Detection of transcriptional triggers in the
dynamics of microbial growth: application
to the respiratorily versatile bacterium
Qasim K. Beg1, Mattia Zampieri2,3, Niels Klitgord2, Sara B. Collins2, Claudio Altafini3,
Margrethe H. Serres4and Daniel Segre `1,2,5,*
1Department of Biomedical Engineering, Boston University, Boston, MA 02215, USA,2Bioinformatics Program,
Boston University, Boston, MA 02215, USA,3SISSA (International School of Advanced Studies), 34136 Trieste,
Italy,4Josephine Bay Paul Center for Comparative Molecular Biology and Evolution, Marine Biological
Laboratory, Woods Hole, MA 02543, USA and5Department of Biology, Boston University, Boston, MA 02215,
Received November 29, 2011; Revised April 27, 2012; Accepted May 1, 2012
The capacity of microorganisms to respond to
variable external conditions requires a coordination
of environment-sensing mechanisms and decision-
making regulatory circuits. Here, we seek to under-
stand the interplay between these two processes
by combining high-throughput measurement of
time-dependent mRNA profiles with a novel compu-
tational approach that searches for key genetic
triggers of transcriptional changes. Our approach
helped us understand the regulatory strategies of a
respiratorily versatile bacterium with promising
Shewanella oneidensis, in minimal and rich media.
By comparing expression profiles across these two
conditions, we unveiled components of the tran-
scriptional program that depend mainly on the
time-dependent data with a previously available
large compendium of static perturbation responses,
we identified transcriptional changes that cannot be
explained solely by internal network dynamics, but
are rather triggered by specific genes acting as key
mediators of an environment-dependent response.
These transcriptional triggers include known and
novel regulators that respond to carbon, nitrogen
and oxygen limitation. Our analysis suggests a
sequence of physiological responses, including a
coupling between nitrogen depletion and glycogen
storage, partially recapitulated through dynamic
flux balance analysis, and experimentally confirmed
by metabolite measurements. Our approach is
broadly applicable to other systems.
depends both on the physico-chemical properties of the
external environment and on the internal state of the cell
(e.g. its growth rate) (1–3). The time-dependent array of
environmental stimuli and the wiring of the underlying
regulatory network jointly determine the dynamical
changes in gene expression (4, 5). Understanding the inter-
play between these two elements is an important open
challenge, especially relevant for metabolically versatile
microbes that may rapidly cycle through different envir-
onments. Given that a microbial cell needs to dynamically
adapt to changing environmental conditions, transient
changes in protein expression must be triggered by
condition-specific external factors, which interact with
the intracellular genetic network. However, other tran-
scriptional programs, such as those involved in cellular
division, must be robust to influences external to the
cell. Thus, it may be possible to discriminate between
*To whom correspondence should be addressed. Tel: +1 617 358 2301; Fax: +1 617 353 4814; Email: firstname.lastname@example.org
Mattia Zampieri, ETH Zurich, 8092 Zurich, Switzerland.
Niels Klitgord, Life Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA.
The authors wish it to be known that, in their opinion, the first two authors should be regarded as joint First Authors.
Nucleic Acids Research, 2012, Vol. 40, No. 15Published online 25 May 2012
? The Author(s) 2012. Published by Oxford University Press.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/
by-nc/3.0), which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
controlling core cellular functions, including growth) and
portions of the genetic networks that have specifically
evolved to mediate environmental sensing (6–8). The inter-
play between genetic circuitry and external variables is
particularly important during growth phase transitions.
Microorganisms are able to sense the depletion of envir-
onmental resources and respond with a variety of
condition-specific strategies, such as sporulation (9,10),
increased motility (11) and the formation or dispersal of
biofilms (12,13). DNA microarrays have been widely used
to monitor the time-dependent response of various micro-
organisms to environmental change, showing that their
fast physiological adaptation is typically accompanied by
a massive transcriptional rearrangement, where selected
genes dynamically adjust their level of expression as a
function of the external stimuli (14–17).
Here, we combine experimental and computational
approaches to study the transcriptional response upon
growth phase transitions for the versatile gram-negative
oneidensis is found mainly in nutrient-rich environments,
such as marine sediments, often in association with fer-
mentative communities (18–20). Shewanella species have
the characteristic ability to reduce a broad spectrum of
metals and organic compounds (in addition to oxygen),
making them very relevant for environmental and
bioenergy-related applications, such as bioremediation of
metal-contaminated waters, biofuel production and mi-
crobial fuel cells (18,19,21–24). Several prior studies on
S. oneidensis cultures have characterized transcriptional
and proteomic responses of this microbe under various
media and stress conditions (25–38). These comparative
studies have mostly focused on ‘binary’ changes, i.e. on
comparing transcriptional states before and after a given
perturbation or environmental shift. However, to our
knowledge, there has been no detailed measurement and
systems level analysis of transcript levels at multiple time
points as S. oneidensis transitions between exponential and
stationary phase of growth.
Our results are organized as follows: first, we introduce
our measurements of time-dependent mRNA expression
levels during batch growth of S. oneidensis MR-1 under
two radically differentgrowth
(minimal lactate and rich LB, i.e. Luria and Bertani’s
Lysogenic Broth) and identify global transcriptional
trends. We then implement a new growth derivative
mapping (GDM) approach to compare transcriptional
profiles across different time-course experiments and to
discriminate genes (and processes) most likely controlled
by growth-associated functions. Next, we dive into a
detailed system-level analysis of the genetic response to
growth phase transitions under the lactate-minimal
approach (dynamic detection of transcriptional triggers
or D2T2) to identify the genes through which environmen-
tal stimuli are expected to affect the internal dynamics.
Our analysis highlights the importance of some specific
pathways, whose metabolic relevance is confirmed by
dynamic flux balance analysis (dFBA) calculations. In
circuits(e.g. dedicated to
someaspects of the
transcriptional response to oxygen limitation, detecting
the activation of genes previously shown to be relevant
for anaerobic respiration.
nitrogen limitation is coupled to storage of glycogen.
Both observations are corroborated by measurement of
relevant intracellular and extracellular metabolites, as
well as by complementary analyses of literature informa-
tion and competitive fitness assay data.
MATERIALS AND METHODS
Chemicals and reagents
DL-Lactate (60% solution) and ammonia analysis kit were
procured from Sigma-Aldrich (St Louis, MO, USA). All
other componentsof the
(Supplementary Table S1 in Supplementary text) were of
highest purity grade and were also procured either from
Sigma-Aldrich or Thermo-Fisher Scientific (Pittsburgh,
PA, USA). Qiagen Inc. supplied the RNA protect
reagent, RNAse easy kit for isolation of RNA, cDNA
purification kit and RNAse-free DNAse enzyme. Other
reagents and chemicals used during isolation and purifica-
tion of RNA and during various steps of Shewanella chips
hybridization (Affymetrix Inc.) were purchased from
several different vendors: Superscript II reverse transcript-
ase, DTT, random hexamers and BSA from Invitrogen
Inc.; Gene chip labeling reagent, One-phor-all buffer and
Biochemicals; MES stock, lysozyme, Goat IgG and 200
proof ethanol from Sigma-Aldrich; Terminal transferase,
Herring sperm DNA and dNTPs from Promega; 0.5M
Streptavidin antibody from Vector laboratories; SSPE,
Streptavidin, SAPE, 10% Tween-20, NaOH and HCl
from Thermo-Fisher Scientific; and TE Buffer (pH 8.0),
Superase 1n, 5M NaCl and nuclease-free water were from
Strain, cultivation and sample collections
Shewanella oneidensis MR-1 ATCC 700550 was used in
this work. The strain was revived from –80?C glycerol
stocks by overnight growth in LB medium. One hundred
microliters of an overnight S. oneidensis MR-1 pre-culture
was inoculated in M4 minimal medium containing lactate
(LAC) (Supplementary Table S1) and rich LB medium
separately for the inoculum, which was used for inocula-
tion in 1.3l working volume Bioreactor vessel (Bioflo110,
New Brunswick Scientific Company) for M4-lactate and
LB media runs, respectively. Various growth parameters,
viz., temperature (30?C), pH (7.2), aeration (1l/m) and
agitation were controlled using microprocessor probes.
The pH was maintained at 7.2 using automatic additions
of 2N NaOH and 10% H3PO4using peristaltic pumps
attached to the bioreactor. The dissolved oxygen (dO2)
probe was calibrated before the inoculation and dO2in
the vessel was maintained at 20% air saturation level
using automatic control of O2 cascade throughout the
experiment. We started collecting biomass samples for
RNA isolation after the optical density of culture was at
least above 0.150. In lactate-minimal medium (LAC), the
Nucleic Acids Research, 2012,Vol.40, No. 157133
time interval for most of the biomass samples between
collections was between 1 and 6h, except after 36h
only two samples were collected at 48 and 50h. For LB
medium, the biomass samples were collected every 30min
between 1.5 and 6h of growth (exponential phase);
however, after late exponential and into stationary
phase, the biomass samples were collected at time intervals
of 1–4h until 36h of growth. Three more samples at 48, 50
and 55h near the end of bioreactor run were also col-
lected. We collected two biomass samples during each col-
lection, and both the samples were processed for RNA
extraction to provide statistical significance in subsequent
Extracellular lactate, acetate, pyruvate, ammonium and
intracellular glycogen quantification
High pressure liquid chromatography (HPLC) was used
for quantifying residual lactate and other organic acids
S. oneidensis MR-1 on lactate. The BioRad column
(HPX 87H 300mm?7.8mm) was used on a HPLC
(HP-Agilent 1090 Series II). The quantification was
carried out by injecting 20ml of sample using 0.008N
H2SO4 in mobile phase at 35?C. The flow rate during
quantification was kept at 0.6ml/h and detection was
done at 210nm. The regression equations from individual
standard curve of all three compounds were used to cal-
culate the amount of individual compound from the test
sample. The retention times for lactate, acetate and
pyruvate were 12.5, 14.9 and 8.9min, respectively.
Residual lactate was also measured using enzymatic
analysis kit (R-Biopharm). Residual ammonium in the
cell-free supernatant and intracellular glycogen were
measured using commercially available kits for ammonium
(Sigma-Aldrich, St Louis, MO, USA) and glycogen
(BioAssay Systems, Hayward, CA, USA), respectively.
RNA isolation and S. oneidensis microarrays
At each time point of interest, 2–30ml (depending on
phase of growth and cell concentration) of live bacterial
culture were sampled, immediately mixed with two
volumes of Qiagen RNA protect solution, vortexed and
centrifuged, and RNA was isolated from biomass pellet
using Qiagen’s RNA easy isolation kit according to manu-
facturer’s recommendation. During isolation procedure,
the RNA sample was treated with Qiagen’s DNAase to
get rid of any genomic DNA contamination. RNA yields
were quantified on Nanodrop and UV 260/280 ratios were
calculated to check purity of each RNA sample. The pre-
viously described protocol (23,31) was used for micro-
arrays on S. oneidensis chips from Affymetrix. Briefly,
?10mg of each RNA sample were used for cDNA synthe-
sis, followed by reverse transcription, cDNA purification
and cDNA fragmentation. This was followed by labeling
of cDNA and 16h of hybridization at 45?C on
S. oneidensis arrays. The labeled and hybridized arrays
were subjected to several cycles of washing and staining
using Affymetrix Wash buffers A and B, Goat IgG,
Streptavidin, Anti-streptavidin and SAPE according to
Affymetrix protocol for prokaryotic arrays. This was
followed by scanning of stained arrays on Affymetrix
Gene Chip Scanner Model 3000. The S. oneidensis
MR-1 platform and microarray data has been submitted
to Gene ExpressionOmnibus
Growth derivative mapping
We developed the GDM method to compare gene expres-
sion profiles during growth under two nutrient conditions.
