Stochastic switching is considered as a cost‐saving strategy for adaptation to environmental challenges. We show here that stochastic switching of a monostable circuit can mediate the adaptation of the engineered OSU12‐hisC Escherichia coli strain to histidine starvation. In this strain, the hisC gene was deleted from the His operon and placed under the control of a monostable foreign promoter. In response to histidine depletion, the OSU12‐hisC population shifted to a higher HisC expression level, which is beneficial under starving conditions but is not favoured by the monostable circuit. The population shift was accompanied by growth recovery and was reversible upon histidine addition. A weak directionality in stochastic switching of hisC was observed in growing microcolonies under histidine‐free conditions. Directionality and fate decision were in part dependent on the initial cellular status. Finally, microarray analysis indicated that OSU12‐hisC reorganized its transcriptome to reach the appropriate physiological state upon starvation. These findings suggest that bacteria do not necessarily need to evolve signalling mechanisms to control gene expression appropriately, even for essential genes.
The fundamental mechanisms underlying adaptations can be divided into responsive switching and stochastic switching (Kussell and Leibler, 2005). Responsive switching is generally considered as resulting from evolved regulatory units, such as operons and regulons, which enable immediate adaptation (Jacob and Monod, 1961). However, as cells are subject to a wide range of both genetic and environmental perturbations that damage the specificity or efficiency of regulatory systems (Carroll, 2005; Crombach and Hogeweg, 2008), the limited number of regulatory units that can evolve and remain functional may not be sufficient to completely protect cell populations from the danger of extinction. Whether and how cells are able to survive external perturbations, when the corresponding regulatory units are absent or have been genetically disrupted, is an open question of great importance.
Recent studies showed the stochastic switching provided cells a huge potential for sustenance under severe conditions via a so‐called ‘bet‐hedging’ strategy. The experimental evidence was generally based on a bistable genetic structure that fixed stochastically appearing fit state thus limiting further random switching (Kussell and Leibler, 2005; Acar et al, 2008). In contrast to bistable gene expression, monostable gene expression is much more common (Newman et al, 2006) and does not rely on a specific complex genetic architecture. Since a monostable structure has no fixation effect, the fit cells that would appear stochastically tend to return to the original steady state (i.e., unfit state). To achieve a population shift from a maladaptive state (but stable) to an adaptive state (but unstable), a significant increase in fitness (i.e., growth rate) of the fit cells is necessary. Otherwise, the random switching will mask occasionally occurring adaptive transitions and lead to an unchanged population at the stable but maladaptive state. Whether adaptation can be achieved by stochastic switching based on a monostable structure is however an open issue.
To address this question, we applied an engineered E. coli strain, OSU12‐hisC, carrying a foreign gene circuit encompassing a physiologically functional gene, hisC, replaced from its native chromosomal locus (Figure 1A). Consequently, hisC in OSU12‐hisC is no longer responsive to the native regulation (His operon) that senses histidine depletion. Instead, the foreign gene circuit provided a monostable structure for hisC's stochastic switching. The green fluorescent protein (gfpuv5) was co‐expressed with hisC for the quantitative evaluation of HisC in single cells. The upstream regulation of TetR, whose expression level was reported by the red fluorescent protein (dsred.T4), was introduced to achieve the inducible GFP (HisC) level. The full induction of TetR by IPTG was applied to avoid any possible upstream noise that caused by the abundance of endogenous LacI.
Microscopic observation revealed that the OSU12‐hisC cells showed stronger green fluorescence after histidine depletion (Figure 1B), which suggested an increased expression level of hisC. Population analysis using flow cytometry showed that the distributions of both GFP concentration and GFP bias (GFP/RFP ratio) in OSU12‐hisC shifted towards a higher level in histidine‐free conditions (Figure 1C and D), whereas, the depletion caused only a slight change in distributions of OSU11, a control strain carrying both the same engineered genetic circuit and an intact His operon, including the hisC gene in its native context. Repeated experiments revealed that the increases in both GFP concentration (∼2.1 folds) and GFP bias (∼1.5 folds) due to histidine depletion were highly significant (P<0.005, N=6) in OSU12‐hisC. In particular, the increased GFP bias strongly suggested that the change in gene expression occurred specifically in the rewired hisC (i.e., GFP) but not in all genes (e.g., RFP). Furthermore, both the growth recovery accompanied population shift and the stress relaxation triggered restoration were clearly observed. It strongly indicated that the adaptation was mediated by stochastic switching of hisC under the monostable control.
