Kinetic Modeling of an Enzymatic Redox Cascade In Vivo Reveals Bottlenecks Caused by Cofactors

Article abstract of DOI:10.1002/cctc.201700573

We describe the development of a kinetic model for the simulation and optimization of an in vivo redox cascade in E. coli in which a combination of an alcohol dehydrogenase, an enoate reductase, and a Baeyer–Villiger monooxygenase is used for the synthesis of lactones. The model was used to estimate the concentrations of active enzyme in the sequential biotransformations to identify bottlenecks together with their reasons and how to overcome them. We estimated adapted Michaelis–Menten parameters from in vitro experiments with isolated enzymes and used these values to simulate the change in the concentrations of intermediates and products during the in vivo cascade reactions. Remarkably, the model indicated that the fastest enzyme was rate-determining because of the unexpectedly low concentration of the active form, which opens up reversible reaction channels towards byproducts. We also provide substantial experimental evidence that a low intracellular concentration of flavin and nicotinamide cofactors drastically decreased the performance of the in vivo cascade drastically.

Full text of DOI:10.1002/cctc.201700573

Accepted Article
Title: Kinetic modeling of an enzymatic redox cascade in vivo reveals
cofactor-caused bottlenecks
Authors: Sofia Milker, Michael J Fink, Nikolin Oberleitner, Anna K
Ressmann, Uwe T Bornscheuer, Marko D Mihovilovic, and
Florian Rudroff
This manuscript has been accepted after peer review and appears as an
Accepted Article online prior to editing, proofing, and formal publication
of the final Version of Record (VoR). This work is currently citable by
using the Digital Object Identifier (DOI) given below. The VoR will be
published online in Early View as soon as possible and may be different
to this Accepted Article as a result of editing. Readers should obtain
the VoR from the journal website shown below when it is published
to ensure accuracy of information. The authors are responsible for the
content of this Accepted Article.
To be cited as: ChemCatChem 10.1002/cctc.201700573
Link to VoR: http://dx.doi.org/10.1002/cctc.201700573
A Journal of
www.chemcatchem.org

ChemCatChem
10.1002/cctc.201700573
FULL PAPER
Kinetic modeling of an enzymatic redox cascade in vivo reveals
cofactor-caused bottlenecks
[
a]
[b]
[a]
[a]
[c]
Sofia Milker, Michael J. Fink, Nikolin Oberleitner, Anna K. Ressmann, Uwe T. Bornscheuer ,
Florian Rudroff*,[a] and Marko D. Mihovilovic[a]
Abstract: This paper describes the development of a kinetic model
for the simulation and optimization of an in vivo redox cascade in
E. coli, using a combination of an alcohol dehydrogenase, an enoate
reductase, and a Baeyer-Villiger monooxygenase for the synthesis of
lactones. The model was used to estimate the concentrations of active
enzyme in the sequential biotransformations to identify bottlenecks
together with their reasons and how to overcome them. We estimated
adapted Michaelis-Menten parameters from in vitro experiments with
isolated enzymes, and used these values to simulate the change in
concentrations of intermediates and products during the in vivo
cascade reactions. Remarkably, the model indicated the fastest
enzyme to be rate-determining due to the unexpectedly low
concentration of the active form, opening up reversible reaction
channels towards side products. We also provide substantial
experimental evidence, that a low intracellular concentration of flavin
and nicotinamide cofactors drastically throttled the performance of the
in vivo cascade.
maximum yield and productivity have been successfully applied
on enzyme cascades.[3] Their success, however, depended on
the straightforward identification of the limiting factor; failing that,
the process involved tremendous experimental work.
Computational approaches, such as kinetic modeling of in vitro or
in vivo enzymatic reactions, have only been explored in few
studies, e.g. cofactor recycling in vitro,[ modeling of the stability
of an aldolase,[5] simulation of a three-enzyme cascade system
for synthesis of 6-hydroxyhexanoic acid,[ or a three-enzyme
cascade for the biotransformation of sucrose to cellobiose.[7]
These studies were important to gain insights into the limiting
factors for biotransformation performance, and largely replaced
experimental effort, since hypotheses could reliably be tested
in silico. So far, most enzymatic kinetic models were based either
4]
6]
[
5, 7]
on known concentrations of enzymes with an in vitro system,
or the enzymes were assumed to be in excess compared to the
substrates in the cell.[ Only few models have taken into account
the intracellular concentration of enzymes, mostly simulating
natural metabolic pathways.[
8]
9]
Here, we developed such a kinetic model based on a well-
characterized, but artificial pathway. This reaction cascade was
previously used to study the sequential operation of three
unrelated redox enzymes with shared cofactors, both in vitro,
and in a whole-cell system[11] (Scheme 1).
Introduction
In recent years, enzymatic cascades have gained popularity due
to the high compatibility of biocatalysts in a one-pot or whole-cell
environment, thereby circumventing the need to purify
[10]
[
1]
intermediates, or to install protective groups. Following the
guidelines of retrosynthesis in designing artificial pathways, the
key transformations can easily be identified, and the connectivity
of reactions can be established.[ This approach, however, does
not consider the performance of the pathway quantitatively. So far,
optimization required considerable experimental effort to identify
problematic steps. Many methods for optimization towards
2]
[
a]
MSc. S. Milker, Dr. N. Oberleitner, Dr. A. K. Ressmann, Prof. Dr. M.
