADH
enzyme.
Besides the enzyme leakage and chemical afnity of membrane
materials, the different fouling mechanism for different mem-
branes was a possible explanationfor diverse conversionproles in
Fig. 7b. For themembranes withdecreasingconversion, i.e. GR61PP,
Biomax 30 and GR51PP, the main fouling mechanism was pore
blocking, while cake layer formation was dominant for the mem-
brane with steady conversion, i.e. PLGC and PLTK. Based on these
results, a hypothesis regarding the effect of fouling mechanism
on substrate ow behavior is proposed in Fig. 8. When enzymes
were entrapped in the membrane pores, some of themwere prone
to hide inside the pores if the foulants (enzymes) went deep
into membrane (this would be in progress during pressure-driven
ltration), and thus the substrate preferred to ow through the
less blocked pores due to the less resistance; hence, the effective
enzyme amount for the reaction decreased, leading to a reduction
in conversion after several cycles (Fig. 7b). When enzymes assem-
bled on the membrane (skin layer), this fouling layer was more
porous and exible, and the substrate was easy to pass through
this catalytic layer and the contact area was increased, resulting in
a high and stable conversion unless this layer sank into membrane.
Fig. 9 illustrates the permeate ux and conversion variations
after 24h storage of enzyme-loaded membranes. Stable and high
conversions were obtained for the PLGC and PLTK membranes, and
partly for the Biomax 30 membrane, while the conversions for
GR61PPmembraneandnotablyfor theGR51PPmembranedropped
signicantlyafter 24hof (cold) storageof theenzyme-loadedmem-
brane. When combining the results fromFigs. 7b and 9b, it can be
concluded that permeation-related enzyme leakage or hiding and
time-related enzyme activity loss were the two main reasons for
the reduction in conversion.
Therefore, a membrane pore size of 30kDa was able to pre-
vent ADHenzyme leakage via the skin layer to the permeate, while
the membrane materials of regenerated cellulose were compatible
with ADH enzymes and the activity was thus well retained in the
PLGC and PLTK membranes.
4.3. Enzyme leakage
The fouling-induced enzyme immobilization in the membranes
may rely on adsorption and entrapment and perhaps in some
cases include hydrophobic or hydrophilic interactions between the
enzyme and the membrane material [18,19]. In any case, enzyme
leakage will invariably be an inevitable problem during the bio-
catalytic reaction and membrane storage [3], and enzyme leakage
through the membrane was also observed in the present study,
even though all membrane cut-off limits were well below the
molecular weight of the ADHenzyme (which is 141kDa). However,
as alreadymentioned, theADHis a proteintetramer containingfour
equal subunits, andthus theaveragemolecular weight of its subunit
is about 35kDa [35]. This lowmolecular weight of the subunits may
explain why the 50kDa membrane could not retain the enzyme
for a long ltration time (Fig. 5b), but enzyme leakage was also
detected in the permeate during the immobilization of the lower
molecular cut-off membranes possibly because of the wide pore
size distribution in the membranes even using a membrane with
a small pore size of 10kDa.
Moreover, when the pressure was released, the enzymes were
apparently able to diffuse back to the bulk solution via leakage
from the support layer. For example, for the PLGC and GR61PP
membranes, 0.036 (32%) and 0.016 (7%) mg cm
2
of enzymes,
respectively, diffused back from the membranes after one night
soakage with buffer. For the GR51PP membrane no enzyme leak-
agewas detectedduringstorage. This might beduetothedifference
in the fouling and in turn in the immobilization mechanism;
hence, for the GR51PP membrane, the enzymes were apparently
immobilizedinside the pores (standardblocking model) andwould
0.15 0.12 0.09 0.06 0.03 0.00
0
50
100
150
200
Permeate flux
Conversion
Enzyme loading (mgcm
-2
)
P
e
r
m
e
a
t
e

f
l
u
x

(
L
m
-
2
h
-
1
)
50
60
70
80
90
100
C
o
n
v
e
r
s
i
o
n

(
%
)
Fig. 10. Effect of enzyme loading on permeate ux and conversion for the PLTK
membrane in PP sandwich mode (Fig. 2f) (operating pressure =1bar).
