Metabolic engineering of heterologous peptide

production in Saccharomyces cerevisiae

Ana Rita Brochado

Resumo alargado da dissertao para a obteno do

Grau de Mestre em Engenharia Biolgica
Jri
Presidente:
Orientadores:
Vogais:

Professora Isabel Maria de S Correia Leite de Almeida (IST)

Professor Jos Monteiro Cardoso de Menezes (IST)
Professor Jens Nielsen (DTU)
Doutor Miguel Nobre Parreira Cacho Teixeira (IST)

Setembro 2007

ABSTRACT
Recent advances in the molecular biology tools for selective DNA manipulation and
emergence of powerful techniques to study cellular physiology at molecular level is allowing the use
of a rational approach for microbial strain improvement. This new approach is referred to as
metabolic engineering which aims at directed genetic modifications in the cell in order to achieve a
desirable phenotype, such as increased yield and productivity of valuable metabolites. The
metabolic capabilities of different microorganisms for producing valuable compounds can therefore
now be more widely exploited in the pharmaceutical and chemical industry. Preferences for
physiological characteristics and other technological factors necessitate the use of certain selected
microorganisms, or cell-factories, for the production of desired compounds. It is therefore often
necessary to retrofit microbial genome by heterologous expression of a pathway in order to achieve
production of desired compounds.
The aim of this project was optimization of a heterologous production of a peptide of
pharmaceutical interest in the widely used cell-factory S. cerevisiae. First part of the study focuses
on a laboratorial approach whereby the metabolic bottleneck in the production of the peptide was
investigated by using supplementation of the precursor metabolites. In the later part, a rational in
silico design approach was used to identify metabolic engineering strategies that may lead to
productivity improvement. A genome-scale stoichiometric model of S. cerevisiae was used together
with an optimization algorithm OptGene and various gene knockout strategies were found. A novel
systematic approach for intuitive visualization and analysis of large number of in silico generated
strategies is presented. Physiological relevance of the identified key strategies is also discussed in
detail. Overall, it is concluded it will be possible to increase the productivity of the peptide by
using a three gene deletion strategy without significant impact on the growth properties of the
strain. Furthermore, the predicted phenotype contains significant alterations on the precursor
metabolism, suggesting that further work in this subject would be profitable.
Keywords: Metabolic engineering, cell factory, Saccharomyces cerevisiae, stoichiometric genomescale models.
Note: This work is part of a confidential project. The extended abstract does not contain
confidential information, thereby important results and conclusions were omitted.

TABLE OF CONTENTS
INTRODUCTION AND BACKGROUND

OPTIMIZATION OF PRECURSOR SUPPLY FOR PEPTIDE PRODUCTION IN S. CEREVISIAE

MATERIALS AND METHODS

MAIN RESULTS

DISCUSSION

DETERMINATION OF THE KM OF A BRANCH POINT ENZYME FROM S. CEREVISIAE

MATERIALS AND METHODS

MAIN RESULTS

DISCUSSION

IN SILICO OPTIMIZATION OF PEPTIDE PRODUCTION IN S. CEREVISIAE - OPTGENE

MATERIALS AND METHODS

MAIN RESULTS AND DISCUSSION

CONCLUSIONS AND FUTURE PERSPECTIVES

12

REFERENCES

13

Metabolic engineering of heterologous peptide production in S. cerevisiae

INTRODUCTION AND BACKGROUND

In metabolic engineering, the metabolic design problem can be approached through the
construction of a mathematical or in silico model of the metabolic network of the microorganism of
interest. This model can be used to simulate the cellular metabolic behaviour under different
conditions and even to design an improved metabolic network by suggesting changes in the
genotype of the micoorganism [3, 8]. In the last years, the availability of complete genome
sequences of several micoorganisms has allowed the development of metabolic models on a
genomic scale. Moreover, the change of the traditional local approach on biological systems to the
new global perspective resulted in a very large number of experimental databases, the so called
omics databases [4, 8].
Stoichiometric models define the metabolic network as a set of stoichiometric equations
representing the biochemical reactions in the system. Although it constitutes a hard task, the
elucidation about the overall metabolic network seems to be very attractive, once that it may
become a platform for in silico biology [8]. The first step to make a stoichiometric model is the
reconstruction of the whole metabolic network of the microorganism in question. A stoichiometric
model includes all known metabolic reactions, therefore, it implies a very exhaustive research in all
annotated genome information, textbooks, recent publications, reviews and pathway databases [3].
After identification of the metabolic network, the next step will be quantifying the fluxes through
different branches the fluxome. The quantification of the fluxes may be achieved by using the
concept of metabolite balancing to build a stoichiometric model. The material balance must be
established for each metabolite assuming stationary state for metabolite concentrations. Then, a
set of algebraic equations relating all the fluxes in the overall network is obtained. The solution of
the system with algebraic equations will provide the metabolic flux vector. This equation system is
often underdetermined, which means that the number of balances around intracellular fluxes is
smaller then the number of fluxes to be determined. Therefore, additional information or
assumptions are required to solve this mathematical problem [3, 6, 7]. Flux Balance Analysis (FBA)
aims to find a particular solution to the flux distribution. A particular flux objective function - is
optimized through linear programming, leading to a solution of the metabolic fluxes distribution.
Naturally, the micoorganisms are evolved for maximizing growth, so optimization of growth rate is
often applied as objective function in FBA [3, 8].
This project aims at the improvement of a heterologous production of a peptide of
pharmaceutical interest in S. cerevisiae. To synthesise this peptide, two genes (geneA and geneB)
are necessary, and the strains containing these genes were previously constructed at BioCentrumDTU laboratories.

