Quantitative modeling of triacylglycerol homeostasis in
yeast – metabolic requirement for lipolysis to promote
membrane lipid synthesis and cellular growth
Ju
¨rgen Zanghellini
1,
*, Klaus Natter
2,
*, Christian Jungreuthmayer
3
, Armin Thalhammer
1
, Christoph
F. Kurat
2
, Gabriela Gogg-Fassolter
2
, Sepp D. Kohlwein
2
and Hans-Hennig von Gru
¨nberg
1
1 Institute of Chemistry, University of Graz, Austria
2 Institute of Molecular Biosciences, University of Graz, Austria
3 Trinity Center of Bioengineering, Trinity College Dublin, Ireland
Triacylglycerols (TAG) are important storage com-
pounds in pro- and eukaryotes. Not only do these
lipids store chemical energy in the form of fatty acids
(FA), they also serve to dispose of excess free FA from
the cellular milieu, thus precluding FA-induced toxicity
[1,2]. Neutral fats, which in yeast consist of TAG and
steryl esters (SE), are stockpiled in lipid droplets (LD)
during periods of cellular growth [3]. In times of star-
vation, esterified FA is then released by lipolysis and
recycled into other lipids, or degraded via b-oxidation
in order to provide the metabolic energy for cellular
maintenance [4].
Recent data have shown that TAG pools in yeast
are filled when growth ceases as a result of carbon
source (typically glucose) limitation, and cells enter
stationary phase [5]. TAG degradation during station-
ary phase occurs rather slowly and the specific activi-
ties involved have not yet been identified clearly.
Surprisingly, on glucose supplementation, quiescent
cells rapidly initiate TAG degradation at a high rate
when they re-enter the cell cycle [5]. Accordingly, tgl3
tgl4 mutants lacking the ability to hydrolyze TAG
show severe growth retardation. These observations
indicate that TAG degradation is an important
Keywords
dynamic flux-balance analysis; lipid
metabolism; Saccharomyces cerevisiae;
systems biology; triacylglycerol degradation
Correspondence
J. Zanghellini, Institute of Chemistry,
University of Graz, Heinrichstraße 28,
A-8010 Graz, Austria
Fax: +43 316 380 9850
Tel: +43 316 380 5421
E-mail: juergen.zanghellini@uni-graz.at
*These authors contributed equally to this
work
(Received 11 July 2008, revised
5 September 2008, accepted 9
September 2008)
doi:10.1111/j.1742-4658.2008.06681.x
Triacylglycerol metabolism in Saccharomyces cerevisiae was analyzed quan-
titatively using a systems biological approach. Cellular growth, glucose
uptake and ethanol secretion were measured as a function of time and used
as input for a dynamic flux-balance model. By combining dynamic mass
balances for key metabolites with a detailed steady-state analysis, we
trained a model network and simulated the time-dependent degradation of
cellular triacylglycerol and its interaction with fatty acid and membrane
lipid synthesis. This approach described precisely, both qualitatively and
quantitatively, the time evolution of various key metabolites in a consistent
and self-contained manner, and the predictions were found to be in excel-
lent agreement with experimental data. We showed that, during pre-loga-
rithmic growth, lipolysis of triacylglycerol allows for the rapid synthesis of
membrane lipids, whereas denovo fatty acid synthesis plays only a minor
role during this growth phase. Progress in triacylglycerol hydrolysis directly
correlates with an increase in cell size, demonstrating the importance of
lipolysis for supporting efficient growth initiation.
Abbreviations
CDP, cytidine diphosphate; DAG, diacylglycerol; DFBA, dynamic flux-balance analysis; FA, fatty acid; FBA, flux-balance analysis; LD, lipid
droplet; MP, membrane particle; PA, phosphatidate; PC, phosphatidylcholine; PE, phosphatidylethanolamine; SE, steryl ester; TAG,
triacylglycerol.