88 J. Luo et al. / Biochemical Engineering Journal 83 (2014) 7989
Biomax 30 PLTK GR61PP PLGC
0
30
60
90
120
P
e
r
m
e
a
t
e

f
l
u
x

(
L
m
-
2
h
-
1
)
Membrane type
"Sandwich" mode
Switch mode
a
Biomax 30 PLTK GR61PP PLGC
0
20
40
60
80
100
C
o
n
v
e
r
s
i
o
n

(
%
)
Membrane type
"Sandwich" mode
Switch mode
b
Fig. 11. Comparison of permeate ux (a) and conversion (b) during reaction in sandwich and switch modes after 24h (operating pressure =2bar).
therefore not readily diffuse back. For the PLGC membrane, the
cake layer formation mainly occurred on the skin layer and the
immobilized enzymes were thus more prone to escape from sup-
port layer. In order to decrease such enzyme leakage, it is obviously
important to immobilize the right amount (or a slight surplus) of
enzyme in the membrane and to select a suitable support layer
with respect to the microporous structure. Therefore, new mem-
brane development or membrane modication for fouling-induced
enzyme immobilization would be novel approaches to diminish
enzyme leakage.
4.4. Flux dilemma
For fouling-induced enzyme immobilization, permeate ux is
inversely proportional to enzyme loading. A high permeate ux
may result in a low conversion due to the short contact time
between the enzyme and the substrate, whereas a (too) low ux
obviously decreases the biocatalytic productivity rate (amount of
product/[amount of enzyme time]) as well as thevolumetric reac-
tor productivity (amount of product/[reactor volume time]). In
the present study, the enzyme loading was excessive in most cases
so the ux during reaction was quite low and the conversion was
high. For example, in Fig. 9, the ux increased by 67% from 15 to
25L m
2
h
1
for the PLGC membrane, due to a 32% enzyme leak-
age, but the conversion did not go down. Thus, a way out of the ux
dilemma is to load the enzyme with just the right enzyme loading
or to balance the pressure to lower the ux so that the conversion
can remain high, and enzyme leakage can be avoided.
Fig. 10 shows the effect of enzyme loading on the permeate ux
andconversionfor thePLTKmembraneruninsandwich modeat a
loweredpressure of 1bar inorder to avoida too highpermeate ux.
As expected, the data clearly showed that the conversion increased
with increased enzyme loading, whereas the permeate ux during
reaction decreased in a decay function in response to the enzyme
loading (Fig. 10). These results also indicated that a low enzyme
loading was sufcient to obtain a satisfactory conversion and ux.
For example, at an enzyme loading of 0.015mg cm
2
, a large ux
of 100L m
2
h
1
was obtained together with a high conversion of
90% (Fig. 10).
Another possible approach for ux improvement is to employ
the switch mode (skin-side down during immobilization while
skin-side up during reaction) as illustrated in Fig. 2g. It is well
known that a reduction of the hydrophilicity of the membrane sur-
face by fouling layer development is one of mechanisms for ux
decline [36]. Here, when operating in the sandwich mode, the
surface of skinlayer was not fouled, but still highly hydrophilic, and
thus the normal mode for the reaction would bring much higher
ux than the sandwich mode. When testing the switch mode,
the permeate ux was elevated for all the membranes, and par-
ticularly high for the PLGC membrane, but the conversion dropped
more signicantly than what was seen for the sandwich mode (at
the same operating pressure of 2bar) (Fig. 11). On the other hand,
there was a large amount of enzymes (0.0150.045mg cm
2
) ee-
ing via the support layer in the switch mode. These data lead to the
conclusion, that there were some enzymes steadily immobilized in
the membrane, via irreversible fouling, and back-washing could
not remove this fouling; secondly, the reduction of the operating
pressure or recirculation of the permeate could be used to balance
the tradeoff of ux andconversion(retentiontime or contact time);
thirdly, high permeate ux and high conversion could be obtained
simultaneously by operating in switch mode if the enzyme leakage
via the support layer could be controlled.
5. Conclusions
Apart from allowing for the reuse of enzymes, the employ-
ment of biocatalytic membranes in enzyme catalyzed conversion
may in many cases reduce the complexity of the production pro-
cess, enabling continuous operation as well as better control of the
catalytic process. This work investigated the effects of membrane
conguration, pore size and materials on ux behavior and enzyme
stability infouling-inducedenzyme immobilizationinmembranes.
Low permeate ux for the membrane conguration in reverse
mode was suggested to be caused by compression of the skin
layer and this membrane compaction could be avoided by pla-
cing an extra support layer beneath the skin layer. A polypropylene
porous support was showntocongure this sandwich membrane
better than a PET support. Operating in switch mode (skin-side
down during immobilization while skin-side up during reaction)
and optimization of the enzyme loading were also alternatives to
increase permeate ux. The reduction in reactive conversion was
mainly caused by enzyme leakage and activity loss. The PLTKmem-
brane could be used to immobilize ADH enzymes in sandwich
mode because the enzyme molecules were retained well by its skin
layer (molecular weight cut-off: 30kDa) and the membrane was
moreover hydrophilic andbiocompatible, producingahighreactive
conversion which was stable after several reaction cycles and 24h
storage. In order to minimize enzyme leakage, membrane devel-
opment or modication, and adding functional reagent to form
combined fouling, are worth doing in the future work.
