Modular microfluidic reactor and inline filtration system for the biocatalytic synthesis of chiral metabolites

Article abstract of DOI:10.1016/j.molcatb.2011.12.010

Biocatalytic synthesis is now well established amongst catalytic methodologies as an extremely useful approach for the industrial synthesis of high-value compounds, due to its many advantages such as high reaction specificity and selectivity. However, engineering a biocatalytic process can be complex and time-consuming. This paper presents a modular microfluidic reactor and in-line filtration system for the rapid and small-scale evaluation of biocatalytic reactions. The system combines a substrate with a biocatalyst in free solution, incubates the two components until full conversion to product has been achieved, before extracting the product. The system has been applied to the transketolase-catalysed reaction of hydroxypyruvate (HPA) and glycolaldehyde (GA) to l-erythrulose, demonstrating complete conversion of substrate to product, complete retention of the enzyme and an overall yield of approximately 65%. The complete conversion of HPA and propanal to (3S)-1,3-dihydroxypentan-2- one with a mutant transketolase further demonstrated the applicability of the microfluidic system for organic synthesis.

Full text of DOI:10.1016/j.molcatb.2011.12.010

Journal of Molecular Catalysis B: Enzymatic 77 (2012) 1–8
Contents lists available at SciVerse ScienceDirect
Journal of Molecular Catalysis B: Enzymatic
journal homepage: www.elsevier.com/locate/molcatb
Modular microfluidic reactor and inline filtration system for the biocatalytic
synthesis of chiral metabolites
a
a
a
a
b
Brian O’Sullivan , Homam Al-Bahrani , James Lawrence , Maria Campos , Armando Cázares ,
a
c
b
a,∗
Frank Baganz , Roland Wohlgemuth , Helen C. Hailes , Nicolas Szita
Department of Biochemical Engineering, University College London, Torrington Place, London WC1E 7JE, UK
Department of Chemistry, University College London, 20 Gordon Street, London WC1H 0AJ, UK
Sigma-Aldrich, Industriestrasse 25, 9470 Buchs, Switzerland
a
b
c
a r t i c l e i n f o
a b s t r a c t
Article history:
Received 9 August 2011
Received in revised form 1 December 2011
Accepted 30 December 2011
Available online 8 January 2012
Biocatalytic synthesis is now well established amongst catalytic methodologies as an extremely use-
ful approach for the industrial synthesis of high-value compounds, due to its many advantages such as
high reaction specificity and selectivity. However, engineering a biocatalytic process can be complex
and time-consuming. This paper presents a modular microfluidic reactor and in-line filtration system
for the rapid and small-scale evaluation of biocatalytic reactions. The system combines a substrate with
a biocatalyst in free solution, incubates the two components until full conversion to product has been
achieved, before extracting the product. The system has been applied to the transketolase-catalysed
reaction of hydroxypyruvate (HPA) and glycolaldehyde (GA) to l-erythrulose, demonstrating complete
conversion of substrate to product, complete retention of the enzyme and an overall yield of approx-
imately 65%. The complete conversion of HPA and propanal to (3S)-1,3-dihydroxypentan-2-one with
a mutant transketolase further demonstrated the applicability of the microfluidic system for organic
synthesis.
Keywords:
Microfluidic reactor
Filtration
Transketolase
Chiral metabolites
Bioprocess microfluidics
Microfluidic downstream processing
©
2012 Elsevier B.V. All rights reserved.
1
. Introduction
synthesis and work-up, the power of modular networks consist-
ing of microfluidic reactor and separator units that can easily be
arranged into different configurations has already been shown
[11,12]. We want to demonstrate that microfluidic devices can
equally provide a modular toolbox for the development of whole
biocatalytic processes, i.e. for both the upstream and downstream
unit operations.
The impact of process integration (i.e. the integration of a bio-
catalytic reaction with the subsequent separation and purification
steps) for efficient product recovery and process intensification has
already been shown in conventional reactors [13,14]. In batch as
well as in continuous-flow systems, there is a need for the sep-
aration of the biocatalyst and the product, either to retain the
biocatalyst for reuse, to purify the product or to quench the reac-
tion (in the case of kinetic studies). This can be achieved either
by immobilisation of the biocatalyst or by ultrafiltration (typi-
cally in tangential or cross-flow filtration mode), separating the
high-molecular weight biocatalyst from the desired low-molecular
weight product. A variety of biocatalyst immobilisation meth-
ods on solid supports as well as biocatalyst crosslinking methods
have been described for use in microfluidic reactors [5,7,10,15–20].
Whilst immobilisation may improve the stability of the biocata-
lyst, it adds further complexity to the fabrication and reduces the
flexibility of easy-to-use microfluidic reactors. On the other hand,
Inspired by the rapid advances and successes of microflu-
idic reactors for chemical synthesis [1] and underpinned by the
development of microfluidic enzyme reactors as analytical tools
for proteomic studies [2], several groups have begun to develop
and explore the use of microfluidic reactors in the field of bio-
chemical synthesis [3–9]. Many advantages have been reported
with the use of microfluidic reactors, including more precise con-
trol over reaction conditions, real time monitoring of reaction
efficiency, more effective thermal transfer due to larger sur-
face area to volume ratios, reduced use of resources, and an
increased throughput via parallelisation [1,10]. These advantages
can potentially enable the rapid evaluation of different reaction
conditions, overcoming the time constraints typically associated
with biocatalytic process development. However, for real imple-
mentation in industrial settings it will be necessary to embed
the reaction (upstream) steps with separation and purification
(
downstream) steps to enable continuous synthesis and isola-
tion of pharmaceutical intermediates or metabolites. For chemical
∗ Corresponding author. Tel.: +44 020 7679 4418; fax: +44 020 7916 3943.