5552 FEBS Journal 275 (2008) 5552–5563 ª2008 The Authors Journal compilation ª2008 FEBS
determinant of rapid growth initiation. As peroxisomes
– the only site of b-oxidation in yeast – are repressed
by glucose, it was hypothesized that, during pre-
logarithmic growth, TAG-derived FA may be used as
a precursor for membrane lipid synthesis rather than
as an energy source [5].
In this study, we used the well-established yeast
model and combined theoretical and experimental
approaches to describe quantitatively the role of TAG
degradation in growing cells and the metabolic flux of
FA. We reconstructed the metabolic pathway of TAG
lipolysis in yeast in silico and specifically addressed the
question of whether FA derived from TAG hydrolysis
in growing cells is channeled into b-oxidation or
towards membrane lipid synthesis by a systems bio-
logical approach [6].
Our theoretical model is based on the well-estab-
lished concept of flux-balance analysis (FBA) [7], a
structural network model that replaces a full kinetic
description which, because of a lack of experimental
parameters, is as yet out of reach. FBA uses stoichi-
ometric information about all possible reactions which
comprise the metabolic network of yeast cells. By
assuming stationarity, FBA allows for the identifica-
tion of the optimum flux distribution to sustain a par-
ticular biological function. However, FBA is unable to
describe the kinetics of individual chemical reactions
and their regulation, as the analysis of the network
behavior is based on steady-state solutions.
Time-dependent effects can be taken into account by
adopting a dynamic extension to conventional, station-
ary FBA (dynamic flux-balance analysis, DFBA). In
brief, DFBA approximates the observed temporal
behavior by a series of steady-state solutions. Based on
technically mature theoretical methods, this systems
biological program has been applied successfully to
simulate a number of complex biological networks
[7,8]. The approach in this study differs from previous
implementations of stationary and dynamic FBA [9–
13] insofar as we experimentally determined the time
dependence of glucose and ethanol concentrations, as
well as of cell mass (growth). These data were used as
constraints to iteratively impose the observed func-
tional behavior on our in silico model in order to
reduce its degrees of freedom. We successively applied
different cellular objectives and locked the resulting
network response. The trained model was then utilized
to predict cellular TAG levels in response to altered
metabolic parameters. To confirm these results, the
average TAG content per cell during growth, and the
cell size, were determined experimentally.
Our study: (a) identifies TAG lipolysis during early
growth as an important, genuine effect; (b) shows that
TAG degradation is most prominent during the initial
lag phase after the inoculation of cells into fresh cul-
ture medium; and, most importantly, (c) yields a quan-
titative description of the utilization of TAG depots
for the production of membrane lipids in order to initi-
ate rapid growth, in accordance with experimental evi-
dence. Taken together, we present, for the first time, a
consistent and accurate quantitative analysis of a lipid
metabolic pathway in yeast.
Results
DFBA satisfactorily models the time-dependent
metabolic behavior of S. cerevisiae
The glucose uptake and growth rate of a wild-type
yeast culture were determined and subjected to DFBA
to predict the time evolution of the maximum possible
ethanol concentration in the medium. As a unique
DFBA solution requires an optimization criterion, we
employed the maximization of ethanol production as
the objective (Table 3, run 1).
As illustrated in Fig. 1 and in accordance with
experiments, ethanol (thin full line) is secreted during
all growth phases up to 35 h. Deviations between
the calculated and measured ethanol concentrations
result from ethanol loss because of evaporation.
Fig. 1. DFBA simulations and experimental data for cell density
(dotted line and open squares, respectively), glucose concentration
(broken line and open circles) and ethanol concentration (full line
and filled diamonds). The input data for the simulation (glucose
uptake and cell density) were first fitted to analytical functions
(dashed and dotted lines) to facilitate easy handling of the data.
The thin full line was obtained by assuming that all available sugar
is converted into ethanol. The shaded area underneath represents
an estimate of the portion of ethanol being evaporated. The thick
full line represents a DFBA calculation, where the maximum etha-
nol secretion rate has been constrained in order to fit the experi-
mentally measured values (filled diamonds).