J. Luo et al. / Biochemical Engineering Journal 83 (2014) 7989 89
Acknowledgement
We thank The Hans Christian rsted Postdoc Program(DTU) for
nancial support.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.bej.2013.12.007.
References
[1] C. Charcosset, K. Fiaty, B. Perrin, R. Couturier, B. Masterrena, Key roles of
enzyme positions and membrane surface potentials in the properties of
biomimetic membranes, Arch. Biochem. Biophys. 424 (2004) 235245.
[2] L. Giorno, E. Drioli, Biocatalytic membrane reactors: applications and perspec-
tives, Trends Biotechnol. 18 (2000) 339349.
[3] A. Sassolas, L.J. Blum, B.D. Leca-Bouvier, Immobilization strategies to develop
enzymatic biosensors, Biotechnol. Adv. 30 (2012) 489511.
[4] P. Jochems, Y. Satyawali, L. Diels, W. Dejonghe, Enzyme immobiliza-
tion on/in polymeric membranes: status, challenges and perspectives in
biocatalytic membrane reactors (BMRs), Green Chem. 13 (2011) 1609
1623.
[5] G.M. Rios, M.P. Belleville, D. Paolucci, J. Sanchez, Progress in enzymatic mem-
brane reactorsa review, J. Membr. Sci. 242 (2004) 189196.
[6] D. Paolucci-Jeanjean, M.P. Belleville, G.M. Rios, N. Zakhia, The effect of enzyme
concentration and space time on the performance of a continuous recycle
membrane reactor for one-step starch hydrolysis, Biochem. Eng. J. 5 (2000)
1722.
[7] L. Donato, C. Algieri, V. Miriello, R. Mazzei, G. Clarizia, L. Giorno, Biocatalytic
zeolite membrane for the production of l-DOPA, J. Membr. Sci. 407408 (407)
(2012) 8690.
[8] D. Sen, A. Sarkar, A. Gosling, S.L. Gras, G.W. Stevens, S.E. Kentish, P.K.
Bhattacharya, A.R. Barber, C. Bhattacharjee, Feasibility study of enzyme
immobilization on polymeric membrane: a case study with enzymatically
galacto-oligosaccharides production from lactose, J. Membr. Sci. 378 (2011)
471478.
[9] L. Mei, R. Xie, C. Yang, X.-J. Ju, J.-Y. Wang, Z. Zhang, L.-Y. Chu, Bio-inspired mini-
eggs with pH-responsive membrane for enzyme immobilization, J. Membr. Sci.
429 (2013) 313322.
[10] G. Ren, H. Yu, Oriented adsorptive immobilization of esterase BioH based on
protein structure analysis, Biochem. Eng. J. 53 (2011) 286291.
[11] Y. Wang, Y.L. Hsieh, Immobilizationof lipase enzyme inpolyvinyl alcohol (PVA)
nanobrous membranes, J. Membr. Sci. 309 (2008) 7381.
[12] L. Giorno, E. Drioli, G. Carvoli, A. Cassano, L. Donato, Study of an enzyme
membrane reactor with immobilized fumarase for production of l-malic acid,
Biotechnol. Bioeng. 72 (2001) 7784.
[13] R. Mazzei, L. Giorno, E. Piacentini, S. Mazzuca, E. Drioli, Kinetic study of a
biocatalytic membrane reactor containing immobilized -glucosidase for the
hydrolysis of oleuropein, J. Membr. Sci. 339 (2009) 215223.
[14] W. Guo, H.-H. Ngo, J. Li, A mini-reviewon membrane fouling, Bioresour. Tech-
nol. 122 (2012) 2734.
[15] C.-C. Ho, A.L. Zydney, Effect of membrane morphology on the initial rate of
protein fouling during microltration, J. Membr. Sci. 155 (1999) 261275.
[16] T.Y. Wu, A.W. Mohammad, J. Md Jahim, N. Anuar, Palmoil mill efuent (POME)
treatment and bioresources recovery using ultraltration membrane: effect of
pressure on membrane fouling, Biochem. Eng. J. 35 (2007) 309317.
