Available online at www.sciencedirect.

com

ScienceDirect
IFAC PapersOnLine 53-2 (2020) 16872–16877

Hybrid Cybernetic Modeling of

Polyhydroxyalkanoate Production in
Cupriavidus necator using Fructose and
Acetate as Substrates
∗
Stefanie Duvigneau Robert Du¨rr ∗∗ Lisa
Carius ∗ Achim Kienle ∗,∗∗
∗
Institute for Automation Engineering, Otto von Guericke University
Magdeburg
∗∗
Process Synthesis and Process Dynamics, Max Planck Institute for
Dynamics of Complex Technical Systems, Magdeburg

Abstract: Due to their advantageous properties, polyhydroxyalkanoates are a promising

alternative to conventional petroleum-based plastics. Currently, high production costs in
upstream and downstream have to be reduced to make the plastic material competitive. As
most of the PHA producing organisms metabolize a wide range of substrates upstream cost can
be reduced using carbon rich waste material. For large scale application sophisticated control
approaches based on mathematical models can help to further increase the productivity. In the
present work, a hybrid cybernetic model approach is presented, which is adapted to
experiments with fructose and acetate feeding, respectively. Furthermore, the validated model
was used to qualitatively predict the effect of co-substrate feeding.
Copyright © 2020 The Authors. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0)
Keywords: Polyhydroxyalkanoates, Hybrid cybernetic modeling, Metabolic yield analysis,
Cupriavidus necator, Fructose, Acetate
variables are introduced to regulate enzyme activity and
1. INTRODUCTION synthesis.
Besides petroleum-based polymers, the group of polyhy- In the present work, we develop an HCM considering fruc-
droxyalkanoates (PHAs) represents promising raw mate- tose and acetate, two common substrates for the PHB
rial for production of plastics. PHA-based materials are pro- duction by combining two metabolic models of C.
biodegradable, non-toxic and bio-based, since these poly- necator from literature (Franz et al., 2011; Yu and Si,
mers can be accumulated by a broad range of microorgan- 2004). Previous work for other microorganisms has
isms caused by limitation of, e.g., nitrogen, phosphorus already shown that HCMs are suitable for studying co-
or oxygen (Raza et al., 2018). Probably the best studied substrates (Song et al., 2009a). This is the first time,
representative from the class of PHAs is polyhydroxybu- where the HCM approach with co-substrate feeding is
tyrate (PHB). PHB is a homopolymer of the short-chain applied to the field of biopolymer production in C.
fatty acid (scl fatty acid) hydroxybutyrate (HB) which necator.
is produced by a variety of microorganisms from organic First, an adaption of the HCM rate constants kr to two
material. The most prominent producer is the bacterium different data sets is performed by using the data
Cupriavidus necator (C. necator ), which was also used in obtained after feeding with single carbon sources fructose
this work. and ac- etate, respectively. An analysis of the theoretically
One major goal to make PHA-based plastics economically pos- sible yield space and experimental yields with respect
competitive, is to reduce the costs of the fermentation to total biomass and HB is provided. Furthermore, the
process by inexpensive substrates (Koller et al., 2010). Fur- time evolutions for single carbon substrate feedings are
thermore, through smart process control and the shown. Finally, simulations w.r.t. the maximum HB
combina- tion of different substrates, productivity of the concentration for the co-feeding are performed. Different
bioprocess can be improved (Carius et al., 2018; Morabito ratios of fruc- tose and acetate are assumed in the in
et al., 2019; Lopar et al., 2013; Sˇ poljari´c et al., 2013). silico study.
These approaches require sophisticated mathematical
process descriptions. Hybrid cybernetic models (HCMs)
combine stoichiometric information of metabolic 2. EXPERIMENTAL METHODS
networks with temporal dynam- ics and are therefore
promising candidates (Ramkrishna and Song, 2018). 2.1 Mircoorganism and cultivation conditions
Thus, more a priori knowledge is used in HCMs in
comparison to simple kinetic models. Further- more, The experiments were performed with C. necator (H16,
cellular regulatory mechanisms are considered with- out DSM 428) obtained from DSMZ GmbH Braunschweig.
modeling them in detail. For that purpose cybernetic For the experiment with acetate as a single carbon
substrate, a 1 L shake flask was filled with 200 mL of
culture. M81 me-
2405-8963 Copyright © 2020 The Authors. This is an open access article under the CC BY-NC-ND license.
Peer review under responsibility of International Federation of Automatic Control.
10.1016/j.ifacol.2020.12.1211
 Stefanie Duvigneau et al. / IFAC PapersOnLine 53-2 (2020) 16872–16877 1687