Analysis on microcolonies’ formation (Figure 4A) showed stochastic behaviour and directionality in individual cells. Variation in cellular GFP level was clearly observed in individual cells. Stochastic switching of hisC was verified according to the random changes in GFP bias along with the cell division under histidine‐rich conditions (Figure 4B). On the other hand, the microcolonies formed under the histidine‐free conditions tended to the higher level of GFP bias were observed (Figure 4B). The directional tendency favoured the high GFP (HisC) level was evidently detected in the first 2 h after histidine depletion, which resulted in a population shift (Figure 4C). In contrast, the distributions of microcolonies grown in histidine‐rich conditions kept steady, due to the randomized directions of stochastic switching (Figure 4C). Further analysis showed that the stochastic fluctuations in the initial state had an important role not only in fate decision (i.e., whether to grow) but also in the directionality of the stochastic switch.
Microarray analysis showed the adaptation of OSU12‐hisC was resulted from the enhanced expression of the structural genes within the native His operon, along with the transcriptional reorganization of a large number of genes. In summary, in contrast to bistable structures, the monostable structure used here did not fix the phenotype but allowed the cells to decide where to go. Taken together, the findings suggest that bacteria do not necessarily need to evolve signalling mechanisms to control gene expression appropriately, even for essential genes.
Stochastic switching of a monostable circuit mediated the adaptation of the engineered OSU12‐hisC Escherichia coli strain to histidine starvation.
The population shift of OSU12‐hisC was accompanied by growth recovery and was reversible upon histidine addition. This is the first report of adaptation mediated by stochastic switching based on a monostable structure.
Weak directionality in stochastic switching initiated the population shift and the fast growth of the occasionally appearing fit cells drove the later stages of adaptation.
Adaptation of OSU12‐hisC was resulted from the enhanced expression of the structural genes within the native His operon, along with the transcriptional reorganization of a large number of genes.
The fundamental mechanisms underlying adaptations can be divided into responsive switching and stochastic switching (Kussell and Leibler, 2005). Responsive switching is generally considered as resulting from evolved regulatory units, such as operons and regulons, which enable immediate adaptation (Jacob and Monod, 1961; Henkin and Yanofsky, 2002; Gama‐Castro et al, 2008). However, as cells are subject to a wide range of both genetic and environmental perturbations that damage the specificity or efficiency of regulatory systems (Carroll, 2005; Crombach and Hogeweg, 2008), the limited number of regulatory units that can evolve and remain functional may not be sufficient to completely protect cell populations from the danger of extinction. Whether and how cells are able to survive external perturbations, when the corresponding regulatory units are absent or have been genetically disrupted, is an open question of great importance.
The studies on persistence, competence and sustenance proposed a complementary mechanism relying on stochastic switching (Balaban et al, 2004; Kussell and Leibler, 2005; Suel et al, 2007; Acar et al, 2008). Through a so‐called ‘bet‐hedging’ strategy, stochastic switching generates population diversity without the need for a specific responsive regulatory system (Slatkin, 1974). Noise in gene expression causes cell‐to‐cell variation and provides a chance for survival of the cells that occasionally are in the fit state (Avery, 2006). As stochastic switching requires neither an evolved regulatory unit nor the maintenance of any signal transduction machinery, it is considered as a universal cost‐saving strategy for adaptation. Such stochastic switching‐mediated survival strategy was greatly supported by the studies employing cells whose default sensing system was deficient or missing (Kashiwagi et al, 2006; Stolovicki et al, 2006; Suzuki et al, 2006; Stern et al, 2007; Isalan et al, 2008).
So far, the experimental evidence on stochastic switching was generally based on a bistable genetic structure (Kussell and Leibler, 2005; Kashiwagi et al, 2006; Stolovicki et al, 2006; Stern et al, 2007; Acar et al, 2008). Bistability allows the cells to stay at a stable state, either fit or unfit, so that the stochastically appearing fit state can be stabilized without further random switching to the unfit state (Supplementary Figure S1; Kussell and Leibler, 2005; Acar et al, 2008). Owing to the fixation effect of bistability, the stochastic switching‐induced population transition was intensively characterized (Kussell and Leibler, 2005; Acar et al, 2008). Nevertheless, the universality of stochastic switching as an adaptation strategy remains open, as bistability is known to be a special case in native genetic structures.
