of 18
*For correspondence:
melowitz@caltech.edu
Competing interests:
The
authors declare that no
competing interests exist.
Funding:
See page 15
Received:
25 October 2017
Accepted:
21 May 2018
Published:
29 May 2018
Reviewing editor:
Martin
Ackermann, ETH Zurich,
Switzerland
Copyright Rosenthal et al. This
article is distributed under the
terms of the
Creative Commons
Attribution License,
which
permits unrestricted use and
redistribution provided that the
original author and source are
credited.
Metabolic interactions between dynamic
bacterial subpopulations
Adam Z Rosenthal
1,2
, Yutao Qi
1,2
, Sahand Hormoz
1,2
, Jin Park
1,2
,
Sophia Hsin-Jung Li
3
, Michael B Elowitz
1,2,4
*
1
Division of Biology and Biological Engineering, California Institute of Technology,
Pasadena, United States;
2
Department of Applied Physics, California Institute of
Technology, Pasadena, United States;
3
Department of Molecular Biology, Princeton
University, Princeton, United States;
4
Howard Hughes Medical Institute, Pasadena,
United States
Abstract
Individual microbial species are known to occupy distinct metabolic niches within multi-
species communities. However, it has remained largely unclear whether metabolic specialization
can similarly occur within a clonal bacterial population. More specifically, it is not clear what
functions such specialization could provide and how specialization could be coordinated
dynamically. Here, we show that exponentially growing
Bacillus subtilis
cultures divide into distinct
interacting metabolic subpopulations, including one population that produces acetate, and another
population that differentially expresses metabolic genes for the production of acetoin, a pH-neutral
storage molecule. These subpopulations exhibit distinct growth rates and dynamic interconversion
between states. Furthermore, acetate concentration influences the relative sizes of the different
subpopulations. These results show that clonal populations can use metabolic specialization to
control the environment through a process of dynamic, environmentally-sensitive state-switching.
DOI: https://doi.org/10.7554/eLife.33099.001
Introduction
Co-utilization of carbon sources was described alongside diauxie by Jacques Monod in his PhD the-
sis (
Monod, 1958
), and is common in many organisms (
Peyraud et al., 2012
). In the Gram-positive
bacterium
Bacillus subtilis,
two preferred carbon sources are co-utilized: glucose and malate
(
Kleijn et al., 2010
). When both of these carbon sources are available they are consumed simulta-
neously, generating growth rates that surpass those achieved with either substrate alone
(
Kleijn et al., 2010
). Under conditions of rapid growth, co-consumption of glucose and malate leads
to the accumulation of high levels of acetate (Kleijn et al., 2010). As a weak organic acid, acetate can
be harmful to cells even in buffered medium (
Rosenthal et al., 2008
). Acetate and related short-
chain fatty acids enter the cell passively in the neutral form and then dissociate intracellularly, releas-
ing a proton and transiently acidifying the cytoplasm (
Russell and Diez-Gonzalez, 1997
;
Roe et al.,
1998
). The intracellular dissociation of acetate also disrupts the cellular anion balance, with negative
effects on metabolism (
Roe et al., 1998
;
Roe et al., 2002
) and transcription (
Rosenthal et al.,
2008
). When extracellular acetate levels rise to toxic levels the growing
Bacillus subtilis
culture con-
sumes the acetate and produces acetoin, a non-toxic pH-neutral overflow metabolite that can be
used as a carbon source in later growth stages (
Speck and Freese, 1973
) (
Figure 1A
).
A biphasic growth strategy, in which acetate is produced to a toxic level and then reabsorbed
and replaced by a non-toxic metabolite (
Wolfe, 2005
), is common to many bacterial species and is
important both for understanding the basic biology of bacterial growth in culture, and for applica-
tions in metabolic engineering (
Papagianni, 2012
). However, it has generally been studied only at
the population level, implicitly assuming a homogeneous progression of the entire culture from
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
1 of 18
RESEARCH ARTICLE
acetate producing to acetate detoxifying states. By contrast, single cell approaches suggest that
bacterial populations can exhibit enormous heterogeneity in functional and gene expression states
(
Eldar et al., 2009
;
Locke et al., 2011
;
Su
̈
el et al., 2006
;
Levine et al., 2012
;
Davidson and Sure-
tte, 2008
;
Dubnau and Losick, 2006
;
Gefen and Balaban, 2009
). This prompts the questions of
whether microbial cells differentiate into metabolically distinct subpopulations, and more specifically,
whether acetate production and detoxification might occur in distinct cells specializing in acetate
production or detoxification, respectively.
Results
To address these questions we constructed a library of strains with reporters for key genes involved
in central carbon metabolism, acetate production, and organic acid detoxification (
Figure 1A
). We
introduced a fluorescent protein (YFP) under the control of promoters for 13 different metabolic
genes and stably incorporated them into the commonly used
sacA
site within the genome
(
Supplementary file 1
), (
Eldar et al., 2009
;
Locke et al., 2011
). We chose to use fluorescent pro-
moter reporters because they allow acquisition of dynamic measurements from individual living cells,
are easy to construct and integrate into the
B. subtilis
genome, allow for analysis of multiple genes
within the same cell, and can be used in fluorescence cell sorting for RNAseq experiments. Using
quantitative single-cell fluorescence microscopy, we analyzed the distribution of expression levels of
these 13 metabolic genes in individual cells at different times along the growth curve in buffered cul-
ture medium containing 22 mM glucose and 50 mM malate. To eliminate oxygen gradients, 10 mL
cultures were grown in 250 mL flasks with rapid shaking (250 RPM).