dia 1 were supplemented with 10 g/L sodium acetate and well as Yu and Si (2004). By combining both, the resulting
1.5 g/L ammonium chloride (both: Carl Roth, Karlsruhe) model was expanded to the two substrate inputs fructose
and inoculated with C. necator H16, precultured in LB and acetate. Furthermore, the reversibility of reaction
media. The initial optical density of the culture was set equations was checked and adjusted using the KEGG
to database.
0.4. The culture was incubated at 30 ◦C and 150 rpm for
120 hours (h). Samples were taken every 8 h and analysed 3.2 Metabolic yield analysis
as described below. In contrast, for fructose as a single
carbon substrate, no new experiments were performed Metabolic yield analysis was performed as described by
but the data of Franz et al. (2011) obtained with the same Song and Ramkrishna (2009). First, the elementary modes
strain were used. of the network have to be calculated. In the present work,
4857 elementary modes were calculated using Metatool. As
2.2 Enzyme Assay suggested by Song and co-authors (Song et al., 2009a), the
The ammonium and fructose concentrations were deter- elementary modes of the metabolic model were classified
mined from supernatant of the samples using an enzy- into different sub-models according to the input
matic test kits (Kit No. 5390 and No. 10139106035, R- substrates in order to preserve functionality of the
Biopharm AG, Darmstadt, Germany) and following the overall model. Subsequently, the elementary modes for
manufactures instructions. each sub-model were reduced to a set of GMs by
designing the convex hull of the elementary modes in
2.3 High pressure liquid chromatography the yieldspace of HB and biomass via the MATLAB
Acetate and HB concentrations were determined with command convhulln. For further information regarding
an Agilent 1100 high performance liquid chromatography the analysis in yield space we refer to the publication of
(HPLC). For acetate 10 µL of the filtered supernatants Song and Ramkrishna (2009). For the presented model, 38
were loaded on reversed phase column (Inertsil 100A relevant GMs were selected to describe the process
ODS-3, 5µm pore size, 250x4.6mm, MZ-Analysentechnik dynamics.
GmbH, Mainz, Germany) and eluted isocratically with GMs were further reduced by selecting active modes rele-
1mL·min−1 and 0.1M NH H PO at pH 2.6 and 40 ◦C. vant to experimental data. In the present work, selection
4 2 4
For the determination of the HB concentration 1 mL was done during the parameter estimation. The set of
with the lowest normalized error square sum (ESS, (2))
culture broth was alkaline digested and prepared as re- was selected as AM set (N=15) shown in Table 1. The AMs
ported in (Satoh et al., 2016). 10µL of the digested and
filtered samples were loaded on the reverese phase col- reflect all possible metabolic states during HB synthesis in
umn (see above) and eluted isocratically with 1mL C. necator. In addition to the AMs that produce biomass,
min−1 · HB or acetate, there are three other modes accounting
◦ for cellular maintenance (AM 2, 11 and 14). Here, HB is
at 60 C. The eluent consists of 92% low concentrated metabolized to residual biomass and energy. This process
H2SO4 (0.025% solution, Carl Roth, Karlsruhe) and 8%
produces CO2 (not shown in Table 1). Maintenance can
acetonitrile (Carl Roth, Karlsruhe). The HB
be considered in the absence of all external carbon
concentration of the samples was determined by crotonic
sources (AM 14) or if the uptake of the substrate is
acid standards (Carl Roth, Karlsruhe). Since the alkaline
reduced (AM 2 and AM 11).
digestion is not a perfect conversion from HB to crotonic
acid, a PHB sample n
(Sigma
must Aldrich, St. Louis) with known concentration exp i sim
ESS i (
additionally
Y be processed to calculate a conversion yield (t ) − x
Σ x (t )
2
HB (Satoh et al., c