In contrast to bistable gene expression, monostable gene expression is much more common (Newman et al, 2006) and does not rely on a specific complex genetic architecture. Stochastic switching based on a monostable genetic structure has, however, not yet been identified as an accessible strategy for adaptation. Since a monostable structure has no fixation effect, the fit cells that appear stochastically tend to return to the original steady state (i.e., unfit state) (Supplementary Figure S1). To achieve a population shift from a maladaptive state (but stable) to an adaptive state (but unstable), a significant increase in fitness (i.e., growth rate) of the fit cells is necessary. Otherwise, the random switching will mask occasionally occurring adaptive transitions and lead to an unchanged population at the stable but maladaptive state. Therefore, in contrast to a bistable structure, the final adaptive state in a monostable system is not determined a priori by the genetic architecture, but is determined by the cellular response.
Whether adaptation can be achieved by stochastic switching based on a monostable structure is an intriguing issue. To address this question, we used an engineered Escherichia coli strain carrying a monostable genetic circuit that allows the experimental observation on stochastic switching. In this strain, the gene hisC was deleted from the His operon and placed under the control of a foreign gene circuit at another chromosomal locus, as previously described (Kashiwagi et al, 2009). The circuit showed a clear monostable behaviour, which was characterized in detail previously (Tsuru et al, 2009). The gene hisC encodes the histidinol‐phosphate aminotransferase, an enzyme essential for histidine biosynthesis, and is regulated by a transcriptional attenuator in the His operon (Keller and Calvo, 1979; Gama‐Castro et al, 2008). In the hisC rewired strain, the responsive switching of hisC mediated by the native transcriptional regulation of the His operon was unavailable. Only the stochastic switching of the engineered hisC monostable circuit can provide the cells a chance of survival from histidine starvation. As a result, stochastic switching‐mediated adaptation was clearly observed in the hisC rewired strain at both population and microcolony levels. Furthermore, transcriptome analysis showed that such adaptation was associated with a global transcriptional reorganization. The stochastic switching of a monostable structure reported here strongly suggests that this stochastic strategy may represent a generic mechanism for adaptation in living organisms.
The hisC rewired strain and its response to histidine depletion
The hisC rewired E. coli OSU12‐hisC strain has been previously described (Kashiwagi et al, 2009). The gene hisC responsible for histidine biosynthesis was deleted from the His operon and placed under the control of the extraneous promoter, PtetA in an engineered gene circuit inserted at a different chromosomal location (Figure 1A). Consequently, hisC in OSU12‐hisC is no longer responsive to the native regulation (His operon) that senses histidine depletion, according to the genomic annotation around this locus. Instead, the foreign gene circuit provided a monostable structure (Supplementary Figure S1) for hisC's stochastic switching. The green fluorescent protein (GFP) (gfpuv5) was co‐expressed with hisC for the quantitative evaluation of HisC in single cells. The upstream regulation of TetR, whose expression level was reported by the red fluorescent protein (RFP) (dsred.T4), was introduced to achieve the inducible GFP (HisC) level. The full induction of TetR by isopropyl β‐d‐1‐thiogalactopyranoside (IPTG) was applied to avoid any possible upstream noise that caused by the abundance of endogenous LacI. The doxycycline‐dependent growth of OSU12‐hisC has been characterized previously (Kashiwagi et al, 2009). Accordingly, the full induction of both Ptrc and PtetA, that is, the addition of 100 μM IPTG and 100 nM doxycycline, was employed here to study how the stochastic switching of hisC allows the cells to grow in histidine‐free conditions.
Source data for Figure 1 [msb201124-sup-0002-SourceData-S2.xls]
Under the described condition, the response of OSU12‐hisC to histidine depletion was examined. Cells initially grown in the presence of 1 mM histidine were inoculated into fresh medium with the same induction but without histidine. Microscopic observation clearly revealed that the cells showed stronger green fluorescence after histidine depletion (Figure 1B), which suggested an increased expression level of hisC. Population analysis using flow cytometry was performed to detect the difference in GFP (HisC) level of the cell populations newly formed in the presence and absence of histidine. It showed that the distributions of both GFP concentration and GFP bias (GFP/RFP ratio) in OSU12‐hisC shifted towards a higher level in histidine‐free conditions (Figure 1C and D), whereas the depletion caused only a slight change in distributions of OSU11, a control strain carrying both the same engineered genetic circuit and an intact His operon, including the hisC gene in its native context (Supplementary Figure S2). Repeated experiments revealed that the increases in both GFP concentration (∼2.1 folds) and GFP bias (∼1.5 folds) due to histidine depletion were highly significant (P<0.005, N=6) in OSU12‐hisC (Supplementary Figure S3). In particular, the increased GFP bias strongly suggested that the change in gene expression occurred specifically in the rewired hisC (i.e., GFP) but not in all genes (e.g., RFP).