The details of GDM are available in the Supplementary
Text. GDM maps gene expression profiles from a
time-dependent to a growth rate-dependent domain,
through the following steps: first, we fit the observed
OD curves to a logistic growth model of bacterial
growth. Then, by matching time points corresponding to
similar instantaneous growth rates in the two conditions,
we generate a non-linear mapping between the rich
and the minimal medium profiles, which we apply system-
atically to each individual gene. Finally, profiles across
conditions can be compared, with the goal of finding
genes whose time-dependent expression levels display
significant correlation or anti-correlation across the two
The D2T2 approach is aimed at integrating gene expres-
sion time courses with steady-state profiles to identify
genes that are primarily involved in the response to
external perturbations. Complete method details can be
found in the Supplementary Text. Here, we provide a
The main input of the D2T2 algorithm, which ultim-
ately provides information about the trigger genes, is a
set of time course data of gene expression. However, the
algorithm requires a preliminary step, which yields an in-
fluence gene network based on steady-state profiles. In this
first step, the influence network model is trained on an
independent large compendium of gene expression meas-
urements taken under the assumption of steady-state con-
ditions (39), in analogy to previous reverse engineering
approaches (40–43). This initial procedure, which we
recently developed and applied to the study of prion in-
fection (44) is further shown here to provide accurate pre-
dictions on an Escherichia coli multiple perturbation
dataset (see Supplementary text). The fundamental sim-
plifying assumption of this model is that the rate of
change in abundance of gene i (i.e. _ xi) can be expressed
as a linear combination of all other gene expression levels.
Under steady state, this assumption reduces our model to
a system of algebraic equations Ax=0, where A is a
sparse n?n matrix (n is the number of genes) and each
non-zero element (i.e. aij) in A represents the influence of
gene j on gene i. Solution to this first preliminary step is
based on a relevance network approach, where pairs of
genes showing a high linear proportionality across
multiple conditions are first selected in order to make
the influence matrix (i.e. A) sufficiently sparse. These re-
lationships are further weighted through a multiple linear
regression scheme, which evaluates the influence of a gene
7134Nucleic Acids Research, 2012,Vol.40, No. 15
node in driving expression changes of the putative inter-
acting ones. A gene network model is then selected on the
basis of a Bayesian criterion, which identifies the best com-
promise between the model complexity and its predictive
ability (See Supplementary text). It is worth noting that
the solution of the regression problems is a weighted
asymmetric matrix, with all diagonal elements identical
to 1 resulting from rescaling each row of A by an undeter-
mined scaling factor (i.e. the diagonal term aii). We named
this rescaled matrix Ac.
In a second step (the core of the D2T2 procedure), we
go beyond the static influence matrix Accomputed earlier
and explicitly incorporate the measured dynamic changes
in the model. This is achieved by modeling the rate of
change in abundance of each gene as a linear combination
of the ‘influencing-gene’ expression levels plus an add-
itional external input (i.e. u). Here, we approximate the
external function u(t) as a unique step function [u(t)=1],
leading to the following formulation of the problem:
_ x ¼ Ax+B. Here, the matrix B represents the influence
that the perturbation u has in driving changes in the tran-
scriptional rate of each gene. Hence, the condition-specific
evolution in time of the system is described by a large-scale
system of linear differential equations (ODEs) for the
time-dependent gene expression profiles. Typically, the
complexity of this type of model is limited by the
sampling frequency and number of time points. In order
to overcome these limitations, we take advantage of the
initial steady-state model to reduce the fitting of the gene
expression profiles to only two parameters (i.e. for gene xi
we fit ciiand bi, see Supplementary text for full details),
and hence we write Dx=CAcx+B, where ?x is the ap-
proximation of the continuous derivative in time using the
Euler scheme and C is a diagonal matrix (n?n) consisting
of normalizing factors cii.
The identification of these two parameters for each
single transcript aims at describing the portion of dynam-
ical changes caused by external factors (i.e. B) from re-
regulatory mechanisms (Ac). In particular, the genes
showing an incoherent behavior relative to the initial
model (e.g. transient evolution) are chosen as the best
candidates for direct targets of environmental changes.
The underlying network and the significantly perturbed
analysis. Once the distribution of empirical P-values has
been generated, the Q-value is used to correct the index of
significance for multiple testing. Categories associated to a
Q-value in the 1% confidence interval are considered to be
significant (similar results are obtained for the 5% confi-
a rigorous statistical
Predicting growth dynamics using a flux balance model
dFBA was implemented as described previously (45),
using a static optimization method. The detailed dFBA
optimization method is described in Supplementary
performs iterations of flux balance analysis (46,47) with
flux bounds computed based on external metabolite con-
centrations updated at each time step. This approach
a brief summary. dFBA
assumes that biomass and environmental metabolite
changes happen at a slower scale relative to the time ne-
cessary for the achievement of intracellular steady state of
metabolite levels. The S. oneidensis MR-1 genome-scale
metabolic model iSO783 (48) was used to predict growth
of the bacteria in the LAC experimental condition. This
model contains 774 reactions and 634 unique metabolites.
Initial medium composition was modeled by assuming an
initial concentration of nutrients that reflects the known
M4 medium components (Supplementary Table S1) with
supplemented D- and L-lactate. The simulated level of
external O2was maintained throughout the dFBA simu-
lation. To test the effect of O2 on overall growth
dynamics, werun thesimulation
constant concentration values, between 2 and 8mM/h.
In order to simulate the production of glycogen, we im-
plemented a secondary objective function which maxi-
mizes a glycogen sink flux (to what could be thought of
as a glycogen storage compartment) when biomass cannot
be produced. This is done by imposing an objective
function which is a linear combination of biomass produc-
tion and glycogen production, in proportions of 99 to 1.
When this objective is maximized, biomass is always
favored, if the appropriate resources are available.
Otherwise, excess glycogen can be observed to accumulate
over time, simulating effectively the internal sequestration
that happens in vivo.
Global transcriptional changes during growth on minimal
and rich media
Shewanella oneidensis MR-1 was cultured in a batch bio-
reactor using nutritionally rich medium (LB) and minimal
M4 medium containing DL-lactate (LAC) (see ‘Materials
and Methods’ section for details). Time-dependent mRNA
expression profiles were obtained using S. oneidensis
Gene Expression Omnibus platform reference number
GPL8434). We collected biomass samples and measured
expression at 21 time points in LB and 19 time points in
LAC, covering the growth curves from early log phase to a
few hours after reaching stationary phase (Figure 1A).
These data sets provide snapshots of the regulatory
programs employed by the organism throughout the tran-
sitions from abundantly available to depleted resources in
two metabolically very different conditions (Supplem-
entary Data set 1). An enrichment analysis for main func-
tional categories in each cluster across the two growth
conditions is presented in Supplementary Data set 2.
Some initial insights into the specific transcriptional
changes occurring throughout the two experiments can
(Supplementary Figure S1 in Supplementary text). In par-
ticular, in the LAC growth experiment, this analysis
reveals how major events in the adaptive response at the
level of the gene transcription are mainly occurring during
the late-exponential phase.
To illustrate the global trends in gene expression
dynamics throughout our experiments, we visualized the
Nucleic Acids Research, 2012,Vol.40, No. 157135
correlation between genome-wide transcriptional profiles
at any two time points (Figure 1B–D). During growth on
LAC medium, while the first consecutive time points are
highly similar, a sudden drastic change marks the transi-
tion from late exponential phase into stationary phase,
between 26 and 30h (Figure 1B). This sharp transition
suggests a fast, highly coordinated system-level response,
likely triggered by environmental changes as explored in
detail later. In contrast, during growth on the LB medium,
the transcriptional dynamics follows a more gradual and
smoother rearrangement (Figure 1C). This is likely a con-
sequence of the much higher number of degrees of
freedom that the complex regulatory network can span
nitrogen sources as they are gradually used up (49,50).
Despite the clear difference between the LAC and LB
identify weak correlations across the two experiments.
As shown in Figure 1D, early log phase gene states in
the LAC and LB growth conditions slightly correlate
with each other, and so do transcription profiles during
entry in stationary phase. This result can be interpreted as
a suggestion that some components of the growth tran-
scriptional program are environment dependent, while
others are environment independent, and likely driven
by the growth process itself. It is important to note that
absolute times in the two experiments are different, sug-
gesting that the LAC versus LB correlation observed
holds with respect to some rescaled timeline as discussed
in the following section.
Comparing the transcriptional response across conditions
In order to perform a systematic and quantitative com-
parative analysis of the LAC and LB time-dependent
courses despite their different timescales, we devised an
approach (GDM) to rescale the transcriptional profiles
based on an analytical estimation of growth-rate (see
‘Materials and Methods’ section and Supplementary
Text for more details) (51), inspired by previous assess-
ments of the dependence of gene expression on growth
rates (52,53). As opposed to a previous method that
compared gene expression across conditions based on
Figure 1. (A) Growth curves in the LB-rich and LAC-minimal medium. Filled symbols (inverted triangle and square) correspond to the time points
at which mRNA was extracted for microarray hybridization, for LAC and LB medium, respectively. The heatmaps in (B–D) indicate the level of
pair-wise similarity in the transcriptome profiles at different phases of growth (B: LAC versus LAC; C: LB versus LB and D: LB versus LAC). These
heatmaps indicate the correlation between genome-wide transcriptional profiles at any two time points.
7136 Nucleic Acids Research, 2012,Vol.40, No. 15
equal optical densities, GDM is based on an analytical
estimation of growth-rate changes. Using the GDM
approach, we compared rescaled transcriptional profiles
for each gene across the two nutrient conditions,
and identified three classes of genes (Supplementary
Data set 3): a first class (Figure 2A) comprising those
genes that have significant positive correlation between
transcriptional profiles across the two conditions (891
genes with a Q?0.01); a second class (Figure 2B)
including 175 genes that are significantly anti-correlated;
and a third class, (the remaining 3164 genes) displaying no
significant correlation or anti-correlation across the two
Among the genes that are most significantly positively
correlated between the two growth conditions, we found
many genes involved in growth-dependent activities, such
as ATP and aminoacyl-tRNA biosynthesis, amino acid
metabolic processes, ribosomal genes, DNA replication
and aerobic respiration. A notable example of correlated
expression profiles in LAC and LB was found for the
transcriptional regulator (TR) rpoD, the primary sD
factor for exponential growth (Figure 2C). Conversely,
rpoS (the sSfactor and master regulator of the general
stress response) displays non-correlated patterns across
the two growth conditions (Supplementary Data set 3).