J. Zanghellini et al. Triacylglycerol mobilization in yeast
FEBS Journal 275 (2008) 5552–5563 ª2008 The Authors Journal compilation ª2008 FEBS 5553
When vaporization was taken into account, all
experimentally measured ethanol concentrations were
in accordance with our calculation. In Fig. 1, the
loss caused by volatilized ethanol is represented by
the shaded area.
We trained our computer model by constraining
the ethanol secretion rate (Table 3, run 2) such that
the experimentally measured concentrations (Fig. 1;
filled diamonds) were best matched using least-
squares fitting. The fitting procedure used was to
reduce the maximum ethanol secretion rate, solve the
corresponding DFBA problem, correct for evapora-
tion and calculate the sum of squares of vertical
deviations. This sequence was repeated until the best
fit was achieved, resulting in a correlation coefficient
of r=98.2%. The maximum ethanol secretion rate
per gram dry weight of biomass was found to be
18.8 mmolÆg
)1
Æh
)1
, which is comparable with the
values reported by Velagapudi et al. [14] (18.2±
1.5 mmolÆg
)1
Æh
)1
) and Duarte et al. [15] (11.98
mmoÆg
)1
Æh
)1
). The data in Fig. 1 illustrate the result-
ing evolution of the ethanol concentration (thick full
line), and confirm that our implementation of DFBA
matches all measured data within the error bounds,
and thus accurately describes the dynamic behavior
of S. cerevisiae.
LD turnover in growing cells cannot solely be
explained by dilution
It has previously been shown that the relative vol-
ume of LD decreases by some 80% when stationary
phase (starving) yeast cells re-enter the cell cycle
after transfer into fresh medium containing glucose
as carbon and energy source (see Fig. 2, left panels)
[5]. One explanation for the time dependence of
cellular neutral lipid content may be simple dilution,
i.e. existing LD is distributed amongst a growing
number of cells, without active degradation. Such a
mechanism can explain the decrease in the relative
LD content per cell as a consequence of the sharing
of a constant amount of LD between an increasing
number of cells.
From our measurements, and in agreement with
published data [16,17], we found that LD typically
consists of 52 mol% SE and 48 mol% TAG. Assum-
ing that the composition of LD does not change dur-
ing hydrolysis, we have focused on the TAG content
of LD. The ‘dilution only’ model was calculated by
assuming the initial, total mass of TAG of the yeast
culture to be constant throughout the subsequent
growth period, m
TAG
(t
0
)X(t
0
)=m
TAG
(t)X(t) = con-
stant. Here, m
TAG
and Xdenote the mass of TAG per
cell and the cell number as a function of time t, respec-
tively, with the initial time t
0
.
In Fig. 2 (top right panel, full line), we show the
expected evolution of TAG levels based on dilution
and the experimentally determined mass levels (dot-
ted line), demonstrating a major deviation of the
observed TAG levels from the content expected as a
result of simple dilution. The difference (bottom
right panel) indeed represents the loss of TAG
caused by lipolytic activity, and shows that LD is
rapidly catabolized, reaching a minimum level after
3 h. After this period, first cell divisions occur, yet
the deviation of TAG levels between calculated dilu-
tion and measured data remains fairly constant
throughout the following 3 h. Figure 2 clearly shows
that the lipolytic activity peaks before the cells enter
exponential growth and continues for several hours
into logarithmic growth.
FA derived from TAG mobilization are not used
for energy production
To simulate LD mobilization, we employed DFBA
based on quantitative data of LD composition
(Table 1). Computationally, we modeled LD by add-
ing a reservoir of various neutral lipids (Table 1) to
our in silico model. Glucose uptake, calibrated etha-
nol production and cellular growth were used as
input values for the calculations. To uniquely define
the internal flux distribution, FBA requires an opti-
mization criterion, which, in biological terms, repre-
sents a certain physiological goal for the cell.