Population shift along with growth recovery and its relaxation
The temporal changes in fluorescent intensity of OSU12‐hisC in response to histidine depletion were investigated. Exponentially growing cells in histidine‐supplied conditions were transferred to fresh medium in the presence or absence of histidine, and the timed sampling was subsequently performed at 2 h intervals. OSU12‐hisC did grow in the absence of histidine, but the growth rate was slower than that in the presence of histidine (Figure 2A). In addition, the growth of OSU11 was approximately equivalent in both nutritional conditions, and the hisC‐deleted strain, OSU11ΔhisC (genotype in Supplementary Figure S2), failed to propagate under conditions of starvation (Supplementary Figure S4). These results verified that the depletion of histidine did bring severe stress to the cells, due to the rewiring of the starvation‐stringent gene, hisC.
The cell population of OSU12‐hisC shifted gradually to higher levels of both GFP concentration and GFP bias in histidine‐free conditions, but maintained the initial level in histidine‐supplied conditions (Figure 2B). The cell growth rates were evaluated according to the cell concentrations between every two measurements (sampling points). Both the growth rate and the GFP level stayed approximately constant, when OSU12‐hisC grown in the presence of histidine (Figure 2C, black cycles). On the contrary, the temporal trajectory of the relation between cell growth and GFP level showed that the growth rate dropped off markedly with histidine depletion, but recovered gradually within a few hours (Figure 2C, green cycles). In the later period, the growth rate recovered accompanied by an increase in GFP (HisC) concentration. The gradual shift along with growth recovery in response to starvation was clearly identified in the independent replications (Supplementary Figure S5). In response to external perturbations, the changes in gene expression relying on a stochastic strategy were reported to be generally slower than the changes due to operon regulation (Keller and Calvo, 1979; Zaslaver et al, 2004; Yamada et al, 2010). The slow recovery in either GFP concentration or growth rate was therefore assumed to be dependent on the stochastic switching of hisC located in the monostable circuit.
Furthermore, the population shift against the environmental transition showed reversibility. The cell population remained steady at the high GFP (HisC) level in the absence of histidine. However, once exponentially growing OSU12‐hisC under starved conditions was supplied with 1 mM histidine, the population distribution returned to a low level of GFP (HisC) within 3 h (Figure 3). That is, the reduction in GFP (HisC) level of the whole population (∼106 cells/ml) took place within only 1–2 generations (i.e., 14–17 h). As described previously, in contrast to bistable structures, monostability did not fix the phenotype but allowed the cells to adapt dynamically. The relaxation to the initial steady state in response to stress release in a stochastic switching‐dependent manner could only be observed only in a monostable structure. Additionally, the restoration also demonstrated that the population shift was neither caused by the mutation in genome nor due to the efficiency in GFP maturation.
Directionality in stochastic switching initiated adaptation
To examine how the stochastic switching occurring in individual cells initiated adaptation, 30–50 microcolonies were characterized in the presence or absence of histidine at time points of 0, 2 and 4 h. Cells in the logarithmic growth phase were placed on glass coverslips and covered with a thin layer of agarose gel pad medium in the presence or absence of histidine, respectively. Microcolony formation from a single cell was observed under the microscope (Figure 4A). Variation in cellular GFP level was clearly observed in individual cells (Figure 4B; Supplementary Figure S6). Stochastic switching of hisC was verified according to the random changes in GFP bias along with the cell division under histidine‐rich conditions. For example, one of four daughter cells showing stronger green fluorescence were born from a single cell 2 h after shifting to histidine‐supplied conditions, but one of seven daughter cells showed weaker green fluorescence 4 h later (Figure 4B, first row, +His). On the other hand, the microcolonies formed under the histidine‐free conditions tended to higher levels of GFP bias (Figure 4B; Supplementary Figure S6). For example, two daughter cells of higher GFP bias appeared 2 h after histidine depletion; subsequently, four daughter cells with stronger or similar GFP level were born (Figure 4B, first row, −His). The directional tendency favoured the high GFP (HisC) level, which was evidently detected in the first 2 h after histidine depletion, resulting in a population shift (Figure 4C). In contrast, the distributions of microcolonies grown in histidine‐supplied conditions kept steady, due to the randomized directions of stochastic switching (Figure 4C). In addition, the restoration to lower GFP bias upon addition of histidine (i.e., stress release) was clearly observed at the microcolony level as well (Supplementary Figure S7). Such dynamic changes in microcolonies were consistent with the population dynamics (Figures 2 and 3).