OPTIMIZATION OF PRECURSOR SUPPLY FOR PEPTIDE PRODUCTION IN S. CEREVISIAE

This chapter will describe the first set of experiments aiming at the improvement of peptide
production through precursors supply optimization. The growth curve of the mutant with
heterologous expressed pathway was obtained, as well as the producing capacity was tested.
Afterwards, the three precursor amino acids were added to the culture medium to study the impact
on the desired metabolite production, as well as on the microorganism growth. Finally, a two-step
adaptation test to galactose as sole carbon source was carried out.
Materials and methods
The genetically engineered microorganism to synthesize the desired peptide was S. cerevisiae
and different strains were provided by BioCentrum-DTU strain collection. All the strains were
previously obtained from the wild type [Link] 113.9D (ura-, trp-) by transformation with the pESCURA cloning vector, comprising genetic information encoding the enzymes of interest and the URA3
selective marker. The strains containing the genes under control of GAL promoters are generically
called pESC-geneA-geneB, while the stains containing the constitutive promoters TEF1/ADH1 are
mentioned as pESC-TA-geneA-geneB.
All growth experiments were carried out in shake-flasks, under aerobic conditions, at 30 C
in an orbital shaker. The shake-flasks were inoculated with pre-cultures that had grown overnight in
the same conditions as the main cultures. All the pre-cultures were grown with SD URA- dropout
medium, while the main cultures were grown with different media depending on the strain and
experiment. The strains expressing the genes under control of GAL promoters were grown in SG
URA- dropout medium containing 1 or 3 g/l of glucose. The strain expressing the genes under control

Metabolic engineering of heterologous peptide production in S. cerevisiae

of the TEF1/ADH1 promoters was grown in SD dropout medium. In all the growth experiments the
biomass concentration was followed by OD600 nm measurement until the stationary-phase was
achieved. One litre of SD/SG dropout medium contains 20 g of glucose/galactose, 6.7 g of yeast
nitrogen base without amino acids, 40 mg of adenine sulphate, 20 mg of L-arginine, 100 mg of Laspartic acid, 100 mg of L-glutamic acid, 20 mg of L-histidine, 60 mg of L-leucine, 30 mg of L-lysine,
20 mg of L-methionine, 50 mg of L-phenylalanine, 375 mg of L-serine, 200 mg of L-threonine, 40 mg
of L-tryptophan, 30 mg of L-tyrosine and 150 mg of L-valine. For agar plates, 20 g of agar was added
to 1 liter of medium. Before sterilization the pH was set to 5.5-6 by adding sodium hydroxide
solution (2N). One experiment was carried out with Delft minimal medium with galactose as main
carbon source. One litre of Delft medium contains 100 ml of a 20 g/l carbon source (galactose)
solution and 900 ml of a solution containing: 75 ml of 0.1 g/l ammonium sulphate - (NH4)2SO4, 120
ml of 0.12 g/l monopotassium phosphate H2KPO4, 10 ml of 0.05 g/l magnesium sulphate heptahydrated MgSO4.7H2O, 2 ml of trace metal solution, 50 l of Sigma 204 antifoam and water. Before
sterilization the pH was set to 6.5 by adding sodium hydroxide solution (2N). After sterilization 0.1
ml of vitamin solution was added.
The external metabolite analysis was preformed by high-performance liquid chromatography
(HPLC) containing an ion exchange column. Detection of glucose, galactose, ethanol and glycerol
was performed by means of a Waters 410 differential refractometer, while detection of acetate an
pyruvate was performed by means of a Waters 490E programmable multiwavelength detector.
Since the peptide of interest is expected to occur mainly as an intracellular metabolite, the
crude extract was prepared by mechanical disruption with an organic solvent and the peptide of
interest concentration was analyzed by liquid chromatography connected with mass spectroscopy
LC-MS. The LC-MS chromatograms were collected from m/z 100 to 900.
Main Results
The strains containing the genes under control of constitutive and inducible promoters were
used to build the growth curve presented on Figure 1.

Ln(OD600 nm)

10

pESC-URA-4
pESC-geneA-geneB-12

0,1

pESC-TA-geneA-geneB-4

0,01
0

10

20

30

40

50

Tim e (h)

Figure 1: Growth curves of the strains containing the genes under control of inducible and constitutive
promoters, as well as of the strain containing the empty cloning vector.

The strains containing the genes under control of GAL promoters has a diauxic curve, since
they were grown in a media containing galactose as main carbon source and glucose to induce
growth. On the other hand, the strain containing the constitutive promoters has a normal, on
growth phase curve. The growth rates obtained for the different strains growing in glucose and
galactose are presented on Table 1.
Table 1: Growth rates and periods of exponential growth phases.
Strains
pESC-URA-4

1st step of exponential growth

-1

2nd step of exponential growth

Growth rate, 1 (h )

Period (h)

Growth rate, 2 (h-1)

Period (h)