[17] D.M. Krsti c, M.G. Antov, D.M. Peri cin, W. Hinger, M.N. Teki c, The possibility
for improvement of ceramic membrane ultraltration of an enzyme solution,
Biochem. Eng. J. 33 (2007) 1015.
[18] S.P. Beier, A.D. Enevoldsen, G.M. Kontogeorgis, E.B. Hansen, G. Jonsson, Adsorp-
tion of amylase enzyme on ultraltration membranes, Langmuir 23 (2007)
93419351.
[19] S. Mazzuca, L. Giorno, A. Spadafora, R. Mazzei, E. Drioli, Immunolocalization
of -glucosidase immobilized within polysulphone capillary membrane and
evaluation of its activity in situ, J. Membr. Sci. 285 (2006) 152158.
[20] A.E. Contreras, A. Kim, Q.L. Li, Combined fouling of nanoltration membranes:
mechanisms and effect of organic matter, J. Membr. Sci. 327 (2009) 8795.
[21] Q.L. Li, M. Elimelech, Synergistic effects in combined fouling of a loose nanol-
tration membrane by colloidal materials and natural organic matter, J. Membr.
Sci. 278 (2006) 7282.
[22] J. Luo, A.S. Meyer, G. Jonsson, M. Pinelo, Fouling-induced enzyme immobiliza-
tion for membrane reactors, Bioresour. Technol. 147 (2013) 260268.
[23] R. Obert, B.C. Dave, Enzymatic conversion of carbon dioxide to methanol:
enhanced methanol production in silica solgel matrices, J. Am. Chem. Soc.
121 (1999) 1219212193.
[24] A. Affandy, E. Keshavarz-Moore, H.K. Versteeg, Applicationof ltrationblocking
models to describe fouling and transmission of large plasmids DNA in sterile
ltration, J. Membr. Sci. 437 (2013) 150159.
[25] A.L. Lim, R. Bai, Membrane fouling and cleaning in microltration of activated
sludge wastewater, J. Membr. Sci. 216 (2003) 279290.
[26] C. Wang, Q. Li, H. Tang, D. Yan, W. Zhou, J. Xing, Y. Wan, Membrane fouling
mechanism in ultraltration of succinic acid fermentation broth, Bioresour.
Technol. 116 (2012) 366371.
[27] F.S. Baskaya, X. Zhao, M. Flickinger, P. Wang, Thermodynamic feasibility of
enzymatic reductionof carbondioxide to methanol, Appl. Biochem. Biotechnol.
162 (2010) 391398.
[28] J.-P. von Wartburg, J.L. Bethune, B.L. Vallee, Human liver-alcohol dehy-
drogenase, kinetic and physicochemical properties, Biochemistry 3 (1964)
17751782.
[29] D.M. Bohonak, A.L. Zydney, Compaction and permeability effects with virus
ltration membranes, J. Membr. Sci. 254 (2005) 7179.
[30] K.M. Persson, V. Gekas, G. Trgrdh, Study of membrane compaction and
its inuence on ultraltration water permeability, J. Membr. Sci. 100 (1995)
155162.
[31] N.A. Li, L. Giorno, E. Drioli, Effect of immobilizationsiteandmembranematerials
on multiphasic enantiocatalytic enzyme membrane reactors, Ann. N.Y. Acad.
Sci. 984 (2003) 436452.
[32] Y.-N. Wang, C.Y. Tang, Protein fouling of nanoltration, reverse osmosis,
and ultraltration membranesthe role of hydrodynamic conditions, solution
chemistry, and membrane properties, J. Membr. Sci. 376 (2011) 275282.
[33] J. Luo, W. Cao, L. Ding, Z. Zhu, Y. Wan, M.Y. Jaffrin, Treatment of dairy efuent by
shear-enhanced membrane ltration: the role of foulants, Sep. Purif. Technol.
96 (2012) 194203.
[34] D. Klemm, B. Heublein, H.-P. Fink, A. Bohn, Cellulose: fascinating biopolymer
and sustainable rawmaterial, Angew. Chem. Int. Ed. 44 (2005) 33583393.
[35] M. Bhner, H. Sund, Yeast alcohol dehydrogenase: SH groups, disulde
groups, quaternary structure, and reactivation by reductive cleavage of disul-
de groups, Eur. J. Biochem. 11 (1969) 7379.
[36] N. Subhi, A.R.D. Verliefde, V. Chen, P. Le-Clech, Assessment of physicochemical
interactions in hollowbre ultraltration membrane by contact angle analysis,
J. Membr. Sci. 403404 (2012) 3240.