E-mail address: n.szita@ucl.ac.uk (N. Szita).
1
381-1177/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.molcatb.2011.12.010

2
B. O’Sullivan et al. / Journal of Molecular Catalysis B: Enzymatic 77 (2012) 1–8
membrane separation processes for ultrafiltration are well estab-
lished downstream operations in industrial production processes
towards the recovery of products [21], and simple ‘clamp-and-play’
type devices for the integration of membranes into microfluidic
systems have previously been established [22].
A loop/recirculation microfluidic enzyme reactor integrated
with a regenerated cellulose membrane for separation of low-
molecular weight pectin hydrolysates from high-molecular weight
pectin has recently been developed [8]. In this configuration,
the membrane was integrated together with the reactor. This
integrated configuration has been useful when applied to high-
molecular weight pectin substrates, where the product removal
occurs due to the large molecular weight difference between the
pectin substrate and the low-molecular weight pectin fragments
Scheme 1. Two transketolase-catalysed reactions studied in this research: the for-
mation of l-erythrulose from glycolaldehyde (A), and the formation of DHP from
propanal (B).
(
product). However, such an integrated configuration is not directly
applicable to the vast number of biocatalytic reactions involv-
ing transformations of low-molecular weight substrates. These
substrates would pass through the membrane before they were
converted to product. We therefore sought to engineer a sys-
tem which would work with low-molecular weight substrates
and where the design of the microfluidic system would allow
for the individual optimisation of the reaction and separation
parameters.
2
. Materials and methods
2.1. Reagents and analysis
Unless otherwise stated, all chemicals and reagents
(Sigma–Aldrich, Gillingham, UK) were used without further
For the efficient separation of high-molecular weight enzymes
from the desired low-molecular weight product, the filtration
method has to be robust, compatible with the microreactor and
resistant to fouling. Tangential flow regimes are preferable as they
help prevent the build up of macromolecules on the membrane
surface [23,24]. This should allow the system to be operated for a
longer period of time without changing or cleaning the membrane,
as well as reducing the amount of biocatalyst lost in the process.
In tangential flow systems, the transmembrane pressure is a cru-
cial parameter for successful operation. It is therefore desirable
to be able to test different membranes and operating conditions
for the tangential flow filtration independent from the reaction
conditions.
The optimisation of the reaction parameters in the microflu-
idic reactor requires a well-defined biotransformation yielding
a single product. We have selected the transketolase-catalysed
two-carbon addition of hydroxypyruvate to glycolaldehyde and
propanal. Transketolases have been shown to be attractive biocat-
alysts in the asymmetric synthesis of chiral metabolites [25–30]
with a wide substrate tolerance and high specificity of product for-
mation. The use of hydroxypyruvate enables the formation of new
carbon carbon bonds with excellent enantioselectivity and allows
the reaction to go to completion because of the release of carbon
dioxide. The coupling of transketolase reactions with either a sub-
sequent transaminase reaction [31] or a preceding triosephosphate
isomerase reaction [32] has on one hand already demonstrated the
opportunities for combining the transketolase reaction step with
other reaction steps, but has on the other hand also shown the
importance of a good selection and control of the reaction param-
eters due to instabilities of reactants.
In this work, a modular microfluidic reactor and a tangential
flow filtration system have been used with the enzyme transke-
tolase (TK) to produce a chiral metabolite, l-erythrulose (Reaction
A, Scheme 1), and to subsequently separate this metabolite from
the biocatalyst, transketolase, producing two distinct outputs –
a purified product solution and a concentrated catalyst solution.
We investigated the conditions required to obtain full conver-
sion and separation of the product. The system was further used
with a more hydrophobic substrate, propanal, to give (3S)-1,3-
dihydroxypentan-2-one (DHP) (Reaction B). To our knowledge, this
is the first study to combine metabolite synthesis using a biocat-
alyst in solution with an inline separation step on a microfluidic
platform.
purification. For HPLC calibration standards, commercially avail-
able l-erythrulose was purified by flash silica chromatography as
described by Smith et al. [33].
Protein concentrations were measured by SDS–PAGE elec-
trophoresis with 12% Tris–glycine resolving gel, using purified TK
standards. 20 ␮g of total protein was applied to each lane and the
samples were stained with Coomassie Blue R-250.
HPLC analysis of products from transketolase-catalysed reac-
tions was performed on an Aminex (Biorad, Hemel Hempstead,
UK) ion-exchange column (HPX-87H, 300 mm × 7.8 mm), mobile
−1
phase: 0.1% (v/v) aqueous trifluoroacetic acid (TFA) at 0.6 mL min
.
HPA and l-erythrulose were quantified from calibration curves of
standard solutions, and detected by UV absorption at 210 nm.
Enantiomeric excess data for the formation of l-erythrulose
were calculated from chiral HPLC analysis of the derivatised
ketodiols compared to racemic standards [33,34]. The enan-
tiomeric excess for (3S)-1,3-dihydroxypentan-2-one was obtained
by derivatisation and chiral HPLC [26].
2.2. Fabrication of microfluidic components
The enzymatic microfluidic reactor and the rigid parts of
the filtration unit were machined from poly(methylmethacrylate)
PMMA) with a micromilling machine (Folken IND, Glendale, USA).
(
Microreactor PMMA components were sealed by thermal bonding
◦
(
1.5 h, 105 C).
Gaskets for the filtration unit were made from
poly(dimethylsiloxane) (PDMS, Sylgard 184, Dow Corning, Mid-
land, USA), prepared according to the manufacturer’s instructions
and cast in PMMA moulds which had been machined as described
above.