Four genes had expression levels that were at or near background and were not considered fur-
ther (
acoA
,
gntZ
,
pycA
,
sdhC
). Most of the genes showed unimodal distributions (
Figure 1—figure
eLife digest
The chemical reactions that occur within a living organism are collectively referred
to as its metabolism. Many metabolic reactions produce byproducts that will poison the cells if they
are not dealt with: fermenting bacteria, for example, release harmful organic acids and alcohols.
How the bacteria respond to these toxins has been most studied at the level of entire microbial
populations, meaning the activities of individual cells are effectively “averaged” together. Yet, even
two bacteria with the same genes and living in the same environment can behave in different ways.
This raises the question: do bacterial populations specialize into distinct subpopulations that play
distinct roles when dealing with metabolic products, or do all cells in the community act in unison?
Rosenthal et al. set out to answer this question for a community of
Bacillus subtilis
, a bacterium
that is commonly studied in the laboratory and used for the industrial production of enzymes. The
analysis focused on genes involved in fundamental metabolic processes, known as the TCA cycle,
which the bacteria use to generate energy and build biomass. The experiments revealed that, even
when all the cells are genetically identical, different
Bacillus subtilis
cells do indeed specialize into
metabolic subpopulations with distinct growth rates.
Time-lapse movies of bacteria that made fluorescent markers of different colors whenever certain
metabolic genes became active showed cells switching different colors on and off, indicating that
they switch between metabolic subpopulations. Further biochemical studies and measures of gene
activity revealed that the different subpopulations produce and release distinct metabolic products,
including toxic byproducts. Notably, the release of these metabolites by one subpopulation
appeared to activate other subpopulations within the community.
This example of cells specializing into unique interacting metabolic subpopulations provides
insight into several fundamental issues in microbiology and beyond. It is relevant to evolutionary
biologists, since the fact that fractions of the population can switch in and out of a metabolic state,
instead of evolving into several inflexible specialists, may provide an evolutionary advantage in
fluctuating natural environments by reducing the risk of extinction. It also has implications for
industrial fermentation processes and metabolic engineering, and may help biotechnologists design
more efficient ways to harness bacterial metabolism to produce useful products.
DOI: https://doi.org/10.7554/eLife.33099.002
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
2 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
Figure 1.
Two genes in central carbon metabolism are heterogeneously expressed in a clonal population of
B. subtilis
. (
A
)
B. subtilis
uses glucose and
malate as preferred carbon sources, and under aerobic culture conditions produces acetate and acetoin as major overflow metabolites. Promoter
reporter strains were made for genes participating in the reactions marked with a yellow dot (
B
) Histograms depict the heterogeneous expression of
the central metabolism genes
sucC
(top panel) and
alsS
(bottom panel). Insets using merged phase and fluorescence images show typical fields of
cells, including cells in the high expressing tail of the distributions. (
C
) The heterogeneous expression of
sucC
(red line) and
alsS
(green line) is maximal
at different timepoints along the growth curve (black line). Black arrows denote the sampling timepoints shown in
Figure 1B
. (
D
) A line graph depicting
the accumulation of extracellular acetate and acetoin in the growth media during exponential and early stationary growth (OD
600
, black line). Acetate
(red line) is released around mid-exponential phase, and is reabsorbed at a later time during which acetoin is produced (green line).
DOI: https://doi.org/10.7554/eLife.33099.003
Figure 1 continued on next page
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
3 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
supplement 1
), with relatively little skew (less than
±
0.7). Two genes,
sucC
and
alsS
, encoding succi-
nate co-A ligase and acetolactate synthase, respectively, were more heterogeneous (
Figure 1B
). We
observed skew values greater than 1 (4.72 and 1.14, respectively) with 3.8% of P
sucC
-YFP and 8.1%
of P
alsS
-YFP cells exhibiting high expression levels (

2 standard deviations above the mean) at
OD
600
~
0.8 (
sucC
) and OD
600
~
2 (
alsS
). In addition, for both genes, we observed cells whose expres-
sion exceeded the mean by >3 fold. While gene expression distribution can be broader immediately
after gene activation than at steady-state (
Shahrezaei and Swain, 2008
), both
sucC
and
alsS
main-
tained heterogeneous expression for several hours after the onset of expression. For these reasons,
we decided to focus on these two genes for further study.
To better understand when this heterogeneity emerges in batch culture, we performed a time
course analysis of the fraction of
sucC
and
alsS
positive cells (cells

2 standard deviations above the
mean were denoted
sucC+
and
alsS+
,
Figure 1C
). We observed that the subpopulation of
sucC+
cells only existed transiently, in mid- to late-exponential phase (
Figure 1C
), coinciding with the time
and culture optical density at which acetate production was observed (when the time derivative in
acetate, that is, the rate of change in acetate concentration, is positive
~
150–300 min,
Figure 1D
).
This observation suggested that
sucC
expression could be involved in acetate production. A parallel
analysis of
alsS
expression revealed the opposite behavior, with
alsS
expression dynamics coinciding
with a decrease in acetate and a concomitant increase in acetoin levels (
Figure 1C,D
). This behavior
is generally consistent with the known role of
alsS
in acetoin production in response to acetate toxic-
ity (
Speck and Freese, 1973
). Together, these results show that a dynamic change in acetate and
acetoin levels in the culture overlaps with changes in the population fraction of
sucC
and
alsS
expressing cells.