D. Mihovilovic, Dr. F. Rudroff
Institute of Applied Chemistry
TU Wien
Scheme 1. Cascade reaction from cyclohexenol 1a to ε-caprolactone 1e.
Getreidemarkt 9/163-OC, 1060 Vienna (Austria)
E-mail: florian.rudroff@tuwien.ac.at
Dr. M.J. Fink
Department of Chemistry and Chemical Biology
Harvard University
It connected the oxidation of a secondary alcohol, a Michael
reduction, and a Baeyer-Villiger oxidation, catalyzed by an alcohol
dehydrogenase (ADH), an enoate reductase (ERED) and a
Baeyer-Villiger monooxygenase (BVMO), respectively.
The ADH (from Lactobacillus kefir)[12] depended on magnesium,
and the other two enzymes required non-covalently attached
[
[
b]
c]
1
2 Oxford St, Cambridge, MA 02138, USA
Prof. Dr. U. T. Bornscheuer
Institute of Biochemistry, Dept. of Biotechnology & Enzyme
Catalysis
Greifswald University
Felix-Hausdorff-Str. 4, 17489 Greifswald (Germany)
flavins as redox-active components (flavin mononucleotide, FMN,
for the ERED from Pseudomonas putida,[
11, 13]
and flavin adenine
Supporting information for this article is given via a link at the end of
the document.
dinucleotide, FAD, for the BVMO from Acinetobacter
[14]
calcoaceticus).
The starting material cyclohexenol 1a was
This article is protected by copyright. All rights reserved.

ChemCatChem
10.1002/cctc.201700573
FULL PAPER
converted to ε-caprolactone 1e via intermediates 1b
ꢀ
ꢁꢂꢃ
(1)
(2)
(3)
(4)
(5)
(
cyclohexanone) and 1c (cyclohexanone) in the particular
ꢅ ꢆꢇꢈ→ꢉ ꢊꢋꢇꢉ→ꢈ
ꢀ
ꢄ
cascade; the transformation was shown to proceed comparably
well on several other examples, including the conversion of
ꢀ
ꢁꢌꢃ
ꢅ ꢇꢈ→ꢉ ꢆ ꢇꢉ→ꢈ ꢆꢋꢇꢉ→ꢍ
ꢀ
ꢄ
(1S,5S)-carveol 2a to the corresponding lactone 2e (Scheme 2).
ꢀ
ꢁꢎꢃ
ꢅ ꢆꢇꢍ→ꢏ ꢊ ꢇꢏ→ꢍ ꢊꢋꢇꢉ→ꢍꢋ ꢆꢋꢇꢍ→ꢐ
The ADH also catalyzed the redox equilibration between
ꢀ
ꢄ
[
10]
cyclohexanone 1c and the side product cyclohexanol 1d. Since
all reactions catalyzed by the ADH were reversible at room
temperature (25 °C), the formation of intermediate 1d merely
extended the reaction time towards the desired product.
ꢀꢁꢀꢃ
ꢅ ꢇꢍ→ꢏ ꢆꢋꢇꢏ→ꢍ
ꢀꢄ
ꢀ
ꢁꢑꢃ
ꢅ ꢇꢍ→ꢐ
ꢀꢄ
We assumed that the concentration of the essential cofactors
+
NADP and NADPH would remain approximately constant over
the course of the reaction in vivo, since viable E. coli cells would
actively regulate the redox balance for homeostasis.[15]
We further assumed that all enzymes would suffice the conditions
for Michaelis-Menten kinetics, possibly with inhibition, and
calculated the individual rates of the reactions based on Eq S1
(see SI). For all ADH-catalyzed reactions, we added terms for
mixed inhibition (Supplementary Eq. S2), since we later observed
that the enzyme was inhibited by the initially used co-solvent
ethanol (see below). Substitution with the enzyme- and substrate-
specific parameters lead to the Michaelis-Menten rate laws,
described by Supplementary Eq. S3-S5, for the productive
reactions of the cascade and the rate laws for the ADH-catalyzed
back and side reactions, described by Supplementary Eq. S6-S8.
These equations were later used for non-linear fitting to the
experimental data from in vivo cascades, after substitution with
the parameters from in vitro experiments.
Scheme 2. The cascade example from starting from (1S,5S)-carveol 2a.
Within the present study, we aimed at identifying the origin for the
reduced productivity, and propose optimization strategies via
simulations with minimum experimental effort. The specific
objectives of the model were (i) to estimate the enzyme
concentrations in the reaction to locate the rate-determining step,
(
ii) to identify possible extrinsic reasons for its reduced velocity
within the cascade, and (iii) to propose an optimization strategy to Determination of Michaelis-Menten
overcome this limitation, and increase the overall performance of parameters
the cascade.
We measured the initial rates of all reactions in the cascade
Table 1), using a standard protocol (see Experimental Section
(
and the Supplementary Information) with one major adaptation:
the nicotinamide cofactors were not supplied in such excess, that
all enzymes were saturated, but rather added at the intracellular
concentration of E. coli (0.12 mM for NADPH, 0.10 mM for
NADP+).[16] The resulting estimate for velocity (kcat) was thus not
directly comparable to values obtained at saturation. That
Results and Discussion
The rate laws for all cascade reactions were derived based on
Michaelis-Menten (MM) theory. We then determined the kinetic
parameters via initial rate measurements, and used them as
constant parameters in non-linear fitting of experimental data from
in vivo cascades. This way, the concentrations of active enzymes
were estimated. The resulting model served to simulate two
cascade examples, and to formulate hypotheses pertaining the
limiting factors to the performance of the cascade. These
hypotheses were then tested both in silico, and by experiments
in vitro.
notwithstanding,
the
adaptation
would
increase
the
meaningfulness of in vitro data for the use in modeling of in vivo
processes.