Membranes for the filtration unit were cut to size from commer-
cially available regenerated cellulose sheets (Durapore, Millipore,
Cork, Ireland) using a CO laser marking head (Synrad Inc., Mukil-
2
teo, WA, USA). The molecular weight cut-off of the membrane was
1
0 kDa.
Interconnect ports were milled from 5 mm PMMA, with two
holes tapped with an M3 thread for attachment to the device, and
an M6 threaded hole to allow standard connection fittings to be
used (P-221, Upchurch Scientific, Oak Harbor, WA, USA). A more
detailed description of the fabrication can be found in the Electronic
Supplementary Information (ESI 1).

B. O’Sullivan et al. / Journal of Molecular Catalysis B: Enzymatic 77 (2012) 1–8
3
2.3. Staggered herringbone mixer studies
2.7. Filtration: back pressure measurements
Two solutions (reverse-osmosis purified water and fluorescein-
conjugated bovine serum albumin in phosphate buffered saline, pH
) were flowed through separate inlet ports of two microreactors
of identical dimensions, one with a passive mixer, one without.
Images were taken at 10× magnification using a microscope fitted
with a CCD camera under fluorescent light with a FITC filter (Nikon
Instruments Europe B.V., Kingston, UK).
Constricting capillaries (internal diameter 50, 75, 100 and
200 ␮m) were attached to the retentate outlet of the filtration
device in cross-flow mode (Fig. 2D). The resulting back pressure
was measured with an inline pressure sensor (40PC100, Honeywell,
NJ, USA), and the data logged with a simple LabVIEW instrument
(National Instruments, TX, USA).
7
2.8. Filtration: proof of concept
2
.4. Preparation of transketolase lysate
The filtration unit was charged with Tris buffer (50 mM, pH 7).
A model stock solution of TK (0.22 mg mL ) and l-erythrulose
50 mM) in Tris buffer (50 mM, pH 7) was pumped through the
retentate inlet with a syringe drive, whilst Tris buffer was pumped
through the permeate inlet. Samples were collected at the retentate
and permeate outlets for analysis by HPLC and SDS–PAGE.
−
1
Transketolase lysates were prepared according to the method
(
of Matosevic et al. [35]. For the wild type TK, overnight cultures
of Escherichia coli BL21gold(DE3) (with transketolase-producing
plasmid pQR791) were grown in 200 mL shake-flasks from inoc-
ulation of 20 mL Lysogeny Broth (LB)–glycerol with a single colony
obtained by streaking out cells from glycerol stocks (25%, v/v, glyc-
For the integration of the microfluidic reactor and the filtration
unit, a hydroxypyruvate/glycolaldehyde solution (both 100 mM)
◦
erol stored at −80 C) on (LB)–agar plates. This was incubated for
−
1
and a TK/cofactors solution (0.68 mg mL TK) were combined and
◦
1
2–16 h at 37 C. The resulting cell suspension was transferred to
−
1
flowed through a microfluidic reactor at 10 ␮L min . Final reagent
and enzyme concentrations were identical to the batch reaction
conditions in Section 2.5. The output was connected to a filtration
unit via a length of tubing to give the desired residence time. A
a conical flask with 200 mL LB–glycerol medium, which was incu-
bated at 37 C until the bacterial growth had reached stationary
phase, as confirmed by optical density measurements. The contents
of the flask were transferred to 50 mL falcon tubes and centrifuged
at 5000 rpm for 10 min. The supernatant was discarded and the cell
◦
5
0 ␮m i.d. capillary was fitted to the retentate outlet of the filtra-
tion unit. The reaction was run for a total of 10 h. Samples taken
from the retentate and permeate outlets were analysed by HPLC.
◦
paste was frozen at −80 C until needed for lysis and purification.
For lysis, the cell paste was resuspended in 5 mL 50 mM Tris
buffer, cooled on ice and sonicated (10 cycles of 10 s on, 10 s off)
with a sonication probe (Soniprep 150, Sanyo, Japan). The suspen-
sion was then centrifuged and filtered with a 0.2 ␮m filter, and
3. Results and discussion
◦
stored at −80 C until required.
3.1. Design considerations for the modular microfluidic system
For the mutant D469T TK, the same procedure was followed
except the pQR791 plasmid had been specifically mutated accord-
ing to the methods of Hibbert et al. [27,28].
3.1.1. Enzymatic continuous flow microfluidic reactor
In previous work, an immobilised enzyme microreactor for
quantitative kinetic analysis of TK-catalysed reactions [16] was
developed. Whilst this microreactor enables the rapid quantifica-
tion of enzyme kinetic parameters, achieving complete substrate
conversion with this microreactor is difficult. For biocatalytic pro-
cess development, however, it is desirable to quickly achieve
complete substrate conversion, and to have a tool that enables the
rapid evaluation of the reaction conditions that lead to complete
conversion. We therefore investigated the use of a microfluidic
reactor with solubilised enzymes, where complete conversion
with carbon carbon bond forming enzymes has previously been
reported [9]. Similar to this and other solubilised enzyme reactors,
in our microfluidic reactor enzyme and substrate are co-flowed
through a T-channel. The residence time is determined by the com-
bined flow rate of the substrate and enzyme solution, and different
reaction conditions are obtained by altering the individual flow
rates and the concentrations. In contrast to other microfluidic reac-
tors, we used wider and deeper channels (width 500 ␮m, depth
250 ␮m, length 89.5 mm, Fig. 1A). These larger structures signifi-
cantly reduce the long transition times that typically occur in very
shallow channels when the flow rate is changed [36]. Shortened
transition times translate to less ‘dead’ time during the evaluation
of different reaction conditions where the flow rate is frequently
changed. Additionally, by using slightly larger channel structures,
it was possible to machine both the reactor and filtration unit from
the same material, and with the same fabrication technology. This
unified approach will simplify further integration of the reactor
and separator units into easily configurable modular synthesis net-
works.