A role for sucC in acetate production has not been studied previously. To understand the rela-
tionship between the subpopulation marked
sucC+
and acetate production, we used fluorescence
activated cell sorting (FACS) of the P
sucC
YFP reporter strain to sort cells expressing YFP from a
SucC
promoter at the time of peak acetate levels, and performed RNAseq to compare gene expression
profiles (
Figure 2A
,
Figure 2—figure supplement 1
). As expected,
sucC
expression was elevated 2-
fold in the
sucC+
sorted subpopulation (blue dot,
Figure 2A
). This is particularly meaningful consid-
ering the fact that the fluorescent marker used for sorting is a stable reporter, making it likely that
some sorted cells may have high level of fluorescent signal even after exiting the transcriptionally
active state. For most genes, we observed a broad correlation in gene expression between the two
populations. However, RNAseq analysis with cuffdiff (
Trapnell et al., 2010
) and gene set enrichment
analysis with GSEA (
Subramanian et al., 2005
) showed that genetic competence genes
(
Berka et al., 2002
) were significantly enriched in the
~
300 upregulated genes in the
sucC+
subpop-
pulation (red dots and inset,
2A and
supplementary file 2
; GSEA p<e-16). The
sucC+
population
also exhibited increased expression of the phosphate acetyltransferase gene,
pta
(green dot,
Figure 2A
), the enzyme that catalyzes the final step in overflow acetate production. Thus,
sucC
expression marks a distinct gene expression state that could be involved in acetate production.
Based on the strong correlation between
sucC
expression and competence gene expression in
the RNAseq results (
Figure 2A
), we next asked whether the
sucC+
population represented the com-
petent state. To analyze the relationship between
sucC
expression and genetic competence in single
cells, we constructed four dual reporter strains, expressing CFP from the
sucC
promoter and YFP
from one of four competence promoters:
comG
,
comK
,
nucA
and
rapH
(
Berka et al., 2002
;
Ogura et al., 2002
). Imaging revealed a clear positive correlation between
sucC
and the compe-
tence genes (
Figure 2B
,
Figure 2—figure supplement 2
). This positive correlation was not general
to all metabolic genes, as
sucC
expression was anti-correlated with
pckA
(
Figure 2C
,
Figure 2—fig-
ure supplement 3
), a gene involved in phosphoenolpyruvate synthesis (
Meyer and Stu
̈
lke, 2013
).
Figure 1 continued
The following figure supplements are available for figure 1:
Figure supplement 1.
Histograms of metabolic promoter reporters made for this study.
DOI: https://doi.org/10.7554/eLife.33099.004
Figure supplement 2.
All histograms in panel B use the same X and Y values, allowing easier cross-comparison of expression levels of each reporter.
DOI: https://doi.org/10.7554/eLife.33099.005
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
4 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
Figure 2.
The heterogeneous expression of
sucC
is correlated with the genetic-competence regulon, and this
metabolic state produces extracellular acetate. (
A
) RNAseq of cells sorted at either a high or moderate
sucC
level
reveals positive correlation between sucC expression, the competence program, and the acetate metabolism
gene
pta.
Cells expressing YFP under the control of the
sucC
promoter were sorted at high or normal expression
Figure 2 continued on next page
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
5 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
We note that
pta
(phosphate acetyltransferase gene) and
sucC
were previously observed to be up-
regulated in the competent state (
Berka et al., 2002
;
Ogura et al., 2002
). Together, these results
suggest that individual cells can exist in at least two distinct metabolic states, one of which repre-
sents genetically competent cells and involves increased expression of
sucC
and
pta
, among other
genes.
We next assessed how competence might be linked to elevated acetate production. The compe-
tence system is controlled by a noise-excitable gene circuit that stochastically initiates transient epi-
sodes of differentiation in individual cells (
Su
̈
el et al., 2007
;
Su
̈
el et al., 2006
;
Cag
̆
atay et al., 2009
;
Maamar et al., 2007
;
Hahn et al., 1994
) To better understand the relationship between compe-
tence and acetate metabolism, we next asked whether activation of the competence system is nec-
essary for increased
sucC
expression and acetate metabolism. Strains in which the competence
master transcription factor
comK
is deleted (
Supplementary File 1
) exhibited greatly reduced ace-
tate production (
Figure 2D
) and a loss of
sucC
as well as
comG
expression (
Figure 2—figure sup-
plement 4
). In addition to the reduced level of acetate itself, expression of
pta
, the key step in
acetate production, is greatly diminished in strains in which
comK
is deleted, both in our conditions
(
Figure 2—figure supplement 4B
) and also in data from previous microarray experiments
(
Ogura et al., 2002
;
Berka et al., 2002
). Although the competent state has been suggested to be
involved in other functions, such as attachment, motility, antibiotic resistance, and DNA metabolism
(
Redfield, 1993
;
Hahn et al., 2015
;
Bakkali, 2013
;
Finkel and Kolter, 2001
), a role in central car-
bon metabolism has not been reported. These results indicate that the
sucC
subpopulation is con-
trolled by the competence system, linking competence both to an alternative metabolic state and to
the control of acetate levels in culture.
To better understand the dynamics with which cells switch into the competent state and later into
the
alsS+
state, we used the ‘Mother Machine’ microfluidic device (
Figure 3A
), to conduct long-
term analysis of individual cells over tens of cell generations under chemostatic conditions
(
Wang et al., 2010
;
Norman et al., 2013
). We set up the Mother Machine as described previously
(
Norman et al., 2013
), but cultured cells with conditioned media obtained from batch growth of
B.
subtilis
cultures at different final optical densities. Specifically, we used media from cultures at OD
600
0.8 and OD
600
2.0, points during the peak of
sucC+
or
alsS+
expression, respectively. This approach
Figure 2 continued
levels. Genes for genetic competence (red) and acetate production (
pta
– green) are higher in cells expressing
high levels of
sucC.