The highest impact of incomplete saturation by the cofactors
was determined for the ADH: the results indicated that it would
operate
(
(
in vivo
[NADP+]E. coli = 3.3×K
[NADPH]E. coli = 1.2×K
at
), and at 0.5×kcat in reductive direction
). The difference in values was
0.8×kcat
in
oxidative
direction
M
M
K
M
beneficial to the productive flux of the cascade, including the
recovery of the side product 1d by re-oxidation. The other two
enzymes would largely remain unaffected by the low
concentration ([NADPH]E. coli = >0.95×kcat for the ERED,
Development of the kinetic model
First, we postulated the generic rate laws for all reactions,
beginning with the starting material
intermediates b (eq. (2)) and c (eq. (3)), the side product d
eq. (4)), and the final product e (eq. (5)).
a (eq. (1)), then the
>0.85×kcat for the BVMO).
(
This article is protected by copyright. All rights reserved.

ChemCatChem
10.1002/cctc.201700573
FULL PAPER
determined the inhibition parameters
i
K = 64.9 mM and
K
ii = 126.9 mM (Supplementary Eq. Fehler! Verweisquelle
konnte nicht gefunden werden., Supplementary Fig. S1–S4).
These constants were then used for all the rate laws that
described ADH-catalyzed reactions.
Table 1. Michaelis-Menten parameters for all enzymatic processes.
ADH
In the first cascade, all ADH-promoted steps had similar rates
and affinity parameters, with the exception of the unproductive
reduction of 1c to 1d, which proceeded with approx. 100-fold
[
a]
[a]
-1
Substrate
K
mM]
M
k
cat,app
k
cat,app
cat M
k K
-1
-1
-1 -1
[
[s ]
[s ]
[mM s ]
oxidation
reduction
M
higher catalytic efficiency (kcat/K ). The high affinity of ADH for 1c
created the possibility of a kinetic dead end, where material was
diverted to 1d quickly, and even at low concentration of 1c. The
side product 1d would then accumulate because of its low affinity,
before being re-oxidized at a slow rate. This unproductive
substrate promiscuity could best be mitigated by an efficient final
step, catalyzed by the BVMO, thus preventing the build-up of 1c.
In the second cascade, the ADH only converted the starting
material, carveol (2a), with no detectable activity for back or side
reactions. The other two enzymes had a higher affinity for their
1
a
b
28.5 ± 2.7
23.6 ± 1.6
2.7 ± 0.3
21.3 ± 1.0
3.6 ± 0.7
–
11.2 ± 0.8
0.39 ± 0.06
0.19 ± 0.02
28.9 ± 5.77
0.7 ± 0.05
0.3±0.09
1
4.4 ± 0.2
1 [
c
b]
b]
78.0 ± 6.9
1 [
d
14.9 ± 0.6
0.7 ± 0.1
2a
2b
2c
2d
No activity detected
No activity detected
4
respective substrates than the ADH (up to 10 -fold). The rates of
the ERED-catalyzed reactions were the slowest of all steps,
whereas the BVMO were the highest in the productive flux.
In summary, the results from in vitro kinetics suggested that,
at approximately equal concentration of all enzymes, bottlenecks
would be caused by the promiscuous activity of the ADH, or by
accumulation of intermediates b due to the inefficient Michael-
addition (ERED).
–
–
Activity towards a mixture of isomers detected
NADP+
0.03[c]
not applicable
not applicable
–
–
–
–
0.14[c]
NADPH
ERED
Modeling of the cascade reactions
Substrate
K
mM]
M
k
cat
k
cat K -1
-1 -1
M
-1
We inserted the experimental Michaelis-Menten values as
[
[s ]
1.53 ± 0.08
0.22 ± 0.04
–
[mM s ]
1.62 ± 0.22
2.54 ± 1.30
–
constant
parameters
into
Supplementary
Eq. Fehler!
1
b
b
0.94 ± 0.08
0.09 ± 0.03
Verweisquelle konnte nicht gefunden werden.–Fehler!
Verweisquelle konnte nicht gefunden werden., and performed
non-linear fitting of this set of differential equations to the
previously obtained in vivo data by iterative least-squares
regression, leaving the concentrations of enzymes as variable
parameters to the fit. That implied the assumption that the
biocatalysts would not degrade over the course of the reaction.
We also omitted any terms for evaporation of the volatile
2
NADPH
BVMO
0.0036 ± 0.0005
cat K -1
Substrate
K
M
k
cat
k
M
-1
-1 -1
[
mM]
[s ]
[mM s ]
compounds, since the cascade reactions were run in closed
vessels.[11] The headspace of the vial was 10×fold higher than the
volume of the biotransformation to avoid oxygen limitation. The
fitting algorithm estimated the following concentrations:
123 ± 10 µM ADH, 51 ± 5 µM ERED and approx. 0.03 µM BVMO
1
c
c
0.0069[d]
22.6[d]
3275.4
2
0.21 ± 0.03
0.018[d]
46.82 ± 3.94
–
225.0 ± 51.0
–
NADPH
(
Figures 1 and 2).