As the preferential direction was evident from 0 to 2 h (weaker after 2 h) (Figure 4; Supplementary Figure S8), the initial cellular status that shaped the subsequent adaptation was further analysed. GFP bias and growth rates were initially partially correlated (∼0.51, P<0.05) for microcolonies growing in the absence of histidine. In comparison, those growing in the presence of histidine showed no correlation (Figure 5A). This observation suggested that the initial state may have a role in determining, to some extent, the growth rate. Moreover, a weak positive correlation (∼0.52, P<0.01) was found between the initial GFP bias and the subsequent changes in cellular status (i.e., relative change in GFP bias) only in histidine‐free conditions (Figure 5B). In particular, under starved conditions, higher initial GFP bias and larger relative changes were detected in the growing cells than in the non‐growing cells. Taken together, these results indicated that the stochastic fluctuations in the initial state had an important role not only in fate decision (i.e., whether to grow) but also in the directionality of the stochastic switch. Note that the unfit cells not only depended on the initial GFP level but are also born stochastically from the growing fit microcolonies (Supplementary Figure S9).
Source data for Figure 5A [msb201124-sup-0003-SourceData-S3.xls]
Transcriptional reorganization for adaptation
As the changes in expression of a single gene, hisC, are generally insufficient for the cells to survive to histidine starvation, transcriptome analysis was additionally performed to investigate global alterations in gene expression in the OSU12‐hisC strain. Cells exponentially growing in the presence or absence of histidine were collected for microarray profiling. The steady expression pattern of OSU12‐hisC was close to that of OSU11 in the presence of histidine, but different in the absence of histidine, according to the scatter plot of total 4467 genes (Figure 6A). A considerably larger number of genes up‐ or downregulated in response to histidine depletion was observed in OSU12‐hisC than in OUS11, and the majority of these genes did not overlap between the two strains (Figure 6B). In total, 225 and 200 genes that were specifically upregulated or downregulated in OSU12‐hisC, respectively, were further characterized according to their categories as previously reported (Riley et al, 2006). The genes in the categories of RNA (n), structural component (s) and transporter (t) were mostly downregulated (Table I). The genes encoding tRNAs, ribosomal proteins and fimbrial proteins were found in these categories, for example, reduced swimming caused by repressed translation as an energy‐saving tactic, which is known as a stringent response (Srivatsan and Wang, 2008; Traxler et al, 2008). The upregulated genes were significantly clustered in the categories of factors (f and pf), unknown function (o) and predicted structural component (ps). These observations strongly suggested that diverse adaptive cellular states were achieved both by the responsive switching that relied on operon regulation and by the stochastic switching of the rewired gene.
Source data for Figure 6A [msb201124-sup-0004-SourceData-S4.xls]
In addition, the transcriptional changes of the structural genes in the His operon were greater in OSU12‐hisC than in OSU11 (Figure 6C; Supplementary Figure S10). In particular, hisG and hisD, located downstream of the promoter, were significantly induced. The levels of expression of genes downstream of the native hisC locus (hisB–hisI) increased as well, in a location‐dependent manner. The enhanced expression of these genes probably compensated for the perturbed native regulation and facilitated sufficient production of histidine. We assume that the transcriptional reorganization improved the expression levels of multiple genes crucial for growth recovery in a synchronized manner with the occasionally increased hisC expression.
Bistable structures are considered as evolved designs for adaptation (Acar et al, 2008), whereas monostability relies on a simpler design, which is much more common in regulatory networks (Newman et al, 2006; Taniguchi et al, 2010). In this study, we showed that adaptation to starvation can be successfully achieved by stochastic switching of hisC under a monostable control (Figure 2). The results showed that the cell population remained at an unstable but adaptive state under starved conditions, but returned to the original stable state once histidine was supplied (Figure 3). It indicated that the selective proliferation of fit cells was sufficiently high to compensate the relaxation of the monostable control structure to the stable but unfit state. This probably resulted from comparable time scales of both the stochastic changes in protein level and the cell growth rate, in accordance with the theoretical proposal that an adaptive state can be fixed by biased growth (i.e., the balance of frequency in protein fluctuation with generation time) in a monostable structure (Sato and Kaneko, 2006). In summary, the present study provides experimental evidence that stochastic switching can contribute to adaptation without a complex responsive or bistable design.
