Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Metabolic Engineering 75 (2023) 12–18

Contents lists available at ScienceDirect

Metabolic Engineering
journal homepage: www.elsevier.com/locate/meteng

Design and application of a kinetic model of lipid metabolism in


Saccharomyces cerevisiae
Shekhar Mishra a, Ziyu Wang b, Michael J. Volk a, Huimin Zhao a, b, c, *
a
Department of Chemical and Biomolecular Engineering, Department of Energy Center for Advanced Bioenergy and Bioproducts Innovation, Carl R. Woese Institute for
Genomic Biology, USA
b
Department of Biochemistry, USA
c
Departments of Chemistry and Bioengineering, University of Illinois at Urbana-Champaign, Urbana, IL, 61801, USA

A R T I C L E I N F O A B S T R A C T

Keywords: Lipid biosynthesis plays a vital role in living cells and has been increasingly engineered to overproduce various
Lipid metabolism lipid-based chemicals. However, owing to the tightly constrained and interconnected nature of lipid biosynthesis,
Kinetic model both understanding and engineering of lipid metabolism remain challenging, even with the help of mathematical
Free fatty acid
models. Here we report the development of a kinetic metabolic model of lipid metabolism in Saccharomyces
Fatty alcohol
cerevisiae that integrates fatty acid biosynthesis, glycerophospholipid metabolism, sphingolipid metabolism,
storage lipids, lumped sterol synthesis, and the synthesis and transport of relevant target-chemicals, such as fatty
acids and fatty alcohols. The model was trained on lipidomic data of a reference S. cerevisiae strain, single
knockout mutants, and lipid overproduction strains reported in literature. The model was used to design mutants
for fatty alcohol overproduction and the lipidomic analysis of the resultant mutant strains coupled with model-
guided hypothesis led to discovery of a futile cycle in the triacylglycerol biosynthesis pathway. In addition, the
model was used to explain successful and unsuccessful mutant designs in metabolic engineering literature. Thus,
this kinetic model of lipid metabolism can not only enable the discovery of new phenomenon in lipid metabolism
but also the engineering of mutant strains for overproduction of lipids.

1. Introduction enhancing our understanding of host metabolism and designing meta­


bolic engineering strategies has already been well documented (Sánchez
Growing concerns of petrochemical resource depletion and climate and Nielsen, 2015). GEMs are, however, restricted in the information
change trends have, in recent times, increased the push towards sus­ they can capture and predict. GEM simulations to predict flux space only
tainable manufacturing of chemicals and fuels via biological routes show relevancy in the metabolic steady state of a microorganism. They
(Nielsen and Keasling, 2016). Numerous studies in recent years have are also unable to account for allosteric and regulatory control on
focused on increasing production of industrial chemicals from the lipid metabolic reactions.
metabolism such as free fatty acids, fatty alcohols, triglycerides and even To address these limitations, kinetic models are employed instead.
oleochemicals (d’Espaux et al. (2017); Runguphan and Keasling, 2014). Kinetic models use ordinary differential equations (ODEs) to capture the
However, due to the highly constrained and regulated nature of the lipid dynamic trajectories of all constituent species included in the model
metabolic network, engineering the lipid metabolism even in (metabolites, proteins, mRNA, etc.). The rate of consumption or pro­
well-studied model organisms such as Saccharomyces cerevisiae has duction of any constituent entity is captured by mathematical formulae,
invariably proved to be non-trivial and often challenging. Mathematical such as mass-action or Michaelis-Menten kinetics. Although no GEM has
modeling can play a significant role in bridging this gap by offering a specifically been utilized towards lipid-related metabolic engineering
framework to integrate prior knowledge of metabolic topology as well as applications, several kinetic models have been constructed to specif­
data from new experiments into a single holistic compendium. For ically focus on the pathways associated with lipid metabolism. For
example, the use of genome-scale stoichiometric models (GEMs) in example, Savoglidis and coworkers developed a lipid kinetic model that

* Corresponding author. Department of Chemical and Biomolecular Engineering, Department of Energy Center for Advanced Bioenergy and Bioproducts Inno­
vation, Carl R. Woese Institute for Genomic Biology, USA.
E-mail address: zhao5@illinois.edu (H. Zhao).

https://doi.org/10.1016/j.ymben.2022.11.003
Received 17 July 2022; Received in revised form 29 October 2022; Accepted 8 November 2022
Available online 9 November 2022
1096-7176/© 2022 International Metabolic Engineering Society. Published by Elsevier Inc. All rights reserved.
S. Mishra et al. Metabolic Engineering 75 (2023) 12–18