+
[
a] maximum kcat at NADP = 0.1 mM, NADPH = 0.12 mM [b] side reaction [c]
Using these values, we obtained a highly accurate model for
[
12]
[14]
taken from ref. . [d] taken from ref.
.
2
both cascades (quantified using adjusted R , Table 2), thus
The kinetic parameters were estimated from non-linear fitting of experimental
data using the MM-equation. For Michaelis Menten plots see supplementary
information (Supplementary Fig. S5-S7). Results estimated with 95%
confidence intervals (CI), N=3.
2
Table 2. Adjusted coefficients of determination (Adj. R ) for the comparison
of modelled vs experimental data.
Cascade
a
b
c
d
e
Furthermore, we investigated the influence of ethanol on the
performance of the cascade, as it had previously been used to
add the substrates to the reaction (0.4 %, approx. 69 mM).[11] Its
1
2
0.99
0.91
0.98
n.a. [a]
-5.00
-1.23
0.60
n.a. [a]
0.98
0.93
technical properties as a solvent (stable, benign, water-miscible,
bio-compatible, volatile)[
17]
advocated its use in enzymatic
[a] n.a = not applicable: the intermediates were not detected in the
cascades,[18] but as a primary alcohol, it might interfere with the
function of ADHs. We observed a mixed type of inhibition, and
experiments.
The equation for calculation of adj. R2 values can be found in the
supplementary information (Supplementary Equation S9).
This article is protected by copyright. All rights reserved.

ChemCatChem
10.1002/cctc.201700573
FULL PAPER
validating our mechanistic assumptions and the adapted
Michaelis-Menten parameters, with some exceptions.
Similarly, the model predicted both cascades to complete within
1–2 h, when only approx. 90 % conversion were reached in the
experiments within that time; it took another 18 h to reach 100 %.
We hypothesized that (i) the equilibration of nicotinamides in
E. coli was too slow to compensate for the artificial pathway
ꢒꢀꢁꢓꢔꢕꢖꢒꢗꢘꢃ/ꢀꢄ ꢙ 0ꢘ , and (ii) the enzymes’ specific activity
decreased significantly over the course of the experiment.
Whereas the prior case would influence all steps, resulting in a
complex effect, the latter scenario would become more apparent
at longer reaction times, and later in the cascade (by count of
steps). The decelerating rate in the last reaction corroborated that
the BVMO was decaying significantly.[19]
Experimentally, the intermediates 1c and 2c only appeared at the
first data point, and could not be modelled correctly (negative adj.
2
R ). Compound 1d was determined less accurately by the model
than the other components, and intermediate 2b was not detected
in the experiment, but was predicted by the model to be present
for the first 6 minutes; these errors likely resulted from insufficient
precision and sampling frequency in the experimental data.
Furthermore, we were curious, why the fitting algorithm had
estimated the concentration of active BVMO at four to five orders
of magnitude lower than of the other two enzymes. Based on the
semi-quantitative protein analysis via SDS-PAGE, we estimated
that the BVMO was present at approx. 10 % of the ERED’s mass
concentration in the soluble protein fraction of the cell.
(Supplementary Fig. S8). That result indicated that the average
specific activity of the BVMO over the course of the reaction was
much lower than that of the other enzymes, since the model
4
1
estimated a ratio of approx. 10 vs. 10 from in vivo analysis.
We speculated that most of the total amount of BVMO was
inactive in the cell. Thus, we simulated the cascade with
increasing titers of BVMO, up to the same order of magnitude as
the other two enzymes (>5 µM). The model predicted a complete
suppression of the accumulation of 1d (Fig. 3), and a much
shorter reaction time (0.1 h vs. 2.2 h; Supplementary Fig. S9),
thereby supporting our hypothesis.
Figure 1. Comparison of experimental (data points) with the simulated (lines)
concentrations for cascade 1. Experimental data were plotted as mean ± 2 SD
(
standard deviation). Conditions for calculations: 4 mM 1a,constant
+
concentration of 0.10 mM NADP and 0.12 mM NADPH, [ADH] = 123 µM,
ERED] = 51 µM, [BVMO] = 0.033 µM.
[
Although the misrepresentation of 1c and 2c detracted only little
from the overall accuracy of the model, it indicated that some of
the assumptions were not entirely correct.
Figure 3. Simulated influence of the BVMO concentration on 1d formation.
+
4
mM 1a, constant concentration of 0.10 mM NADP and 0.12 mM NADPH,
[ADH] = 123 µM, [ERED]= 51 µM, [BVMO] varied from 0.033–10 µM, Michaelis-
Menten parameters can be found in Table 1.
Figure 2. Comparison of experimental (data points) with the simulated (lines)
concentrations for cascade 2. Experimental data were plotted as mean ±
Next, we searched for possible reasons for the low specific activity
of the BVMO. We amended our previous hypothesis – that
cofactors would not be recycled sufficiently fast – with the
assumption that FAD and its precursors would not be in adequate
supply to maintain all BVMO in its active, FAD-bound form.
2
SD. Conditions for calculations: 4 mM 2a,constant concentration of 0.10 mM
+
NADP and 0.12 mM NADPH, [ADH] = 123 µM, [ERED] = 51 µM,
BVMO] = 0.033 µM.
[
This article is protected by copyright. All rights reserved.