YHB = ·D (1) To calculate the ESS,i the difference between n simulated

cHB and experimental data points (xsim and xexp) at time ti
Here, the dilution ratio is D = 2 (1:2 v/v) and cHB was calculated, weighted with the maximum experimental
is the known HB concentration of the test sample. The value and summed up. The variable x is defined as x =
concentration of crotonic acid cCA is determined by the [xfru, xace, xN , xHB, xc ].
processed crotonic acid standard.
For both procedures peaks were detected with a photodiode- 3.3 State equation
array detector (G7115A, Agilent, Waldbronn, Germany).
State equations for the given input substrates fructose
(xfru), acetate (xace) and ammonium (xN ) are defined
2.4 Determination of Cell Dry Weight as follows:
For the determination of total cell dry weight, 1 ml culture d xxfru
broth was centrifuged in pre-weighed plastic tubes for 10 ace = SsZrMc (3)
minutes at 13000 x g and 4 ◦C. Subsequently, the cell dt x
N
pellet
was dried for 1 h at 99 ◦C and weighed. models of Franz and co-authors (Franz et al., 2011) as
1
Recipes for the Medium 81 can be found in Franz et al. (2011) or
3. MATHEMATICAL MODEL on the web page of the DSMZ.
3.1 Metabolic model
The present metabolic model (Appendix B) is based on
 Stefanie Duvigneau et al. / IFAC PapersOnLine 53-2 (2020) 16872–
16 stoichiometric information and selected AMs of the
The
metabolic network are included by the AM matrix in yield
space SsZ. Furthermore, the influence of various
substrate concentrations for each AM is modeled with
Monod-type kinetic in rM,i (9).
In addition to the equations for the substrates,
differential equations for internal metabolites with slow
dynamics compared to the remaining internal
metabolites can also be described (Franz et al., 2011).
This assumption holds
 Stefanie Duvigneau et al. / IFAC PapersOnLine 53-2 (2020) 16872–16877 1687