Typically, the maximization of cellular growth is
chosen as an objective [15,18,19]. As the time-depen-
dent growth behavior of our system is already deter-
mined by the input data, we were especially
interested in identifying conditions with high lipolytic
activity in silico to explain the experimental data.
Therefore, maximum LD mobilization was chosen as
an objective (Table 3, run 3).
The calculation revealed that, in the absence of addi-
tional metabolic fluxes, no change in TAG levels, and
thus no LD mobilization, takes place. The inability to
catabolize TAG under these conditions clearly indicates
that the release of FA and their degradation by peroxi-
somal b-oxidation are not possible. To confirm this
result, we simulated growth with the objective of maxi-
mizing acetyl-CoA generated by FA degradation
(Table 3, run 4). Yet, even under these conditions, a
negligible amount of TAG was mobilized (3 ·10
)5
mmolÆg
)1
Æh
)1
). We therefore conclude that peroxisomal
b-oxidation does not contribute to the experimentally
observed LD mobilization. This inability to break down
Triacylglycerol mobilization in yeast J. Zanghellini et al.
5554 FEBS Journal 275 (2008) 5552–5563 ª2008 The Authors Journal compilation ª2008 FEBS
free FA indicates that the cell transfers FA from TAG
to another acceptor molecule, as a balanced flux
distribution is otherwise unachievable. Accumulation of
free FA can be excluded due to their lipotoxic effects
and hence, free FA have to be processed further.
TAG are hydrolyzed exclusively to provide
precursors for membrane lipid synthesis
It has been suggested that, during pre-logarithmic
growth, FA released from TAG and SE may be used
as precursors for membrane lipid synthesis [4,5]. To
test this hypothesis, we simulated TAG mobilization
by DFBA under the assumption that the production
and storage of excess membrane material is possible by
including a pool of membrane lipids in our model.
Computationally, we introduced virtual membrane
particles (MP), which contain glycerophospholipids
and membrane sterols in a single entity that reflects
the typical lipid composition of cellular membranes.
The chemical composition of MP is listed in Table 2.
Fig. 2. Measured LD mobilization during early growth in comparison with LD kinetics caused by dilution. Top left panel: cellular growth X(t)
in complete medium. Bottom left panel: time profile of the TAG content per cell: m
TAG
(t). Top right panel: measured (filled circles) and calcu-
lated (open squares) normalized mass of TAG per cell as a function of time. The calculation assumes that, during the growth period, LD is
not metabolized, but shared between mother and daughter cells, hence diluting the initial LD concentration in the cell culture. Bottom right
panel: deviation Dbetween the measured and calculated normalized LD mass, defined as m
TAG
(t)⁄m
TAG
(t
0
)–X(t
0
)⁄X(t). Note that the largest
deviation occurs approximately 4 h before the TAG content reaches its minimum.
Table 1. LD components and their FA composition as obtained from mass spectroscopy.
Compound w ⁄w (%)
Fatty acid (mol%)
10 : 0 12 : 0 14 : 0 16 : 0 16 : 1 18 : 0 18 : 1
Ergosterol 19.1 – – 0.2 54.3 0.1 43.3 2.1
Episterol 8.1 – – 1.8 8.6 63.1 22.2 4.3
Fecosterol 6.1 – – 1.8 8.6 63.1 22.2 4.3
Lanosterol 1.0 – – 0.7 20.8 38.1 33.3 7.1
Zymosterol 12.2 – – 0.4 4.3 50.1 41.7 3.5
Triacylglycerol 53.5 2.0 6.0 19.9 39.2 17.0 8.4 27.0
Table 2. Composition of virtual membrane particle.