ChemCatChem
10.1002/cctc.201700573
FULL PAPER
To test this hypothesis, we investigated the effect of excess
NADP /NADPH and flavin nucleotides (FMN/FAD) on the
In the second experiment (Fig. 4B), which included FMN/FAD,
we additionally observed a much shorter reaction time (3 h vs.
24 h in vivo); both flavin-dependent enzymes had a higher activity
than in vivo, with a larger increase for the BVMO than for the
ERED.
We speculated that the unequal improvement resulted from
the sequence in the biosynthesis of flavin nucleotides: FAD is only
produced via FMN as a direct precursor. Bound FMN (e.g. to the
ERED) would not be available for further conjugation with adenine
to form FAD.[20] The amount of available FAD in the cell would
thus be lowered by, and proportional to, the ERED, consequently
reducing the specific activity of the BVMO. Hence, an excess of
FAD should have a stronger influence on the activity of the last
enzyme, as seen experimentally. Since our kinetic model did not
account for inactive enzyme or stability, the initially fitted value did
not reflect the estimated total concentration of BVMO, but only the
active fraction.
+
performance of the cascade in vitro. Therefore, we obtained cell-
free extracts of the cascade enzymes, simultaneously produced
in the same host, and under the same conditions as for the in vivo
biotransformations. Nicotinamide cofactors were added in
stoichiometric excess to the substrates; flavins were added sub-
stoichiometric to the ERED, but in excess to the BVMO. We
performed two sets of experiments, starting from 1a: (i) with
+
+
NADP /NADPH only (Fig. 4A), and (ii), with NADP /NADPH and
FMN/FAD (Fig. 4B). The concentrations of the enzymes were at
the same ratio as the estimates in the model (calculated: 1845 µM
ADH, 750 µM ERED, 0.5 µM BVMO). In both experiments, the
substrate was added in acetonitrile instead of ethanol to exclude
the effects of inhibition.
An immediate solution for a flux improvement would be the
addition of cofactors to the cellular system. This is not feasible
due to the lack of their cell permeability. Another strategy would
be the addition of cofactor precursors like riboflavin which are cell
permeable. The group of Stewart[21] investigated the influence of
riboflavin addition to a whole-cell BVMO oxidation but no influence
on the overall cascade flux was observed. Furthermore the
introduction of a more stable BVMO,[ as recently discovered by
Romero et al. would be beneficial for the overall flux through the
cascade.
22]
In summary we assume, that further flux improvements of our
cascade can be achieved by changing the host organism, which
+
provides increased cofactor availability (NADP /NADPH and
FMN/FAD) and applying a more stable CHMO enzyme.
Effect of ethanol on cascade performance
Ultimately, we used the kinetic model to investigate mixed
inhibition of ethanol on the ADH (see section on MM parameters).
In detail, we tested (i) if ethanol had a significant effect on the
performance of the cascade, and if so, (ii) how large that effect
was. Our hypothesis was that ethanol would slow down the
formation of the final product. Therefore, we simulated the
cascade starting from 1a with a range of ethanol concentrations
(0 % to 10 %, Supplementary Fig. S10). At the experimentally
used level (0.4 % ethanol), the first reaction was estimated to be
slightly slower than without cosolvent (80 mM h-1 vs. 53 mM h-1)
(Supplementary Fig. S11). Still, the productivity of the whole
Figure 4. Influence of the cofactors on the in vitro cascade biotransformation:
cascade was not affected at all tested concentrations, since the
ADH-catalyzed step was not rate-determining in our system.
+
A: Stoichiometric amounts of NADP and NADPH for the reaction (5 mM each),
4
mM 1a; B: Stoichiometric amounts of NADP+ and NADPH for the reaction
(5 mM each) and addition of FMN and FAD (0.1 mM each) to the cell free
extracts, 4 mM 1a; n.d. not determined.
Conclusions
In the first experiment (Fig. 4A) the byproduct 1d was detected at
a lower concentration, and for a shorter period of time, than in the
in vivo experiments. That strongly suggested that its
accumulation was not only promoted by the suspected low BVMO
concentration in vivo, but was also caused by an unfavorable
balance of the nicotinamides.
We successfully developed an accurate kinetic model for an
artificial in vivo redox cascade, using adapted Michaelis-Menten
parameters from in vitro experiments. The concentrations of
active enzyme in vivo were estimated by non-linear fitting of the
rate laws to the experimental data. This way, we could simulate
This article is protected by copyright. All rights reserved.

ChemCatChem
10.1002/cctc.201700573
FULL PAPER
the change in the formation of by-products, and the change in
productivity upon varying the levels of the rate-determining
catalyst (large influence), or of the co-solvent (no influence).
More importantly, the model identified a bottleneck resulting
from an unexpectedly low specific activity of the last catalyst
(BVMO), which was not evident from in vitro kinetics and SDS-
PAGE analysis of in vivo protein titers. We thus argue that the
estimation of intracellular enzyme concentrations is an important
parameter for the optimization of cascades.
Furthermore we tested our assumption of stable and
continuous supply of cofactors (FAD & NADPH) by E. coli,
suspecting that a lack thereof would have caused the low activity
of the BVMO.
An in vitro experiment with an excess of cofactors was
compatible with this hypothesis, reducing the overall reaction time
from 24 to 3 h.