It is known that the relative concentrations of sSand
sDplay a crucial role in dictating the growth strategy of
a bacterium, balancing optimal growth and survival to
stress events (54,55). Our data support the possibility
that rpoD transcription is associated with growth-rate
processes irrespective of media composition, while other
regulators (e.g. rpoS) or post-transcriptional control may
modulate environment-specific changes (55,56). As a
further hint to the complex regulation of these two
Figure 2. The approach of GDM is used to compare individual gene time-courses across the two conditions. In particular, one can identify genes
whose expression (on X-axis of all panels) time-courses are correlated (A) or anti-correlated (B) between LAC and LB growth conditions, in a
statistically significant way. The solid red (LAC) and blue (LB) curves in (A and B) are the mean expression value of all genes from each data set.
Dashed lines represent expression values that are one standard deviation above or below the average. The time points R0, R1,..., R7 on the X-axis
correspond to labels for the rescaled and interpolated time obtained through the GDM approach across two growth conditions (see ‘Materials and
Methods’ in Supplementary Text). For the LAC experiment, R0=15hr, R2=20hr,..., R7=50hr; for the LB experiment. R0=1.5hr,
R2=7.6hr,..., R7=55hr. The Q values were obtained as described in the ‘Materials and Methods’ section. The other panels show selected
examples of important correlated genes, rpoD (C) and relA (E), with Q values of 0 and 0.004, respectively; and anti-correlated genes, rsd (D) and csrA
(F) with Q values of 0.0025 and 0.0007, respectively.
Nucleic Acids Research, 2012,Vol.40, No. 15 7137
factors, relA (57), a key enzyme in the biosynthesis of the
global stringent response regulatory molecule (p)ppGpp
(responding to nutrient and energy starvation), shows
positively correlated profiles in the LAC and LB condi-
tions (Figure 2E). Also, rsd, an anti-sigma factor that se-
questers RpoD from the RNA polymerase (58,59) displays
anti-correlated expression patterns across the two condi-
tions (Figure 2D), revealing an overall dichotomous
behavior of regulators. An additional anti-correlated
gene worth highlighting is csrA (Figure 2F), which is
upregulated in the LAC data set and encodes a known
regulator of carbon and glycogen metabolism (20,60). As
evidencedlater, this finding
during LAC growth (see section on Metabolite measure-
ments and dFBA observations).
Thus, using GDM, we showed that it is possible to
identify genes whose expression changes are likely
associated with the growth process, and are only weakly
affected by the drastically different nutritional parameters
in the LAC and LB conditions. Conversely, in the next
section, we ask what genes are influenced the most by en-
vironmental changes, and could be considered the main
entry points for downstream transcriptional regulation.
D2T2: identifying genes that mediate the response to
Is it possible to use the measured expression profiles to
understand environment-dependent genetic interactions
(61,62), and in particular to identify the gene nodes
(triggers) that mediate the response to changing environ-
mental conditions? A lot of work has been devoted to
clustering microarray data and reconstructing regulatory
maps (5,42,43,63–65). Here, we focus on a rather different
time-course data to infer key genes that serve as interface
between the internal network and the external environ-
ment. To overcome issues typically encountered in
time-course analyses (such as small number of time
samples compared to a large number of transcripts), we
propose a new method, which capitalizes on the availabil-
ity of a large compendium of S. oneidensis transcriptional
response data to a variety of environmental perturbations
(M3D) (39) in addition to our time-dependent profiles.
Specifically, our approach, named D2T2 (see Supplem-
entary text for complete details), performs reverse engin-
eering of gene expression time courses through integration
with steady-state profiles.
Briefly, the D2T2 algorithm operates in two main steps
(Figure 3). A gene network model is first trained on a large
compendium of gene expression measurements taken
under the assumption of steady state, using a relevance
network approach (see ‘Materials and Methods’ section
and Supplementary text for full details). These relation-
ships are further weighted through a multiple linear
regression scheme, and refined to reach the best comprom-
ise between the model complexity and its predictive
ability. The time-dependent gene expression profile,
describing the condition-specific dynamic behavior of the
system, is then integrated with the previously determined
static network, leading to the predictions of the key
responsive genes. More specifically, the genes showing
an incoherent behavior relative to the initial model (e.g.
transient response) are chosen as the best candidates for
being the mediators (or triggers) of the overall transcrip-
tional rearrangement. Both the underlying network and
the trigger genes are determined based on statistical infer-
ence, with confidence quantifiable through Q-values
(Supplementary Data set 4).
D2T2 validation: the time-dependent response of
E. coli to antibiotics
S. oneidensis data, we tested its performance on the
time-dependent transcriptional profiles in E. coli under
the effect of a quinolone antibiotic (Norfloxacin) (43),
and we asked whether D2T2 could recover the molecular
targets and the mechanisms of action of this antibiotic,
which had been well characterized before (66–68). In
Table 1, we compare the performance of D2T2 with the
approach of Time Series Network Identification (TSNI)
(43). Full details of the analysis of the results from
applying D2T2 to E. coli can be found in Supplementary
Our ranking of confirmed Norfloxacin targets was con-
sistently and significantly improved relative to the TSNI
results, giving e.g. recA and gyrA among the top hits (see
Table 1 for details). Our method also identified two genes,
tisA and tisB, so far not suspected to be implicated in
quinolone action, but which could explain the interplay
applyingtheD2T2 algorithmtothe new
Figure 3. Our new method for reverse engineering time-series gene ex-
pression data, D2T2, acts through the integration of 245 gene expres-
sion profiles of S. oneidensis MR-1 from the M3D database (http://
m3d.bu.edu) (39). A connectivity map is extrapolated from this
training set and edges are weighted performing two linear regressions.
The time-dependent data from our experiments are integrated with the
steady-state network, to provide an estimate of what genes have ex-
pression changes that cannot be well explained by the internal network
itself, but rather require the presence of external triggers.
7138Nucleic Acids Research, 2012,Vol.40, No. 15
agreement with a recent report linking TisB to proton
motive force and ATP levels and highlighting the role of
this gene in persistency and drug tolerance (70).
Applying D2T2 to S. oneidensis during growth
in LAC medium
In applying the D2T2 approach to our S. oneidensis tran-
scriptional data, we had no discrete perturbation events
(as opposed to the E. coli antibiotics case described
above). However, we envisaged that nutrient availability
might be similarly linked to transcriptional triggers whose
signals propagate through the internal genetic networks.
The D2T2 approach allowed us to focus on what we refer
to as perturbed, or ‘trigger’ genes (290 in the LAC and 430
in LB condition, at Q?0.01, Supplementary Data set 4).
Among these trigger genes, 17 are known or putative TRs
in the LAC and 24 in the LB medium (Table 2). These
regulators can be thought of as the responsive triggers that
dictate the overall response to the environmental changes
during both growth conditions.
In Table 2 and Figure 4A, we showed 17 significantly
perturbed TRs that are identified by D2T2 as the major
environmental triggers during growth in LAC minimal
medium. Based on available literature, we confirmed
that most of these identified TRs are known to have an
important role as environmental sensors in different or-
ganisms (Table 3). Many of these TRs are related to
carbon and nitrogen metabolism or depletion, most
notably lldR (SO3460), a gene recently shown to regulate
the transcription of the L-LDH operon (74), and rpoN, a
well-characterized sigma factor associated with nitrogen
limitation (92). Their perturbed signals seem to propagate
to the downstream nodes: lldR is likely responsible for the
subsequent repression of SO0827 (L-lactate permease of
the LctP family) and lldE (subunit of L-lactate dehydro-
genase) (74,88). It is worth noting that both lldE and
SO0827 show a significant reduction in expression near
the end of the exponential growth phase, while lldR
tends to increase, suggesting that lldR acts as a repressor
of lldE and SO0827 (Figure 5A). Alternatively, these two
genes might be coregulated by other TRs or other
signalling circuits, as corroborated by the fact that both
score very high in the D2T2 algorithm (Supplementary
Data set 4). The potential complexity of the regulatory
circuit determining the fate of lactate utilization is sup-
ported by the fact that in E. coli, lldR acts both as a
repressor and an activator of the lldPRD operon (73).
Other TRs that appear to act as interfaces for environ-
mental sensing include EtrA/Fnr (electron transport regu-
lator) and ArcA (TR for aerobic respiration). In our data,
both TRsare up-regulated
and down-regulated upon entry into late-log phase
(Figure 5B). EtrA, a homolog of FNR in E. coli (18,81),
has recently been shown to be involved in fine tuning the
expression of anaerobic metabolism genes in S. oneidensis
(82). Another report (83) showed that the expression of
either ArcA or EtrA/Fnr was not influenced by growth
conditions. Our D2T2 analysis suggests that both regula-
tors play an important role during aerobic growth,
responding to high oxygen demand during fast growth
(as supported by data in the section on metabolite
profiles and dFBA discussed later), and possibly to the
initial sensing of carbon and nitrogen starvation.
A third example of TRs whose transcriptional changes
are best explained by external factors using our D2T2 al-
gorithm is the pair of nitrogen-related global regulators lrp
(leucine responsive protein) and rpoN (s54). These regula-
tors affect a number of genes relevant to nitrogen utiliza-
tion (ntrB, ntrC, amtB_1, amtB_2, glnA, glnK_1, glnK_2).
In addition, the network of interdependencies obtained by
D2T2 shows that these downstream genes are also highly
connected to several glycogen-related genes (glgA, glgB,
glgC, glgP, glgX, see Figure 5C and Supplementary Figure
S8). Both RpoN and Lrp, as well as the two-component
glutamine regulatory system (NtrBC) are known to be
involved in a complex regulatory network that plays a
central role during growth on limited nitrogen in E. coli
(92, 93). One interesting pattern that emerges from our
data is a tight relationship between genes associated with
the nitrogen starvation response and genes involved in
glycogen metabolism, which are up-regulated between 28
and 34h, with highest peak at ?31h (Figure 5C). This
pattern can be visually observed by analyzing the
synchronized expression profiles of the two categories.
In addition, the network of gene interdependencies
obtained from the D2T2 algorithm places the nitrogen
and glycogen related genes into a single highly intercon-
nected cluster (Supplementary Figure S8). Detection of
the gluconeogenesis genes (pckA, phosphoenolpyruvate
carboxykinase and ppc, phosphoenolpyruvate carboxyl-
ase) by D2T2 further supports the notion that the
upstream flow of carbon from lactate to glycogen via
gluconeogenesis is regulated by the cell through environ-
mental signals (Figure 5D). These observations are
reported for the first time in Shewanella species and are
consistent with prior knowledge of glycogen production as
a storage compound in other bacterial species during
exponential and/or stationary phase growth (94–97).
Glycogen biosynthesis has been reported to be enhanced
during acid stress in S. oneidensis (34). A correlation
Table 1. Escherichia coli test of D2T2
We report here the comparison of D2T2 with an existing computa-
tional method (TSNI) for identifying key mediators of Norfloxacin
antibiotic in E. coli, based on data and analysis reported in (43).
Known key transcriptional mediator (recA) and molecular target
(gyrA) are among the most significant hits recovered by D2T2, outper-
forming TSNI. It is worth noting that D2T2 takes advantage of inde-
pendent data set, while other methods, such as TSNI, rely exclusively
on the time samples. This is a crucial step in the entire procedure
because it reduces the underdetermined nature of reverse engineering
time course gene expression data set, and enables the method to work
on a genome scale level, while for example TSNI can be applied to a
pre-selected subset of genes of the order of hundreds.