0.264 0.008

0-14

0.071 0.009

20-38

pESC-geneA-geneB-12

0.277 0.008

0-14

0.089 0.008

20-30

pESC-TA-geneA-geneB-4

0.30 0.01

0-18

Metabolic engineering of heterologous peptide production in S. cerevisiae

The cell lysates from the strains pESC-geneA-geneB-12 and pESC-TA-geneA-geneB-4 were
analysed in acetonitrile or methanol by LC-MS. Moreover, the supernatant resulting from the
centrifugation of the cell culture of these two strains was analysed too, in order to confirm that
there was no loss of peptide of interest to the extracellular broth. A small peak with an elution time
of ~9 min, most probably representing the peptide was found in the cell lysate of the strain pESCgeneA-geneB-12 with both solvents. The peptide peak was not detected neither in the cell lysate of
the strain pESC-TA-geneA-geneB-4, neither in any of the supernatant samples.
Five different organic solvents were tested to carry out the cell lyses and peptide recovery:
acetonitrile, methanol, ethanol, acetone and isopropanol. All the solvents had very similar results
and the peptide eluted at ~9 min. Apparently it is indifferent which organic solvent might be the
best to the recovery process based on the product extraction, all of them seemed to be equally
suitable. Thus, with no special motive, acetonitrile and methanol were selected to proceed for
further analysis. However, three months later, the same standards with acetonitrile or methanol
stored at -20 C were run on LC-MS. It was found that, in acetonitrile, the peak corresponding to the
peptide (retention time of ~9 min) had a pronounced decrease, while another peak with double m/z
value and retention time of ~12.5 min had a pronounced increase. This result suggests that
acetonitrile is not a good solvent to the recovery process, probably because it may promote the
formation of dimmer bis-peptide. The same phenomenon was verified with methanol, but in less
extent, which may indicate methanol as a better solvent to the peptide recovery process.
To perform the amino acids supplying experiments, the strain pESC-geneA-geneB-12 was
grown in SG URA- dropout and Delft minimal media and after 20/24 hours different combinations of
the three precursor amino acids were added to the cell cultures. Although some impact on the
growth was detected after amino acids addition, the LC-MS results did not reported formation of
peptide in any of analysed the samples.
Finally, a two-step adaptation test was carried out with the strains pESC-geneA-geneB-3, -5,
-10 and -12. A 1st pre-culture was grown in SD URA- dropout medium and was used to inoculate the
2nd pre-culture, growing in SG URA- dropout medium supplied with 1 g/l glucose. After this
transition medium, the main cultures were grown in SG URA- dropout medium with no glucose
supplementation. Generally, the strains showed better growth performance in galactose (~0.1 h-1),
but no peptide formation was detect on LC-MS analysis.
Discussion
Generally, galactose was the carbon source used to grow the strains caring the genes under
control of the GAL1/GAL10 promoters. Unexpectedly, these strains were unable to grow with
galactose as sole carbon source, thus a low amount of glucose was supplied to the culture medium.
All the cultures growing with glucose and galactose had a diauxic growth, as was expected. These
results are supported by the external metabolite measurements by HPLC, the glucose is being
consumed during the first exponential growth phase, while the galactose concentration is
decreasing during the second exponential growth phase.
In general, the values obtained in this work for the growth rate in SD dropout medium
(glucose) are higher than the values obtained for the wild type by Villas-Boas et al. (2005) using
minimal medium, but they agree with the values previously determined in BioCentrum laboratories.
Concerning the growth rate in galactose, in all the experiments where the microorganisms were
grown in SG URA- supplied with glucose, the growth rate in galactose was very similar, between 0.08
and 0.09 h-1. These values are slightly higher than the values suggested by Villas-Boas et al. (2005)
[11]. When four strains where adapted to galactose as sole carbon source, the growth rate obtained
was higher than in the rest of the experiments (~0.1 h-1), suggesting that adapting the strains gives
better growth performance. However, it is unexpected that the strain caring the empty plasmid
shows a lower growth rate in galactose than the strains containing the plasmid with the genes, a
fact that occurred in most of the experiments. This tendency was only outdated after the
glucose/galactose adapting test, the obtained growth rate for the strain containing the empty
cloning vector was ~0.11 h-1.
During these experiments, different recovery processes for the peptide of interest were
tested, namely different organic solvents were used. The chromatograms obtained by LC-MS analysis
indicate that all the solvents were equally applicable and the peptide eluted in all the cases at ~9
minutes. Since no significant differences were found between the extraction process, acetonitrile
and methanol where the selected solvents with any special motive. However, three months later
the same standards containing acetonitrile or methanol were analysed and the results suggested
that acetonitrile may favour the formation of bis-peptide. Although it may be a good solvent to the

Metabolic engineering of heterologous peptide production in S. cerevisiae

recovery process, acetonitrile seems not to be suitable for long term storage of the samples. The
results obtained for methanol, suggested that with this solvent formation of bis-peptide also occurs,
although much more attenuated compared to acetonitrile.
The peptide of interest production was analysed for all the strains grown during this work.
In the first experiment, no peptide was detected in the strain containing the genes under control of
constitutive promoters and a very small peak from LC-MS analysis suggested that the strain
containing the GAL promoters produced some peptide. These results agree with results previously
obtained in BioCentrum laboratories. However, in the next two experiments no production was
detected. Although the strains had reacted differently to different amino acids additions, any
impact was reflected in the peptide of interest production. Even the strain growing without amino
acid addition, thus with the same conditions as before, was not able to produce. Concerning the
galactose adaptability test, although the strains had a good growth reaction, also here no peptide
was detected in the samples analysed by LC-MS.
Previous work with these strains also suggested that in the strain containing the GAL
promoters, the expression level of the two heterologous genes was higher than in the strains caring
the constitutive promoters. Furthermore, the expression of one gene in both strains was verified by
RT-PCR and Northern-blot. The expression of the other gene was confirmed by RT-PCR, but no
hybridization bands where shown in the northern-blot experiment, what may be explained by
possible mRNA degradation during the procedures. Although the gene expression has been
confirmed before, the peptide production by these strains was much lower in this work than in the
previous test. Furthermore, some peptide was detected in the first experiment, but no more after
that, which support the suggestion that these strains may be loosing not only growth performance,
but also peptide production capacity. There are some reasons that may be contributing to the low
(or no) peptide production by these strains:

geneB is a considerably big gene (> 11000 bp) and, even if it is being expressed, the mRNA
may be unstable, susceptible to mRNAases;

the two genes used in this project belong to two different microorganisms. In spite of the
homology between the genes form the different used microorganisms, the enzymes may be
not be working properly together;

although, the expression was confirmed, it does not mean that the proteins are being
actually produced, furthermore, there is no guarantee that the proteins are being produced
with the correct folding, or if they are stable in [Link];

there is no information about the possible degradation of product by the host strain.
These are just few reasons that may be contributing for no heterologous peptide of interest
production by the studied strains of S. cerevisiae.