was used with inverse metabolic control analysis (IMCA) to predict the for 30 min in a Fisherbrand MultiTube vortexer (Thermo Fisher Scien­
fold-change in gene expressions as a consequence of a single gene tific, Waltham, MA). Cell lysate equivalent to 1 OD600 unit was trans­
deletion, solely from the changes in metabolite concentrations (Savo­ ferred to a new tube and ABC buffer was added to a total volume of 200
glidis et al., 2016). Recently, Tsouka and coworkers developed a lipid μL. To this, 10 μL of the internal lipid standard mix was added along with
metabolic reconstruction that integrates with a core metabolic GEM and 1 mL of 15:1 chloroform-methanol. The tube was vortexed at 2500 rpm
was systematically reduced around the lipid biosynthetic core (Tsouka for 2 h. Phase separation was achieved by centrifugation at 4000 rpm for
and Hatzimanikatis, 2020). However, no lipid kinetic model has been 10 min. The lower organic layer was transferred into a new vial and
used to predict the effects of lipid metabolic engineering on the pro­ dried in a SpeedVac (Thermo Fisher Scientific, Waltham, MA). To the
duction of value-added compounds. remaining chloroform-methanol mix in the tube, 1 mL of 2:1
In this work, we describe the construction of a kinetic model of lipid chloroform-methanol was added and the tube vortexed at 2500 rpm for
metabolism in S. cerevisiae that was designed and trained to rationalize another 2 h. After a second phase separation step, the lower organic
and predict viable lipid metabolic engineering strategies. The model layer was transferred into a new vial and dried in a SpeedVac. The
connects all submodules of the lipid network that are known to play a non-polar dried lipids from the first extraction were resuspended in 100
role towards the success or failure of metabolic engineering strategies μL of 4:2:1 isopropanol-methanol-chloroform with 7.5 mM ammonium
for lipid overproduction. The use case of the model was demonstrated by formate, while the polar dried lipids from the second extraction resus­
comparing in silico predictions of metabolic engineering strategies from pended in 5:1 methanol-chloroform containing 0.2 mM methylamine.
prior studies. The construction of mutants for fatty alcohol over­ The lipid resuspensions were then analyzed using a Q-Exactive Mass
production led to the discovery of a futile cycle in TAG biosynthesis, Spectrometer (Thermo Fisher Scientific, Waltham, MA) operated in the
which was hypothesized by model simulations and validated by 13C direct infusion mode via a Triversa Nanomate chip-based robotic infu­
labeling experiments. Thus, the model can be used to design new mu­ sion nanospray (Advion, Ithaca, NY). Details of mass spectrometric
tants for overproduction of lipid-based biochemicals in future studies. analysis, including Q-Exactive method parameters and Nanomate infu­
sion parameters, are provided in the Supplementary Information.
2. Methods and materials
2.4. Lipidomics data analysis
2.1. Strains, plasmids, and mutant construction
Identification and quantification of lipid species and classes from the
The reference yeast strain used in this study was BY4741 (MATa lipidomic data generated by the mass spectrometer was performed by an
his3Δ1 leu2Δ0 met15Δ0 ura3Δ0). All mutant strains were constructed in-house Python script that employed the pymzML library (Bald et al.,
by introducing mutations into the BY4741 genome. Genetic engineering 2012). Raw data files were input to the pipeline after conversion to the
to construct mutants was performed using the pCRCT plasmid reported open source mzML format using the tool MSConvert from the software
in the HI-CRISPR methodology (Bao et al., 2015). The plasmid contains Proteowizard (Chambers et al., 2012). After identification of lipid peaks
a variant of the SpCas9 gene, called iCas9 that was used to generate in the data, absolute quantification of each lipid species was performed
double stranded breaks in the yeast genome. To direct the iCas9 protein by normalizing to the intensity of the lipid internal standard from its
to the target region, the plasmid contains a 20 bp crispr-RNA (crRNA) corresponding class using a single-point calibration method. The Python
that was designed using the CRISPR tool provided by Benchling. The tool files associated with this pipeline are available in the GitHub repository.
uses two scoring systems for on-target (Doench et al., 2016) and
off-target analysis (Hsu et al., 2013). In appropriate mutations, a 50 bp 2.5. Model construction
linear homologous recombination (HR) donor was added along with the
crRNA-carrying pCRCT plasmid. For yeast transformation, the Model construction was performed in a systematic manner, where
commonly used lithium acetate heat shock protocol was used with a 1 h the topology of the lipid network was first established by referring to
heat shock (Gietz and Schiestl, 2007). The details of mutant construction databases such as BRENDA (https://www.brenda-enzymes.org/) and
including guide RNA sequences and list of all strains studied can be LipidMaps (https://www.lipidmaps.org/). Then, each enzymatic reac­
found in the Supplementary Information. tion was assigned a mathematical rate expression in the form of
Generalized Michaelis-Menten kinetics, also known as convenience ki­
2.2. Yeast cultivation netics (Liebermeister and Klipp, 2006). The Systems Biology Markup
Language (SBML) format was employed to store the model parameters,
All yeast cultures were carried out in a similar manner. A yeast metabolites and reaction rate laws (Bornstein et al., 2008). The Python
colony was chosen from a freshly streaked agar plate and inoculated in 2 library AMICI was used for integrating the kinetic model’s ODEs over
mL YPD media (1% yeast extract, 2% peptone, 2% dextrose) and grown time (Fröhlich et al., 2021).
for 24 h in a 30 ◦ C incubator with shaking at 250 rpm. At the 24 h mark,
the optical density (OD600) of the cell culture was measured and a new 2.6. Parameter estimation scheme
culture with initial OD600 of 0.2 was started in either 5 mL (round-
bottom tube) or 30 mL (baffled shake flask) of synthetic complete (SC) Lipidomic data generated from perturbation mutants was used to
media with no dropouts (0.17% yeast nitrogen base, 0.5% ammonium train the kinetic model by estimating parameters that fit in silico pre­
sulphate, 2% dextrose). The cells were then sampled at a pre-defined dictions of the model to the training dataset. The parameter estimation
timepoint or multiple timepoints of this SC growth culture based on process closely followed a multi-start gradient-based local optimization
the nature of the experiment (timeseries or steady state sampling). method as described here (Raue et al., 2013). Briefly, parameter esti­
mation was performed by generating a pre-determined number of guess
2.3. Lipidomic analysis vectors. From each guess vector, a gradient-based local search was
carried out to solve an optimization problem. The optimization problem
Lipidomics analysis of both the reference and mutant yeast strains formulated in this case was the minimization of the negative
was performed using a two-step chloroform-methanol extraction as log-likelihood function. The gradient searches from each start are in­
described elsewhere (Ejsing et al., 2009; Klose and Tarasov, 2016). dependent of each other and hence, could be performed in a parallel
Briefly, harvested cells were washed with 150 mM ammonium bicar­ manner, resulting in a faster estimation procedure. The Python libraries
bonate (ABC). Cell lysis was performed by resuspending the cells in ABC of AMICI (Fröhlich et al., 2021), PEtab (Schmiester et al., 2021) and
buffer, adding 200 μL zirconium glass beads and vortexing at 2500 rpm pyPESTO (Schälte et al., 2021) were employed to formulate the