Nucleic Acids Research, 2012,Vol.40, No. 157139
between glycogen accumulation and low nitrogen in the
plained to be a result of reduced carbon flux (98). The
use of lipid as a storage compound under nitrogen-limiting
conditions has also been observed (99). Our study suggests
a possible link between glycogen biosynthesis and nitrogen
starvation in S. oneidensis, though the timing of these
effects relative to the carbon starvation response in our
experiment remains to be understood.
While we have been focusing here on genes that mediate
the response to metabolically relevant processes, we
should emphasize that D2T2 also identified other genes
associated with the entry into stationary phase, such
as TRs and target genes associated with flagellar biosyn-
thesis, pili, chemotactic functions and biofilm formation
(see Supplementary Figures S9 and S10; Supplementary
Data set 4).
echinospora, was ex-
Another subtle aspect emerging from the GDM and
D2T2 analyses is that the cells may respond to transcrip-
tional triggers that are internal (i.e. associated to the status
of the cell) rather than external (i.e. associated with envir-
onmental stimuli). Our initial expectation was that the
D2T2 approach would only identify extracellular (i.e. en-
vironmental) entry points into the regulatory network.
Our results suggest, however, that some intracellular
growth-phase sensing may feed back into the system as
well, effectively ‘communicating’ to the cell its growth
status, and triggering a specific environment-independent
transcriptional response. This phenomenon is specifically
suggested by the fact that some genes (e.g. rpoD, which is
regulated by cAMP/CRP in E. coli) appear both as
correlated in the GDM analysis (i.e. it is regulated simi-
larly across the two conditions, and hence is weakly de-
pendent on nutrient availability), and as major potential
Table 2. List of TRs identified by the D2T2 approach
SO number TROntologyPerturbation
Transcriptional repressor of iron–sulfur cluster assembly genes, IscR
DNA polymerase II, PolB
TR, LysR family
TR, TetR family
TR, AraC family
TR, LysR family
TR, LysR family
TR, TetR family
RNA polymerase sigma-54 factor, RpoN
Oxygen-sensitive electron transport regulator A, EtrA
TR, GntR family
Two-component TR for aerobic respiration, ArcA
DNA polymerase III, beta subunit, DnaN
Activator of ProP osmoprotectant transporter, ProQ
RNA polymerase sigma-70 factor, RpoD
RNA polymerase-binding protein, DksA
Leucine-responsive regulatory protein, Lrp
Branched chain amino acid metabolism regulator, LiuR
Two-component TR for aerobic respiration, ArcA
RNA polymerase sigma-24 factor, RpoE
Regulatory protein, RecX
Rra-like regulator of RNAse E
Regulator of ppGpp phosphohydrolase, CgtA
TR, MarR family
Sigma-54-specific TR, Fis family
Pseudogene: TR, TetR family, degenerate
Excisionase/response regulator inhibitor-like protein
Two-component response regulator for KtrAB potassium uptake, KtrE
Protein phosphatase with response regulator receiver modulation
Two-component Sigma-54-specific TR of C4-dicarboxylate transport, DctD
TR, LysR family
Two-component TR for periplasmic stress, CpxR
PHP (Polymerase and Histidinol Phosphatase) domain protein
Regulator for small RNA, YhbJ
DNA polymerase III, beta subunit, DnaN
Transcriptional activator of cys regulon, CysB
DNA-directed RNA polymerase, omega subunit, RpoZ
Activator of ProP osmoprotectant transporter, ProQ
TR for histidine utilization, HutC
TRs for which a significant ‘perturbing stimulus’ (e.g. Q?0.01) have been identified by the D2T2 procedure are listed here, with
the corresponding putative annotation and estimated perturbation intensities (i.e. b values). The full list of significant genes
identified by D2T2 in the LAC and LB conditions can be found in the Supplementary Data set 4 (N/A: name not available).
7140Nucleic Acids Research, 2012,Vol.40, No. 15
trigger from the D2T2 analysis (i.e. its transcriptional
changes cannot be explained simply by the internal regu-
latory network architecture). These genes (a total of 107)
are listed in Supplementary Data set S6. Our results
suggest that post-transcriptional regulation may supple-
ment core regulatory processes providing a feed-forward
control of environment on growth strategies (6). An
intimate and delicate interplay between growth and envir-
onmental control might become a crucial requisite to
balance the internal state of the cell and achieve optimal
conditions for growth or long-term survival.
Understanding resource limitation using dFBA
The high-throughput mRNA profile analyses using GDM
and D2T2 have provided us with insight into genes whose
expression is apparently associated with growth phase
changes, as well as genes whose regulation is likely
influenced by environmental factors, such as changes in
nutrient availability. The microarray analyses presented
earlier indicate that at different times along the growth
curve, changes in the availability of resources (carbon,
nitrogen, oxygen) trigger major transcriptional events. In
order to better understand the interplay between growth-
dependent processes and the effects of resource limitations
on the physiologyofS.
time-dependent version of the widespread genome-scale
stoichiometric approach of FBA. Specifically, we set to
simulate the experimentally measured growth curve and
metabolite depletion using a dFBA approach (45) (see
‘Materials and Methods’ section for details), based on a
recently published stoichiometric reconstruction of the S.
oneidensis metabolic network (48). We initialized dFBA by
using the known initial concentrations of available nutri-
ents, and applied consecutive rounds of flux balance
updates at equally spaced time intervals. As shown in
oneidensis, we useda
Figure 4. Expression changes (A) over time of 17 TRs identified by D2T2 approach (red-low expression, blue-high expression). The size of circle
represents level of expression. The shaded region between 22 and 33hr corresponds to the phase of growth in LAC-minimal medium involving major
changes in expression of 17 TRs identified by D2T2 and shown in (A). HPLC was used for quantifying acetate and pyruvate (B), whereas ammonium
(C) was quantified using commercially available kit (see ‘Materials and Methods’ section). Lactate (D) was quantified using HPLC and a commer-
cially available kit (see ‘Materials and Methods’ section). Optical density (E) of ?1ml culture samples was measured during growth of S. oneidensis
at 600nm using a spectrophotometer, and the external O2feed data was collected using the BioCommand software for Bioflo110 bioreactor from
New Brunswick Scientific Company, Edison, NJ.
Nucleic Acids Research, 2012,Vol.40, No. 157141
Figure 6, the dFBA simulation predicts a number of im-
portant features of the growth curve consistent with the
transcriptional signals described earlier. In particular, the
dFBA predicts that: (i) the major nitrogen source
(ammonium) should be depleted from the minimal
medium approximately 2h before lactate depletion. This
would explain the large number of nitrogen utilization
genes identified in the D2T2 analysis, including the TRs
Lrp, RpoN and ProQ (Figure 5C and Supplementary
Figure S8); (ii) abundant oxygen availability (8mM)
may not be enough for satisfying the demand of
log-phase growing cells. Hence, it is conceivable that
during log phase, the cells sense a reduction in oxygen
availability. This would be compatible with the transcrip-
tional signals identified with D2T2, in particular TRs
EtrA/FNR and ArcA (Figure 5B). Notably, the dFBA
model predicts (Figure 6) that under oxygen limitation,
S. oneidensis will produce acetate at approximately
t=12h (O2=2mM) or t=18h (O2=6mM); and (iii)
if indeed ammonium runs out before lactate, there is po-
tentially a time window during which cells would not be
able to produce biomass, but would potentially be able to
store the available carbon for survival during starvation
periods. This hypothesis may be related to the spike of
Table 3. List of regulatory interactions (input signals and targets) of some selected identified TRs during growth in LAC medium
TR Gene productInput signal (Signal class) Target genes in
S. oneidensis MR-1a
Known target functions
MexR Antibiotic resistance
HypR TR of proline
Oxidative state (stress) mexF, mexE, mexRMulti-drug efflux transporter
Lactate transport and utiliza-
tion (E. coli, S. oneidensis)
FeS cluster assembly, anaer-
obic respiration (E. coli)
Pseudomonas aeruginosa (71),
S. oneidensis (none)
Sinorhizobium meliloti (72),
S. oneidensis (none)
Proline (nutrient source) SO0639, putA
LldRTR of lactate
TR of FeS cluster
Lactate (carbon source,
FeS cluster level (cofactor) erpA, iscR, iscS, iscU,
lldG, lldF, lldEEscherichia coli (73),
S. oneidensis (74)
Escherichia coli (75,76),
S. oneidensis (none)
iscA, hscB, hscA,
fdx, dnrN, nfuA
ybgT, cydB, cydA
regulator for aerobic
Small proteins: HptA,
ArcS (oxygen tension)
Cytochrome oxidase, DMSO
reductase (S. oneidensis)
Escherichia coli (77),
S. oneidensis (78,79)
Oxygen (oxygen tension) ybgT, cydB, cydA,
ccmF, ccmG, otr,
coxB, coxC, ccmH,
SO0269, frdC, frdC,
frdA, frdB, lpdA,
SO0581, tpmT, fbpC,
fbpB, fbpA, napB,
napH, napG, napA,
napD, bfr2, bfr1,
SO1250, cyaC, dldD,
lldP, ompW, mtrF,
mtrE, mtrD, feoA,
feoB, tig, SO2005,
hypF, hyaE, hyaD,
hyaC, hyaB, hyaA,
ydaO, uspE, fnr,
SO2357, ccoS, ccoI,
ccoH, ccoP, ccoQ,
ccoO, ccoN, nrdG,
nrfA, narQ, narP,
hupE, moaE, moaD,
moaC, moaA, etfQ,
SO4512, fdhA, fdhB,
fumarate reductase, nitrate/
nitrite reductase, Fe trans-
port storage, adenylate
cyclase, lactate transport/
oxidase, formate/NiFe hy-
drogenase (S. oneidensis)
Escherichia coli (80), S.
SO1415TROxygen (oxygen tension) Anaerobic respiration
Shewanella oneidensis (86,87)
The identified input signals, documented by specific references for different organisms, include oxygen concentration, osmolarity, oxidative stress,
ppGpp (signal for nutrient and energy limitation), carbon (amino acid, lactate) and nitrogen metabolism (via DksA). Pathway genome database
group feature was used to extract direct targets of regulators in S. oneidensis MR-1 (88–91).
aybgT, cydB, cydA are target genes for both ArcA and EtrA/Fnr in S. oneidensis MR-1.
7142Nucleic Acids Research, 2012,Vol.40, No. 15
Figure 5. Expression profiles for the genes of selected pathways during batch growth in LAC-minimal medium. Genes responsible for uptake and
conversion of lactate into acetate via pyruvate (A); know oxygen sensors, EtrA/Fnr and ArcA (B); nitrogen and glycogen-related genes (C); and two
gluconeogenesis genes, ppc and pckA (D).
Figure 6. dFBA was used to simulate bacterial growth in minimal medium with lactate (LAC) using a genome scale model of S. oneidensis MR-1 for
various constant concentrations of oxygen: (A) 2mM O2, (B) 4mM O2, (C) 6mM O2and (D) 8mM O2. In each instance, a 2% death rate was
implemented. Also, the lactate uptake rate and ATP maintenance cost were constrained to known experimental values. A dual objective function was
also implemented to maximize biomass and glycogen production. Acetate production is inversely related to oxygen abundance. In low oxygen
environments (e.g. 2mM in A), abundant acetate is excreted and later utilized, and in high oxygen environments (8mM in D), acetate production is
severely diminished until it is no longer detected.