DETERMINATION OF THE KM OF A BRANCH POINT ENZYME FROM S. CEREVISIAE

The availability of precursors in the cell to synthesise a certain metabolite is essential. In
this case, the precursors are amino acids, cellular building blocs that will be drained form the
anabolic reactions. The capacity to drain these metabolites from the anabolic reactions is strongly
dependent on the affinity of the branch point enzymes that split the normal route from the
new route [5]. The goal of this chapter is to obtain the kinetic parameters of one enzyme from a
branch point of the peptide biosynthesis.
Materials and Methods
The strain used to determine the KM value of the enzyme was [Link] 113 7D without any
auxotrophy marks, also provided by BioCentrum-DTU strains collection. The cells were grown in SG
dropout media (20 g/l galactose) and the inoculum was grown in SD dropout medium (20 g/l of
glucose). The media composition was already mentioned on previous topic.
The strain was grown aerobically with similar conditions as mentioned above. The cells were
grown until stationary phase and were harvested, centrifuged, washed and the pellet was kept at 20 C until further utilization. To perform the crude extraction the pellet was defrosted on ice and
resuspended in a volume of solution containing buffer (10 mM Tris-HCl pH 8) and protease inhibitor
(protease inhibitor cocktail from Sigma-Aldrich). The cells were broken with glass beads. To avoid
high temperatures (>10 C) the samples were vortexed in four rounds of 30 seconds (speed 5) and 40
seconds on ice. The lysate was centrifuged and the supernatant was recovered, frozen and kept at 20 C until further utilization. The protein concentration in the crude extract was calculated by the

Metabolic engineering of heterologous peptide production in S. cerevisiae

BCA method. To build the standard curve, protein (bovine serum albumine, BSA) solutions of known
concentration were used.
It was assumed that the enzyme follows the Michaelis-Menten kinetics and the objective of
this chapter is to determine the Michaelis-Menten constant, KM. To obtain the KM of the enzyme the
velocity of the catalyzed reaction was experimentally determined at different substrate
concentrations. The total volume for each reaction mixture was 1 ml. The initial velocity
corresponding to each different substrate concentration was determined by measuring the product
formation during the first 10 minutes of reaction at 30 C. The reaction was terminated by addition
of 1 ml of 2 % chemical compound A in a specific solvent and the product formation is detected by
formation of a compound with yellow colour absorbing at 460 nm.
Main results
The cells were grown in SG dropout medium (20 g/l galactose) until stationary phase. The
average value obtained for the cell concentration was 4.5 mgDW/ml of cell suspension and the
protein concentration was ~19.5 mg/ml of crude extract.
The enzymatic assay was performed and the absorbance of the samples was measured
immediately after the addition of the stop solution. However, the expected trend was not detected,
which means that was not possible to see higher product formation after longer time of incubation.
Furthermore, the absorbance of the samples was very instable and difficult to precise the exact
value. Thus, the absorbance of the samples was measured periodically during 44 hours. The
absorbance of all the samples increased sharply during 2 hours after the reaction had been stopped.
After that, the absorbance of all the samples increased reasonably at similar rate, independently of
the incubation time. Curiously, the sample containing higher substrate concentration but with no
incubation at 30 C, showed the highest absorbance values. Besides the instability of the absorbance
values, it is very difficult to decide which would be the right period to consider the best values.
The development of the absorbance with time is different depending on the concentration and time
of incubation of each sample. Furthermore, the differences between the absorbance of different
samples are too small to exclude the possibility of interference of another compound or to ignore
the instability of the measurement. Thus, if no correlation was found between the product
formation and the incubation time (reaction time), it was not possible to obtain the reaction rate
for different substrate concentrations.
In order to attenuate the possible interference of molecules with low molecular weight with
the absorbance or even with the reaction itself, the assay was partially repeated after dialysis of
the crude extract. The crude extract was filtrated in a size exclusion column, Microcon from
Millipore, model YM-10 with 10 000 of nominal molecular weight limit in Dalton. Moreover, to
reduce solvent evaporation, after incubation the samples were kept on ice and to ensure that the
reaction temperature was better controlled, the reactions were carried out in water bath at 30 C.
Once more, it was noticed that the absorbance was increasing, even if the samples were on ice.
After one hour and half, the experiment was interrupted based on the persistent instability of the
absorbance measurements and not concluding results. Furthermore, the absorbance of the diluted
crude extract was measured at 460 nm and the obtained value was very near the range of the
samples absorbance, which means that there is a possibility of the crude extract is interfering with
the absorbance results.
Discussion
Unfortunately, it was not possible to determine the kinetic parameters of the enzyme
applying the colorimetric assay suggested above. As mentioned during the Results, the expected
trend was not detected, it was not possible to see higher product formation after longer time of
incubation. Actually, it seemed that the incubation time had no effect at all on the absorbance of
the samples, since any pattern correlating the incubation time with absorbance was detected for
any substrate concentration. On the other hand, samples containing the same substrate
concentration showed similar absorbance increasing rate after incubation, independently of the
incubation time. Thus, this assay seems to be more sensitive to the substrate concentration rather
than the reaction time.
Along this experiment, it was verified that even the reaction with no substrate addition and
no incubation showed yellow colour with comparable intensity to the other samples. Indeed, this
result may suggest that the crude extract may contain other compounds absorbing in the same
wavelength after addition of the stop solution. Thus, if the reaction that was supposed to be the
negative control is showing analogous behaviour to the rest of the reaction, another negative
control should be used. Using crude extract of a knockout strain might be an attractive solution,

Metabolic engineering of heterologous peptide production in S. cerevisiae

since the possible lacking specificity of the colorimetric reaction would be detected and taken into
account.