The inset histogram shows a histogram of log
2
fold change for all genes (grey) and the
competence program genes (red) (
B
) A scatter plot shows that the expression of
sucC
reporter is positively
correlated with the
comG
reporter, a marker for the competence program. Each dot represents a single cell
centered on the mean fluorescence of reporters for
sucC
and
comG
. The right panel shows fluorescent
microscopy images taken from a typical field of cells (
C
) A scatter plot shows that the expression of the
sucC
reporter is negatively correlated with expression of the metabolic gene
pckA
. Each dot represents a single cell
centered on the mean fluorescence of reporters for
sucC
and
pckA
. Right panel shows fluorescent microscopy
images taken from a typical field of cells. (
D
) The competence gene expression program is necessary for the
buildup of high levels of extracellular acetate. Growth curves demonstrate only a small difference in growth of
wildtype strain (solid black line) or the competence-null
D
comK
strain (dashed black line). However, maximal
acetate buildup is approximately five fold higher in the wildtype strain (solid red line) than in a strain that is unable
to produce the competent cell population (
D
comK
dashed red lines).
DOI: https://doi.org/10.7554/eLife.33099.006
The following figure supplements are available for figure 2:
Figure supplement 1.
FACS sorting parameters for sorting
sucC+
and normal cells for RNAseq experiments.
DOI: https://doi.org/10.7554/eLife.33099.007
Figure supplement 2.
Competence genes are positively correlated with the TCA cycle gene
sucC
.
DOI: https://doi.org/10.7554/eLife.33099.008
Figure supplement 3.
Competence genes are negatively correlated with the metabolic gene
pckA
.
DOI: https://doi.org/10.7554/eLife.33099.009
Figure supplement 4.
The heterogeneous expression of
sucC, comG
, and
pta
are dependent on the competence
master regulator comK.
DOI: https://doi.org/10.7554/eLife.33099.010
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
6 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
Figure 3.
Cells switch in and out of the slower growing
sucC+
and
alsS+
states based on media conditions. (
A
) A schematic of the Mother Machine
microfluidic experiment. Cells are loaded into growth channels that are capped on one end and surrounded by flowing media. A ‘mother’ cell settles at
the capped end, and produces daughters. The daughters at the uncapped end of the growth channel are washed away by the current of media.
Positions were filmed for up to 4 days, and for visualization purposes the images from each growth channel were cropped and aligned to generate a
kymograph depicting time on the x-axis. (
B
) Filmstrip kymographs from representative mother-machine experiments using conditioned media at OD
600
0.8 and
sucC
reporter strain (left panel) or conditioned media at OD
600
2.0 using
alsS
reporter strain (right panel). Dashed red lines show the trend of
growth of two daughters, a
sucC+
and
sucC-
pair on the left panel and a
alsS+
and
alsS-
pair on the right panel. As seen from the slope of the trend
lines and as indicated below the kymographs, the elongation rates of both
sucC+
and
alsS+
cells are slower than their counterparts. All values reported
(events per 100 hr, elongation rate, and dwell time) were averaged from at least three separate movies per condition. Representative mother machine
movies are available in the supplement. (
C
) Extracellular acetate levels activate heterogeneous expression of the
alsS
promoter. A histogram shows the
population of cells expressing different levels of YFP under control of the
alsS
promoter. When no acetate is added (green line), practically all cells are
in the low expressing portion of the histogram. When extracellular acetate levels are added to mimic the maximal amount produced in the growth
Figure 3 continued on next page
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
7 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
provides the simplicity of long-term chemostatic analysis with the ability to compare cellular behavior
at different culture time-points.
Using the Mother Machine, we analyzed cell lineages for up to 4 days (approximately 60 genera-
tions) for a total of 1,400 cell generations (
Figure 3B
). We tabulated the number of activation events
per hour occurring in the terminal (top) cell of each channel, as well as the length of time that cells
remained in an activated state (dwell time). Finally, we also measured the length of cells through suc-
cessive frames of the movie, to obtain the mean relative elongation rates of
alsS+
and
alsS-
cells
(Materials and Methods).
With
sucC
-inducing media (conditioned at OD 0.8), we observed rare episodes of
sucC
activation
in some cells, lasting for approximately four hours each (252
±
89 min, mean
±
standard deviation for
n = 31 events) (
Figure 3B
, left). Consistent with previous analysis of competence dynamics
(
Su
̈
el et al., 2006
),
sucC+
cells divided less frequently and grew more slowly than other cells in the
same movies (elongation rates of 47.4
±
2.7 %/hr and 67.7
±
2.3 %/hr, respectively). Cells in the acti-
vated state could switch out of the
sucC+
state and resume normal growth rates (
Figure 3—figure
supplement 1
,
Video 1
). Under these conditions,
we did not observe activation of
alsS
expression.
By contrast, in the OD 2.0 conditioned media we
Figure 3 continued
curve (orange and red lines), some cells remain in the low expressing portion of the histogram, but a correspondingly larger number of cells are in the
long tail of high
alsS
expression.
DOI: https://doi.org/10.7554/eLife.33099.011
The following figure supplement is available for figure 3:
Figure supplement 1.
Filmstrip kymographs from mother-machine experiments using conditioned media at OD
600
0.8 and
sucC
reporter strain.
DOI: https://doi.org/10.7554/eLife.33099.012
Video 1.
A representative microscopy time-lapse video
of
sucC
fluorescent reporter expression experiment
demonstrates that a subset of cells switches
sucC
expression on and off. Bright green cells, expressing
high levels of
sucC
(
sucC
+) have a slower growth and
division rate, but can still divide.
DOI: https://doi.org/10.7554/eLife.33099.013
Video 2.