Nucleic Acids Research, 2012,Vol.40, No. 157143
glycogen-associated genes (Figure 5C) that we observed in
conjunction with major changes in nitrogen utilization
processes. To verify whether the global stoichiometry of
the system would be compatible with the storage of
glycogen, we modified the dFBA model to include an
outflow (‘storage’) of glycogen as a potential flux in the
objective function. Upon this modification, the dFBA
model predicts that glycogen production is compatible
with other constraints in the time window between
nitrogen and carbon depletion (Figure 6).
Metabolite measurements corroborate transcriptional
and dFBA observations
Both the gene expression analyses (Figures 4A and 5) and
the dFBA calculations (Figure 6) indicate that carbon,
nitrogen and oxygen limitations seem to underlie specific
physiological transitions, to which the cells respond with
transcriptional control of relevant processes. In order to
verify the dFBA-based interpretation of the observed
transcriptional changes, we measured at different time
points the supernatant concentrations of lactate, acetate
and ammonium, and the intracellular concentration of
glycogen (Figure 4B–D and ‘Materials and Methods’
section). Since pyruvate secretion had been reported in
previous growth experiments (100,101), we included
pyruvate measurement in our experiment as well. In
addition, to assess potential oxygen limitation, we also
analyzed the time-dependent rate of external oxygen
feeding monitored throughout the growth experiment
Consistent with the array data and with the onset of
stationary phase (Figure 1B), lactate becomes depleted
from the culture vessel between 30 and 32h (Figure 4D).
It turns out that the concentration of ammonium drops to
zero at ?30h (Figure 4C), placing the cells into a small (up
to 2h) window in which carbon is still available, but cells
are experiencing depletion in nitrogen levels. The nitrogen
depletion is consistent with the observed increase in tran-
script levels of genes involved in nitrogen starvation
response (Figure 5C). As suggested also by the dFBA
model, this window of excess carbon coupled with
nitrogen limitation could explain the observed activation
of glycogen-related genes (Figure 5C), similar to previous
reports in yeast (102,103). The cells, unable to produce
biomass because of the lack of nitrogen, instead turn
their metabolic activity to storing the available carbon.
Indeed, upon measuring intracellular glycogen over time,
we observed an increase from zero up to ?3mg/ml (cor-
responding to ?5.5mg/g dried biomass), peaking at ?30h,
exactly when cells enter the window of ammonium deple-
tion in presence of carbon (Figure 4B–D).
The transcriptional profiles and flux balance predictions
also suggested that oxygen limitation occurs during fast
exponential growth. In addition, the dFBA model pre-
dicted that upon oxygen limitation, acetate would be
secreted as a byproduct. While the level of dO2in our
LAC growth experiment was kept constant (?20%), the
rate of external oxygen feed used to maintain such level of
dO2in the bioreactor vessel underwent significant changes
(Figure 4E). In particular, one can see a steep increase in
O2 demand at different time intervals between mid-
exponential to stationary phase of growth. At ?24h,
one can observe the occurrence of a sharp increase in
the oxygen uptake signal. This corresponds to the time
window (24–27h) where acetate is excreted into the
medium and subsequently consumed (Figure 4B) and is
consistent with the up- and down-regulation of genes that
convert acetaldehyde to acetate (aldA) and acetate to
acetyl-CoA, respectively (Figure 5A and Supplementary
Data set 4 of perturbed genes in LAC growth condition).
Another sudden peak in the demand of oxygen (?27–28h)
corresponds to the onset of nitrogen limitation in the cul-
ture, and corresponds to a major peak in the transcription
of the oxygen sensors ArcA and EtrA/FNR (Figures 4A
and 5B). Note that an alternative explanation for the se-
cretion of acetate into the external environment between
23 and 26h of growth (Figure 4B) is the sudden change in
becoming limiting, at ?22–23h (Figure 4C). The rate of
nitrogen utilization decreases sharply at this time, poten-
tially causing a sudden overflow of carbon flux, that is
relieved into the medium as acetate before the glycogen
biosynthesis mechanism is ready to dissipate it.
In addition to the accumulation and consumption of
acetate, we also detected accumulation and re-consump-
tion of pyruvate (Figure 4B). A possible explanation is
that the rate of lactate uptake and subsequent conversion
to pyruvate outpaces the cells ability to process the
pyruvate, leading to its secretion. Alternatively, the accu-
mulation of this and other metabolites in the supernatant
may be due to the break-down of lactate as a by-product
of extracellular detoxification of peroxides in the environ-
ment by the bacteria. This aspect of the physiology of S.
oneidensis is not caught in a straightforward manner by
the dFBA model. This discrepancy may be attributed to
limitations of the stoichiometric model (e.g. missing or
unusually regulated reactions) or of the flux balance
framework (e.g. the use of an inadequate objective
function), and will require further investigation.
Independent validation of the global role of triggers
through phenotypic data analysis
Complementary evidence that the D2T2 trigger genes are
indeed involved in sensing external perturbations related
to the three main environmental signals relevant to our
experimental setup (lactate, nitrogen and oxygen), can
be obtained from the analysis of recently published pheno-
type data (104) (Supplementary Text). This data set
consists of fitness assays for 3355S. oneidensis non-
essential mutants (including several of the D2T2-identified
genes) in 121 diverse conditions. For each mutant, we used
a two-sample t-test to test whether fitness changes were
preferentially observed upon the perturbations of nutri-
tional parameters associated with the aforementioned
environmental signals (i.e. anaerobic versus aerobic,
lactate versus other carbon source, diverse nitrogen
sources). Several of the genes identified by D2T2 under
LAC growth show a significant (P?0.01) phenotypic sig-
(Supplementary Table S2).
7144Nucleic Acids Research, 2012,Vol.40, No. 15
In addition, in order to test whether our list of trigger
genes was significantly enriched for transcripts showing
phenotypic patterns upon oxygen, lactate or nitrogen
signals, we performed a permutation test, where we itera-
tively randomly selected an equal subset of putative trigger
genes and counted how many times we could find a sig-
nificant phenotypic pattern in each of these conditions.
The results of this analysis indicate that there is a signifi-
cant enrichment in our predicted gene list for transcripts
showing phenotypes under these three environmental
conditions (P=0.068, 0.091 and 0.025, respectively).
In particular, there seems to be a prevalence of genes
responding to nitrogen availability (last row of Sup-
plementary Table S2), pointing again to the fundamental
role of this nutrient depletion in our analysis.
We have presented a new set of high-throughput tran-
scriptional data useful for understanding growth phase
physiology in S. oneidensis MR-1. In order to decode
the information embedded in such data, we devised a
mapping procedure (GDM) that allows us to compare
transcriptional programs across different growth media.
Most importantly, we developed a novel method (D2T2)
for interrogating time-dependent gene expression about
the entry points in the genetic network, and interpreted
some of our results in light of flux balance model predic-
tions and metabolite measurements.
S. oneidensis have focused on mRNA changes upon
several discrete environmental or genetic perturbations,
with no time resolution (25–30,32–35,37,38). Similar
studied before, but without probing transcriptional
profiles (100,101). Our data, which spans two fundamen-
tally different environmental conditions, a minimal lactate
and a rich medium (both aerobic), adds a new dimension
to the existing knowledge of S. oneidensis physiology
(39,88,89,105). Our new data are particularly timely,
given the increasingly relevant role of Shewanella species
for studying microbial sediment communities and in
bioenergy-related research (18,19), such as bioremediation
(106), microbial fuel cells and biohydrogen production
A large portion of the computational methods pre-
sented in this work is dedicated to the analysis of
time-dependent gene expression. In contrast to other
analyses of gene expression data, our focus here is inten-
tionally not on obtaining a putative regulatory network.
Rather, we sought to take advantage of our unique data
sets to propose a novel way of interpreting dynamic gene
expression. Our goal was to infer the genes whose dynam-
ical changes cannot be explained by internal gene–gene
influences, but likely to depend on other factors such as
changes in metabolites in the environment.
By developing a time-rescaling approach (GDM), based
on growth-rate estimates derived from measured optical
density data, we formulated a direct comparison between
time-course profiles for individual genes across the two
in Shewanellahave been
conditions, revealing transcripts mainly responding to
growth-rate variations and weakly dependent on whether
simple carbon (lactate) or a complex mixture of nutri-
tional resources are available in the medium. Our
current analysis can only tell us which genes, upon
rescaling, behave similarly in the two conditions con-
sidered. In principle, the same approach could be
applied to multiple conditions and can help identify
genes and pathways whose regulatory dynamics is truly
independent of environmental conditions and changes
mainly as a function of the growth phase of the cell.
Although the GDM approach allowed us to tease out
transcriptional pattern that are correlated between the two
conditions, a different method was necessary in order to
highlight the specific environmental cues and transcrip-
tional gates allowing cells to respond to changing
nutrient availability. To address this question, we took
advantage of the availability of a unique large compen-
dium of S. oneidensis microarray data for response to
multiple perturbations (39). Specifically, we devised a
new algorithm (D2T2) that integrates our new time-series
data presented in this manuscript with this compendium’s
static measurements to decipher how environmental
signals feed into the regulation of gene expression and
Our metabolite measurements and literature analyses
(Figure 4, Table 3 and Supplementary Table S2) support
the D2T2 predictions. However, a direct testing of the
‘trigger’ role of predicted genes will require complemen-
tary experimental approaches. One possible avenue would
be themeasurement of
S. oneidensis strains under different conditions. For
example the deletion of a TR predicted to mediate the
response to a given external metabolite should lead to
higher phenotypic changes under conditions in which
such metabolite is present than in all other conditions.
Along this line, we performed a statistical analysis of
recently published data on gene deletion phenotypes
under multiple conditions (104), showing overall agree-
ment with our predictions (Supplementary Table S2 and
Supplementary Text). In future, more targeted experi-
ments could further help to validate specific D2T2 predic-
tions. For example the activity of a TR could be
monitored in a high time-resolution manner using GFP
constructs (109), reporting the expression level of its
targets upon pulses of the specific putative sensed signal.
In addition, one could directly evaluate the sensing role of
a TR through protein–metabolite binding assays (110).
Interpreted through the lens of a dFBA simulation, the
key transcriptional changes and entry points detected with
D2T2 point to specific metabolic events that apparently
dictate some of the major decisions taken by S. oneidensis
throughout the growth curve. Our analysis reveals that
both ammonium (nitrogen) and lactate (carbon) limitation
constitute key transcriptional triggers. As suggested by
transcriptional spikes and dFBA predictions, nitrogen
limitation turns out to be a major metabolic determinant
in our analysis, and ammonium is predicted to run out
slightly before lactate does in dFBA simulations. This is
ammonium in the culture supernatant. At the same time,
the responseof mutant
of lactate and
Nucleic Acids Research, 2012,Vol.40, No. 15 7145
the transcriptional response, as well as a modified dFBA
calculation, point to a glycogen-related activity that
seemed tightly coupled with the nitrogen starvation
response. As verified through direct measurement of intra-
cellular glycogen, this pattern indicates that S. oneidensis
integrates information about carbon and nitrogen avail-
ability in a decision-making process that leads to the ac-
cumulation of storage carbon rather than biomass due to
lack of nitrogen. This type of behavior, previously
observed in other organisms, is to our knowledge
reported here for the first time in S. oneidensis, suggesting
how this lake sediment organism is able to survive under
different modes of nutrient availabilities. The dFBA simu-
lations presented in this work allowed us to interpret tran-
scriptional data in light of global resource allocation
constraints inherent to metabolism. In this case, the meta-
bolic objective function, a key parameter of constraint-
based models of metabolism, was assumed to be the
same throughout the growth process. Future research in
constraint-based models could take better advantage of
data from dynamic processes to try and infer time- or
condition-dependent objective functions, such as a shift
from maximal growth to minimal energy turnover, in an
attempt to mimic the global metabolite management deci-
sions of the cell. Moreover, in integrated models, tran-
scriptional data could be used to directly constrain
Finally, the pipeline of experimental and computational
approaches applied and developed for this work could be
extended to other microbes and additional conditions. In
Shewanella, for example, it would be highly interesting to
use similar methods to explore what transcriptional
changes and environmental triggers accompany the
switching between different electron acceptors, or other
physical factors e.g. gradual switching between aerobic
and anaerobic conditions, as well as between different an-
aerobic acceptors including metals/metal oxides.