13
S. Mishra et al. Metabolic Engineering 75 (2023) 12–18

parameter estimation problem and perform the parallelized gradient reductase (FAR) reaction was added. Fatty acid elongation and ergos­
searches. The scripts employed in the above parameter estimation terol biosynthesis were modeled as lumped reactions. This was carried
scheme as well as the PEtab files can be found in the GitHub repository out to minimize uncertainty in parameter estimation in these pathways
associated with this study. as quantitative data of the intermediates within these pathways could
not be obtained from the lipidomic analysis. To model the kinetic rate
2.7. Isotope labeling experiments for futile cycle analysis expressions for each reaction, the convenience rate law was employed
(Liebermeister and Klipp, 2006). Convenience kinetics also allows for
Experimental confirmation of futile cycles in the fatty alcohol mu­ the addition of regulatory interactions such as activation or inhibition.
tants was performed via stable isotope labeling experiments using 13C6 Such details were borrowed from earlier models of lipid metabolism that
–glucose containing media. Briefly, the control and mutant strains were included regulatory information (Alvarez-Vasquez et al., 2005, 2011).
cultured in 2 mL YPD for 24 h and transferred to 5 mL SC media at a
starting OD600 of 0.2. After 24 h of growth in SC media, the cells were 3.2. Parameter estimation with perturbation data
washed in water and resuspended in labeled isotope SC media con­
taining 2% 13C6 – glucose. At timepoints of 1, 2, 3, 4, 8 and 24 h, OD600 The first round of parameter estimation on the kinetic model was
values of the cultures were measured, and 5 OD600 units of cells were performed using lipidomic data obtained from the reference strain of
harvested. After washing and cell lysis with glass beads, lipid extraction S. cerevisiae, BY4741, grown in synthetic complete media. 19 lipid
was performed on 2 OD600 units of cell lysate, followed by lipidomic classes were quantified using internal standards along with free fatty
analysis. Steps for lipid extraction and lipidomic analysis were followed acid measurements in the extracellular environment (supernatant).
as described above. Parameter estimation on the model using the collected lipidomic data
from the reference strain suggested that this dataset alone could not
2.8. Code availability suitably be used to identify all the parameters in the reference model. To
improve estimates of parameters, mutations associated with lipid
The Python scripts accompanying this study can be found in the overproduction were introduced into the BY4741 strain. Specifically,
following GitHub repository: https://github.com/HuiminZhao/Mishra- three strains associated with increased flux towards either free fatty
Saccharomyces-lipid-kinetic-model. acids, fatty acyl-CoA or fatty alcohol were constructed, hereby referred
to as overproduction reference strains. These three strains were created
3. Results with mutations that have been reported in literature as successful stra­
tegies towards increasing titers of the corresponding products – i) triple
3.1. Model construction - network topology and rate kinetics knockout of faa1Δ, faa4Δ and pox1Δ for free fatty acid overproduction
(scFA-4) (Runguphan and Keasling, 2014), ii) triple overexpression of
The construction of a kinetic model of lipid metabolism in fatty acid biosynthesis genes via knock-in of TEF1 promoter (pTEF1) as
S. cerevisiae began with a tabulation of all the important pathways pTEF1:ACC1, pTEF1:FAS1 and pTEF1:FAS2 for fatty acyl-CoA over­
within the lipid network that are reported to play a role in either lipid production (scFA-5) (d’Espaux et al., 2017), and iii) the pTEF1 knock-in
homeostasis or lipid overproduction (Klug and Daum, 2014). The list of strain scFA-5 carrying a chromosomally-integrated heterologous copy of
pathways incorporated in the model include fatty acid synthesis and the MmFAR1 gene for fatty alcohol overproduction (scFA-7) (d’Espaux
elongation, biosynthesis of glycerophospholipids, sphingolipids, storage et al., 2017). The MmFAR1 expression cassette was integrated at the
lipids (diacylglycerol, triacylglycerol and ergosterol esters), ergosterol chromosomal site X-4 identified in an earlier study (Mikkelsen et al.,
and free fatty acid synthesis and degradation (which include activation 2012). Fig. S6 shows a lipid metabolic map with all the targeted genes
as well as β-oxidation) as depicted in Fig. 1. A separate reaction for the highlighted.
conversion of fatty acyl-CoA to fatty alcohol via the fatty acyl-CoA Network perturbation effects on lipid profiles were probed by

Fig. 1. The lipid metabolic map captured in the ki­


netic model and the essential submodules that are
known to contribute towards lipid overproduction
phenotypes - Fatty acid elongation pathway (white),
Glycerophospholipid biosynthesis (cyan), Sphingoli­
pid biosynthesis (green), Storage lipids (brown), Ste­
rols (gray) and Sterol esters (red). A-CoA, acetyl
coenzyme A; M-CoA, malonyl coenzyme A; F-CoA,
fatty acyl coenzyme A; DHAP, dihydroxyacetone
phosphate; Glyc-3-P, glycerol 3-phosphate; LPA,
lysophosphatidic acid; PA, phosphatidic acid; CDP-
DAG, cytidine diphosphate diacylglycerol; PS, phos­
phatidylserine; PE, phosphatidylethanolamine; PME,
phosphatidylmethylethanolamine; PDME, phosphati­
dyldimethylethanolamine; PC, phosphatidylcholine;
PI, phosphatidylinositiol; DAG, diacylglycerol; TAG,
triacylglycerol; VLCF-CoA, very long chain fatty acyl
coenzyme A; DHS, dihydrosphingosine; DHS-P,
dihydrosphingosine phosphate; FFA, free fatty acid;
Ethn-P, ethanolamine phosphate; F-ald, fatty alde­
hyde; IPC, inositolphosphorylceramide; MIPC, man­
nosyl inositolphosphorylceramide; M(IP)2C,
mannosyl-diinositolphosphoceramide. Solid lines
indicate the general direction of flux, dashed lines
indicate the presence of recycle reactions or regula­
tory interactions.