2
.5. Batch transketolase reaction
For the reaction of HPA with glycolaldehyde, 30 ␮L wild type TK
−1
lysate (4.47 mg mL ) was incubated with 170 ␮L cofactor solution
23.1 mM MgCl , 5.7 mM thiamine diphosphate in 50 mM Tris–HCl
(
2
buffer, pH 7) for 20 min at room temperature. 150 ␮L of this solution
was added to 150 ␮L substrate solution (100 mM hydroxypyru-
vate, 100 mM glycolaldehyde in 50 mM Tris–HCl buffer, pH 7) in a
9
6-well plate to initiate the reaction (final concentrations: 50 mM
−1
HPA, 50 mM GA, 0.34 mg mL wild type TK, 2.4 mM ThDP, 9.8 mM
MgCl ). Aliquots (20 ␮L) were removed at the required time inter-
vals, quenched with 180 ␮L 0.1% (v/v) aqueous TFA, centrifuged
2
(
5000 rpm, 5 min) and the supernatant analysed by HPLC.
The same protocol was used for the mutant TK-catalysed reac-
tion of HPA with propanal (final reaction conditions: 50 mM HPA,
7
MgCl , room temperature, pH 7).
−1
0 mM PA, 0.33 mg mL TK mutant D469T, 2.4 mM ThDP, 9.8 mM
2
2
.6. Continuous flow microfluidic transketolase reaction
TK lysate solutions and substrate solutions were prepared at
the concentrations described above. Using a twin syringe drive, the
solutions were pumped into separate inlets of the microfluidic reac-
tor at the same flow rate for at least 1.5 times the residence time
before collecting the output in pre-weighed sample vials containing
4
00 ␮L 0.1% TFA. The quenched reaction mixture was centrifuged
To make valid comparisons between the kinetics of the microflu-
idic reactor and those of the batch, it is essential that the reagents
are thoroughly mixed very quickly after being brought together.
and analysed by HPLC as above. Residence times were calculated
by dividing the reactor volume (60 ␮L) by the flow rate.

4
B. O’Sullivan et al. / Journal of Molecular Catalysis B: Enzymatic 77 (2012) 1–8
Fig. 1. (A) Exploded view of microfluidic T-junction reactor: channel dimensions are 0.5 mm (w) × 89.5 mm (l) × 0.25 mm (d) and the total reactor volume, including tubing
◦
is 60 ␮L. (B) A staggered herringbone micromixer (SHM) was integrated into the reaction channel. The asymmetric chevron grooves were 100 ␮m apart, and set at 45 to the
−
1
direction of flow. (C) Fluorescein-tagged bovine serum albumin (BSA, 0.3 mg mL ) was used to illustrate the effectiveness of SHM reactor compared with a control (straight)
channel of the same size.
However, in the absence of turbulent fluid flow, mixing in microflu-
idic reactors is diffusion limited, which can be problematic when
dealing with large protein molecules with low diffusivities. To
overcome this, passive mixing structures based on a modified ver-
sion of Stroock’s [37] “staggered herringbone micromixer” (SHM),
described by Ansari and Kim [38] were machined into the “floor” of
the channel. The dimensions of these asymmetric chevron grooves,
prevent the use of faster flow rates in the reactor, reducing its
versatility.
3.1.2. Filtration device
The filtration device was designed to seal membranes in place
by clamping, rather than bonding, in order to allow the integration
of any membrane type (regardless of material) and so maximise
the flexibility of the system. The device consisted of two PMMA
plates, both with a straight channel (width 1 mm, depth 1 mm,
length 45 mm) and two 1 mm holes milled through at either end to
allow fluidic connection (Fig. 2A–C). A recess (1 mm wide, 0.75 mm
deep) around the channel on one plate held a PDMS gasket in
place, and the filter membrane sat in the resulting enclosure. The
plates were clamped together by means of M3 screws along the
edges, creating a reversible seal around the membrane that did
not require the use of adhesives. In this way, the filter could be
quickly repurposed to other applications. This, combined with the
standardised interconnects used on both the filter and the reac-
tor, allows the system to be reconfigured to test different process
conditions.
To characterise the burst pressure of the filtration unit, the
ports on the permeate side of the membrane were capped off,
whilst air was pumped into the retentate side with a gas-tight
syringe (Fig. 2D). Pressure was logged with a pressure sensor, and
showed steady increases with increasing volume of air pumped
into the system; a sudden pressure drop indicated the bursting
point (Fig. 3A). Burst pressures varied with the torque applied to
the screws on both the main assembly and the ports, but some-
what counter-intuitively, higher applied torques resulted in lower
burst pressures. This may be due to deformation of the rigid frames
of the assembly, leading to poor contact between the frame and the
PDMS gasket.
◦
set at 45 to the direction of flow, were 100 ␮m (l) × 120 ␮m (d),
across the full width of the channel (500 ␮m), arranged in 7 × 10
cycles (Fig. 1B).
To demonstrate the efficiency of mixing in our microreactor
when large protein molecules are involved, T-channel reactors
of identical size were fabricated, with and without passive mix-
ing structures. Flow visualisation was carried out with two
input solutions: reverse osmosis purified water and a solution of
−
1
fluorescein-conjugated bovine serum albumin (BSA, 0.3 mg mL ).
These two solutions were run through the inlets at various flow
rates, and images were taken at the mid-point of the reactor chan-
nel, 4.5 cm from the T-junction (Fig. 1C).