Supplementary Data are available at NAR Online:
Supplementary Tables 1–2, Supplementary Figures 1–10,
The authors are grateful to members of the Segre ` lab for
helpful feedback and discussions, to Timothy S. Gardner
for providing laboratory facilities during the initial phase
of this project, to Jennifer Reed for sharing early stoichio-
metric models of S. oneidensis and to Norman Lee,
Director, CIC facility at Boston University, for use of
Office of Science (BER), U.S. Department of Energy
08ER64511 to M.H.S.]; National Aeronautics and Space
[NNA08CN84A to D.S.]. Funding for open access
charge: Boston University.
Conflict of interest statement. None declared.
1. Airoldi,E.M., Huttenhower,C., Gresham,D., Lu,C., Caudy,A.A.,
Dunham,M.J., Broach,J.R., Botstein,D. and Troyanskaya,O.G.
(2009) Predicting cellular growth from gene expression signatures.
PLoS Comput. Biol., 5, e1000257.
2. de Lichtenberg,U., Jensen,L.J., Brunak,S. and Bork,P. (2005)
Dynamic complex formation during the yeast cell cycle. Science,
3. Klumpp,S., Zhang,Z. and Hwa,T. (2009) Growth rate-dependent
global effects on gene expression in bacteria. Cell, 139,
4. Goelzer,A., Bekkal Brikci,F., Martin-Verstraete,I., Noirot,P.,
Bessieres,P., Aymerich,S. and Fromion,V. (2008) Reconstruction
and analysis of the genetic and metabolic regulatory networks of
the central metabolism of Bacillus subtilis. BMC Syst. Biol., 2, 20.
5. Lemmens,K., De Bie,T., Dhollander,T., De Keersmaecker,S.C.,
Thijs,I.M., Schoofs,G., De Weerdt,A., De Moor,B.,
Vanderleyden,J., Collado-Vides,J. et al. (2009) DISTILLER: a
data integration framework to reveal condition dependency of
complex regulons in Escherichia coli. Genome Biol., 10, R27.
6. Levy,S. and Barkai,N. (2009) Coordination of gene expression
with growth rate: a feedback or a feed-forward strategy? FEBS
Lett., 583, 3974–3978.
7. Martinez-Antonio,A., Janga,S.C., Salgado,H. and Collado-Vides,J.
(2006) Internal-sensing machinery directs the activity of the
regulatory network in Escherichia coli. Trends Microbiol., 14,
8. Yi,T.M., Huang,Y., Simon,M.I. and Doyle,J. (2000) Robust
perfect adaptation in bacterial chemotaxis through integral
feedback control. Proc. Natl Acad. Sci. USA, 97, 4649–4653.
9. Bischofs,I.B., Hug,J.A., Liu,A.W., Wolf,D.M. and Arkin,A.P.
(2009) Complexity in bacterial cell-cell communication: quorum
signal integration and subpopulation signaling in the Bacillus
subtilis phosphorelay. Proc. Natl Acad. Sci. USA, 106, 6459–6464.
10. Singh,A.H., Wolf,D.M., Wang,P. and Arkin,A.P. (2008)
Modularity of stress response evolution. Proc. Natl Acad. Sci.
USA, 105, 7500–7505.
11. Amsler,C.D., Cho,M. and Matsumura,P. (1993) Multiple factors
underlying the maximum motility of Escherichia coli as cultures
enter post-exponential growth. J. Bacteriol., 175, 6238–6244.
12. Nadell,C.D., Xavier,J.B., Levin,S.A. and Foster,K.R. (2008) The
evolution of quorum sensing in bacterial biofilms. PLoS Biol., 6,
13. Wagner,V.E., Bushnell,D., Passador,L., Brooks,A.I. and
Iglewski,B.H. (2003) Microarray analysis of Pseudomonas
aeruginosa quorum-sensing regulons: effects of growth phase and
environment. J. Bacteriol., 185, 2080–2095.
14. Balazsi,G., Barabasi,A.L. and Oltvai,Z.N. (2005) Topological
units of environmental signal processing in the transcriptional
regulatory network of Escherichia coli. Proc. Natl Acad. Sci.
USA, 102, 7841–7846.
15. Gasch,A.P., Spellman,P.T., Kao,C.M., Carmel-Harel,O.,
Eisen,M.B., Storz,G., Botstein,D. and Brown,P.O. (2000)
Genomic expression programs in the response of yeast cells to
environmental changes. Mol. Biol. Cell., 11, 4241–4257.
16. Tao,H., Bausch,C., Richmond,C., Blattner,F.R. and Conway,T.
(1999) Functional genomics: expression analysis of Escherichia coli
growing on minimal and rich media. J. Bacteriol., 181,
17. Chechik,G., Oh,E., Rando,O., Weissman,J., Regev,A. and
Koller,D. (2008) Activity motifs reveal principles of timing in
transcriptional control of the yeast metabolic network. Nat.
Biotechnol., 26, 1251–1259.
18. Fredrickson,J.K., Romine,M.F., Beliaev,A.S., Auchtung,J.M.,
Driscoll,M.E., Gardner,T.S., Nealson,K.H., Osterman,A.L.,
7146Nucleic Acids Research, 2012,Vol.40, No. 15
Pinchuk,G., Reed,J.L. et al. (2008) Towards environmental
systems biology of Shewanella. Nat. Rev. Microbiol., 6, 592–603.
19. Hau,H.H. and Gralnick,J.A. (2007) Ecology and biotechnology of
the genus Shewanella. Annu. Rev. Microbiol., 61, 237–258.
20. Heidelberg,J.F., Paulsen,I.T., Nelson,K.E., Gaidos,E.J.,
Nelson,W.C., Read,T.D., Eisen,J.A., Seshadri,R., Ward,N.,
Methe,B. et al. (2002) Genome sequence of the dissimilatory
metal ion-reducing bacterium Shewanella oneidensis. Nat.
Biotechnol., 20, 1118–1123.
21. Bretschger,O., Obraztsova,A., Sturm,C.A., Chang,I.S.,
Gorby,Y.A., Reed,S.B., Culley,D.E., Reardon,C.L., Barua,S.,
Romine,M.F. et al. (2007) Current production and metal oxide
reduction by Shewanella oneidensis MR-1 wild type and mutants.
Appl. Environ. Microbiol., 73, 7003–7012.
22. Meshulam-Simon,G., Behrens,S., Choo,A.D. and Spormann,A.M.
(2007) Hydrogen metabolism in Shewanella oneidensis MR-1.
Appl. Environ. Microbiol., 73, 1153–1165.
23. Rosenbaum,M.A., Bar,H.Y., Beg,Q.K., Segre,D., Booth,J.,
Cotta,M.A. and Angenent,L.T. (2011) Shewanella oneidensis in a
lactate-fed pure-culture and a glucose-fed co-culture with
Lactococcus lactis with an electrode as electron acceptor.
Bioresour. Technol., 102, 2623–2628.
24. Sani,R.K., Peyton,B.M. and Dohnalkova,A. (2008) Comparison
of uranium(VI) removal by Shewanella oneidensis MR-1 in flow
and batch reactors. Water. Res., 42, 2993–3002.
25. Beliaev,A.S., Klingeman,D.M., Klappenbach,J.A., Wu,L.,
Romine,M.F., Tiedje,J.A., Nealson,K.H., Fredrickson,J.K. and
Zhou,J. (2005) Global transcriptome analysis of Shewanella
oneidensis MR-1 exposed to different terminal electron acceptors.
J. Bacteriol., 187, 7138–7145.
26. Beliaev,A.S., Thompson,D.K., Khare,T., Lim,H., Brandt,C.C.,
Li,G., Murray,A.E., Heidelberg,J.F., Giometti,C.S., Yates,J. III
et al. (2002) Gene and protein expression profiles of Shewanella
oneidensis during anaerobic growth with different electron
acceptors. OMICS, 6, 39–60.
27. Bencheikh-Latmani,R., Williams,S.M., Haucke,L., Criddle,C.S.,
Wu,L.Y., Zhou,J.Z. and Tebo,B.M. (2005) Global transcriptional
profiling of Shewanella oneidensis MR-1 during Cr(VI) and U(VI)
reduction. Appl. Environ. Microbiol., 71, 7453–7460.
28. Brown,S.D., Martin,M., Deshpande,S., Seal,S., Huang,K., Alm,E.,
Yang,Y.F., Wu,L.Y., Yan,T.F., Liu,X.D. et al. (2006) Cellular
response of Shewanella oneidensis to strontium stress. Appl.
Environ. Microbiol., 72, 890–900.
29. Brown,S.D., Thompson,M.R., VerBerkmoes,N.C., Chourey,K.,
Shah,M., Zhou,J.Z., Hettich,R.L. and Thompson,D.K. (2006)
Molecular dynamics of the Shewanella oneidensis response to
chromate stress. Mol. Cell. Proteomics, 5, 1054–1071.
30. Chourey,K., Thompson,M.R., Morrell-Falvey,J.,
VerBerkmoes,N.C., Brown,S.D., Shah,M., Zhou,J.Z., Doktycz,M.,
Hettich,R.L. and Thompson,D.K. (2006) Global molecular and
morphological effects of 24-hour chromium(VI) exposure on
Shewanella oneidensis MR-1. Appl. Environ. Microbiol., 72,
31. Driscoll,M.E., Romine,M.F., Juhn,F.S., Serres,M.H.,
McCue,L.A., Beliaev,A.S., Fredrickson,J.K. and Gardner,T.S.
(2007) Identification of diverse carbon utilization pathways in
Shewanella oneidensis MR-1 via expression profiling. Genome
Inform., 18, 287–298.
