Development and characterization of Escherichia coli triple reporter strains for investigation of population heterogeneity in bioprocesses

Background Today there is an increasing demand for high yielding robust and cost efficient biotechnological production processes. Although cells in these processes originate from isogenic cultures, heterogeneity induced by intrinsic and extrinsic influences is omnipresent. To increase understanding of this mechanistically poorly understood phenomenon, advanced tools that provide insights into single cell physiology are needed. Results Two Escherichia coli triple reporter strains have been designed based on the industrially relevant production host E. coli BL21(DE3) and a modified version thereof, E. coli T7E2. The strains carry three different fluorescence proteins chromosomally integrated. Single cell growth is followed with EmeraldGFP (EmGFP)-expression together with the ribosomal promoter rrnB. General stress response of single cells is monitored by expression of sigma factor rpoS with mStrawberry, whereas expression of the nar-operon together with TagRFP657 gives information about oxygen limitation of single cells. First, the strains were characterized in batch operated stirred-tank bioreactors in comparison to wildtype E. coli BL21(DE3). Afterwards, applicability of the triple reporter strains for investigation of population heterogeneity in bioprocesses was demonstrated in continuous processes in stirred-tank bioreactors at different growth rates and in response to glucose and oxygen perturbation simulating gradients on industrial scale. Population and single cell level physiology was monitored evaluating general physiology and flow cytometry analysis of fluorescence distributions of the triple reporter strains. Although both triple reporter strains reflected physiological changes that were expected based on the expression characteristics of the marker proteins, the triple reporter strain based on E. coli T7E2 showed higher sensitivity in response to environmental changes. For both strains, noise in gene expression was observed during transition from phases of non-growth to growth. Apparently, under some process conditions, e.g. the stationary phase in batch cultures, the fluorescence response of EmGFP and mStrawberry is preserved, whereas TagRFP657 showed a distinct response. Conclusions Single cell growth, general stress response and oxygen limitation of single cells could be followed using the two triple reporter strains developed in this study. They represent valuable tools to study population heterogeneity in bioprocesses significantly increasing the level of information compared to the use of single reporter strains.

robust high-yielding bioprocesses for a biotechnological product of interest, ideally, all cells should exhibit the same constant and high productivity. However, although single cells in a population in bioreactors on an industrial scale originate from isogenic cultures, they can be metabolically diverse and are known to respond differently to environmental fluctuations [2][3][4][5]. This population heterogeneity leads to loss in process efficiency especially when, due to non-ideal mixing, gradients of process variables arise inside the reactor creating, various local microenvironments [3]. Cells experience these while moving through the reactor, each following an individual path. Additionally to this environmentally induced, extrinsic heterogeneity, further heterogeneity is added by intrinsic perturbations due to the stochasticity of reactions of intracellular molecules that is caused by their low abundance in a cell [3]. Consequently, it is essential to understand single cell physiology to keep population heterogeneity induced yield reduction in industrial scale bioprocesses to a minimum.
Even though population heterogeneity is well-studied as appearing in bioprocesses and especially in response to gradients or different kind of stress [1,3,6], mechanistic understanding of this phenomenon is poor. So far, it is only suspected that cells in a bioreactor adapt strategies they also apply in natural habitats to cope with changes in their environment. One of these strategies is noise in gene expression, which is even apparent in a stable environment and was found to affect bioprocess efficiency [6][7][8][9]. Thereby, cells diversify their expression profile of certain proteins to have a broader range of response possibilities to environmental changes [7,10,11]. Noise influences essential cellular functions like regulation, survival, differentiation and the stress response by a global regulatory feedback system [7,[12][13][14]. In studies of noise in gene expression an inverse correlation between gene expression and noise could be established [8]. From these findings and the discovery that stress-related genes are noisier than essential genes, the hypothesis was raised that noise in gene expression is used as a regulatory strategy to adjust the level of population heterogeneity depending on whether it is beneficial for the cell population or not [15,16].
Population heterogeneity in a bioprocess can be conveniently visualized using reporter strains [1,17]. Reporter strains carry genes for reporter proteins, whose expression can be easily monitored, integrated in their genome, so that they are expressed together with a protein of interest. This enables the collection of snapshots of single cell dynamics following a bioprocess. Different types of reporter strains exist [17,18]. However, for investigation of population heterogeneity in bioprocesses mostly promoter-based reporter strains are applied.