At higher flow rates (1.5–3 mL min−1), at the mid-point of the
channel, there was a distinct boundary between the labelled and
non-labelled fluid streams in the non-SHM channel, whilst the
streams in the SHM channel appear to be completely mixed. At
−1
lower flow rates (0.03–0.3 mL min ) the two streams appeared
to be well mixed in both of the reactors, with or without SHM
structures, the residence times (at this point of the channel) being
sufficiently long to allow diffusion of the protein across the width
of the channel. The mixing in the SHM reactor was further charac-
terised with confocal microscopy (see ESI 2).
From these results, one could conclude that the SHM might
not be necessary, at least for ‘slow’ reaction systems where
longer residence times (and thus lower flow rates) are required
to achieve complete conversion. However, co-flowing enzyme and
substrate into a microfluidic channel without rapid and induced
mixing will initially lead to partially mixed and unmixed zones
Control of back pressure on the retentate output was achieved
by constricting fluid flow in capillaries of specific length and diam-
eter. Capillaries with lengths of 30 cm and internal diameters of
50, 75, 100 and 200 ␮m (Upchurch) were connected to the out-
put of the retentate channel, purified water (reverse osmosis) was
run through both the retentate and permeate channels at various
flow rates, and the back pressure recorded using a pressure sensor
[
39]. It will be difficult therefore to judge where the two solu-
tions are intermixed enough to form the ‘final’ concentrations
for the reaction and hence impossible to accurately measure the
effective residence time. A lack of SHM structures would also

B. O’Sullivan et al. / Journal of Molecular Catalysis B: Enzymatic 77 (2012) 1–8
5
Fig. 2. (A) Exploded view of tangential-flow filter – a membrane fits inside the space created by the gasket and is sealed in place when permeate/retentate plates are clamped
together. This ‘clamp and play’ principle does not rely on adhesives, and allows the testing of different membrane types. Additionally, the filter uses the same interconnects as
the reactor, allowing the reactor/filter system to be easily reconfigured to test different process conditions; (B) photograph of assembled filter after assembly with M3 screws;
(
C) schematic of setup used to evaluate the robustness of the sealed filter under pressure; (D) schematic of setup used to measure the effect of outlet capillary dimension on
back pressures.
(
Honeywell, NJ, USA) with a LabView interface (National Instru-
significantly different in conversion time or e.e. (87%). Data are
accurate to ± 4%.
ments, TX, USA).
It was found that using capillary diameters of 50 ␮m gave the
highest permeate yield at the operating flow rates required by the
Similar to the HPA reaction with GA, the reaction of HPA with
propanal (Fig. 4B), catalysed by the mutant TK, showed no sig-
nificant difference between the batch and microfluidic reaction
rates (these were faster than the erythrulose reaction, approaching
completion after 60 min). Enantiomeric excess values of 66% and
65% (± 4%) were determined for the DHP product in the batch and
microfluidic reactors respectively, consistent with earlier reports
[26,33].
−1
reactor. At flow rates of 10–30 ␮L min , with capillary diameters
of 50, 75, 100 and 200 ␮m, the pressures achieved were in the range
of 1–15 psi (Fig. 3B). Back pressure was found to increase linearly
with increasing flow rate, and with decreasing capillary diame-
ter. Overall, control of back pressure with capillaries was found
to be more reproducible than with commercially available micro-
metering valves, with the added advantages of being considerably
cheaper, and having no moving parts.
Since the mode of enzyme/substrate interaction is solution-
based for both the batch and continuous flow reactors, we would
expect similar reaction rates and e.e. data for each system.
Microwell platforms such as 96-well plates are standard equip-
ment for the study of bioprocesses, offering rapid data generation
and process optimisation [40]. However, the continuous operation
of the microfluidic system allows us to move beyond probing and
optimising reaction conditions and switch to small-scale produc-
tion without further changes to our experimental set-up. At condi-
tions where 100% conversion is achieved, the microreactor is capa-
3
.2. Transketolase reactions
In order to demonstrate the general applicability of the microflu-
idic reactor for biocatalysis, two model TK reactions were studied:
the wild-type TK-catalysed reaction of hydroxypyruvate (HPA)
with glycolaldehyde (GA), and the mutant D469T TK-catalysed
reaction of HPA with propanal. Using HPA as a substrate produces
carbon dioxide as a by-product, which drives the reaction to com-
pletion. The reaction profiles from the batch (96-well plate) and the
microfluidic reactors are shown in Fig. 4.
Under standard conditions, the erythrulose batch reaction was
found to go to completion in approximately 120 min (Fig. 4A).
Reaction data were quantified from the calibration curves derived
from standard solutions of purified erythrulose, and concentrations
were measured and found to be at 50 mM, within the margins
of error. The time to full conversion is comparable to the work
of Matosevic et al. [16]. There was good mass balance between
the initial HPA concentration, and the sum of HPA depletion and
erythrulose formation throughout, and the enantiomeric excess
−
1
−1
ble of producing 0.72 mg h of erythrulose, or 0.36 mg h DHP.
Previous work on enzymatic microfluidic reactors has tended to
focus on supported enzymes [3,7,16–18,41]. Although these exam-
ples show reduced enzyme activity and may involve complicated
immobilisation techniques, they are favoured because there is less
risk of clogging microchannels with cell debris. We found that the
use of partial clarification techniques, where the crude lysate was
centrifuged to spin out the cell debris and the supernatant sepa-
rated off and filtered, gave solutions with high enzymatic activity
which did not clog the channels.