32. Gao,H., Wang,Y., Liu,X., Yan,T., Wu,L., Alm,E., Arkin,A.,
Thompson,D.K. and Zhou,J. (2004) Global transcriptome analysis
of the heat shock response of Shewanella oneidensis. J. Bacteriol.,
33. Gao,H.C., Yang,Z.M.K., Wu,L.Y., Thompson,D.K. and
Zhou,J.Z. (2006) Global transcriptome analysis of the cold
shock response of Shewanella oneidensis MR-1 and mutational
analysis of its classical cold shock proteins. J. Bacteriol., 188,
34. Leaphart,A.B., Thompson,D.K., Huang,K., Alm,E., Wan,X.F.,
Arkin,A., Brown,S.D., Wu,L.Y., Yan,T.F., Liu,X.D. et al. (2006)
Transcriptome profiling of Shewanella oneidensis gene expression
following exposure to acidic and alkaline pH. J. Bacteriol., 188,
35. Qiu,X.Y., Daly,M.J., Vasilenko,A., Omelchenko,M.V.,
Gaidamakova,E.K., Wu,L.Y., Zhou,J.Z., Sundin,G.W. and
Tiedje,J.M. (2006) Transcriptome analysis applied to survival of
Shewanella oneidensis MR-1 exposed to ionizing radiation. J.
Bacteriol., 188, 1199–1204.
36. Tang,Y.J.J., Hwang,J.S., Wemmer,D.E. and Keasling,J.D. (2007)
Shewanella oneidensis MR-1 fluxome under various oxygen
conditions. Appl. Environ. Microbiol., 73, 718–729.
37. Thompson,M.R., VerBerkmoes,N.C., Chourey,K., Shah,M.,
Thompson,D.K. and Hettich,R.L. (2007) Dosage-dependent
proteome response of Shewanella oneidensis MR-1 to acute
chromate challenge. J. Proteome Res., 6, 1745–1757.
38. Yang,Y., Harris,D.P., Luo,F., Wu,L., Parsons,A.B.,
Palumbo,A.V. and Zhou,J. (2008) Characterization of the
Shewanella oneidensis Fur gene: roles in iron and acid tolerance
response. BMC Genomics, 9(Suppl 1), S11.
39. Faith,J.J., Driscoll,M.E., Fusaro,V.A., Cosgrove,E.J., Hayete,B.,
Juhn,F.S., Schneider,S.J. and Gardner,T.S. (2008) Many microbe
microarrays database: uniformly normalized affymetrix compendia
with structured experimental metadata. Nucleic Acids Res., 36,
40. di Bernardo,D., Thompson,M.J., Gardner,T.S., Chobot,S.E.,
Eastwood,E.L., Wojtovich,A.P., Elliott,S.J., Schaus,S.E. and
Collins,J.J. (2005) Chemogenomic profiling on a genome-wide
scale using reverse-engineered gene networks. Nat. Biotechnol., 23,
41. Faith,J.J., Hayete,B., Thaden,J.T., Mogno,I., Wierzbowski,J.,
Cottarel,G., Kasif,S., Collins,J.J. and Gardner,T.S. (2007)
Large-scale mapping and validation of Escherichia coli
transcriptional regulation from a compendium of expression
profiles. PLoS Biol., 5, e8.
42. Gardner,T.S., di Bernardo,D., Lorenz,D. and Collins,J.J. (2003)
Inferring genetic networks and identifying compound mode of
action via expression profiling. Science, 301, 102–105.
43. Bansal,M., Della Gatta,G. and di Bernardo,D. (2006) Inference
of gene regulatory networks and compound mode of action
from time course gene expression profiles. Bioinformatics, 22,
44. Zampieri,M., Legname,G., Segre,D. and Altafini,C. (2011) A
system-level approach for deciphering the transcriptional response
to prion infection. Bioinformatics, 27, 3407–3414.
45. Mahadevan,R., Edwards,J.S. and Doyle,F.J. III (2002) Dynamic
flux balance analysis of diauxic growth in Escherichia coli.
Biophys. J., 83, 1331–1340.
46. Kauffman,K.J., Prakash,P. and Edwards,J.S. (2003) Advances in
flux balance analysis. Curr. Opin. Biotechnol., 14, 491–496.
47. Price,N.D., Reed,J.L. and Palsson,B.O. (2004) Genome-scale
models of microbial cells: evaluating the consequences of
constraints. Nat. Rev. Microbiol., 2, 886–897.
48. Pinchuk,G.E., Hill,E.A., Geydebrekht,O.V., De Ingeniis,J.,
Zhang,X., Osterman,A., Scott,J.H., Reed,S.B., Romine,M.F.,
Konopka,A.E. et al. (2010) Constraint-based model of
Shewanella oneidensis MR-1 metabolism: a tool for data
analysis and hypothesis generation. PLoS Comput. Biol., 6,
49. Baev,M.V., Baev,D., Radek,A.J. and Campbell,J.W. (2006)
Growth of Escherichia coli MG1655 on LB medium: determining
metabolic strategy with transcriptional microarrays. Appl.
Microbiol. Biotechnol., 71, 323–328.
50. Beg,Q.K., Vazquez,A., Ernst,J., de Menezes,M.A., Bar-Joseph,Z.,
Barabasi,A.L. and Oltvai,Z.N. (2007) Intracellular crowding
defines the mode and sequence of substrate uptake by Escherichia
coli and constrains its metabolic activity. Proc. Natl Acad. Sci.
USA, 104, 12663–12668.
51. Aach,J. and Church,G.M. (2001) Aligning gene expression time
series with time warping algorithms. Bioinformatics, 17, 495–508.
52. Bremer,H., Dennis,P. and Ehrenberg,M. (2003) Free RNA
polymerase and modeling global transcription in Escherichia coli.
Biochimie, 85, 597–609.
53. Liang,S.T., Xu,Y.C., Dennis,P. and Bremer,H. (2000) mRNA
composition and control of bacterial gene expression. J Bacteriol,
54. Dong,T. and Schellhorn,H.E. (2009) Control of RpoS in global
gene expression of Escherichia coli in minimal media. Mol. Genet.
Genomics., 281, 19–33.
Nucleic Acids Research, 2012,Vol.40, No. 157147
55. Ferenci,T. (2003) What is driving the acquisition of mutS and
rpoS polymorphisms in Escherichia coli? Trends Microbiol., 11,
56. Battesti,A., Majdalani,N. and Gottesman,S. (2011) The
RpoS-mediated general stress response in Escherichia coli (*).
Annu. Rev. Microbiol., 65, 189–213.
57. Potrykus,K. and Cashel,M. (2008) (p)ppGpp: still magical? Annu.
Rev. Microbiol., 62, 35–51.
58. Costanzo,A., Nicoloff,H., Barchinger,S.E., Banta,A.B.,
Gourse,R.L. and Ades,S.E. (2008) ppGpp and DksA likely
regulate the activity of the extracytoplasmic stress factor sigmaE
in Escherichia coli by both direct and indirect mechanisms. Mol.
Microbiol., 67, 619–632.
59. Piper,S.E., Mitchell,J.E., Lee,D.J. and Busby,S.J. (2009) A global
view of Escherichia coli Rsd protein and its interactions. Mol.
Biosyst., 5, 1943–1947.
60. Baker,C.S., Morozov,I., Suzuki,K., Romeo,T. and Babitzke,P.
(2002) CsrA regulates glycogen biosynthesis by preventing
translation of glgC in Escherichia coli. Mol. Microbiol., 44,
61. Ernst,J., Vainas,O., Harbison,C.T., Simon,I. and Bar-Joseph,Z.
(2007) Reconstructing dynamic regulatory maps. Mol. Syst. Biol.,
62. Smith,E.N. and Kruglyak,L. (2008) Gene-environment interaction
in yeast gene expression. PLoS Biol., 6, e83.
63. Bar-Joseph,Z. (2004) Analyzing time series gene expression data.
Bioinformatics, 20, 2493–2503.
64. Ernst,J., Beg,Q.K., Kay,K.A., Balazsi,G., Oltvai,Z.N. and Bar-
Joseph,Z. (2008) A semi-supervised method for predicting
transcription factor-gene interactions in Escherichia coli. PLoS
Comput. Biol., 4, e1000044.
65. Tegner,J., Yeung,M.K., Hasty,J. and Collins,J.J. (2003) Reverse
engineering gene networks: integrating genetic perturbations with
dynamical modeling. Proc. Natl Acad. Sci. USA, 100, 5944–5949.
66. Dwyer,D.J., Kohanski,M.A. and Collins,J.J. (2009) Role of
reactive oxygen species in antibiotic action and resistance. Curr.
Opin. Microbiol., 12, 482–489.
67. Dwyer,D.J., Kohanski,M.A., Hayete,B. and Collins,J.J. (2007)
Gyrase inhibitors induce an oxidative damage cellular death
pathway in Escherichia coli. Mol. Syst. Biol., 3, 91.
68. Kohanski,M.A., Dwyer,D.J., Hayete,B., Lawrence,C.A. and
Collins,J.J. (2007) A common mechanism of cellular death
induced by bactericidal antibiotics. Cell, 130, 797–810.
69. Kohanski,M.A., Dwyer,D.J., Wierzbowski,J., Cottarel,G. and
Collins,J.J. (2008) Mistranslation of membrane proteins and
two-component system activation trigger antibiotic-mediated cell
death. Cell, 135, 679–690.
70. Dorr,T., Vulic,M. and Lewis,K. (2010) Ciprofloxacin causes
persister formation by inducing the TisB toxin in Escherichia coli.
PLoS Biol., 8, e1000317.
71. Chen,H., Yi,C., Zhang,J., Zhang,W., Ge,Z., Yang,C.G. and
He,C. (2010) Structural insight into the oxidation-sensing
mechanism of the antibiotic resistance of regulator MexR. EMBO
Rep., 11, 685–690.
72. Maclean,A.M., White,C.E., Fowler,J.E. and Finan,T.M. (2009)
Identification of a hydroxyproline transport system in the legume
endosymbiont Sinorhizobium meliloti. Mol. Plant. Microbe.
Interact., 22, 1116–1127.
73. Aguilera,L., Campos,E., Gimenez,R., Badia,J., Aguilar,J. and
Baldoma,L. (2008) Dual role of LldR in regulation of the lldPRD
operon, involved in L-lactate metabolism in Escherichia coli.
J. Bacteriol., 190, 2997–3005.
74. Pinchuk,G.E., Rodionov,D.A., Yang,C., Li,X., Osterman,A.L.,
Dervyn,E., Geydebrekht,O.V., Reed,S.B., Romine,M.F.,
Collart,F.R. et al. (2009) Genomic reconstruction of Shewanella
oneidensis MR-1 metabolism reveals a previously uncharacterized
machinery for lactate utilization. Proc. Natl Acad. Sci. USA, 106,
75. Schwartz,C.J., Giel,J.L., Patschkowski,T., Luther,C., Ruzicka,F.J.,
Beinert,H. and Kiley,P.J. (2001) IscR, an Fe-S cluster-containing
transcription factor, represses expression of Escherichia coli genes
encoding Fe-S cluster assembly proteins. Proc. Natl Acad. Sci.
USA, 98, 14895–14900.
76. Yeo,W.S., Lee,J.H., Lee,K.C. and Roe,J.H. (2006) IscR acts as
an activator in response to oxidative stress for the suf operon
encoding Fe-S assembly proteins. Mol. Microbiol., 61, 206–218.
77. Compan,I. and Touati,D. (1994) Anaerobic activation of arcA
transcription in Escherichia coli: roles of Fnr and ArcA. Mol.
Microbiol., 11, 955–964.
78. Lassak,J., Henche,A.L., Binnenkade,L. and Thormann,K.M.
(2010) ArcS, the cognate sensor kinase in an atypical Arc system
of Shewanella oneidensis MR-1. Appl. Environ. Microbiol., 76,
79. Gralnick,J.A., Brown,C.T. and Newman,D.K. (2005) Anaerobic
regulation by an atypical Arc system in Shewanella oneidensis.