14
S. Mishra et al. Metabolic Engineering 75 (2023) 12–18

introducing single gene deletions in the reference strain BY4741 as well Feng et al., 2015). The two studies reported opposing effects on fatty
as the 3 overproduction reference strains. The genes targeted belong to alcohol titers from the knockout of this genes. OPI1 is an inhibitor of the
the glycerophospholipid, sphingolipid, and fatty acyl transfer sub­ INO2–INO4 heterodimer, known to activate various lipid biosynthetic
networks. Quantitative lipidomic data from the overproduction refer­ genes, while an increase in PA levels suppresses the effect of OPI1 by
ence strains and the single gene deletion mutants constructed in these tethering to it (Henry et al., 2012). In our model, we implemented the
strains resulted in a more comprehensive dataset to train the kinetic action of OPI1 as indirect activation of lipid biosynthetic genes by PA, as
model (Tables S4–S7). Parameter estimation was then performed with suggested elsewhere (Ferreira et al., 2018). Thus, modeling the
this dataset using the procedure described in the Methods section above. knockout of OPI1 simply involved removing the regulation terms by PA.
A multi-start approach was employed, and 1000 guess vectors We first implemented the base strains reported in these two studies
generated with gradient optimization starting from each of the guesses. and then enforced the effects of knockouts of OPI1 to observe the pre­
The initial conditions of observed variables were fixed to the measured dicted flux distributions as well as fatty alcohol titers (Fig. 3). According
lipid values of the reference strain BY4741, thus reducing the number of to the model, the success of OPI1 knockout in the study by Feng and
parameters to be estimated. From these optimization trajectories, the 10 coworkers (24) could be traced to an increase in phospholipid biosyn­
best results (lowest objective function value) were collected to create a thesis, which was accompanied by a commensurate increase in flux of
population of models for further analysis. Goodness-of-fit plots were fatty acyl-CoA biosynthesis that could be converted to fatty alcohol. In
constructed for this population of models to assess how close the model contrast, the model attributed the failure of the same strategy in the
matched experimental data. The goodness-of-fit plot for the lipid mea­ study by d’Espaux and coworkers (24) to the lack of increase in fatty
surements of reference strain BY4741 and the 3 constructed strains acyl-CoA flux. The authors of the latter study employed the constitutive
(scFA-4, scFA-5 and scFA-7) are shown in Fig. 2. The mean for each lipid TEF1 promoter to increase the flux in the fatty acyl-CoA biosynthetic
class was calculated by integrating each of the 10 models from the steps. This effectively decoupled expression of ACC1, FAS1 and FAS2
chosen population of models. As can be seen in the plot, most predicted from any lipid regulation as the TEF1 promoter lacks the inositol-
values closely match the actual measurements. The plot has an R2 value sensitive upstream activating sequence (UASINO) characteristic to
of 0.963 and a mean squared error (MSE) of 1.43 × 10− 3. many promoters in lipid metabolism that enables regulation at the gene
expression level (Henry et al., 2014). Thus, knockout of OPI1 led to an
increase in expression of other lipid biosynthetic genes without a similar
3.3. Analysis of known mutants using the kinetic model
increase in fatty acyl-CoA flux. This imbalance in fluxes at the fatty
acyl-CoA node causes a bottleneck resulting in lowered fatty alcohol
The kinetic model was first used to perform post-hoc analysis
titers.
comparing the strategies and outcomes presented in previous studies of
lipid-related metabolic engineering in S. cerevisiae reported in literature.
The model predicts a flux distribution of the lipid biosynthetic reactions 3.4. Lipid model as a hypothesis generator: discovery of a futile cycle in
thus providing a possible reason for the success or failure of mutants TAG biosynthesis
reported in literature. Briefly, we compared the performance of similar
mutations constructed in the two studies focusing on fatty alcohol Next, the model was used to assist in designing mutants for fatty
overproduction in S. cerevisiae, specifically the knockout of a tran­ alcohol overproduction. A strategy of accumulating triacylglycerol
scriptional regulator of lipid biosynthesis, OPI1 (d’Espaux et al., 2017; (TAG) via knockouts in the scFA-7 (BY4741 acc1:PTEF1-ACC1, fas1:
PTEF1-FAS1, fas2: PTEF1-FAS2, X4: PTEF1-MmFAR1-TCYC1) over­
production platform strain (producing fatty alcohol) followed by chan­
neling the accumulated carbon towards fatty alcohol was implemented.
Based on simulation predictions, we constructed various knockout mu­
tants in the scFA-7 strain aimed at increasing the triacylglycerol (TAG)
amounts in the biomass. Knockout of the genes OPI3 and INO1 in scFA-7
showed the highest increase in TAG accumulation when normalized by
OD600 units (Supplementary Fig. S1). The TAG accumulating mutants
were then combined with overexpression of FAA1, FAA4 and TGL3

Fig. 2. Goodness-of-fit plot of predictions from the population of models


plotted versus the lipid values for the reference strain BY4741 (red) as well as 3
platform strains, scFA-4 (green), scFA-5 (blue) and scFA-7 (yellow). Mean Fig. 3. Prediction of fatty alcohol titers on implementation of OPI1 knockouts
values for each lipid class were calculated from 10 model predictions belonging in each of the production strains described in either Feng et al. (Feng et al.,
to the population of models. R2 and mean squared error (MSE) are marked on 2015) or d’Espaux et al., 2017. Prediction mean and standard deviations were
the figure. calculated from simulating the population of models.