Synthetic pathways of many high-value compounds involve
multiple steps. The full conversion (without the need for
recirculation) observed in the microfluidic reactor is of great
(
e.e.) was 90%. The corresponding microfluidic reaction was not

6
B. O’Sullivan et al. / Journal of Molecular Catalysis B: Enzymatic 77 (2012) 1–8
Fig. 3. (A) To test the sealing strength of the ‘clamp and play’ principle, the filtra-
tion unit was assembled with various screw torques (ST = torque on main assembly
screws; PT = port attachment screws; units are Ncm for both). Pressure profiles were
generated by pumping air into the unit using the burst pressure set-up shown in
Fig. 2C. Sudden drop of pressure indicates burst pressure; (B) to generate the back
pressure necessary for optimum permeate yield, capillaries of various inner diame-
ters were connected to the filtration unit (cf. Fig. 2D) and analysed for back pressure
at various flow rates.
Fig. 4. (A) Reaction profiles in batch and microfluidic reactors for the TK-catalysed
reaction of HPA (᭹ = microfluidic, ꢀ = batch reactor) with GA to produce l-
erythrulose (ERY, ꢀ = microfluidic, ꢁ = batch). Reaction conditions: 50 mM HPA,
−1
50 mM GA, 0.34 mg mL
E. coli TK wild type, 2.4 mM thiamine diphosphate
(
ThDP), 9.8 mM MgCl2, room temperature, pH 7. Batch data show average of two
experiments run in triplicate (error bars are half-min/max of the two averages),
microfluidic data is from a single experiment run in triplicate (error bars are stan-
dard deviation). (B) Reaction profiles for mutant TK-catalysed reaction of HPA
(
ꢁ
᭹ = microfluidic, ꢀ = batch reactor) with propanal to produce DHP (ꢀ = microfluidic,
−
1
= batch). Reaction conditions: 50 mM HPA, 70 mM propanal, 0.33 mg mL E. coli
significance, as the product of one reactor may become a contin-
uous source of a required substrate for another, when integrated
into a larger modular platform.
transketolase mutant D469T, 2.4 mM ThDP, 9.8 mM MgCl2, room temperature, pH
7. Error bars represent standard deviation from three experiments.
3.3. Filtration experiments, “model reactor output”
If we wish to connect reactor modules in series for a multi-step
reaction, it is desirable to have a purification step for the prod-
uct of the first reactor before it enters the next one. To this end,
an inline tangential flow filtration was investigated, using erythru-
lose and lysate solutions designed to mimic the expected output
from the reactor. SDS–PAGE studies from a representative selec-
tion of flow conditions confirmed that a membrane with 10 kDa
molecular weight cut-off was sufficient to ensure the full retention
of transketolase along with all other protein in the lysate (Fig. 5).
Similar concentrations of l-erythrulose were found in the perme-
ate and retentate solutions (as measured by HPLC), indicating that
the compound is able to permeate the membrane freely, without
adsorption or other interactions.
To improve the yield of the product, therefore, it is essential
that the volume of permeate fluid is maximised, which can be done
by increasing the back pressure on the retentate outlet, forcing
more fluid across the membrane. To investigate this, 30 cm cap-
illaries with internal diameters of 100 ␮m, 75 ␮m and 50 ␮m were
connected to the retentate outlet, and the permeate and retentate
volume outputs measured under various flow conditions (Table 1).
Fig. 5. SDS–PAGE of samples taken before and after filtration. The samples from
the permeate stream (P1, P2, P3) show no proteins present, implying that the TK
is completely retained by the filter. Pre = pre-filtration sample, R = retentate sample,
and P = permeate sample. Total input to filter was calculated at 15.2 mg, output from
retentate 11.7 mg.
−
1
Interestingly, increasing the flow rate from 10 to 30 ␮L min
had little effect on the permeate yield for any given capillary, but

B. O’Sullivan et al. / Journal of Molecular Catalysis B: Enzymatic 77 (2012) 1–8
7
Table 1
Comparison of permeate volumetric output as a percentage of the total filter output, and the back pressures generated, under various conditions of capillary diameter and
flow rate.
−
Flow rate/␮L min
1
Capillary diameter/␮m
10
20
30
5
7
1
0
5
00
% permeate (pressure/psi)
% permeate (pressure/psi)
% permeate (pressure/psi)
89(5.3)
85 (5.3)
31(2.7)
56 (14)
41 (10)
24(4.6)
67 (20)
36 (15)
25(6.9)
The long-term performance of the filter could be improved
in several ways. Modifying the reaction conditions to reduce the
residence time (for example, by increasing TK concentration) would
also reduce the amount of time required to prime (and equili-
brate) the system, thereby minimising the material lost during
this stage of the process. The filter itself could be also modified to
improve the surface area available for product transmission. Finally,
the back-washing process could be incorporated into the system
and automated with the help of electronically actuated valves and
pumps to reduce the build-up of cellular debris on the membrane
surface.
4. Conclusion
In this work we have successfully demonstrated the use of a
modular continuous flow microfluidic reactor with in-line sepa-
ration of the biocatalyst for the transketolase-catalysed coupling
of the irreversible donor hydroxypyruvate to the acceptors glyco-
laldehyde and propanal. Complete or nearly complete conversion
has been achieved rapidly in the microfluidic reactor for two model
reaction systems which use substrates of different hydrophobic
properties. The rate of conversion to product is comparable to
batch reactors, thus facilitating high-throughput evaluation of bio-
catalytic process conditions and rapid process development. In
addition, the reaction times can be shortened by increasing the
activities of the biocatalysts utilised.
This is especially interesting for the design and reaction develop-
ment of biocatalytic carbon carbon bond formation systems with
labile substrates or products [30,32,42] leading to unwanted non-
enzymatic side-reactions.
We also report the complete separation of the transketolase
enzyme from the product stream by means of a tangential flow fil-
tration device, affording yields of approximately 65%. These yields
could be improved by relatively simple modifications to the reac-
tion conditions and the design of the system.