Mol. Microbiol., 56, 1347–1357.
80. Kiley,P.J. and Beinert,H. (1998) Oxygen sensing by the global
regulator, FNR: the role of the iron-sulfur cluster. FEMS
Microbiol. Rev., 22, 341–352.
81. Saffarini,D.A. and Nealson,K.H. (1993) Sequence and genetic
characterization of etrA, an fnr analog that regulates anaerobic
respiration in Shewanella putrefaciens MR-1. J. Bacteriol., 175,
82. Cruz-Garcia,C., Murray,A.E., Rodrigues,J.L., Gralnick,J.A.,
McCue,L.A., Romine,M.F., Loffler,F.E. and Tiedje,J.M. (2011)
Fnr (EtrA) acts as a fine-tuning regulator of anaerobic
metabolism in Shewanella oneidensis MR-1. BMC Microbiol., 11,
83. Gao,H., Wang,X., Yang,Z.K., Chen,J., Liang,Y., Chen,H.,
Palzkill,T. and Zhou,J. (2010) Physiological roles of ArcA, Crp,
and EtrA and their interactive control on aerobic and anaerobic
respiration in Shewanella oneidensis. PLoS One, 5, e15295.
84. Beliaev,A.S., Thompson,D.K., Fields,M.W., Wu,L.Y., Lies,D.P.,
Nealson,K.H. and Zhou,J.Z. (2002) Microarray transcription
profiling of a Shewanella oneidensis etrA mutant. J. Bacteriol.,
85. Maier,T.M. and Myers,C.R. (2001) Isolation and characterization
of a Shewanella putrefaciens MR-1 electron transport regulator
etrA mutant: reassessment of the role of EtrA. J. Bacteriol., 183,
86. Yang,Y., Harris,D.P., Luo,F., Xiong,W., Joachimiak,M., Wu,L.,
Dehal,P., Jacobsen,J., Yang,Z., Palumbo,A.V. et al. (2009)
Snapshot of iron response in Shewanella oneidensis by gene
network reconstruction. BMC Genomics, 10, 131.
87. Charania,M.A., Brockman,K.L., Zhang,Y., Banerjee,A.,
Pinchuk,G.E., Fredrickson,J.K., Beliaev,A.S. and Saffarini,D.A.
(2009) Involvement of a membrane-bound class III adenylate
cyclase in regulation of anaerobic respiration in Shewanella
oneidensis MR-1. J. Bacteriol., 191, 4298–4306.
88. Novichkov,P.S., Laikova,O.N., Novichkova,E.S., Gelfand,M.S.,
Arkin,A.P., Dubchak,I. and Rodionov,D.A. (2010) RegPrecise: a
database of curated genomic inferences of transcriptional
regulatory interactions in prokaryotes. Nucleic Acids Res., 38,
89. Karpinets,T.V., Romine,M.F., Schmoyer,D.D., Kora,G.H.,
Syed,M.H., Leuze,M.R., Serres,M.H., Park,B.H., Samatova,N.F.
and Uberbacher,E.C. (2010) Shewanella knowledgebase:
integration of the experimental data and computational
predictions suggests a biological role for transcription of
intergenic regions. Database, 2010, baq012.
90. Novichkov,P.S., Rodionov,D.A., Stavrovskaya,E.D.,
Novichkova,E.S., Kazakov,A.E., Gelfand,M.S., Arkin,A.P.,
Mironov,A.A. and Dubchak,I. (2010) RegPredict: an integrated
system for regulon inference in prokaryotes by comparative
genomics approach. Nucleic Acids Res., 38, W299–W307.
91. Keseler,I.M., Collado-Vides,J., Santos-Zavaleta,A., Peralta-Gil,M.,
Gama-Castro,S., Muniz-Rascado,L., Bonavides-Martinez,C.,
Paley,S., Krummenacker,M., Altman,T. et al. (2011) EcoCyc: a
comprehensive database of Escherichia coli biology. Nucleic Acids
Res., 39, D583–D590.
92. Tintut,Y., Wang,J.T. and Gralla,J.D. (1995) A novel bacterial
transcription cycle involving sigma 54. Genes Dev., 9, 2305–2313.
93. Reitzer,L. (2003) Nitrogen assimilation and global regulation in
Escherichia coli. Annu. Rev. Microbiol., 57, 155–176.
94. Belanger,A.E. and Hatfull,G.F. (1999) Exponential-phase glycogen
recycling is essential for growth of Mycobacterium smegmatis.
J. Bacteriol., 181, 6670–6678.
7148Nucleic Acids Research, 2012,Vol.40, No. 15
95. Leckie,M.P., Ng,R.H., Porter,S.E., Compton,D.R. and Download full-text
Dietzler,D.N. (1983) Regulation of bacterial glycogen synthesis.
Stimulation of glycogen synthesis by endogenous and exogenous
cyclic adenosine 3’:5’-monophosphate in Escherichia coli and the
requirement for a functional CRP gene. J. Biol. Chem., 258,
96. Preiss,J., Yung,S.G. and Baecker,P.A. (1983) Regulation of
bacterial glycogen synthesis. Mol. Cell Biochem., 57, 61–80.
97. Seibold,G.M. and Eikmanns,B.J. (2007) The glgX gene product
of Corynebacterium glutamicum is required for glycogen
degradation and for fast adaptation to hyperosmotic stress.
Microbiology, 153, 2212–2220.
98. Hoskisson,P.A., England,R., Sharples,G.P. and Hobbs,G. (2004)
Modulation of glycogen and trehalose levels in Micromonospora
echinospora (ATCC 15837). Antonie Van Leeuwenhoek, 86,
99. Hernandez,M.A., Mohn,W.W., Martinez,E., Rost,E.,
Alvarez,A.F. and Alvarez,H.M. (2008) Biosynthesis of storage
compounds by Rhodococcus jostii RHA1 and global
identification of genes involved in their metabolism. BMC
Genomics, 9, 600.
100. Elias,D.A., Tollaksen,S.L., Kennedy,D.W., Mottaz,H.M.,
Giometti,C.S., McLean,J.S., Hill,E.A., Pinchuk,G.E.,
Lipton,M.S., Fredrickson,J.K. et al. (2008) The influence of
cultivation methods on Shewanella oneidensis physiology and
proteome expression. Arch. Microbiol., 189, 313–324.
101. Tang,Y.J.J., Meadows,A.L. and Keasling,J.D. (2007) A kinetic
model describing Shewanella oneidensis MR-1 growth, substrate
consumption, and product secretion. Biotechnol. Bioeng., 96,
102. Hazelwood,L.A., Walsh,M.C., Luttik,M.A., Daran-Lapujade,P.,
Pronk,J.T. and Daran,J.M. (2009) Identity of the growth-limiting
nutrient strongly affects storage carbohydrate accumulation in
anaerobic chemostat cultures of Saccharomyces cerevisiae. Appl.
Environ. Microbiol., 75, 6876–6885.
103. Schulze,U., Liden,G., Nielsen,J. and Villadsen,J. (1996)
Physiological effects of nitrogen starvation in an anaerobic batch
culture of Saccharomyces cerevisiae. Microbiology, 142(Pt 8),
104. Deutschbauer,A., Price,M.N., Wetmore,K.M., Shao,W.,
Baumohl,J.K., Xu,Z., Nguyen,M., Tamse,R., Davis,R.W. and
Arkin,A.P. (2011) Evidence-based annotation of gene
function in Shewanella oneidensis MR-1 using
genome-wide fitness profiling across 121 conditions. PLoS
Genet., 7, e1002385.
105. Syed,M.H., Karpinets,T.V., Leuze,M.R., Kora,G.H.,
Romine,M.R. and Uberbacher,E.C. (2009) ShewregDB: database
and visualization environment for experimental and predicted
regulatory information in Shewanella oneidensis MR-1.
Bioinformation, 4, 169–172.
106. Gadd,G.M. (2010) Metals, minerals and microbes: geomicrobiology
and bioremediation. Microbiol.-SGM, 156, 609–643.
107. Rosenbaum,M., Cotta,M.A. and Angenent,L.T. (2010) Aerated
Shewanella oneidensis in continuously fed bioelectrochemical
systems for power and hydrogen production. Biotechnol. Bioeng.,
108. Rosenbaum,M., He,Z. and Angenent,L.T. (2010) Light energy to
bioelectricity: photosynthetic microbial fuel cells. Curr. Opin.
Biotechnol., 21, 259–264.
109. Zaslaver,A., Mayo,A.E., Rosenberg,R., Bashkin,P., Sberro,H.,
Tsalyuk,M., Surette,M.G. and Alon,U. (2004) Just-in-time
transcription program in metabolic pathways. Nat. Genet., 36,
110. Li,G., Ruan,X., Auerbach,R.K., Sandhu,K.S., Zheng,M.,
Wang,P., Poh,H.M., Goh,Y., Lim,J., Zhang,J. et al. (2012)
Extensive promoter-centered chromatin interactions provide a
topological basis for transcription regulation. Cell, 148, 84–98.
111. Everitt,B. and Dunn,G. (2001) Applied Multivariate Data
Analysis, 2nd edn. Arnold; Oxford University Press, London,
112. Zheng,Q. and Wang,X.J. (2008) GOEAST: a web-based software
toolkit for Gene Ontology enrichment analysis. Nucleic Acids
Res., 36, W358–W363.
113. Storey,J.D. (2002) A direct approach to false discovery rates.
J. R. Stat. Soc.: Series B (Statistical Methodol.), 64, 479–498.
114. Bansal,M. and di Bernardo,D. (2007) Inference of gene networks
from temporal gene expression profiles. IET Syst. Biol., 1,
115. Cosgrove,E.J., Zhou,Y., Gardner,T.S. and Kolaczyk,E.D. (2008)
Predicting gene targets of perturbations via network-based
filtering of mRNA expression compendia. Bioinformatics, 24,
116. Balaji,S., Babu,M.M., Iyer,L.M., Luscombe,N.M. and
Aravind,L. (2006) Comprehensive analysis of combinatorial
regulation using the transcriptional regulatory network of yeast.
J. Mol. Biol., 360, 213–227.
117. Chang,D.E., Smalley,D.J. and Conway,T. (2002) Gene
expression profiling of Escherichia coli growth transitions: an
expanded stringent response model. Mol. Microbiol., 45,
118. Thormann,K.M., Duttler,S., Saville,R.M., Hyodo,M., Shukla,S.,
Hayakawa,Y. and Spormann,A.M. (2006) Control of formation
and cellular detachment from Shewanella oneidensis MR-1
biofilms by cyclic di-GMP. J. Bacteriol., 188, 2681–2691.
119. Rakshe,S., Leff,M. and Spormann,A.M. (2011) Indirect
modulation of the intracellular c-Di-GMP level in Shewanella
oneidensis MR-1 by MxdA. Appl. Environ. Microbiol., 77,
120. Saville,R.M., Dieckmann,N. and Spormann,A.M. (2010)
Spatiotemporal activity of the mshA gene system in Shewanella
oneidensis MR-1 biofilms. FEMS Microbiol. Lett., 308, 76–83.
Nucleic Acids Research, 2012,Vol.40, No. 15 7149