15
S. Mishra et al. Metabolic Engineering 75 (2023) 12–18

genes, which are expected to increase the flux of fatty acyl chain hy­
drolysis from TAG and convert them back to fatty acyl-CoA. However,
this had no noticeable effect on the fatty alcohol titers, and on further
inspection, the TAG levels showed no appreciable decrease in their
levels (Supplementary Fig. S1). On the other hand, the triple over­
expression strain of FAA1, FAA4 and TGL3 showed an increase in the
levels of diacylglycerol (DAG) and ergosterol ester (Supplementary
Fig. S2).
The lack of noticeable change in TAG levels were inspected through
in silico analysis using the trained kinetic models. The predicted flux
values from the kinetic model showed a concomitant increase in both
the in and out flux of TAG pools, i.e. in the TAG synthesis and TAG
degradation rates. These flux trends suggested the action of a futile cycle
in TAG biosynthesis. A futile cycle is said to exist when a reaction con­
verting a substrate to a product is counteracted by a reaction in the
opposite direction replenishing the substrate, leading to a net con­
sumption of ATP without any change in metabolite levels (Stephano­
poulos et al., 1998).
Experimental validation of the increased flux within the TAG
biosynthesis – degradation cycle was performed by a labeled isotope
experiment with dynamic sampling. At 24 h of growth in SC media, the
control and mutant strains were washed and resuspended in SC media
containing 2% 13C6 – glucose and the cultures sampled at timepoints of
1, 2, 3, 4 and 8 h. Lipidomic analysis of the intracellular contents
coupled with HPLC analysis of remaining 13C6 – glucose in the media
were used to relate consumed carbon to produced TAGs. The biomass-
normalized amounts of 13C TAG accumulated showed a roughly linear
trend in both strains from 1 to 4 h of sampling. The absolute rate of 13C
incorporation into TAG pools was used as a proxy for rate of synthesis of
TAG and contrasted with the biomass-predicted flux of TAG requirement
(Fig. 4A). Specific growth rates of both the control strain and the mutant
strain, required for the predicted TAG requirement, were estimated from
the exponential phase of the growth curves (Fig. 4B). As can be seen in
Fig. 4C, the mutant strain showed a significant increase in actual TAG
incorporation over the predicted TAG requirement, whereas the control
strain showed nearly equally matched rates. The mismatch in the mutant
strain incorporation rates indicates the presence of a futile cycle in TAG
synthesis-degradation, suggesting that the mutations in this strain may
have weakened the control architecture implemented by the host strain
to prevent futile cycles in TAG biosynthesis.
A single exponential enrichment model (Buescher et al., 2015) was
assumed for all the measured lipids within the TAG
biosynthesis-degradation cycle, namely PA, DAG and TAG. The enrich­
ment data along with pool size and time of sampling was used to
calculate the flux of 13C carbon entering the pools. The calculated flux
values were upregulated in all 3 lipids in the mutant strain compared to
the control (Fig. 5). The concomitant increase of flux in all the major
lipids within the TAG biosynthesis cycle establishes the presence of a
futile cycle in the mutant around TAG biosynthesis and degradation.
Fig. 4. A) Time series plot of incorporation of 13C into the TAG pools of both
4. Discussion control and mutant strains. B) Growth curves of both control and mutant strains
during the 13C labeling experiment, where labeled substrate was introduced at
In this work, we developed a lipid kinetic model that was designed t = 0 h. C) Comparison of TAG requirements as predicted from biomass cal­
and trained to guide metabolic engineering of S. cerevisiae for lipid culations and actual synthesis of TAG, as estimated by rate of 13C incorporation.
overproduction. The model was constructed to include all essential The comparison shows a clear increase in TAG incorporation in the mutant
submodules of the lipid metabolic network of S. cerevisiae that are strain over its predicted requirements.
known to contribute towards lipid overproduction phenotypes. The
training dataset for the model included intracellular and extracellular parameters estimated in this study specifically capture the phenotypes of
lipidomic profiles of the reference S. cerevisiae strain BY4741 as well as lipid overproduction (Alvarez-Vasquez et al., 2005, 2011; Savoglidis
mutants constructed to have lipid overproduction phenotypes. This et al., 2016). This enabled a closer scrutiny of flux through the lipid
enabled the model to both capture the trends of lipidomic shifts occur­ metabolic map when the reactions carried a higher amount of flux for
ring in overproduction mutants and predict the necessary genetic ma­ overproduction. Specifically, the model was trained on data from the
nipulations required for further overproduction mutants to be reference strain BY4741 as well strains that overproduce free fatty acids,
constructed. The use of overproduction strains and perturbation muta­ fatty acyl-CoA and fatty alcohols.
tions constructed on these reference strains sets the lipid kinetic model Designing a training dataset that includes single gene knockout
presented here apart from earlier published models as kinetic mutants constructed on the original reference strain BY4741 and the

16
S. Mishra et al. Metabolic Engineering 75 (2023) 12–18

Fig. 5. A) A reduced map of the TAG biosynthesis


cycle which is initiated by the successive acylation of
glycerol 3-P into LPA and PA. PA is then dephos­
phorylated to DAG, and DAG is finally acylated to
synthesize TAG. The action of lipases convert TAG
back to DAG along with the release of FFA. The
activation of FFA to fatty acyl-CoA completes the
cycle of reactions in TAG biosynthesis. To obtain es­
timates of flux of 13C incorporation into different
lipids in the TAG biosynthesis cycle, a single expo­
nential enrichment model was used. The flux values
of three lipid classes are as follows, B) PA, C) DAG,
and D) TAG.