IN SILICO OPTIMIZATION OF PEPTIDE PRODUCTION IN S. CEREVISIAE - OPTGENE

This chapter aims to design a strategy to improve the peptide of interest yield and
productivity in S. cerevisiae by using a genome-scale stoichiometric model with a genetic algorithm
OptGene - and thereby identifying key target genes for knockout.
Materials and Methods
The genome-scale reconstructed metabolic network constitutes de basis of all the work and
it was accomplished by Forster (2002) and is properly described in his PhD thesis [34]. Since the
objective here is to produce a heterologous peptide, it is important to add the heterologous
pathway before performing the simulations. Thus, the peptide biosynthetic pathway was added to
the in silico network, as well as an artificial reaction to excrete the peptide from the cell. If only
the first reaction is added to the model, no production will be obtained, due to the steady state
assumption by FBA approach. Consequently, mathematically it is impossible to accumulate an
intracellular metabolite such as the peptide of interest. If there is no further utilization of this
metabolite, it will be a dead-end reaction and the resulting flux is zero. After this in silico synthesis
step, the new S. cerevisiae strain that was obtained in the laboratory is reproduced by this updated
metabolic network.
OptGene algorithm was developed by Patil et al. (2005) and applies evolutionary
programming to rapidly identify gene deletion strategies aiming at the optimization of a desired
phenotypic objective function [9]. It derivers from OptKnock which searches for a set of gene
(reaction) deletions that maximizes the flux towards a desired product, while the internal flux
distribution is still operated such that growth is optimized. Within this approach, the identified
gene deletion will force the microorganism to produce the desired product in order to achieve
maximal growth [0, 9]. However, OptKnock is highly computational time demanding, which is a
problem for solving big problems. By formulating the in silico design problem as a Genetic Algorithm
(GA) optimization, OptGene can be faster and less computational time demanding then OptKnock.
Genetic algorithms use the principle of Darwinian evolution to search (evolve through mutations and
reproduction) for identifying the global optimal solution that is, individual with a maximum fitness
score. Concerning the use of a genetic algorithm, some parameters were established, namely the
size of the initial population (200 individuals), the crossover probability (0.95), the mutation rate
(0.25) and a stop criterion (2000 generations). The fitness score for each individual was obtained by
using FBA approach. For further investigation about OptGene, Patil et al. (2005) [9] is
recommended.
As already mentioned, Optgene was applied to find a knockout strategy to maximize in silico
yield and productivity of the desired peptide in S. cerevisiae. The yield of a metabolite is defined as
the gproduct/gsubstrate, while the substrate-specific productivity is defined as the gproduct/gsubstrate.h.
Note that the models based only on stoichiometry can not predict rates without an assumption of a
fixed substrate uptake rate. Since the substrate uptake rates for deletion mutants may change and
is very difficult to predict a priori, in general it is very difficult to optimize productivity by using
stoichiometric models. Thus, a possible alternative to go around this restriction is to optimize
another variable which is named Biomass-Product-Coupled-Yield (BPCY), which is obtained by
multiplying the product yield by the growth rate.
Main results and discussion
Before performing the predictive simulations, the maximum theoretical yield of biomass (or
growth rate) and peptide was estimated by using FBA. The maximum theoretical product yield is
obtained for a unit uptake rate of substrate by maximizing the sum of all reaction fluxes producing
minus those consuming the target metabolite, weighted by the stoichiometric coefficients. To study
the relationship between the growth rate and the product yield, the biomass formation constraints
were gradually set to different values from zero to the maximum theoretical value, and the product
pathway flux was maximized (Figure 2).

Metabolic engineering of heterologous peptide production in S. cerevisiae

3,5

0,25

PY (gprod/gsubst)

0,15

2
1,5

0,1

BPCY (gprod/psubst.h)

0,2
2,5

0,05
0,5
PY
0
0

0,2

0,4

0,6

0,8

1,2

1,4

0
1,6

BPCY

( h 1 )

Figure 2: Maximum theoretical values of peptide of interest yield and productivity in S. cerevisiae. Note
that the values presented on this figure are not real.

When the growth rate is fixed to its maximum value there is no production of peptide. On
the other hand, when the peptide production is maximized, the biomass growth is zero. Thus, it is
stoichiometrically impossible for the cell to grow and produce the peptide at the maximum rate.
This implies that there is a direct competition between biomass and product formation.
Consequently it is necessary to obtain a solution that strikes the correct balance between these two
competing variables. It is usually the economics of the process and the possible difficulties in
obtaining the predicted mutant that determines such point of balance. Thus, both yield and
productivity are important decision variables and the corresponding metabolic engineering
strategies are often different for these two different objectives. The obtained results were based
on an assumption of a fixed substrate uptake rate, which may change depending on the conditions.
Thus, all the remaining calculations presented in this work should be treated as relative values to
the maximum theoretical product yield and BPCY.
The maximum theoretical yield and productivity of peptide were calculated based on the
stoichiometric model, however it is difficult to predict a priori how many deletions would be
necessary to achieve these values (no algorithms have reported such methodology to date). Thus, to
optimize the peptide yield or productivity, the OptGene was applied to find the best set of 3, 4, 5,
6, 7 and 10 deletions. Note that optimization of yield or productivity are two different objectives
with two different criteria of fitness evaluation, which means that two different model formats
are necessary to optimize these two variables. In order to have a general overview on the obtained
phenotypes, three performance parameters were compared for the mutants with different number
of deletions for both approaches, namely optimization of yield and BPCY (Figure 3).
0,8

0,5

0,4

0,6
BPCY/BPCYmax

0,3

0,4

Productyield/[Link]
/max

0,2

0,2
0,1

0
3

5
6
Number of deletions

10

10

Number od deletions

Figure 3: Performance parameters of the obtained mutants for BPCY (right) optimization with 3, 4, 5, 6, 7
and 10 deletions.