A representative microscopy time-lapse video
of
alsS
fluorescent reporter expression
experimentdemonstrates that individual cells switch
alsS
expression on and off. Bright green cells,
expressing high levels of
alsS
(
alsS
+) are able to divide.
DOI: https://doi.org/10.7554/eLife.33099.014
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
8 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
did not observe activation of
sucC
expression, but did observe frequent pulses of
alsS
gene expres-
sion.
alsS+
cells grew at a slightly reduced elongation rate (63
±
2.6 %/hr increase compared to 74
±
2.3 %/hr for alsS- cells,
Figure 3B
, right panel,
Video 2
). Together, these results provide rates of
transitions into the
sucC+
(competent) and
alsS+
gene expression states, and show that these states
have altered growth rates and respond to medium composition.
These results further suggested the possibility that acetate predominantly produced by
sucC+
cells early in the growth could induce cell switching to the
alsS+
state in later growth stages, when it
accumulates to toxic levels. However, many media components could differ between the OD
600
0.8
and OD
600
2.0 cultures. To determine whether acetate was sufficient to affect
alsS
expression, we
cultured reporter cells in varying levels of acetate, in unconditioned liquid medium, and quantified
the fraction of
alsS+
cells. We observed both a systematic increase in the distribution of
alsS
expres-
sion levels, and in the fraction of cells in the high expressing ‘tail’ of the distribution (
Figure 3C
).
The Mother Machine is ideal for analyzing cells over multiple generations in a relatively constant
environment but not ideal for analyzing responses to environmental changes that happen as a conse-
quence of growth. We therefore designed microcolony pad experiments in which acetate was added
to standard microcolony medium (
Eldar et al., 2009
;
Locke et al., 2011
;
Young et al., 2011
) to 20
mM, the acetate concentration present in mid-exponential phase (
Figure 1D
) (
Speck and Freese,
1973
). In these experiments (
Figure 4
), all cells started with a low growth rate, likely owing to the
initial acetate present in the growth media. As cells divided, approximately half of the population
switched on high levels of
alsS
expression within 7 to 10 hr (
Figure 4
,
Figure 4—figure supplement
1
,
Videos 3
5
). As growth progressed, these
alsS+
cells exhibited a reduced growth rate, similar to
that of the original culture. However, a distinct subpopulation with approximately 2.5-fold lower
alsS
expression emerged (
alsS-
), becoming greater than 70% of the population. These cells exhibited a
faster division rate (
Figure 4B
,
Figure 4—figure supplement 1
) and faster elongation rate
(
Figure 4C
,
Figure 4—figure supplement 2
).
The fast growing
alsS-
cells that appeared late in pad growth experiments had a large growth
advantage compared to the slow growing
alsS+
cells (median elongation rates of 65 %/hr and 20 %/
hr, respectively). In general, growth in the chemostatic Mother Machine using conditioned media is
faster than on a pad with non-conditioned media containing acetate. However, the difference
between
alsS-
and
alsS+
cells is much smaller in
the microfluidic condition (74 %/hr vs 63 %/hr for
alsS-
and
alsS+
). This finding is consistent with
the established role of acetoin as a molecule
Video 3.
One representative agarose-pad microcolony
microscopy experiment of
alsS
fluorescent reporter
demonstrates that individual cells switch
alsS
expression on and off. Bright green cells, expressing
high levels of
alsS
(
alsS
+) divide and elongate more
slowly, but are able to divide.
DOI: https://doi.org/10.7554/eLife.33099.019
Video 4.
A second representative agarose-pad
microcolony microscopy experiment of
alsS
fluorescent
reporter demonstrates that individual cells switch
alsS
expression on and off. Bright green cells, expressing
high levels of
alsS
(
alsS
+) divide and elongate more
slowly, but are able to divide.
DOI: https://doi.org/10.7554/eLife.33099.020
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
9 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
Figure 4.
alsS +cells have slower division and elongation rate than alsS- cells. (
A
) A filmstrip of a representative timelapse experiment. Cells were
grown on agarose pads containing acetate at a level that mimics mid-exponential phase. Additional filmstrips are added in supplementary movies (
B
)
AlsS+
cells divide more slowly. A genealogy tree depicts cell division events in the experiment shown in panel A.
alsS
levels are color coded by the
heatmap on the right. Cells switch in and out of high
alsS
expression levels. Cells expressing high
alsS
levels (red and orange) divide more slowly than
cells with low
alsS
levels (blue). Similar genealogy trees are provided for three separate experiments in
Figure 4—figure supplement 1
. (
C
)
AlsS-
cells
in the end of the experiment have faster elongation rates. Cells in the last 7 frames of the experiment which had arbitrary fluorescence levels greater
than 5400 were designated
alsS+
and those expressing less were designated
alsS-
. The elongation rate of each group of cells was determined and
plotted as a histogram.
alsS-
cells (blue line) had a median elongation rate of 65.2 %/hr while
alsS+
cells (red line) had a median elongation rate of
19.96 %/hr. Similar analysis was performed for three separate experiments. (
D
) Summary
sucC
and
alsS
interactions (schematic): In early growth stages a
subset of cells become
sucC+
. These cells secrete acetate, which accumulates to toxic levels. High acetate levels in turn activate some cells in the
population to preferentially express metabolic genes for the production of acetoin, a non-toxic pH-neutral metabolite. Finally, acetoin replaces acetate
in the media.
DOI: https://doi.org/10.7554/eLife.33099.015
The following figure supplements are available for figure 4:
Figure supplement 1.
Cells expressing high levels of alsS divide more slowly than cells expressing lower levels.