e.g. for cellular storage materials such as PHB. The dy-

namics of the HB content mHB can be described as
dmHB
= S HB ZrM —µmHB , (4)
dt
where SHBZ describes the AM matrix with respect to HB
in yield space and µ the cellular growth rate.
A special feature of the hybrid cybernetic approach is the
use of enzymes for each theoretical AM:
de
= α + rEMb − βe − µe, (5)
where dt
ei αi + ke,i
erel = with emax = . (6)
i max i
i β i + kr,i(ScZ)i Fig. 1. Yield space of carbon-normalized HB and
e
The increase of the enzyme level occurs via the total biomass. Experimental yields are divided into
constitutive enzyme synthesis rate α and the catalytically different phases (see text) for the carbon sources
active part of the total biomass b. The catalytically active fructose (FRU) and acetate (ACE). The grey squares
biomass frac- tion is controlled by Monod-type kinetics are the selected active modes (AMs) and the red
(10). Enzymes are degraded by the rate β and diluted due crosses are those AMs, that are used to adapt on
to cell growth (µ). The state equation for the total biomass experiments with single carbon source (scs).
concentration c can be described as follows:
dc 3.5 Numerical solution and parameter estimation
= µc. (7)
dt The model was implemented and solved numerically in
The growth rate µ is defined by the rate vector rM and MATLAB2016b. Parameter adjustment of rates kr were
the AM matrix for the biomass in yield space ScZ: performed using the MATLAB routine fmincon with a
µ = ScZrM. (8) lower bound of zero on all model parameters.
In general, the rates rM and rEM for the ith AM are Furthermore, as the optimization problem is highly
defined as: nonlinear, a multi- start optimization approach with n =
100 random initial parameter estimates was
rM,i = vikr,ierelrcore,i, (9) implemented to reduce the risk of ending up in a local
i
rEM,i = uike,ircore,i. (10) extremum when minimizing the cost function described
in Equation (2).
The rate rcore of the ith AM is a multiplied Monod-type
kinetic: 4. RESULTS
r = X1 · ... · Xn . (11)
In the present work an HCM was developed, which can
core,i KX1 + KXn + Xn describe the growth and the HB accumulation with either
X1
The substrate index n (fructose, acetate, ammonium) fructose or acetate as carbon source. Figure 1 shows the
depend on the number of negative yield coefficients for yield space for carbon-normalized HB and total biomass
each AM given in Table 1. Besides the Monod-type kinetic, as well as the experimental yields. Two data sets were
cybernetic control variables u and v which regulate used to adjust the parameters of the model: one data set
enzyme synthesis and enzyme activity, respectively, are with fructose as carbon source from literature (Franz et
calculated by cybernetic control laws (Young and al., 2011) and another with acetate as carbon source
Ramkrishna, 2007): from
p p our own shake flask experiment. Both data sets have been
u= v= . (12) divided into different phases for the calculation of yields:
ǁpǁ1 ǁpǁ∞ • Phase I: exponential growth
Here, p is the return on investment which can be calcu-
lated as follows: • Phase II: HB accumulation
p = diag(fc)diag(erel)diag(kr)rcore • Phase III: HB consumption
(13)
The vector of uptaken carbon units fc is normalized with Gonzalez and De Wever, 2018). Therefore, the saturation
respect to the highest carbon uptake. The vector is shown constant KACE is assumed as KFRU/2. All parameter values
in Table 1. are given in Table 2.
3.4 Parameters
Since the system was extended by enzyme equations for
the AMs and an adapted Monod kinetic for each mode on
the basis of our previous work (Franz et al., 2011), only
the constants kr,i in the definition of the rate rM,i are
free model parameters to be estimated from experimental
data. All other parameters of the model were kept as
given in Franz et al. (2011). The uptake of the carbon
source acetate is known to accelerate metabolism (Garcia-
 Stefanie Duvigneau et al. / IFAC PapersOnLine 53-2 (2020) 16872–
16 position of experimental yields
The in Figure 1 gives an
indication about the model fit to describe the data. Yields
for data sets with fructose as carbon source are inside or
on the theoretical hull spanned by metabolic model. Due
to the high uncertainty in the measurement of small dry
biomass and the broad sampling time interval for the
data set with acetate as the carbon source, the yields for
phase I and III are on the edge or slightly outside the
hull.
In the following, a parameter estimation was
performed for the constants kr,i, whose Monod-type
kinetics in rM,i can describe the growth on single
carbon substrates. All remaining constants kr,i have
been set to zero, as the co- substrate feeding was not
investigated experimentally. The AMs that were
included in the adjustment are marked in the yield
space (Figure 1, crosses in gray squares) and in Table 1
(FRU, ACE).
 Stefanie Duvigneau et al. / IFAC PapersOnLine 53-2 (2020) 16872–16877 1687