Fig. 6. Permeate output as a percentage of the total from a filtration unit connected
to an enzymatic reactor for long-term continuous production of l-erythrulose. A
back-flush of the membrane with purified water at 420 min partially restored the
permeate yield.
the yield notably improved as capillaries with decreasing internal
diameter were used, regardless of the flow rate operated. It is pos-
sible that although an increased flow rate would increase the back
pressure, and thus the permeate yield, under these conditions more
material would be pushed into the membrane and fouling would
be increased, negating any increase in the driving force.
Optimum conditions were found to hold with a 50 ␮m capillary,
−1
operating at a flow rate of 10 ␮L min ; this gave a permeate yield
of 95%, and the flow rate was compatible with an optimal residence
time for an upstream microfluidic reactor.
3.4. Integration of microfluidic reactor with filtration device
In order to investigate the continuous synthesis of erythru-
With significant advances in bioreactor miniaturisation made,
the next logical step for micro-scale bioprocessing is the integra-
tion of downstream unit operations [43]. Whilst such efforts have
been made in batch reactors, there has been little work on the
development of equivalent microfluidic systems. We believe that
our integration of a microreactor for synthesis and in-line down-
stream processing on a microfluidic platform is a significant step
towards the modular cascading of reaction and separation units,
and thus towards the realisation of complete “lab on chip” devices
for the development of whole biocatalytic processes [44,45]. The
modular aspect of the system facilitates the easy integration of
new features, in particular analytical devices, which are increas-
ingly important in microfluidic bioprocess development [46]. It also
opens new operational opportunities for laboratory-scale devel-
opment and production of a number of metabolites. The system
could be used to rapidly optimise operational conditions (for exam-
ple, enzyme concentration) to make industrial synthesis processes
more economically viable. Furthermore, beyond the simple transfer
of unit operations from conventional to microfluidic formats, this
development is also of interest to solve fundamental challenges in
lose with our system, the output of the microfluidic enzymatic
reactor was connected in series with a filtration unit and run
for 10 h. Conditions had been optimised for the reactor, with GA
−
1
and HPA concentrations of 50 mM, 3 mg mL TK, and a flow rate
−1
of 10 ␮L min , corresponding to a residence time of 120 min. A
capillary of 50 ␮m diameter was fitted to the retentate outlet to
maximise the permeate yield.
HPLC analysis showed that complete conversion of the GA and
HPA to erythrulose occurred throughout the experiment – there
was no loss of activity of the TK. The yield of erythrulose recovery
was therefore entirely dependant on the partition of fluid between
the permeate and retentate. It was found (Fig. 6) that the permeate
output was stable at almost 100% for 2 h (once the remaining Tris
required for the priming of the system had been flushed out), after
which it declined sharply, presumably due to build-up of lysate
material on surface of membrane, reducing the permeability and
therefore the flux. After 7 h, the filtration unit was back-washed
with purified water in an attempt to remove this material, and the
permeate yield was restored to 70–80%.

8
B. O’Sullivan et al. / Journal of Molecular Catalysis B: Enzymatic 77 (2012) 1–8
the molecular and engineering aspects of interfacing two and more
reaction steps, which would be incompatible in the same reaction
compartment, for the important biocatalytic multi-step synthesis
of metabolites [47].
[16] S. Matosevic, G.J. Lye, F. Baganz, Biotechnol. Prog. 26 (2010) 118–126.
[
17] K.F. Schilke, K.L. Wilson, T. Cantrell, G. Corti, D.N. McIlroy, C. Kelly, Biotechnol.
Prog. 26 (2010) 1597–1605.
[
18] R. Moser, T. Chappuis, E. Vannioli, S. Crefier, O. Naef, Chimia 64 (2010)
799–802.
[
[
19] M.A. Holden, S.Y. Jung, P.S. Cremer, Anal. Chem. 76 (2004) 1838–1843.
20] A. Srinivasan, H. Bach, D.H. Sherman, J.S. Dordick, Biotechnol. Bioeng. 88 (2004)
Acknowledgements
528–535.
[
[
[
[
21] R. van Reis, A. Zydney, J. Membr. Sci. 297 (2007) 16–50.
22] J. de Jong, R.G.H. Lammertink, M. Wessling, Lab Chip 6 (2006) 1125–1139.
23] M. Kalyanpur, Mol. Biotechnol. 22 (2002) 87–98.
24] I.S. Ngene, R.G.H. Lammertink, M. Wessling, W. van der Meer, J. Membr. Sci.
346 (2010) 202–207.
The authors would like to acknowledge the BBSRC, EPSRC
and CONACYT for providing funding for studentships to James
Lawrence, Homam Al-Bahrani and Armando Cázares respectively.
The authors would also like to thank David Steadman for his help
in purifying l-erythrulose for HPLC standards and Marcel Reichen
for his support on fabrication.
[25] R. Wohlgemuth, M.E.B. Smith, P.A. Dalby, J.M. Woodley, in: M.C. Flickinger (Ed.),
Encyclopaedia of Industrial Biotechnology: Bioprocess, Bioseparation and Cell
Technology, Wiley, Hoboken, USA, 2010, pp. 4746–4752.
[26] A. Cázares, J.L. Galman, L.G. Crago, M.E.B. Smith, J. Strafford, L. Ríos-Solís, G.J.
Lye, P.A. Dalby, H.C. Hailes, Org. Biomol. Chem. 8 (2010) 1301–1309.
[
[
[
[
27] E.G. Hibbert, T. Senussi, S.J. Costelloe, W. Lei, M.E.B. Smith, J.M. Ward, H.C. Hailes,
Appendix A. Supplementary data
P.A. Dalby, J. Biotechnol. 131 (2007) 425–432.