In summary, our findings suggested that the optimization of
cascades in a cellular environment should also strongly consider
the availability of redox cofactors, their fluxes upon operation of
the artificial pathway as well as the amount and stability of all
enzymes involved.
This article is protected by copyright. All rights reserved.

ChemCatChem
10.1002/cctc.201700573
FULL PAPER
Experimental Section
Determination of kinetic parameters
Activity measurements were performed spectrophotometrically with the
Shimadzu UV-1800 photometer at 340 nm by monitoring the initial rates of
Materials
NADPH
formation
or
depletion
in
700 µL
cuvettes
(ε
NADPH = 6.22 mM-1 cm ). The Michaelis-Menten parameters of the ADH
-1
All chemicals were purchased from Sigma (Steinheim, Germany) or VWR
were measured in Tris-HCl buffer (50 mM, pH 7.5, 50 mM MgCl
with different substrate concentrations in triplicates and 0.1 mM for NADP
2
) at 25 °C
(
Hannover, Germany) and were used without further purification unless
+
otherwise specified.
(
for oxidations) and 0.12 mM for NADPH (for reductions). The enzyme
concentration was between 0.8 and 10 µM depending on the velocity of
the reaction. The total volume was 500 µL and the measurement was
started by addition of cofactor.
Bacterial strains, plasmids and culture conditions
The enzymatic cascade consists of an alcohol dehydrogenase (ADH) from
Lactobacillus kefir (LK-ADH[12]), an enoate reductase (ERED) from
Pseudomonas sp. (XenB[11, 13]) and a Baeyer-Villiger monooxygenase
For the ERED, the Michaelis-Menten parameters were measured in Tris-
HCl buffer (50 mM, pH 7.5) at 25 °C in triplicates. The measurement was
started with 0.12 mM NADPH. The enzyme was applied in a concentration
of 4 µM and total reaction volume was 500 µL.
(BVMO), precisely cyclohexanone monooxygenase from Acinetobacter sp.
CHMOAcineto[14]). All three enzymes were expressed separately in E. coli
(
BL21 (DE3). LK-ADH was encoded on a pET22b(+) vector, the XenB on a
pGASTON vector and the CHMO on the pET22b(+) vector. For protein
production, LB medium (for LK-ADH and CHMO) and TB medium (for
XenB) was inoculated from a LB overnight preculture (1:100) with the
appropriate antibiotic. The main culture was treated according to the
respective expression protocol (media compositions, antibiotics and
expression conditions as given in the supplementary information).
For the BVMO, the Michaelis-Menten parameters were determined in Tris-
HCl buffer (50 mM, pH 7.5, 0.1 mM FAD) at 25° C in triplicates. The
measurement was started with 0.12 mM NADPH. The enzyme was applied
in a concentration of 0.05 µM and the total reaction volume was 500 µL.
All substrate stocks were prepared in acetonitrile. For determination of the
ethanol effect on the ADH the ethanol was added to the measurements
prior to starting the reaction.
Purification of enzymes
After expression the culture was centrifuged (10 min, 5000 × g, 4 °C), the
supernatant was discarded and the cells were resuspended in 40× less
volume Tris-HCl buffer (50 mM, pH 7.5) containing phenylmethylsulfonyl
fluorid (0.1 mM PMSF). Cells were placed on ice and sonicated using a
Bandelin KE76 sonotrode connected to a Bandelin Sonoplus HD 3200 in
The specific activity was calculated according to
volumetricꢋactivity
ꢚꢛꢑꢎꢜꢝꢜꢎꢋꢂꢎꢄꢜꢇꢜꢄꢞꢋꢟꢍꢈꢠ
ꢅ
ꢁ
ꢡꢓꢢꢃ
9
cycles (5 s pulse, 55 s break, amplitude 50 %). Precipitates were
removed by centrifugation (15000 × g, 15 min, 4° C).
The fitting of the Michaelis-Menten curves to determine kinetic parameters
was performed with Origin 2015G.
For the ADH, enrichment was performed by loading the clear supernatant
into a centrifugal device with a 50 kDa cut-off and centrifuged to wash all
proteins below 50 kDa into the flowthrough. The retained solution was
Mixed inhibition studies
washed twice with Tris-HCl (50 mM, pH 7.5, 50 mM MgCl
concentrated to a concentration of around 500 µM.
2
) and
See supplementary information (Supplementary Figures S2-S4).
Biocatalytic screenings
For poly-His-tag purification (ERED and BVMO), the clear supernatants
containing the polyhistidine-tagged enzymes were loaded on a Ni2+
-
sepharose HP affinity column (1 mL, GE Healthcare bioscience)
equilibrated with Tris-HCl buffer (50 mM, pH 7.5, 0.5 M NaCl)
Unspecifically bound enzymes were eluted with a binding buffer containing
For the in vivo reactions of cyclohexenol or carveol as substrate, the E. coli
BL21 (DE3) with all three enzymes (ADH, ERED and BVMO) in one cell
was expressed in TB, supplemented with ampicillin and kanamycin,
according to the expression protocol (supplementary information). To the
expressed culture 1 % glucose was added and the culture was transferred
into vials with 10× higher volume as the volume of the biotransformation.
The glass vials were sealed during the reaction, evaporation would be
4
0 mM imidazole in 7 column volumes. The enzyme of interest was eluted
with elution buffer containing 250 mM imidazole in 3 column volumes.