In contrast to previous gene rewiring studies using native regulation (Stolovicki et al, 2006; Stern et al, 2007; Isalan et al, 2008), a foreign regulation strategy was adopted here, to minimize disturbances from the native regulatory system. An inducible cascade circuit was employed here to taken into account stochastic noise in gene expression, which depends markedly on the gene function and its regulation (Newman et al, 2006; Taniguchi et al, 2010). This genetic structure has been previously widely used (Pedraza and van Oudenaarden, 2005; Rosenfeld et al, 2005; Kashiwagi et al, 2009; Tsuru et al, 2009), and its monostability and stochasticity were clearly characterized by varying the concentrations of inducers (Pedraza and van Oudenaarden, 2005; Rosenfeld et al, 2005; Tsuru et al, 2009). Under induced conditions, a gradual change in both GFP concentration and GFP bias was observed together with a unimodal distribution characteristic of a monostable structure (Pedraza and van Oudenaarden, 2005; Rosenfeld et al, 2005; Tsuru et al, 2009). In the present study, the full induction of TetR (i.e., addition of 100 nM doxycycline) was used to break the cascaded structure. The stochastic expression of hisC was only due to the promoter PtetA without any upstream contribution, as the full induction avoided amplification of the noise in the upstream promoters, that is, both the foreign Ptrc and the endogenous PlacI.
As the cascaded gene circuit comprised both GFP and RFP, it was used as a two‐colour system to identify the specific changes in GFP (HisC) level according to the ratio between GFP and RFP (i.e., GFP bias). The addition of 100 μM IPTG not only masked the intrinsic LacI to maximize the TetR level, but also intensified the red fluorescence to reach a steady RFP level, which facilitated the precise estimation of GFP bias. In contrast to the GFP concentration (GFP/FS) that maintained approximately steady independent on IPTG, the GFP bias increased considerably due to the absence of IPTG (Supplementary Figure S11). Nevertheless, the shift of the OSU12‐hisC population to higher GFP (HisC) levels in response to histidine depletion was independent of the addition of IPTG (Supplementary Figure S11). Additionally, the steady distributions of OSU12‐hisC and OSU11 were quite different in GFP levels even under the same inductions (Figure 1C and D; Supplementary Figures S3 and S11). Actually, it remains unclear why gene expression levels changed even under identical regulation, an observation which was reported previously as well (Stolovicki et al, 2006; Stern et al, 2007; Isalan et al, 2008). The increase in GFP (HisC) levels in response to starvation occurred in OSU12‐hisC but not in OSU11, in spite of the fact that both strains had reached their maximal levels of GFP (HisC) before histidine depletion. In conclusion, the increase in GFP (HisC) was a phenomenon unique to the adaptation of the engineered OSU12‐hisC strain.
We propose that the adaptation of OSU12‐hisC consists of two scenarios: (1) the directional response of individual cells towards high level of HisC was initiated in the early period (0–2 h); (2) rapid growth of stochastically appearing cells with high HisC level overtook the population in the late period (after 2 h). These scenarios were also supported by the population dynamics (Figure 2). For instance, in the initial period of adaptation (0–2 h), the population increase, which represents both the growth of the cells with fixed expression level and the transition of the cells from other expression levels, dropped to negative at low GFP level, but increased markedly at high GFP level (Supplementary Figure S12). It suggested that cells with low GFP level switched high GFP levels and accelerated the population shift. In addition, the population increase remained constant independent on GFP concentration in the presence of histidine, but largely varied in the absence of histidine (Supplementary Figure S12). It indicated that only cells with high GFP (HisC) level could proliferate.
Moreover, the transcriptome analysis showed that a large number of genes were involved in the stochastic switching‐induced adaptation (Figure 6). The global transcriptional change must be driven somewhat directionally; otherwise an extremely long time would be required for all of these genes to reach the appropriate state in a stochastic manner. Such directionality was suggested by the stochastic switching of hisC in response to starvation (Figure 4). Since adaptation by stochastic switching is initiated by the proliferation of a very small number of occasionally appearing fit cells, the process tends to be slow (Koch, 1988; Nguyen et al, 1989; Dong et al, 1995; Kussell and Leibler, 2005). The directionality observed here may have had a role in increasing the initial number of fit cells and improved the efficiency of stochastic switching‐mediated adaptation. It may also result from evolving mechanisms that are intermediate between stochastic switching and signal‐dependent regulation. Regulatory mechanisms that respond to a specific environmental condition, for example, signal transduction, may be a consequence of evolutionary canalization (Waddington, 1959; Eldar et al, 2009; Raj et al, 2010) and eventual fixation of the primitive multireactivity of heterogeneous cell populations, once such conditions become frequent.