OptGene suggested large number of strategies that will probably lead to the optimal
peptide production. The mutants resulting form yield optimization have very low growth rates,
while the phenotypes resulting from BPCY optimization show higher growth rates. On the other
hand, the maximum yield reached by the mutants form BPCY maximization is lower then the
maximum yield reached by the mutants form yield maximization. From 4 up to 10 deletions there is
no improving of the performance parameters of the mutants from yield maximization and the
highest yield is almost 50% of the maximum. In the mutants form BPCY maximization, the highest
yield (almost 40 % of the maximum) is reached with 4 deletions and any significant improvement is
verified up to 10 deletions. The highest BPCY of these mutants is more than 60 % of the theoretical
maximum.

Metabolic engineering of heterologous peptide production in S. cerevisiae

Generally, mutants obtained form BPCY maximization are more promising than the mutants
resulting from the yield optimization due to the better growth performance. Moreover, it was found
a 3 deletions strategy, which gives already few improvement of yield (~10 %) and a very attractive
growth preformance (almost 80 % of the theoretical maximum). The number of deletions is a very
important issue to select possible candidates for laboratorial implementation of the mutants, since
the mutants will probably loose robustness and growth performance with the high number of
deletions. Moreover, the cell behaviour will become more difficult to predict with the higher
number of deletions, which may lead to poor model predictions. For these reasons, the 3 deletions
strategy resulting in the phenotype showed on Figure 3 for BPCY optimization was selected as a
good candidate for laboratory work. Moreover, this strategy can be a start point for other strategies
also suggested by OptGene with higher number of deletions (4 or 5). The mutants lacking each one
of the three suggested genes were checked for viability and the uniqueness of the flux distribution
was verified.
In order to better understand the impact of the 3 suggested gene deletions on the metabolic
network and how it will affect the peptide production flux, the flux distribution within the overall
metabolic network of the mutant was compared with the flux distribution of the wild type (Equation
1).

Normalized diff =

flux n mut flux nWT

flux nWT

Equation 1

If the flux n in WT is zero, but it is different from zero in the mutant, the normalized
difference will assume a very big value without physical meaning. To avoid this type of
inconsistencies, it was assumed an Off-On switch for these reactions. Furthermore, the silenced
reactions (normalized diff = -1) were also labelled as On-Off reactions, as well as the inverted fluxes
were labelled depending on the direction of the reaction in the wild type. The labels were no
more than numbers (Table 2) that made possible the representation of the whole network in a plot,
showing the most affected fluxes and the type of affection. The fluxes highly increased
(normalized diff > 25) were also labelled with a smaller number to make their representation
possible on the same plot.
Table 2: Labels of the normalized differences between the fluxes of the mutant and the wild type used to
plot the overall metabolic network.
Description
Off-On - turned on fluxes
On-Off - silenced reactions
Highly increased fluxes
Inverted - Negative in WT
Inverted - Positive in WT

Value
20
-10
25
22.5
-7.5

Additionally, each reaction of the phenotype was assigned with a number and the
normalized differences were plotted in a graph: reaction number versus normalized difference or
the corresponding label (Figure 4).

10

Metabolic engineering of heterologous peptide production in S. cerevisiae

25

20

Normalized difference

15

10

0
0

100

200

300

400

500

-5

-10
Flux number

Carbohydrate metabolism

Energy metabolism

Lipid metabolism

Nucleotide metabolism

Amino acids metabolism

Complex carbohydrate metabolism

Complex lipid metabolism

Cofactors, vitamins, etc

Membrane transport

Artificial reactions

Peptide production

Figure 4: Graphic representation of the normalized differences between the fluxes of the 3 deletions
mutant and the wild type.

Figure 4 constitutes a possible way to represent the comparison of the phenotype of a