three overproduction reference strains assisted in estimating parameters engineering. The utility of the model was demonstrated while designing
for a model that closely matched the training dataset (Fig. 2). Methods mutants for fatty alcohol overproduction, during which a futile cycle
such as Design of Experiments can be successfully employed in the was discovered in the TAG biosynthesis pathways. The ability of the
future to design optimal perturbation experiments that expand the model to explain successful and unsuccessful mutant designs in meta­
prediction space of a kinetic model. As demonstrated in this study, the bolic engineering literature also showcases its potential in assisting the
model can also be used to interpret and hypothesize mechanistic reasons design of future mutant strains for overproduction of lipid-based target
for the success or failure of mutant designs in metabolic engineering chemicals.
literature. Model simulations comparing the lipid profiles of knockouts
of OPI1, a transcriptional regulator of lipid biosynthesis, provided a Author contributions
reasonable explanation for the discrepancies noticed in their impact on
fatty alcohol titers (d’Espaux et al., 2017; Feng et al., 2015). Shekhar Mishra: Conceptualization, Data curation, Formal analysis,
The presence of cycles in the lipid network makes the emergence of Investigation, Methodology, Software, Writing – Original draft,
futile cycles a real possibility to contend with during lipid metabolic Visualization.
engineering. Combining insights from in silico analysis of the model and Ziyu Wang: Investigation.
13
C labeling data, we studied the emergence of a futile cycle in the tri­ Michael J. Volk: Investigation.
acylglycerol biosynthesis and degradation pathways within a mutant Huimin Zhao: Conceptualization, Writing – Original draft.
constructed for fatty alcohol overproduction. In humans, studies have
reported futile cycles around TAG biosynthesis that were activated Data availability
under interventions such as drug treatment (Guan et al., 2002). Even in
S. cerevisiae, a futile cycle with TAG biosynthesis has been reported Data will be made available on request.
under various conditions deviating from wild-type behavior, such as
mutations in the enzyme glycerol-3-phosphate acyltransferase (GPAT) Acknowledgements
which catalyzes the first committed reaction for TAG synthesis (Kiegerl
et al., 2019), or the removal of an endoplasmic reticulum membrane The authors wish to thank Dr. Jay Keasling for generously providing
protein required for efficient utilization of diacylglycerol (Markgraf the MmFAR1 gene. The authors also thank Dr. Yihui Shen for helpful
et al., 2014). In silico methods have also been developed to identify futile discussions on 13C data analysis. The authors wish to thank Dr. Alex­
cycles in genome-scale models (Gebauer et al., 2012). To investigate a ander Ulanov and Dr. Michael La Frano at the Metabolomics Center of
potential TAG futile cycle, we combined an in silico analysis with 13C the Roy J. Carver Biotechnology Center at the University of Illinois,
labeling experiments. We utilized kinetic model simulations to formu­ Urbana-Champaign. This work was supported by the U.S. Department of
late a hypothesis for the cycling of carbon flux in the TAG biosynthesis Energy (DE-SC0018260) (H.Z.).
and degradation pathways. Higher incorporation rates of 13C in TAG and
precursor intermediates demonstrated the existence of such a cycle
Appendix A. Supplementary data
(Figs. 4 and 5). In contrast with previously reported TAG futile cycles,
the cycle in this study relied on an upregulated TAG degradation and
Supplementary data to this article can be found online at https://doi.
FFA activation pathway. Our findings demonstrate the fine balance that
org/10.1016/j.ymben.2022.11.003.
exists between TAG synthesis and degradation pathways and suggest the
need for more strategies that can effectively channel carbon from stor­
References
age lipids towards valuable products without creating an imbalanced
system. Insights into the TAG synthesis-degradation balance may enable Alvarez-Vasquez, F., Riezman, H., Hannun, Y.A., Voit, E.O., 2011. Mathematical
design of future strains presenting oleaginous phenotypes. modeling and validation of the ergosterol pathway in Saccharomyces cerevisiae.
In conclusion, we developed a kinetic model of lipid metabolism in PLoS One 6, e28344. https://doi.org/10.1371/journal.pone.0028344.
Alvarez-Vasquez, F., Sims, K.J., Cowart, L.A., Okamoto, Y., Voit, E.O., Hannun, Y.A.,
S. cerevisiae that was trained on phenotypes relevant to metabolic
2005. Simulation and validation of modelled sphingolipid metabolism in