DOI: https://doi.org/10.7554/eLife.33099.016
Figure 4 continued on next page
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
10 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
secreted to counter the pH and anion producing toxic effect of secreted short chain fatty acids,
including acetate (
Xiao and Xu, 2007
;
Speck and Freese, 1973
). In the pad environment, transient
activation of genes for the production of the pH-protective acetoin in the
alsS+
cells has the poten-
tial to produce a milder growth environment which may enable other cells to grow faster
(
Figure 4D
). By contrast, in the mother-machine experiments (
Figure 3
), while cells switch in and out
of the
alsS
metabolic state, the chemostatic nature of the device minimizes their ability to impact the
growth rates of their neighbors.
To test the idea that
alsS+
cells can enhance the growth of
alsS-
cells by influencing the extracel-
lular environment, we seeded petri plates with equal numbers of cells from either wild-type or
D
alsS
B. subtilis
strains. The wildtype strain produced larger colonies than the deletion strain (
Figure 4—
figure supplement 3
). However,
D
alsS
colonies plated next to wildtype colonies exhibited a distinct
growth advantage compared to the same colonies plated further away from the wildtype strain (
Fig-
ure 4—figure supplement 3
), indicating that
alsS+
cells (in the wildtype colony) alter the growth
environment to benefit
D
alsS
cells. This may occur through detoxification by
alsS+ cells
(consistent
with the role of alsS in the literature). However, our data do not rule out alternative scenarios in
which
alsS+
cells provide this benefit through a distinct mechanism, for example by the secretion of
a shared metabolite or co-factor.
Discussion
The natural metabolic niches in soil occupied by
B. subtilis
are not well understood, and likely het-
erogeneous. In general, growth of soil organisms
is typically highest near deposits of organic mate-
rials which serve as carbon and nitrogen sources,
including deposits of rotting fruits and vegeta-
tion. In many fruits and vegetables both organic
acids (including malic acid) and sugars are very
high, providing a natural setting where
B. subtilis
has access to its two preferred carbon sources
simultaneously.
B subtilis
is commonly enriched
(almost to levels of purity) in traditional fermenta-
tion of certain foods, including fermented soy-
bean foods (eg natto) that are common in Asia
(
Kubo et al., 2011
). In these beans sugar is pres-
ent, and malic acid is also present in the root sys-
tem of legumes and in soybean exudate, where it
plays a role in iron homeostasis as a metal chelat-
ing agent (
Tiffin and Brown, 1962
).
The ‘acetate switch’ refers to the accumulation
and reabsorption of acetate. It is a classic hall-
mark of bacterial growth in aerobic conditions,
common to many bacteria including
B. subtilis
and
E. coli
(
Wolfe, 2005
). It allows for rapid initial
growth until acetate levels and associated acidity
reach toxic levels, at which point acetate is reab-
sorbed and replaced with pH-neutral overflow
Figure 4 continued
Figure supplement 2.
Cells expressing high levels of alsS elongate more slowly than cells expressing lower levels in the end of the pad culture
experiment.
DOI: https://doi.org/10.7554/eLife.33099.017
Figure supplement 3.
Cells able to express
alsS
enhance the growth of neighboring cells unable to express
alsS
by improving the extracellular growth
environment.
DOI: https://doi.org/10.7554/eLife.33099.018
Video 5.
A third representative agarose-pad
microcolony microscopy experiment of
alsS
fluorescent
reporter demonstrates that individual cells switch
alsS
expression on and off. Bright green cells, expressing
high levels of
alsS
(
alsS
+) divide and elongate more
slowly, but are able to divide.
DOI: https://doi.org/10.7554/eLife.33099.021
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
11 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
metabolites such as acetoin (
Wolfe, 2005
;
Speck and Freese, 1973
). Growth strategies in which a
preferred toxic overflow metabolite is produced under aerobic conditions are also used by other
organisms that expel and control different fermented toxic overflow metabolites, including ethanol
fermentation by yeast (
Otterstedt et al., 2004
) and lactic acid in lactobacillus species (
Borch and
Molin, 1989
). Interestingly, in the fermentation of ethanol by the budding yeast
Saccharomyces cer-
evisiae
, ethanol is produced in dynamic bursts in which some cells switch in and out of fermentative
metabolism. These bursts can be synchronized in chemostat growth (
Tu et al., 2005
), but also
appear in batch culture (
Silverman et al., 2010
). In yeast, the single cell dynamics, mechanisms, and
role of these bursts have not been fully elucidated. However, the presence of metabolically special-
ized subpopulations of cells in both eukaryotes and bacteria suggests that segregating different fer-
mentative or respiratory pathways into individual cells may be a general strategy. It could function to
avoid metabolic incompatibilities (
Brandriss and Magasanik, 1981
;
Ackermann, 2015
;
Kumar et al., 2010
), controlling cellular challenges such as reducing potential (
Liu et al., 2017
), or
to optimize enzyme and substrate scaling, in which locally high concentrations of enzymes and sub-
strates may be needed for efficient enzymatic conversion to occur (
Nikel et al., 2014
;
Acker-
mann, 2015
). Better understanding of the principles that govern segregation of metabolic activities
will facilitate metabolic engineering and industrial fermentation approaches. This is especially true
for commonly used industrial strains which naturally produce multiple fermentation products. For
example,
E. coli
strains simultaneously produce five different fermentation products during mixed-
acid fermentation (
Clark, 1989
).
The rapid release of toxic overflow products also has a role in the context of competition within a
multi-species environment. In such environments, a quick buildup of toxic products can be advanta-
geous to ward off competing species. In the case of human infectious disease, the buildup of
byproducts such as lactic acid from normal microbiota limits infection by pathogens that are not lac-
tic acid specialists (
O’Hanlon et al., 2013
). Likewise, in industrial fermenters and microbial food fer-
mentation secreted overflow metabolites, including acetate and ethanol, limit contamination.