Table 1. Values for the yields in g/gC of the active mode matrix subdivided into
different subsections: substrates SSZ (1, upper white entries), HB SHBZ (2, light grey
entries) and total biomass ScZ (3, grey entries). Furthermore the normalized vector of
uptaken carbon units fc and division of AMs for the submodels fructose (FRU) and acetate
(ACE) are shown.
AM 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Yfru -2.50 -0.28 -2.50 -2.44 -1.00 -1.29 -2.50 -1.56 -2.50 0.00 0.00 0.00 0.00 0.00 -2.50
Yace (1) 0.00 0.00 0.00 -0.06 -0.01 -1.19 0.30 0.01 0.06 0.43 -0.01 -2.46 -2.46 0.00 0.00
YN -0.64 -0.73 0.00 -0.75 -0.72 0.00 -0.70 -0.76 0.00 -0.18 -0.33 -0.51 0.00 -0.001 -0.77
YHB (2) 0.31 -1.59 1.19 0.00 -1.07 0.25 0.00 -0.68 0.00 -1.79 -1.78 0.00 1.19 -0.021 0.00
Yc (3) 1.66 -0.05 1.19 1.58 0.47 0.25 1.49 0.93 0.00 -1.42 -1.08 1.08 1.19 -0.018 1.63
fc 1.00 0.75 1.00 1.00 0.83 0.99 1 0.89 1.00 0.72 0.72 0.98 0.98 0.01 1.00
Submodel FRU FRU FRU ACE ACE ACE ACE FRU FRU

Table 2. Parameters of the HCM 10 2

ConstantValue
8
kr [1/h] 1 [0.26 3.71 0.07 01x6 3.44
1.51 0.44 0.44 0.44 0.25]T 6
ke [1/h] 0.115x1 1
α [1/h] 0.01ke 4
β [1/h] 515x1
KN [g/L] 0.01 2
KF RU [g/L] 0.06
0 0
KACE [g/L] 0.03 0 20 40 60 80 100 120
KHB [g/L] 0.05
1
estimated values, arranged according to Table 1 4 0.8

20 4
3 0.6

15 3
2 0.4

10 2
1 0.2

5 1
0 0
0 20 40 60 80 100 120
0 0
0 10 20 30
Fig. 3. Dynamic behaviour using 10 g/L acetate as
12 4 carbon source. Experimental (circles and crosses)
and model kinetics (solid and dashed lines) of acetate
10
3 (upper black line and circles), ammonium (upper
8 dashed blue line and crosses), total biomass (bottom
6 2
black line and circles) and hydroxybutyrate (HB,
bottom dashed blue line and crosses) concentrations
4 are shown.
1
2 approximately 15 h. The linear phase in the model starts
0 0 earlier, as the exponential decay of the substrate in the
0 10 20 30 kinetics is coupled to ammonium, that is consumed after
15 h. The delayed start of linear substrate uptake in the
experiment indicates that other metabolites such as or-
Fig. 2. Dynamic behaviour using 20 g/L fructose as ganic acids are also generated at the beginning of the PHB
carbon source. Experimental (circles and crosses)
production phase. In the future, additional measurements
and model kinetics (solid and dashed lines) of
of the latter should be taken into account and included in
fructose (upper black line and circles),
an extended model formulation.
ammonium (upper dashed blue line and crosses),
In case of acetate as a carbon source, the dynamics can
total biomass (bottom black line and circles) and
be reproduced very well (Figure 3). Solely, the simulated
hydroxybutyrate (HB, bottom dashed blue line and
residual ammonium concentration gives evidence that the
crosses) concentrations are shown.
linear relationship between biomass growth and ammo-
Figure 2 and 3 show the simulated dynamics after pa- nium consumption should be adjusted in later model
rameter adjustment for both data sets. In the case of variants by adjusting the stoichiometry in the biomass
fructose as single carbon substrate (Figure 2), the model equation.
can reproduce the data very well. An exception is the Finally, the adapted HCM was used to analyze the effect of
fructose concentration, which decreases exponentially up fructose and acetate as co-substrates (Figure 4). Therefor,
to approximately 18 h, but in the simulation only up to different carbon ratios of acetate/fructose are simulated
16 Stefanie Duvigneau et al. / IFAC PapersOnLine 53-2 (2020) 16872–