17
S. Mishra et al. Metabolic Engineering 75 (2023) 12–18

Saccharomyces cerevisiae. Nature 433, 425–430. https://doi.org/10.1038/ Henry, S.A., Gaspar, M.L., Jesch, S.A., 2014. The response to inositol: regulation of
nature03232. glycerolipid metabolism and stress response signaling in yeast. Chem. Phys. Lipids
Bald, T., Barth, J., Niehues, A., Specht, M., Hippler, M., Fufezan, C., 2012. pymzML- 180, 23–43. https://doi.org/10.1016/j.chemphyslip.2013.12.013.
Python module for high-throughput bioinformatics on mass spectrometry data. Henry, S.A., Kohlwein, S.D., Carman, G.M., 2012. Metabolism and regulation of
Bioinformatics 28, 1052–1053. https://doi.org/10.1093/bioinformatics/bts066. glycerolipids in the yeast Saccharomyces cerevisiae. Genetics 190, 317–349. https://
Bao, Z., Xiao, H., Liang, J., Zhang, L., Xiong, X., Sun, N., Si, T., Zhao, H., 2015. doi.org/10.1534/genetics.111.130286.
Homology-integrated CRISPR-cas (HI-CRISPR) system for one-step multigene Hsu, P.D., Scott, D.A., Weinstein, J.A., Ran, F.A., Konermann, S., Agarwala, V., Li, Y.,
disruption in Saccharomyces cerevisiae. ACS Synth. Biol. 4, 585–594. https://doi. Fine, E.J., Wu, X., Shalem, O., Cradick, T.J., Marraffini, L.A., Bao, G., Zhang, F.,
org/10.1021/sb500255k. 2013. DNA targeting specificity of RNA-guided Cas9 nucleases. Nat. Biotechnol. 31,
Bornstein, B.J., Keating, S.M., Jouraku, A., Hucka, M., 2008. LibSBML: an API library for 827–832. https://doi.org/10.1038/nbt.2647.
SBML. Bioinformatics 24, 880–881. https://doi.org/10.1093/bioinformatics/ Kiegerl, B., Tavassoli, M., Smart, H., Shabits, B.N., Zaremberg, V., Athenstaedt, K., 2019.
btn051. Phosphorylation of the lipid droplet localized glycerol-3-phosphate acyltransferase
Buescher, J.M., Antoniewicz, M.R., Boros, L.G., Burgess, S.C., Brunengraber, H., Clish, C. Gpt2 prevents a futile triacylglycerol cycle in yeast. Biochim. Biophys. Acta Mol. Cell
B., DeBerardinis, R.J., Feron, O., Frezza, C., Ghesquiere, B., Gottlieb, E., Hiller, K., Biol. Lipids 1864, 158509. https://doi.org/10.1016/j.bbalip.2019.08.005.
Jones, R.G., Kamphorst, J.J., Kibbey, R.G., Kimmelman, A.C., Locasale, J.W., Lunt, S. Klose, C., Tarasov, K., 2016. Profiling of yeast lipids by shotgun lipidomics. In: Devaux, F.
Y., Maddocks, O.D., Malloy, C., Metallo, C.M., Meuillet, E.J., Munger, J., Nöh, K., (Ed.), Yeast Functional Genomics, Methods in Molecular Biology. Springer New
Rabinowitz, J.D., Ralser, M., Sauer, U., Stephanopoulos, G., St-Pierre, J., Tennant, D. York, New York, NY, pp. 309–324. https://doi.org/10.1007/978-1-4939-3079-1_17.
A., Wittmann, C., Vander Heiden, M.G., Vazquez, A., Vousden, K., Young, J.D., Klug, L., Daum, G., 2014. Yeast lipid metabolism at a glance. FEMS Yeast Res. 14,
Zamboni, N., Fendt, S.-M., 2015. A roadmap for interpreting 13 C metabolite 369–388. https://doi.org/10.1111/1567-1364.12141.
labeling patterns from cells. Curr. Opin. Biotechnol. 34, 189–201. https://doi.org/ Liebermeister, W., Klipp, E., 2006. Bringing metabolic networks to life: convenience rate
10.1016/j.copbio.2015.02.003. law and thermodynamic constraints. Theor. Biol. Med. Model. 3, 41. https://doi.org/
Chambers, M.C., Maclean, B., Burke, R., Amodei, D., Ruderman, D.L., Neumann, S., 10.1186/1742-4682-3-41.
Gatto, L., Fischer, B., Pratt, B., Egertson, J., Hoff, K., Kessner, D., Tasman, N., Markgraf, D.F., Klemm, R.W., Junker, M., Hannibal-Bach, H.K., Ejsing, C.S., Rapoport, T.
Shulman, N., Frewen, B., Baker, T.A., Brusniak, M.-Y., Paulse, C., Creasy, D., A., 2014. An ER protein functionally couples neutral lipid metabolism on lipid
Flashner, L., Kani, K., Moulding, C., Seymour, S.L., Nuwaysir, L.M., Lefebvre, B., droplets to membrane lipid synthesis in the ER. Cell Rep. 6, 44–55. https://doi.org/
Kuhlmann, F., Roark, J., Rainer, P., Detlev, S., Hemenway, T., Huhmer, A., 10.1016/j.celrep.2013.11.046.
Langridge, J., Connolly, B., Chadick, T., Holly, K., Eckels, J., Deutsch, E.W., Mikkelsen, M.D., Buron, L.D., Salomonsen, B., Olsen, C.E., Hansen, B.G., Mortensen, U.
Moritz, R.L., Katz, J.E., Agus, D.B., MacCoss, M., Tabb, D.L., Mallick, P., 2012. H., Halkier, B.A., 2012. Microbial production of indolylglucosinolate through
A cross-platform toolkit for mass spectrometry and proteomics. Nat. Biotechnol. 30, engineering of a multi-gene pathway in a versatile yeast expression platform. Metab.
918–920. https://doi.org/10.1038/nbt.2377. Eng. 14, 104–111. https://doi.org/10.1016/j.ymben.2012.01.006.
d’Espaux, L., Ghosh, A., Runguphan, W., Wehrs, M., Xu, F., Konzock, O., Dev, I., Nielsen, J., Keasling, J.D., 2016. Engineering cellular metabolism. Cell 164, 1185–1197.
Nhan, M., Gin, J., Reider Apel, A., Petzold, C.J., Singh, S., Simmons, B.A., https://doi.org/10.1016/j.cell.2016.02.004.