Fractions containing the enzymes were identified by SDS-PAGE analysis,
pooled and concentrated by ultrafiltration by using ultracentrifugal tubes
with a cut-off of 10 kDa (Amicon Ultra, Millipore). For the ADH purification
reduced to a minimum. The reaction was started by substrate addition (1 M
stock in ethanol) to a final concentration of 4 mM. [11]
2
all buffers contained 50 mM MgCl and for the BVMO purification, all
buffers contained 100 µM FAD.
For the in vitro screening of cyclohexenol (1a), the expressed E. coli BL21
(DE3) with ADH, ERED and BVMO was centrifuged, resuspended in 15×
less Tris-HCl (50 mM, pH 7.5) and divided in two different fractions. One
fraction was only supplemented with 0.1 mM PMSF and the other fraction
was supplemented with 0.1 mM PMSF and additionally with 0.1 mM FMN
and FAD. Both fractions were sonicated on ice using a Bandelin KE76
sonotrode connected to a Bandelin Sonoplus HD 3200 in 9 cycles (5 s
pulse, 55 s break, amplitude 50 %). Precipitates were removed by
centrifugation (15000 × g, 15 min, 4° C). The supernatant was applied in a
biotransformation with 90 % of the total biotransformation volume
Determination of the protein concentration
Protein concentrations were determined by the dye-binding method of
Bradford at 595 nm using a pre-fabricated assay (BioRad) and bovine
serum albumin as the calibration standard.
This article is protected by copyright. All rights reserved.

ChemCatChem
10.1002/cctc.201700573
FULL PAPER
(
19 mg mL-1 enzyme, which is approx. 15 × more concentrated than in the
whole-cell reaction mixture). The other 10 % of the biotransformation
volume were cofactors, added to the final concentration of 5 mM NADP+
and 5 mM NADPH. The reaction was started by addition of the substrate
cyclohexenol (1a).
Acknowledgements
In both cases, the reactions were performed at 20° C with shaking and
samples were taken and analysed via GC-FID.
We thank the FWF (grant no. P-24483-B20 and grant no. I723-
N17) and the DFG (grant no. Bo1862/6-1) for financial support.
We are also grateful to Prof. W. Hummel (University of
Düsseldorf) for supplying the gene encoding LK-ADH.
GC-analysis
1
00 µL sample were extracted with 500 µL ethyl acetate supplemented
Keywords: kinetic modeling • biocatalytic cascade • enzyme
estimations in vivo • cascade optimization • cofactor limitation
with 1 mM methyl benzoate as internal standard. Conversion was
determined by GC using a ThermoQuest Trace GC 2000 with a FID
detector equipped with a BGB175 (30 m x 0.25 mm ID, 0.25 µm film).
[
1]
a) E. Ricca, B. Brucher, J. H. Schrittwieser, Adv. Synth. Catal. 2011, 353,
239-2262; b) J. Muschiol, C. Peters, N. Oberleitner, M. D. Mihovilovic,
2
Model development
U. T. Bornscheuer, F. Rudroff, Chem. Commun. 2015, 51, 5798-5811; c)
S. P. France, L. J. Hepworth, N. J. Turner, S. L. Flitsch, Acs Catal 2017,
The model was developed on the basis of Michaelis-Menten equations for
every reaction step of the cascade, including possible back and side
reactions for the ADH from initial rate measurements. The model was built
in MATLAB 2013b Simbiology Toolbox. The parameter [ENZ] for
enzymatic concentration in the different reactions was fitted with nonlinear
regression using iterative least squares estimation. [ADH] was fitted to the
same value with several different initial values [0.1, 1, 10, 100 µM] in the
first round of fitting. In the second round, [ADH] was fixed to the determined
value and the other two parameters [ERED]; [BVMO] were fitted with
different initial values [0.1 to 100 µM], which always converged on the
same results.
7
, 710-724.
[
2]
3]
a) S. G. Warren, P. Wyatt, Organic synthesis : the disconnection
approach, 2nd ed., John Wiley & Sons Ltd., Chichester, UK, 2008; b) N.
J. Turner, E. O'Reilly, Nat. Chem. Biol. 2013, 9, 285-288.
[
a) R. Agudo, M. T. Reetz, Chem. Commun. 2013, 49, 10914-10916; b)
A. M. Kunjapur, Y. Tarasova, K. L. Prather, J. Am. Chem. Soc. 2014, 136,
11644-11654.
[
4]
5]
M. C. Hogan, J. M. Woodley, Chem. Eng. Sci. 2000, 55, 2001-2008.
M. Sudar, Z. Findrik, Đ. Vasić-Rački, P. Clapés, C. Lozano, J. Biotechnol.
[
2013, 167, 191-200.
[
6]
C. Scherkus, S. Schmidt, U. T. Bornscheuer, H. Gröger, S. Kara, A. Liese,
Biotechnol. Bioeng. 2017, 10.1002/bit.26258.
Synthesis of (2R,5S)-dihydrocarvone
[
7]
8]
C. Zhong, P. Wei, Y.-H. P. Zhang, Chem. Eng. Sci. 2017, 161, 159-166.
D. Machado, L. R. Rodrigues, I. Rocha, Bio. Systems. 2014, 125C, 16-
21.