and compared regarding their maximum amount of HB. parametric bootstraps or profile likelihoods (Raue et al.,
Our model predicts an benefical effect of co-feeding that 2009), is planned.
is increased with an evaluated acetate concentration in After successful experimental validation, acetate can also
the substrate mixture. A ratio of 1/10 acetate/fructose in be used in combination with low cost substrates (for
the initial concentration predicts an improvement of 6% exam- ple, regional waste products) to maximize PHA
com- pared to the summed HB concentration for feeding production and make the bio-based, biodegradable plastic
with only one carbon source material economically competitive to conventional plastic
(max(HB)FRU+max(HB)ACE). material. In addition, the hybrid model approach can be
With an 1/1 acetate/fructose ratio, the increase is already integrated into model-based control approaches to better
45%. This effect is a valuable information for the combi- control con- tinuous bioprocesses for PHA production
nation of waste materials as carbon sources with respect (Morabito et al., 2019).
to a cheap and productive process.
2.5
ACKNOWLEDGEMENTS
2

1.5 We would like to acknowledge the EU-programme ERDF

(European Regional Development Fund) for the funding
1 of Stefanie Duvigneau as part of the project DIGIPOL.
Furthermore we would like to thank Anja Julius for the
0.5
support in laboratory.
0
1/10 1/5 1/2 1/1
Appendix A. METABOLITE ABBREVIATIONS
Fig. 4. Relative maximum concentration of HB with 3PG 3-phosphoglycerate
different acetate/fructose (ACE/FRU) ratios. αKG alpha-ketoglutarate
The HB concentrations are normalized w.r.t. the ACE acetate
maximum HB concentration in the simulation with AcACE acetoacetate
fructose as the single carbon substrate. AcAcCoA acetoacetyl CoA
AcCoA acetyl CoA
ADP adenosine diphosphate
5. CONCLUSION ATP adenosine triphosphate
In the present work, a HCM was developed and adapted AMP adenosine monophosphate
to experimental data with fructose and acetate as a single AMC ammonium chloride
BIO residual biomass
carbon source. For this purpose, a metabolic yield analysis
CO2 carbon dioxide
was performed in which the relevant AMs were E4P erythrose-4-phosphate
determined (Song and Ramkrishna, 2009). This set of AMs F16P fructose-1,6-bisphosphate
is used to describe the rates of all state variables. Thus, F6P fructose-6-phosphate
the model has the advantage of being able to use the a FAD flavin adenin dinucleotide, oxidized
priori informa- tion from the metabolic network for the FADH flavin adenin dinucleotide, reduced
simulation of the dynamics. In addition to the use of FRU fructose
metabolic information, the HCM approach uses cybernetic G3P glyceraldehyde-3-phosphate
variables. These cyber- netic variables make it possible to G6P glucose-6-phosphate
GLN glutamine
describe the growth on several substrates in a proper
GLU glutamate
way (Song et al., 2009a). Due to the lack of a suitable GOX glyoxylate
experimental data set for co-feeding, the HCM adapted on HB hydroxybutyrate
data with single feeding strategies was used for an in silico ISC isocitrate
study with different acetate/fructose ratios. Therein, a MAL malate
positive effect on the HB yield at a given co-substrate NAD nicotinamide adenine dinucleotide (ox.)
feeding with fructose and acetate is predicted. This effect NADH nicotinamide adenine dinucleotide (red.)
is already well-known for different experimental setups NADP nicotinamide adenine dinucleotide
phosphate (ox.)
and organisms (Garcia- Gonzalez and De Wever, 2018;
NADPH nicotinamide adenine dinucleotide
Karmann et al., 2019). In contrast to our predictions, phosphate (red.)
Garcia-Gonzalez and De Wever (2018) also observed an NH3 ammonia
inhibitory effect on biomass production when feeding 4 g/l O2 oxygen
or even higher concentra- tions of a feed containing only OXA oxaloacetate
acetate. Future work will focus on experimental PEP phosphoenol pyruvate
investigations and corresponding extension of our HCM to PYR pyruvate
take such inhibitory effects into account. R5P ribose-5-phosphate
Furthermore, the simulation results should be experimen- Rl5P ribulose-5-phosphate
S7P sedoheptulose-7-phosphate
tally validated under controlled conditions. With such a
SUC succinate
validation experiment the remaining AMs considering the SucCoA succinyl-CoA
uptake of both, fructose and acetate, could be included in SUCx succinate, external
the HCM, as an expansion of the model. In addition, a X5P xylulose-5-phosphate
sophisticated analysis of parametric inference, e.g. using
 Stefanie Duvigneau et al. / IFAC PapersOnLine 53-2 (2020) 16872–16877 1687