mutant and the wild type in terms of whole relevant metabolic reactions. Indeed, much information
about the mutant and the predicted changes in the metabolome can be extracted from Figure 4.
The reactions on the plot were assigned with different colours depending on the part of the
metabolism they belong, thus the connection of each reaction with the corresponding part of the
metabolism is immediate. Figure 4 enables to conclude that the main changes on the phenotype of
this mutant will be, most probably, on the central carbon and amino acids metabolisms. In the
central carbon metabolism there are some silenced fluxes and some fluxes are being slightly
increased, e.g. reactions from the pentose phosphate pathway, probably due to the high demand of
NADPH for amino acids biosynthesis. The most affected reactions seem to be from the amino acids
metabolism. Besides some soft changes (norm diff < 5), the most increased reactions, as well as a
big number of turned on or turned of fluxes are detected on the this part of the metabolism. The
rest of the metabolic reactions seem to be less affected, specially the lipid and nucleotides
metabolism. Some changes were also detected on membranes transport reactions, probably to
guarantee the correct balance of the metabolites without accumulation. To better understand the
results from Figure 4, the most affected parts of the metabolism (central carbon and amino acids
metabolisms) may be used to build a metabolic map where the most affected fluxes may be
distinguished according to their change.
Finally, a cost-profit analysis was performed to check the impact of protein overexpressing
on the mutants generated by OptGene. A great increase of a flux is a complex phenomenon as it
depends on several factors, viz., enzyme levels, reactant and product concentrations,
inhibitor/activator concentrations. Assuming that one reaction is being limited by enzyme
concentration, increasing the reaction flux, will necessitate increasing in the enzyme concentration.
In the model used in this study, enzymes are considered to be part of the biomass composition as
proteins, implying that the biomass composition of the mutants needs to be different relative to the
wild type. Moreover, high increase of a certain protein concentration may not be achievable and
may seriously affect the cell viability or growth performance. If a benefit of over-expression of a
certain protein can be predicted by using FBA and its production cost as the stoichiometric cost of
synthesizing it from amino acids, it is possible to add additional constraints in the OptGene
formulation. As a first step in developing this idea, the impact of change in the protein content on
the growth and production capacity of the yeast was tested in silico. The protein content was
increased by 5 and 10 %, while the lipid content was duplicated and triplicated. Since the biomass

11

Metabolic engineering of heterologous peptide production in S. cerevisiae

composition is mostly protein, it was already expected that increasing the protein content would
affect more the maximum yield and BPCY than the lipid content modification. The results showed
that until 10 % of protein increase, the performance parameters of the mutants will not suffer great
decrease. The concept of cost-profit analysis needs further development at individual protein level
and results presented here should serve only as guide-lines.
Along this entire chapter, different deletion strategies were suggested by OptGene to
improve peptide production by S. cerevisiae. Furthermore, one possible method to carefully analyze
the output of OptGene results was established here. Indeed, how to analyze and extract all the
relevant information from a pos-genomic tool as OptGene is always a hard task. The phenotype
discussed with more detail along this study was obtained with OptGene using FBA maximizing
growth and peptide production. The reasons leading to the selection of this strategy as good
candidate to proceed with mutant construction through gene deletion were mentioned above, but
its growing performance coupled with peptide production improvement were decisive. The flux
distribution predicted for this mutant was analysed and, as it was already expected, the main
changes in the metabolic network relatively to the wild type belong to the amino acids and central
carbon metabolisms. Although some changes in the network were object of special attention, other
important issues might not be noted here, even the interpretation of the metabolic changes of this
phenotype constitutes no more than possibilities. The implementation of this strategy could,
definitely, bring much information capable to be compared with the predictions. However, the
implementation of a mutant like the one suggested before could be very difficult, due to the
important pathways being directly affected. Instead of completely delete the reactions suggested
by OptGene, these reactions could be, in first approach, attenuated. This attenuation could be
achieved by downregulation of the expression of the target gene, for example, by changing its
promoter. Alternatively, if there are many isoenzymes possibly catalysing the same reaction, the
flux of this reaction can be reduced by deletion of one gene encoding just for one isoenzyme . After
a softer approach for the suggested strategy implementation, more reliable conclusions about the
viability and producing capacity of this mutant could be taken.

CONCLUSIONS AND FUTURE PERSPECTIVES

The initial approach to optimize a peptide of pharmaceutical interest heterologous
production by S. cerevisiae, was by direct supplementation of the precursors into the culture
medium. S. cerevisiae strains were previously constructed at BioCentrum-DTU laboratories and were
also verified to express the heterologous genes. Both, inducible - GAL7 and GAL10 and constitutive
ADH1 and TEF1 promoters were used to construct two different groups of strains. During this
project, the producing capacity of these strains was tested in different media with galactose or
glucose, as main carbon source. Peptide quantification was carried out by LC-MS technology. The
strains containing the genes under control of the GAL7/GAL10 promoters had shown low producing
capacity and any peptide production was detected for the strains with constitutive promoters.
During these experiments different organic solvents were tested for the recovery process of the
peptide. Although all the solvents showed to be equally suitable for the recovery process, it was
found that acetonitrile is not indicated to long term storage of the samples because it seems to
favour bis-peptide formation. On the other hand, methanol was also tested as a long term storage
solvent, and the results were much more suitable, showing less bis-peptide formation. The strain
showing the best producing capacity was used to test the effect of precursor supplementation.
Different media were test and no peptide production was detected. During the first two set of
experiments, the strains containing the genes under control of GAL7/GAL10 promoters showed
progressive loss of producing and growing capacities. Indeed, they were unable to grow with
galactose as sole carbon source from the beginning and few amount of glucose needed to be added
to induce growth. Thus, to avoid the glucose initial supplementation, a two-step adaptation test to
galactose as sole carbon source was carried out. Unfortunately, also here, pepetide production was
not verified, which suggests that different kind of approaches could be tried to find a solution for
this problem. Further work at the protein level, namely performing immunoassays to detect the
enzymes, codon optimization, fusion of the enzymes with GFP or test enzymes from different
microorganisms are just some ideas that could help to find out what can be done to produce the
peptide of interest in S. cerevisiae.
The role of the kinetic parameters of the enzymes from the branch points in the network
was already mentioned along this work. A colorimetric assay was performed to obtain the KM value
of an enzyme from a branch point between the peptide and a precursor synthesis. Unfortunately, no