[
To six different batches of non-baffled Erlenmeyer flasks (500 mL)
containing 100 mL resting cells expressing XenB, 75 µL of a S-(+)-carvone
stock (50 mg mL-1 in acetonitrile) were added (expression details in the
supplementary information). The Erlenmeyer flasks were shaken at 25 °C
and 300 rpm. The progress of the reaction was monitored using GC. After
[9]
a) L. Kuepfer, M. Peter, U. Sauer, J. Stelling, Nat. Biotech. 2007, 25,
1001-1006; b) J. Hassan, L. L. Bergaust, L. Molstad, S. de Vries, L. R.
Bakken, Environ. Microbiol. 2016, 18, 2964-2978.
[10] N. Oberleitner, C. Peters, F. Rudroff, U. T. Bornscheuer, M. D.
Mihovilovic, J. Biotechnol. 2014, 192, 393-399.
3
0 min GC showed full conversion and 75 µL of the carvone stock were
added. After 1.5 h GC showed full conversion and another 75 µL of the
substrate stock were added resulting in a total addition of 700 mg
S-(+)-carvone.
[11] N. Oberleitner, C. Peters, J. Muschiol, M. Kadow, S. Saß, T. Bayer, P.
Schaaf, N. Iqbal, F. Rudroff, M. D. Mihovilovic, U. T. Bornscheuer,
ChemCatChem 2013, 5, 3524-3528.
[
[
[
[
[
12] A. Weckbecker, W. Hummel, Biocatal. Biotransformation 2006, 24, 380-
389.
After 2.5 h total reaction time all batches were extracted three times with
diethyl ether and the combined organic layers were dried over Na SO ,
2 4
filtered off and concentrated in vacuo. The crude product was purified
using 90 g silica and pentane/diethyl ether (gradient, 0-10 % ether). A
colorless oil (88 % yield) was obtained as cis:trans 8:1 mixture.
13] C. Peters, R. Kölzsch, M. Kadow, L. Skalden, F. Rudroff, M. D.
Mihovilovic, U. T. Bornscheuer, ChemCatChem 2014, 6, 1021-1027.
14] N. A. Donoghue, D. B. Norris, P. W. Trudgill, Eur. J. Biochem. 1976, 63,
1
75-192.
15] S. Spaans, R. Weusthuis, J. Van Der Oost, S. Kengen, Front. Microbiol.
015, 6, 742.
16] a) K. B. Andersen, K. von Meyenburg, J. Biol. Chem. 1977, 252, 4151-
156; b) B. D. Bennett, E. H. Kimball, M. Gao, R. Osterhout, S. J. Van
Dien, J. D. Rabinowitz, Nat. Chem. Biol. 2009, 5, 593-599.
2
GC samples were taken after 30 min of each batch separately: 100 µL of
the reaction mixture were added to a 1.5 mL Eppendorf tube containing
4
2
00 µL of ethyl acetate (containing 1 mM methyl benzoate as internal
standard) and vortexed for 30–35 s and centrifuged for 1 min. The organic
layer was desiccated with Na SO analyzed by achiral GC. For NMR
[17] F. J. Weber, J. A. M. de Bont, Biochim. Biophys. Acta. Rev. Biomembr.
1996, 1286, 225-245.
2
4
[
18] Y. L. Khmelnitsky, V. V. Mozhaev, A. B. Belova, M. V. Sergeeva, K.
Martinek, Eur. J. Biochem. 1991, 198, 31-41.
spectrum see supplementary information (Supplementary Fig. S12). NMR
code: major cis isomer:
[19] H. L. van Beek, H. J. Wijma, L. Fromont, D. B. Janssen, M. W. Fraaije,
FEBS Open Bio 2014, 4, 168-174.
1H NMR (400 MHz, CDCl
) δ 1.09 (d, J = 6.9 Hz, 3H, CH3), 1.56 –1.63 (m,
), 1.79 – 1.90 (m, 3H), 2.35 – 2.46 (m, 2H), 2.51
2.66 (m, 2H), 4.69 (m, 1H, C=CH ), 4.83 (m, 1H, C=CH ).
) δ = 15.8, 21.7, 26.5, 30.8, 44.1, 44.3, 44.7,
3
1
H), 1.73 (s, 3H, C=CCH
3
[
[
[
20] C. A. Abbas, A. A. Sibirny, Microbiol. Mol. Biol. Rev. 2011, 75, 321-360.
21] A. Z. Walton, J. D. Stewart, Biotechnol. Progr. 2004, 20, 403-411.
22] E. Romero, J. R. G. Castellanos, A. Mattevi, M. W. Fraaije, Angew. Chem.
Int. Ed. 2016, 55, 15852-15855.
–
2
2
13C NMR (100 MHz, CDCl
11.6, 147.0, 214.2.
Analytical data was in accordance with literature values.[23]
3
C
1
[23] D. H. Hua, S. Venkataraman, J. Org. Chem. 1988, 53, 1095-1097.
This article is protected by copyright. All rights reserved.

ChemCatChem
10.1002/cctc.201700573
FULL PAPER
This article is protected by copyright. All rights reserved.

ChemCatChem
10.1002/cctc.201700573
FULL PAPER
Entry for the Table of Contents (Please choose one layout)
FULL PAPER
Sofia Milker, Michael J. Fink, Nikolin
Oberleitner, Anna K. Ressmann, Uwe T.
Bornscheuer, Florian Rudroff* and
Marko D. Mihovilovic
Page No. – Page No.
Kinetic modeling of an enzymatic
redox cascade in vivo reveals
cofactor-caused bottlenecks
This article is protected by copyright. All rights reserved.