Additionally, if a particular metabolic niche is transient, as in the case of acetate production in
B.
subtilis
colony growth or batch culture, a strategy in which cells can switch in and out of metabolic
states can be advantageous relative to an alternative scenario in which multiple strains are evolution-
arily ‘locked’ into distinct specialist roles. This is especially true if the metabolic niche (e.g. acetate) is
short-lived, because a ‘locked’ specialist strain would be at a disadvantage during periods of growth
when acetate is absent.
In this study, we linked the presence of extracellular acetate with the activation of
alsS
in a subset
of cells. In the case of competence and
sucC
, which are both controlled by the master regulator
comK
, quorum sensing plays a critical role in the activation of the competence program (
Dub-
nau, 1991
). Maximizing competence requires specific media conditions (
Dubnau, 1982
), raising the
possibility that alongside quorum sensing peptides, secreted metabolic byproducts also play a role
in this process.
Our results reveal that several metabolic processes may be affiliated with a given subpopulation.
For instance, it is not clear at this point how
sucC
is related to acetate metabolism. One possible
connection between
sucC
expression and acetate production might be that during co-consumption
of glucose and malate, the TCA cycle may be overloaded with incoming material, causing accumula-
tion of intermediates that can impact the glycolytic pathway, including acetate secretion. Another
possibility is that
sucC
is used to regulate coenzyme-A levels at a time when a large amount of coen-
zyme-A is used and freed when pyruvate is converted to acetyl-coA and then acetate. Additionally,
alongside the roles of the
sucC
and
pta
metabolic genes, arginine metabolism genes are also differ-
entially regulated by competence, as seen in our results and in prior microarray experiments
(
Berka et al., 2002
;
Ogura et al., 2002
), further complicating our understanding of the full physio-
logical role of this subpopulation. Tools to systematically study metabolites in single cells would help
to map the actual metabolic states of each subpopulation in a given condition and will allow us to
address these questions.
Going beyond microbial systems, cell-cell heterogeneity can be advantageous as a ‘bet-hedging’
strategy both for microbial and cancer cells (
Veening et al., 2008
). In such cases, cell-cell heteroge-
neity enables the population as a whole to withstand unforeseen challenges, such as antibiotic or
chemotherapeutic drugs (
Sharma et al., 2010
;
Rotem et al., 2010
), or metabolic shifts
(
Solopova et al., 2014
). By contrast to simple bet-hedging, the emergence of
alsS+
acetoin
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
12 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
producing populations described here arises as a response to an anticipated challenge that is part
of the growth progression in conditions favoring weak organic acid production. Thus, unlike in bet-
hedging, metabolic state switching could provide a predictable benefit in a more deterministic
dynamic environment.
Materials and methods
Plasmid design
Plasmids for the integration of fluorescent reporters were made as previously reported (
Eldar et al.,
2009
). YFP promoter reporters were cloned into the ECE174 backbone plasmid which uses sacA
integration site and encodes chloramphenicol resistance (R. Middleton, obtained from the Bacillus
Genetic Stock Center). CFP promoter reporters were cloned into the pDL30 backbone which uses
amyE integration sites and encodes spectinomycin resistance (obtained from the Bacillus Genetic
Stock Center). A constitutive RFP reporter, using a minimal sigA promoter, was used for image seg-
mentation as previously reported (
Locke et al., 2011
). pAZR1: (Pcggr:alsS/D) is a plasmid for the
integration of a constitutive promoter (cggr promoter) to drive the constitutive expression of the
alsS/D regulon from its native site. The plasmid was constructed by Gibson cloning (
Gibson et al.,
2009
). A markerless deletion of alsS/D was made using the alsS/D strain of the BKE collection and
the pdr244 plasmid, both obtained from the BGSC, followed by selection.
Bacterial strains
All strains were made by genomic integration into the genome. Fluorescent reporters were inte-
grated into either the
sacA
(YFP) or the
amyE
(CFP) loci as previously described (
Locke et al., 2011
).
A constitutive RFP color was utilized, relying on constitutive expression of a partial ptrpE promoter
reporter driving mCherry expression, which was inserted into the ppsB locus as previously
(
Locke et al. 2011b
). Non chaining strains for microfluidic mother-machine experiments used a lytF
overexpression construct as previously reported. Strain information is included as
Supplementary file 1
.
Growth conditions
Strains were started from glycerol stocks and grown in M9 minimal media prepared according to the
directions of the manufacturer (BD – difco, Franklin Lakes NJ). Base media was supplemented with
0.4% glucose (22 mM) and a cocktail of trace metals (
Leadbetter et al., 1999
) 50 mM Malate
(0.66%) was added to growing cultures at OD 0.4–0.5 as per previous publications (
Buescher et al.
2012
;
Kleijn et al., 2010
) to facilitate comparison of results between our studies and other pub-
lished work. Samples for fluorescence microscopy were prepared using agarose pads for either
snapshot analysis (timepoint measurements) or pad movies, as previously described by our labora-
tory (
Young et al. 2011b
).
Microscopy
Images were acquired using a Nikon inverted TI
-
E microscope via a coolsnap HQ2 camera. Commer-
cially available software (Metamorph) controlled the stage, microscope, camera, and shutters. Fluo-
rescent illumination was provided by a Sola Light Engine LED source (Lumencor). Temperature was
kept at 37
̊
C using an enclosed microscope chamber (Nikon) attached to a temperature sensitive
heat exchanger. All experiments used a Phase 100x Plan Apo (NA 1.4) objective. Filter sets used
were Chroma #41027 (mCh), Chroma #41028 (YFP), and Chroma #31044 v2 (CFP).