Compound w ⁄w (%)
Ergosterol 1.9
Phosphatidate 4.3
Phosphatidylcholine 29.6
Phosphatidylethanolamine 23.2
Phosphatidylinositol 27.2
Phosphatidylserine 9.9
Zymosterol 3.9
J. Zanghellini et al. Triacylglycerol mobilization in yeast
FEBS Journal 275 (2008) 5552–5563 ª2008 The Authors Journal compilation ª2008 FEBS 5555
The basic metabolic pathways involved in the produc-
tion of membrane lipids and MP, and their interaction
with TAG mobilization, are illustrated in Fig. 3. By
permitting or forbidding a flux from TAG degradation
products to virtual MP, in DFBA, we are able to dis-
sect the contribution of lipolysis to membrane lipid
synthesis.
Indeed, these DFBA calculations confirmed the
hypothesis that LD are only degraded if cells are
able to generate membrane material which utilizes
products of TAG hydrolysis (Table 3, runs 4 and 5,
respectively). Figure 4 illustrates that the predicted
lipolytic activity (degradation rate of 1.5·10
)2
mmolÆg
)1
Æh
)1
; thick full line) is in excellent quantita-
tive agreement with experimental observations (filled
circles) if the production of MP is permitted. If MP
production is disabled in the simulation, no lipolysis
occurs (LD degradation rate of 3 ·10
)5
mmolÆ
g
)1
Æh
)1
). Figure 4 plots the resulting time dependence
of the TAG concentration per cell for these two
cases. The difference between the simulation with
and without enabled membrane production (differ-
ence between the full and broken lines in Fig. 4) is
indeed dramatic, and MP production is increased by
three orders of magnitude if metabolically accessible
TAG pools are provided (inset in Fig. 4). In both
simulations, the impact of lipolytic activity was
found to be restricted to the production of mem-
brane material, as we could not detect any signifi-
cant changes in other metabolite concentrations. On
the basis of these results, we suggest that TAG
Fig. 3. Schematic representation of FA, neutral, and phospholipid
metabolism implemented in our reconstructed yeast network. Bro-
ken arrows mark the Kennedy pathway, which is turned off in our
calculations. Full arrows indicate the direction of metabolic fluxes
according to simulation 5 listed in Table 3. The circular areas repre-
sent the relative amount of FFA derived from LD mobilization (large
circle) and de novo synthesis (small circle). For further details, see
text. DAG, diacylglycerol; FFA, free fatty acid; LD, lipid droplet;
MAG, monoacylglycerol; MP, (virtual) membrane particle; PA, phos-
phatidate; PC, phosphatidylcholine; PE, phosphatidylethanolamine;
PI, phosphatidylinositol; PS, phosphatidylserine; SE, steryl esters.
Table 3. Summary of the simulation arrangements together with their main features. Additional parameters used in the simulations are
listed in Table 4.
Run
no.
Input (time
dependent) Constraint Objective function
Output
(time dependent) Comment
1 Glucose uptake Maximum ethanol production Ethanol concentration Overestimates data
Growth rate Fig. 1, thin full line
2 Glucose uptake Ethanol secretion £CMaximum ethanol production Ethanol concentration Fitting ethanol concentration
Growth rate C[12, 20] mmolÆg
)1
Æh
)1
Fig. 1, thick full line
3 Glucose uptake Excess MP production = 0 Maximum LD mobilization TAG concentration Inconsistent with experiment
Growth rate Fig. 4, broken line
Ethanol secretion
4 Glucose uptake Excess MP production = 0 Maximum acetyl-CoA production TAG concentration Inconsistent with experiment
Growth rate Fig. 4, broken line
Ethanol secretion
5 Glucose uptake Maximum LD mobilization TAG concentration Consistent with experiment
Growth rate MP concentration Fig. 4, full line
Ethanol secretion
6 Glucose uptake Maximum MP production TAG concentration Consistent with experiment
Growth rate MP concentration
Ethanol secretion
7 Glucose uptake LD mobilization = 0 Maximum MP production MP concentration Growth retardation
Growth rate Fig. 5, broken line
Ethanol secretion
Triacylglycerol mobilization in yeast J. Zanghellini et al.
5556 FEBS Journal 275 (2008) 5552–5563 ª2008 The Authors Journal compilation ª2008 FEBS