28] E.G. Hibbert, T. Senussi, M.E.B. Smith, S.J. Costelloe, J.M. Ward, H.C. Hailes, P.A.
Dalby, J. Biotechnol. 134 (2008) 240–245.
Supplementary data associated with this article can be found, in
the online version, at doi:10.1016/j.molcatb.2011.12.010.
29] F. Charmantray, V. He ˇı laine, B. Legeret, L. Hecquet, J. Mol. Catal. B: Enzym. 57
(
2009) 6–9.
30] R. Wohlgemuth, J. Mol. Catal. B: Enzym. 61 (2009) 23–29.
References
[31] C.U. Ingram, M. Bommer, M.E.B. Smith, P.A. Dalby, J.M. Ward, H.C. Hailes, G.J.
Lye, Biotechnol. Bioeng. 96 (2007) 559–569.
[
[
[
[
32] J. Shaeri, I. Wright, E. Rathbone, R. Wohlgemuth, J.M. Woodley, Biotechnol.
Bioeng. 101 (2008) 761–767.
33] M.E.B. Smith, E.G. Hibbert, A.B. Jones, P.A. Dalby, H.C. Hailes, Adv. Synth. Catal.
[
[
[
[
1] C. Wiles, P. Watts, Eur. J. Org. Chem. 10 (2008) 1655–1671.
2] P.L. Urban, D.M. Goodall, N.C. Bruce, Biotechnol. Adv. 24 (2006) 42–57.
3] M.S. Thomsen, P. Polt, B. Nidetzky, Chem. Commun. 24 (2007) 2527–2529.
4] M. Ti sˇ ma, B. Zeli c´ , Ð. Vasi c´ -Ra cˇ ki, P. Zˇ nidar sˇ i cˇ -Plazl, I. Plazl, Chem. Eng. J. 149
3
50 (2008) 2631–2638.
34] M.E.B. Smith, K. Smithies, T. Senussi, P.A. Dalby, H.C. Hailes, Eur. J. Org. Chem.
(2006) 1121–1123.
35] S. Matosevic, M. Micheletti, J.M. Woodley, G. Lye, F. Baganz, Biotechnol. Lett. 30
2008) 995–1000.
(
2009) 383–388.
5
[
[
5] P. He, G. Greenway, S.J. Haswell, Process Biochem. 45 (2010) 593–597.
6] H.R. Luckarift, B.S. Ku, J.S. Dordick, J.C. Spain, Biotechnol. Bioeng. 98 (2007)
(
7
01–705.
[
[
36] H.A. Stone, A.D. Stroock, A. Ajdari, Annu. Rev. Fluid Mech. 36 (2004) 381–411.
37] A.D. Stroock, S.K.W. Dertinger, A. Ajdari, I. Mezic, H.A. Stone, G.M. Whitesides,
Science 295 (2002) 647–651.
[
[
[
7] B. Ngamsom, A.M. Hickey, G.M. Greenway, J.A. Littlechild, P. Watts, C. Wiles, J.
Mol. Catal. B: Enzym. 63 (2010) 81–86.
8] M.N.H.Z. Alam, M. Pinelo, K. Samanta, G. Jonsson, A. Meyer, K.V. Gernaey, Chem.
Eng. J. 167 (2011) 418–426.
9] K. Koch, R.J.F. van den Berg, P.J. Nieuwland, R. Wijtmans, H.E. Schoemaker, J.C.M.
van Hest, F.P.J.T. Rutjes, Biotechnol. Bioeng. 99 (2008) 1028–1033.
[38] M.A. Ansari, K.Y. Kim, Chem. Eng. Sci. 62 (2007) 6687–6695.
[39] W.D. Ristenpart, J. Wan, H.A. Stone, Anal. Chem. 80 (2008) 3270–3276.
[40] G.J. Lye, P. Ayazi-Shamlou, F. Baganz, P.A. Dalby, J.M. Woodley, Trends Biotech-
nol. 21 (2003) 29–37.
[
[
[
10] M. Miyazaki, H. Maeda, Trends Biotechnol. 24 (2006) 463–470.
11] H.R. Sahoo, J.G. Kralj, K.F. Jensen, Angew. Chem. Int. Ed. 119 (2007) 5806–5810.
12] R.L. Hartman, J.R. Naber, S.L. Buchwald, K.F. Jensen, Angew. Chem. Int. Ed. 49
[
[
41] M.B. Kerby, R.S. Legge, A. Tripathi, Anal. Chem. 78 (2006) 8273–8280.
42] J. Shaeri, R. Wohlgemuth, J.M. Woodley, Org. Proc. Res. Dev. 10 (2006)
6
05–610.
43] M.P.C. Marques, J.M.S. Cabral, P. Fernandes, Recent Pat. Biotechnol. 3 (2009)
24–140.
(
2010) 899–903.
[
[
[
13] R. Wohlgemuth, Chem. Biochem. Eng. Q. 25 (2011) 125–134.
14] R. Wohlgemuth, in: M. Moo-Young, M. Butler, C. Webb, A. Moreira, B.
Grodzinski, Z.F. Cui, S. Agathos (Eds.), Comprehensive Biotechnology, Elsevier,
Amsterdam, Netherlands, 2011, pp. 591–601.
1
[
[
[
[
44] G.M. Whitesides, Nature 442 (2006) 368–373.
45] A.J. DeMello, Nature 442 (2006) 394–402.
46] J.M. Bolivar, J. Wiesbauer, B. Nidetzky, Trends Biotechnol. 29 (2011) 333–342.
47] R. Wohlgemuth, Biotechnol. J. 4 (2009) 1253–1265.
[
15] A. Schwarz, M.S. Thomsen, B. Nidetzky, Biotechnol. Bioeng. 103 (2009)
8
65–872.