Appendix B. METABOLIC REACTIONS a novel option to produce poly(3-hydroxybutyrate)

(PHB). Frontiers in Bioengineering and Biotechnology,
No.Reaction 7(59), 1–11.
1 FRU + PEP + ATP −→ F16P + PYR + ADP Katoh, T., Yuguchi, D., Yoshii, H., Shi, H.D., and Shimizu,
2 F16P −→ F6P
3 F16P ←→ 2 G3P
K. (1999). Dynamics and modeling on fermentative pro-
4 AMC −→ NH3 duction of poly(β-hydroxybutyric acid) from sugars
5 G6P + 2 NADP −→ Rl5P + CO2 + 2 NADPH via lactate by a mixed culture of Lactobacillus
6 Rl5P ←→ R5P delbrueckii and Acaligenes eutrophus. Journal of
7 Rl5P ←→ X5P Biotechnology, 67(2-3), 113–134.
8 X5P + R5P ←→ S7P + G3P Koller, M., Atli, A., Dias, M., Reiterer, A., and Braunegg,
9 S7P + G3P ←→ E4P + F6P G. (2010). Microbial PHA Production from Waste Raw
10 X5P + E4P ←→ G3P + F6P Materials, volume 14. Springer-Verlag Berlin Heidel-
11 F6P ←→ G6P
berg.
12 G3P + NAD + ADP ←→ 3PG + NADH + ATP
Lopar, M., Vrana Sˇ poljari´c, I., Atli´c, A., Koller,
M.,
13 3PG ←→ PEP batch cultivations of rhodospirillum rubrum under mul-
14 PEP + ADP ←→ PYR + ATP tiple nutrient-limited growth conditions on syngas as
15 OXA + ATP ←→ PEP + ADP + CO2
16 PYR + NAD ←→ AcCoA + NADH + CO2
17 AcCoA + OXA ←→ ISC
18 ISC + NADP ←→ αKG + NADPH + CO2
19 αKG + NAD −→ SucCoA + NADH + CO2
20 SucCoA + ADP ←→ SUC + ATP
21 SUC −→ SUCx
22 SUC + FAD ←→ MAL + FADH
23 MAL + NAD ←→ OXA + NADH
24 MAL + NADP ←→ PYR + CO2 + NADPH
25 PYR + ATP −→ OXA + ADP
26 ISC −→ SUC + GOX
27 AcCoA + GOX −→ MAL
28 NH3 + αKG + NADPH ←→ GLU + NADP
29 GLU + NH3 + ATP ←→ GLN + ADP
30 2 AcCoA ←→ AcAcCoA
31 AcAcCoA + NADPH −→ HB + NADP
32 ACE + ATP ←→ AcCoA + AMP
33 HB + NAD ←→ AcACE + NADH
34 ACE + SucCoA ←→ AcAcCoA + SUC
35 AcACE + ATP −→ AcCoA + AMP
36 2 NADH + O2 + 4 ADP −→ 2 NAD + 4 ATP
37 ATP + AMP ←→ 2 ADP
38 2 FADH + O2 + 2 ADP −→ 2 FAD + 2 ATP
39 0.21 G6P + 0.07 F6P + 0.9 R5P + 0.36 E4P + 0.13 G3P
+ 1.5 3PG + 0.52 PEP + 2.83 PYR + 3.74 AcCoA + 1.79
OXA + 8.32 GLUT + 0.25 GLUM + 41.1 ATP + 8.26
NADPH + 3.12 NAD −→ BIO + 7.51 αKG + 2.61 CO2 +
41.1 ADP + 8.26 NADP + 3.12 NADH (Katoh et al.,
1999)