Mukhopadhyay, A., García Martín, H., Keasling, J.D., 2017. Engineering high-level Raue, A., Schilling, M., Bachmann, J., Matteson, A., Schelke, M., Kaschek, D., Hug, S.,
production of fatty alcohols by Saccharomyces cerevisiae from lignocellulosic Kreutz, C., Harms, B.D., Theis, F.J., Klingmüller, U., Timmer, J., 2013. Lessons
feedstocks. Metab. Eng. 42, 115–125. https://doi.org/10.1016/j. learned from quantitative dynamical modeling in systems biology. PLoS One 8,
ymben.2017.06.004. e74335. https://doi.org/10.1371/journal.pone.0074335.
Doench, J.G., Fusi, N., Sullender, M., Hegde, M., Vaimberg, E.W., Donovan, K.F., Runguphan, W., Keasling, J.D., 2014. Metabolic engineering of Saccharomyces
Smith, I., Tothova, Z., Wilen, C., Orchard, R., Virgin, H.W., Listgarten, J., Root, D.E., cerevisiae for production of fatty acid-derived biofuels and chemicals. Metab. Eng.
2016. Optimized sgRNA design to maximize activity and minimize off-target effects 21, 103–113. https://doi.org/10.1016/j.ymben.2013.07.003.
of CRISPR-Cas9. Nat. Biotechnol. 34, 184–191. https://doi.org/10.1038/nbt.3437. Sánchez, B.J., Nielsen, J., 2015. Genome scale models of yeast: towards standardized
Ejsing, C.S., Sampaio, J.L., Surendranath, V., Duchoslav, E., Ekroos, K., Klemm, R.W., evaluation and consistent omic integration. Integr. Biol. 7, 846–858. https://doi.
Simons, K., Shevchenko, A., 2009. Global analysis of the yeast lipidome by org/10.1039/C5IB00083A.
quantitative shotgun mass spectrometry. Proc. Natl. Acad. Sci. U.S.A. 106, Savoglidis, G., da Silveira dos Santos, A.X., Riezman, I., Angelino, P., Riezman, H.,
2136–2141. https://doi.org/10.1073/pnas.0811700106. Hatzimanikatis, V., 2016. A method for analysis and design of metabolism using
Feng, X., Lian, J., Zhao, H., 2015. Metabolic engineering of Saccharomyces cerevisiae to metabolomics data and kinetic models: application on lipidomics using a novel
improve 1-hexadecanol production. Metab. Eng. 27, 10–19. https://doi.org/ kinetic model of sphingolipid metabolism. Metab. Eng. 37, 46–62. https://doi.org/
10.1016/j.ymben.2014.10.001. 10.1016/j.ymben.2016.04.002.
Ferreira, R., Teixeira, P.G., Siewers, V., Nielsen, J., 2018. Redirection of lipid flux toward Schälte, Yannik, Fröhlich, Fabian, Stapor, Paul, Vanhoefer, Jakob, Weindl, Daniel,
phospholipids in yeast increases fatty acid turnover and secretion. Proc. Natl. Acad. Jost, Paul Jonas, Wang, Dantong, Lakrisenko, Polina, Raimúndez, Elba,
Sci. U.S.A. 115, 1262–1267. https://doi.org/10.1073/pnas.1715282115. Pathirana, Dilan, Schmiester, Leonard, Städter, Philipp, Contento, Lorenzo,
Fröhlich, F., Weindl, D., Schälte, Y., Pathirana, D., Paszkowski, Ł., Lines, G.T., Stapor, P., Merkt, Simon, Dudkin, Erika, Grein, Stephan, Hasenauer, Jan, 2021. pyPESTO -
Hasenauer, J., 2021. AMICI: high-performance sensitivity analysis for large ordinary Parameter EStimation TOolbox for python. https://doi.org/10.5281/
differential equation models. Bioinformatics 37, 3676–3677. https://doi.org/ ZENODO.6606687.
10.1093/bioinformatics/btab227. Schmiester, L., Schälte, Y., Bergmann, F.T., Camba, T., Dudkin, E., Egert, J., Fröhlich, F.,
Gebauer, J., Schuster, S., de Figueiredo, L.F., Kaleta, C., 2012. Detecting and Fuhrmann, L., Hauber, A.L., Kemmer, S., Lakrisenko, P., Loos, C., Merkt, S.,
investigating substrate cycles in a genome-scale human metabolic network: substrate Müller, W., Pathirana, D., Raimúndez, E., Refisch, L., Rosenblatt, M., Stapor, P.L.,
cycles in the human metabolic network. FEBS J. 279, 3192–3202. https://doi.org/ Städter, P., Wang, D., Wieland, F.-G., Banga, J.R., Timmer, J., Villaverde, A.F.,
10.1111/j.1742-4658.2012.08700.x. Sahle, S., Kreutz, C., Hasenauer, J., Weindl, D., 2021. PEtab-Interoperable
Gietz, R.D., Schiestl, R.H., 2007. High-efficiency yeast transformation using the LiAc/SS specification of parameter estimation problems in systems biology. PLoS Comput.
carrier DNA/PEG method. Nat. Protoc. 2, 31–34. https://doi.org/10.1038/ Biol. 17, e1008646 https://doi.org/10.1371/journal.pcbi.1008646.
nprot.2007.13. Stephanopoulos, G.N., Aristidou, A.A., Nielsen, J., 1998. Review of cellular metabolism.
Guan, H.-P., Li, Y., Jensen, M.V., Newgard, C.B., Steppan, C.M., Lazar, M.A., 2002. In: Metabolic Engineering. Elsevier, pp. 21–79. https://doi.org/10.1016/B978-
A futile metabolic cycle activated in adipocytes by antidiabetic agents. Nat. Med. 8, 012666260-3/50003-0.
1122–1128. https://doi.org/10.1038/nm780. Tsouka, S., Hatzimanikatis, V., 2020. redLips: a comprehensive mechanistic model of the
lipid metabolic network of yeast. FEMS Yeast Res. 20 https://doi.org/10.1093/
femsyr/foaa006 foaa006.

18

You might also like