Materials and methods
The E. coli strains OSU11, OSU11ΔhisC and OSU12‐hisC (Supplementary Figure S2) were described previously (Kashiwagi et al, 2009). These strains commonly carry a core gene circuit, PtetA–gfpuv5 and Ptrc–dsred.t4–tetR–PEM7–zeo, on the genome, located at galK and intC sites, respectively. PtetA–gfpuv5 consists of the promoter PtetA and the mutated gfp gene (gfpuv5) encoding the GFP, and Ptrc–dsred.t4–tetR–PEM7–zeo is comprising the promoter Ptrc, dsred.t4 encoding RFP, tetR encoding the repressor protein, and the independent expression unit of the Zeocin resistance gene, zeo, with its promoter PEM7. OSU11ΔhisC is the hisC knockout mutant of OSU11 carrying the rewired hisC gene downstream of gfpuv5.
E. coli cells were grown in minimal medium, modified M63 (mM63: 62 mM K2HPO4, 39 mM KH2PO4, 15 mM (NH4)2SO4, 2 μM FeSO4·7H2O, 15 μM thiamine hydrochloride, 203 μM MgSO4·7H2O and 22 mM glucose) (Kashiwagi et al, 2009) in the presence or absence of 1 mM histidine. OSU12‐hisC was grown in mM63 medium supplied with 25 μg/ml kanamycin and 1 mM histidine. Cells were cultured at 37°C for several passages until the growth rate became stable. Full expression of RFP was induced in the presence of 100 μM isopropyl IPTG. Exponentially growing cells were subsequently transferred to fresh medium supplemented with 100 nM doxycycline hydrochloride (Dox) to induce the expression of GFP. The initial cell concentration was ∼104 cells/ml.
Bacterial cells were grown in mM63 medium supplemented with 25 μg/ml kanamycin, 1 mM histidine, 100 μM IPTG and 100 nM Dox for 18 h as described above. Aliquots of 300 μl of exponentially growing cells (about 106 cells/ml) were harvested by centrifugation at 5000 r.p.m. (or 2300 g) for 30 s at 37°C using spin columns (Ultrafree‐MC Centrifugal Filter Units, 0.2 μm; Millipore). After discarding the flow‐through fraction, the cells were washed with 300 μl of the same medium minus histidine. After repeating centrifugation and washing processes twice, the concentration of the cell suspension was measured by flow cytometry and inoculated into 5 ml of the same medium without histidine, where the initial cell concentration was ∼104 cells/ml.
Gene expression (fluorescence intensity) and cell volume were evaluated using a flow cytometer (FACSAria cell sorter; Becton Dickinson) with a 488‐nm argon laser, a 515–545 nm emission filter (GFP) and a 563–589 nm emission filter (RFP). The following PMT voltage settings were applied: forward scatter (FSC), 203; side scatter (SSC), 440; GFP, 615; and RFP, 593. The flow data were converted to TXT format. The cellular GFP concentration was calculated by dividing the green fluorescence value by the FSC value using MATLAB (The MathWorks Inc.), as described previously (Tsuru et al, 2009). Autofluorescence was subtracted. Systematic errors resulting from events that occurred at the bottom or top of the instrument's range were eliminated. Data from 1000 to 10 000 cells were collected for each measurement. Cell samples mixed with fluorescent beads (Floresbrite YG Microspheres, 2 μm; Polysciences Inc.) were loaded for calculation of cell concentration.
Microscopic observation and image acquisition
OSU12‐hisC was grown overnight in mM63 medium supplemented with 25 μg/ml kanamycin, 1 mM histidine, 100 μM IPTG and 100 nM Dox, at 37°C. Cells (4 μl of cell culture) in the logarithmic growth phase were placed on glass coverslips and covered with a thin layer of agarose gel (1.5%) pad medium containing 25 μg/ml kanamycin, 100 μM IPTG and 100 nM Dox, with or without 1 mM histidine. The cover glasses with the inoculated cells were incubated at 37°C and observed every 2 h. The fluorescence images were acquired using a fluorescence microscope (IX70; Olympus) and recorded using a CCD camera (VB‐6010; Keyence).