12

Metabolic engineering of heterologous peptide production in S. cerevisiae

conclusive results were obtained. However, this is an important issue to future work and alternative
assays could be applied to find out the kinetic parameters of this enzyme.
The goal of the second part of this study was to find a metabolic engineering strategy
suggesting a set of gene knockouts leading to improvement of peptide production in yeast. Assuming
a S. cerevisiae strain which already has the peptide biosynthetic pathway, a genome-scale
stoichiometric model OptGene - was used to find the mentioned strategy. Several phenotypes
were generated and different gene deletion strategies were suggested to maximize two different
objectives: peptide yield or Biomass Product Coupled Yield (BPCY). Generally, the phenotypes
obtained for yield maximization showed very poor growth performance, due to the total drain of
cellular resources to peptide production. On the other hand, maximization of BPCY showed very
promising mutants, with attractive prediction for growth rate and product yield. Most affected
reaction belong to the central carbon and amino acids metabolisms. The flux distribution of
phenotypes resulting from the OptGene simulations constitutes a huge challenge. An original
method to analyse these phenotypes was proposed here, based on the comparison between the new
mutant and the wild type, the reference. Within this method, each mutant can be plotted and the
most affected reaction in the metabolic network can be immediately detected. This tool was very
useful in this study and provided already the opportunity, not only to analyse the flux distribution of
the mutants, but also to detect some parts of the metabolic network that could be improved. Based
in these observations, few modifications were made in the metabolic network and new, and
probably better, predictions were generated. The energy sources of the cell were also taken into
account in the last part of this project, since to overproduce a peptide, the building block amino
acids must be drained from the normal metabolism growth. Although no significant effects are
detected on the cell properties after 10 % increase the overall protein cellular content, a detailed
cost-profit analysis to the most affected enzymes in the network is suggested.
The results obtained along this study to optimize in silico peptide production by yeast were
very promising. Actually, more that one strategy capable to improve peptide production in yeast
was found through OptGene. However one must be aware that there are some limitations inherent
to the use of a stoichiometric model. The assumption of steady state for the metabolic network and
the lack of other kind of regulation but metabolic, constitute two delicate subjects. Furthermore,
to predict the flux distribution of mutants with three deletions (minimal number of deletions with
positive results in this work) is a difficult task, due to the natural unpredictability of the cell
behaviour after genetic manipulation. The permanent existence of unbalanced metabolites, which
is inevitable due to the incomplete knowledge of the metabolic network, will be always an
investigation field, since it is necessary to keep the model and the metabolic network always
updated. The co-factors mesh is another delicate topic, since it is very difficult to reproduce in
silico, due to its complexity. Evolution in stoichiometric models provided by advances in
biotechnology will be valuable, enabling the improvement of model predictions and intensifying the
stoichiometric models as strong metabolic engineering tools [2, 6, 7, 10].
Along this entire project, basic ideas about the principles of metabolic engineering have
been given, as well as some examples and how it may be useful to achieve the objective of this
project. Notwithstanding the fact that much work has been developed in metabolic engineering,
much more work needs to be done to explore the potential of this emergent field. In the future, the
rapid developments in genomics and functional genomics will lead to the discovery of many new
signal transduction pathways and new transcription factors. These may be very attractive targets
for metabolic engineering, since in many cases modulating transcription factors is more successful
than over-expressing individual structural genes because in this way a balanced up-regulation of a
complete pathway may be obtained. Furthermore, new biosynthetic pathways may be constructed
or identified in a certain organism, which may lead to new natural products to be used in different
fields. The continuous evolution in transcriptomics, proteomics, and fluxomics providing great
enhancement in the analysis step of the metabolic engineering cycle will also be very helpful to
develop metabolic engineering [7].

REFERENCES
1.

2.

Burgard AP, Pharkya P and Maranas CD. 2003. OptKnock: a bilevel programming framework for identifying
gene knockout strategies for microbial strain optimization. Biotechnology and Bioengineering. Vol. 84,
No. 6, p. 647-657.
Covert MW and Palsson BO. 2002. Trancrisptional regulation in constraints-based metabolic models of
Escherichia coli. Journal of Biological Chemistry. Vol. 277, p. 28058-28064.

13

Metabolic engineering of heterologous peptide production in S. cerevisiae

3.
4.
5.

6.
7.

Forster J. 2002. Pathway analysis of the metabolic network of Saccharomyces cerevisiae [PhD Thesis].
Technical University of Denmark.
Forster J, Famili L, Fu P, Palsson BO and Nielsen J. 2003. Genome-scale reconstruction of the
Saccharomyces cerevisiae metabolic network. Genome Research. Vol.13, p. 244-253.
Gunnarsson N, Eliasson A and Nielsen J. 2004. Control of fluxes towards antibiotics and role of primary
metabolism in production of antibiotics. Advances in Biochemical Engineering /Biotechnology. Vol 88, p.
137-178.
Koffas M, Roberge C, Lee K, Stephanopoulos G. 1999. Metabolic engineering. Annual Reviews of
Biomedical Engineering. Vol 1, p. 535-557.
Nielsen J. 2001. Metabolic Engineering. Applied Microbial Biotechnology. Vol. 55, p. 263-283.

8. Patil KR, Akesson M and Nielsen J. 2004. Use of genome-scale microbial models for metabolic
engineering. Current Opinion in Biotechnology. Vol. 15, p. 64-69.
9.

Patil KR, Rocha I, Forster J and Nielsen, J. 2005. Evolutionary programming as a platform for in silico
metabolic engineering. BMC Bioinformatics. Vol. 6, p. 308.
10. Pharkya P, Burgard AP and Maranas CD. 2004. OptStrain: A computational framework to redesign of
microbial production systems. Genome Research. Vol. 14, p. 2367-2376.

11. Villas-Boas SG, Akesson M and Nielsen J. 2005. Biosynthesis of glyoxylate from glycine in
Saccharomyces cerevisiae. FEMS Yeast Research. Vol. 5, p. 703-709.
12. Saccharomyces genome database: [Link]

14