Measurements of secreted acetate and acetoin
Media was collected from growing cultures by centrifuging 500 ul culture samples at 5000 g for 2
min and filtering the supernatant in 0.2 uM syringe filters. Clarified conditioned media samples were
placed into glass sample vials and run at the Caltech environmental analysis center using an Aminex
HPX-87H column (Bio-Rad, Rockville NY) in an Agilent 1100 HPLC with UV and Refractive Index
detectors with elution using 0.013 N H
2
SO
4
at ambient temperature and 36 ml/hour flow as
described in (
Leadbetter et al., 1999
). Standards of acetate and acetoin were prepared in uninocu-
lated growth media, and diluted to produce a standard curve.
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
13 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease
Microfuidic mothermachine experiments
Microfluidic experiments used the mothermachine devices described in (REFS
Wang Jun 2010
;
Norman Losick 2013
). SU80 wafers were made based on masks provided by the Losick lab. PDMS
devices were prepared by pouring degassed Sylgard 184 PDMS silicone (corning corporation, Corn-
ing NY) onto wafers and curing the molds for a minimum of 8 hr at 65
̊
C. Cured PDMS devices were
bonded onto microscopy coverslips (60

22 mm, Gold Seal coverslips Thermo Fisher Scientific, Wal-
tham MA) by plasma cleaning. Plasma bonding was done in a PDC32G plasma cleaner (Harrick
Plasma Ithaca, NY) set to chamber pressure between 600–700 microns. Coverslips were cleaned sep-
arately for 1 min, and then the devices and coverslips were cleaned jointly for 20 s. Device bonding
was immediately done by inverting the plasma treated device onto the treated coverslips. After
bonding the devices were cured for an additional 4 hr at 65
̊
C. Devices were kept for up to 2 weeks
in the dark at room temperature. Before use, holes for inlet and outlet were punched using a biopsy
punch and each device was passivated by loading the channels using growth-media containing 1
mg/ml BSA using a handheld micro-pippete and a 20 uM tip. Cells were loaded by flowing a concen-
trated cell culture (OD 2.0) and letting cells reach the growth chambers by waiting for 30 min. Devi-
ces were placed on an inverted Nikon TiE microscope and growth media was flowed using syringe
pumps set to a flow rate of 50–100 ul per hour. Media used in microfluidic mother machine experi-
ments was conditioned media taken from batch growth cultures. Media for the sucC/competence
experiments contained media conditioned by growth on glucose/malate media until OD 0.8. Condi-
tioned media used for alsS movies was from OD 2.0. Fluorescent Images were captured using a
CoolSnap HQ2 and analyzed with custom MATLAB software (
Rosenthal, 2018
; copy archived at
https://github.com/elifesciences-publications/Schnitzcells_2018
) and in imageJ.
Events were followed for the terminal ‘mother’ cell position in each channel of each device, and
were manually counted to provide the events per hour. Dwell time was determined by counting the
number of frames from activation (passing the fluorescence threshold) and deactivation, and con-
verted to hours and minutes by multiplying the number of frames by the frame rate of the movie.
Elongation rates were measured by quantifying the length of at least 50 cells in either the activated
or non-active state per each movie (at least three movies per condition), and calculated as previously
determined (
Young et al., 2011
).
RNAseq
Cultures of cells expressing YFP under the control of
sucC
(strain AZRE1) were grown in M9 glucose-
malate media. Cells in mid log phase (OD 0.8) were fixed in 4% formaldehyde for 10 min at room
temperature. Fixed cells were washed twice in Tris pH 7, and gently filtered using a 5.0 uM filter to
remove clumps and chains. Cells were sorted on either a MoFlo astrios cell sorter or a BSfacsARIA in
the USC medical school sorting facility. Cells sorted for either high YFP fluorescence or regular fluo-
rescence (a minimum of 200,000 cells) were collected into tubes containing RNA protect (Qiagen,
Hilden, Germany). Sorted samples were centrifuged and cells were rehydrated in 240 ul qiagen PKD
buffer (FFPE miRNEASY kit, Qiagen). Cells were lysed by the addition of 10 ul lysosome solution for
10 min followed by bead beating for 2 min in high setting. Samples were further processed using
the qiagen FFPE miRNA kit. Libraries were prepared using the Epicentre Scriptseq V2 kit, follow the
directions for highly fragmented DNA. Libraries were sequenced at the Caltech Millard and Muriel
Jacobs sequencing facility. Analysis followed the standard Galaxy RNAseq workflow (grooming, trim-
ming, bowtie mapping, and cuff-diff and cuff-links) (
Afgan et al., 2016
).
Agarose pad timelapse experiments
Agarose pad experiments were done as previously described (
Young et al., 2011
), with the follow-
ing exceptions: Cells were spotted on agarose pads made with standard pad movie media
(
Young et al., 2011
) which is a Spizizen’s minimal media with 0.4% glucose to which acetate was
added to a final concentration of 20 mM, to mimic acetate concentrations at mid exponential phase.
Cells were allowed to acclimate to the agarose pad growth condition for
~
2 hr, before the start of
imaging. Images were acquired from multiple fields every 12 min for a total of 22 hr. Movie analysis
was performed in Matlab using the Schnitzcells analysis package (
Young et al., 2011
) with slight
edits. The current version of this analysis package is available at
http://www.elowitz.caltech.edu
Rosenthal
et al
. eLife 2018;7:e33099.
DOI: https://doi.org/10.7554/eLife.33099
14 of 18
Research article
Computational and Systems Biology
Microbiology and Infectious Disease