Time series images of the green and red fluorescence were used to analyse the fluorescence (green for GFP FI and red for RFP FI) and shape of cells. Phase contrast images were used for identification and segregation of cells. Shapes of camera‐captured cells were outlined using ImageJ (http://rsbweb.nih.gov/ij/) automatically and manually, respectively. The area of cells was calculated by summing the number of fluorescent pixels. Growth rates were determined by an exponential fit to the cells’ area over sampling time intervals. Green and red fluorescence of each microcolony or each cell were determined by averaging pixel intensity. GFP bias (GFP FI/RFP FI) was calculated by dividing green fluorescence by red fluorescence. The relative change in GFP bias was calculated over sampling time intervals, and 30–50 microcolonies (up to ∼270 cells) were measured.
Sample preparation and microarray
Exponential cell growth was interrupted by putting the cell culture into cold phenol–ethanol solution (1 g of phenol in 10 ml of ethanol) prepared in advance. The cells were collected by centrifugation at 16 000 g for 5 min at 4°C, and the pelleted cells were stored at −80°C before use. Total RNAs were extracted using an RNeasy mini kit (Qiagen) in accordance with the manufacturer's instructions. The quantity of the purified RNA was carefully determined by the absorbance at 260 nm using a NanoDrop ND‐1000 (Thermo Fisher Scientific Inc.), an Agilent 2100 Bioanalyzer with an RNA 600 Nano kit (Agilent Technologies) and electrophoresis on an agarose gel under reducing conditions. High‐density oligonucleotide microarray gene expression analysis was performed using an Affymetrix GeneChip system. The cDNA synthesis, fragmentation, labelling and hybridization of cDNA were carried out according to the Affymetrix GeneChip Expression Analysis Technical Manual. To improve the efficiency of labelling, the incubation time was increased to 2 h. Three to five independent experiments (sample preparations) were performed for each condition.
A high‐density DNA microarray (Suzuki et al, 2007; Ono et al, 2008) covering the whole genome of E. coli W3110 strain (GenoBase; http://ecoli.naist.jp/GB6/search.jsp) was utilized. Both strands of the genome were alternately tiled at single‐base resolution using 21‐mer perfectly matching and three mismatching probes. A library consisting of a total of 18.6 million probes was arranged on three GeneChips for microarray analysis, covering four types of probe throughout the 4.6‐Mb E. coli genome. One of three GeneChips was used here for transcriptional analysis.
Data extraction and analysis
Microarray data were extracted using custom scripts written in R (Ihaka and Gentleman, 1996) based on the finite hybridization model (Ono et al, 2008) and the thermodynamic model of non‐specific binding on short nucleotide microarrays (Furusawa et al, 2009). Three data sets from 3 to 5 repeated microarray assays were used for mRNA profiling analysis. To avoid any significant error due to the diverse signal intensities in GeneChips, the array results only showing the identical distribution of probe fluorescence intensity were subjected to subsequent expression analysis. A total of 12 array results for two strains under two conditions (in triplicate) were finally used. The microarray raw data sets have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE28210 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE28210). In all, 4467 genes were used for the analysis and global normalization was performed for 12 data sets of mRNA transcriptional level. The gene names and the gene categories are based on the genome information of W3110 and the genes in common to strains W3110 and MG1655 (Riley et al, 2006). The whole data set of gene names and categories is from GenoBase, Japan (http://ecoli.naist.jp/gb6/Download.html).
We thank Natsuko Yamawaki, Junko Asada and Ruriko Ogawa for technical assistance; and Naoaki Ono and Chikara Furusawa for analytical assistance. We thank Junya Ichinose and Kunihiko Kaneko for helpful discussions and comments. This work was partially supported by the Grants‐in‐Aid for Challenging Exploratory Research 22657059 (to BWY) and the ‘Global COE (Centers of Excellence) program’ of the Ministry of Education, Culture, Sports, Science and Technology, Japan.
Author contributions: ST, NY and YM performed the experiments; ST, NY, YM, BWY and TY analysed the data; JU, KM, AK and SS provided materials and experimental tools; BWY, ST and TY wrote the paper.
Conflict of Interest
The authors declare that they have no conflict of interest.
Supplementary figures S1–12 [msb201124-sup-0001.pdf]
Source data for Supplementary figure S3 [msb201124-sup-0001-SourceData-S1.xls]
This is an open‐access article distributed under the terms of the Creative Commons Attribution License, which permits distribution, and reproduction in any medium, provided the original author and source are credited. This license does not permit commercial exploitation without specific permission.
- Copyright © 2011 EMBO and Macmillan Publishers Limited