All stoichiometric coefficients are given in mmol, except

BIO is given in g.
REFERENCES
Carius, L., Pohlodek, J., Morabito, B., Franz, A.,
Mangold, M., Findeisen, R., and Kienle, A. (2018).
Model-based State Estimation Based on Hybrid
Cybernetic Models. IFAC-PapersOnLine, 51(18),
197–202.
Franz, A., Song, H.S., Ramkrishna, D., and Kienle, A.
(2011). Experimental and theoretical analysis of poly(β-
hydroxybutyrate) formation and consumption in Ralsto-
nia eutropha. Biochemical Engineering Journal, 55(1),
49–58.
Garcia-Gonzalez, L. and De Wever, H. (2018). Acetic acid
as an indirect sink of CO2 for the synthesis of polyhy-
droxyalkanoates (PHA): Comparison with PHA produc-
tion processes directly using CO2 as feedstock. Applied
Sciences (Switzerland), 8(9). doi:10.3390/app8091416.
Karmann, S., Panke, S., and Zinn, M. (2019). Fed-
16Braunegg, G., and Horvat, P. Stefanie Duvigneau et al. / IFAC PapersOnLine 53-2 (2020) 16872–
(2013). Five-step con-
tinuous production of PHB analyzed by elementary
flux, modes, yield space analysis and high structured
metabolic model. Biochemical Engineering Journal, 79,
57–70.
Morabito, B., Kienle, A., Findeisen, R., and Carius, L.
(2019). Multi-mode Model Predictive Control and
Estimation for Uncertain Biotechnological Processes.
IFAC-PapersOnLine, 52(1), 709–714.
Ramkrishna, D. and Song, H.S. (2018). Cybernetic
Model- ing for Bioreaction Engineering. Cambridge
University Press.
Raue, A., Kreutz, C., Maiwald, T., Bachmann, J.,
Schilling, M., Klingmu¨ller, U., and Timmer, J.
(2009). Structural and practical identifiability analysis
of par- tially observed dynamical models by exploiting
the pro- file likelihood. Bioinformatics, 25(15), 1923–
1929.
Raza, Z.A., Abid, S., and Banat, I.M. (2018). Poly-
hydroxyalkanoates: Characteristics, production, recent
developments and applications. International Biodete-
rioration and Biodegradation, 126, 45–56.
Satoh, H., Sakamoto, T., Kuroki, Y., Kudo, Y., and
Mino,
T. (2016). Application of the Alkaline-Digestion-HPLC
Method to the Rapid Determination of Polyhydrox-
yalkanoate in Activated Sludge. Journal of Water
and Environment Technology, 14(5), 411–421.
Song, H.S., Morgan, J.A., and Ramkrishna, D. (2009a).
Systematic development of hybrid cybernetic models:
application to recombinant yeast co-consuming
glucose and xylose. Biotechnology and Bioengineering,
103(5), 984–1002.
Song, H.S. and Ramkrishna, D. (2009). Reduction of a
set of elementary modes using yield analysis.
Biotechnology and Bioengineering, 102(2), 554–568.
Sˇ poljari´c, Ivna, V., Lopar, M., Koller, M., Muhr,
A.,
Salerno, A., Reiterer, A., Malli, K., Angerer, H.,
Strohmeier, K., Schober, S., Mittelbach, M., and Hor-
vat, P. (2013). Mathematical modeling of poly[(R)- 3-
hydroxyalkanoate] synthesis by Cupriavidus necator
DSM 545 on substrates stemming from biodiesel pro-
duction. Bioresource Technology, 133, 482–494.
Young, J.D. and Ramkrishna, D. (2007). On the
Matching and Proportional Laws of Cybernetic
Models. Biotech- nology Progress, 23(1), 83–99.
Yu, J. and Si, Y.T. (2004). Metabolic carbon fluxes
and biosynthesis of polyhydroxyalkanoates in Ralstonia
eutropha on short chain fatty acids. Biotechnology
Progress, 20(4), 1015–1024.