Their fluorescent protein is fused to a promoter responsive to a cellular event of interest. Consequently, fluorescence intensity can be correlated to promoter activity [3,19]. Possible targets for integration of fluorescent proteins can be confirmed with transcriptomics data, as they allow to predict if a promoter is activated under the conditions that are to be tracked in a bioprocess, e.g. shifts in environmental conditions, exposure to stress of different kind or cellular growth [20][21][22][23]. Reporter strains have already been proven to be useful tools in numerous studies of population heterogeneity in lab-scale simulating industrial scale bioprocesses. Examples are investigation of nutrient and oxygen limitation, study of the redox balance or identification of the most robust and best producing subpopulation of cells by integrating a fluorescent protein whose fluorescence output is proportional to product formation [24][25][26][27].
Apart from carefully choosing the promoter, the choice of the appropriate fluorescent protein is crucial. A bright spectrum of fluorescence proteins exist and have been employed for generation of reporter strains. The most common is green fluorescent protein (GFP), as its excitation and emission maxima (484/507 nm) fit well and result in high fluorescence intensity in the standard setting of a flow cytometer with a blue laser [28]. But also derivatives of GFP in different colours are applied like blue fluorescent protein (BFP), red fluorescent protein (RFP) and yellow fluorescent protein (YFP) as well as fluorescent proteins of the fruit series, e.g. mStrawberry, mCherry or mPlum (reviewed in [28][29][30]). When choosing fluorescent proteins, some characteristics should be considered like its maturation time as it is, together with its photostability, crucial when fast changes are to be monitored. For instance, while wildtype GFP has a maturation time of around 60 min in E. coli, the maturation time of EmeraldGFP (EmGFP), a modified version thereof, is just 12 min [31]. Furthermore, also a possible dependence of the fluorescence signal on environmental factors, like temperature, oxygen and pH has to be considered.
When investigating population heterogeneity in bioprocesses with reporter strains, the resulting fluorescence distributions are most often visualized with flow cytometry (for reviews see [32,33]). This high-throughput technique enables visualization of different subpopulations in a reactor, while at the same time also collecting information about single cells in a sample. As the instruments gradually advance, it is possible to simultaneously measure more and more fluorescence parameters with high resolution [3]. Hence, fluorescent proteins have to also be fitted to not interfere with each other, exhibit spectral interaction or distributions that are superimposed. Additionally, their fluorescence characteristics (excitation/emission) have to be compatible with the flow cytometers' lasers and filters [1]. However, so far mostly single reporter strains, expressing one fluorescent protein to follow one cellular event of interest, are used when studying population heterogeneity [3,12]. Then, other cellular characteristics remain in the dark, if one does not deploy global omics-methods. Nevertheless, even these are only partly available with single cell resolution or possible with big invest in labour and time [34,35].
Our goal was to construct and characterize triple reporter strains based on one of the most important industrial production hosts E. coli BL21(DE3) and a modified version thereof, E. coli T7E2, which enable to follow three essential cellular characteristics simultaneously to raise the information content gained from studies of population heterogeneity in bioprocesses. To track growth rate of single cells, the strains express EmGFP together with the ribosomal promoter rrnB. Furthermore, mStrawberry is integrated downstream of the alternative sigma factor 38, whose expression is correlated with the general stress response of single cells. The third marker is integrated to follow oxygen limitation of single cells. For that purpose, TagRFP657, a variant of RFP, is placed in the narGHIJ gene cluster. Since the expression of these three characteristics are interlinked, the triple reporter strains should also enable to observe correlations between the different markers. All markers have been previously employed as single reporters of population heterogeneity in bioprocesses, but never combined in one experimental setup that enables the direct correlation of the three cellular characteristics on single cell level [22,36,37]. This is to our knowledge the first time to employ a triple reporter strain to investigate population heterogeneity in bioprocesses. The only other approach we are aware of, was built to study promoter activities related to quorum sensing, biosynthesis of galactoglucan and viability with three different fluorescent proteins [38].
In E. coli the number of ribosomes is proportional to the growth rate to fulfil the requirements for protein synthesis [39]. The ribosomal promoter rrnB is part of one of two operons compromising in total seven copies of rrn in the cell. Since the rrnCABE-operon is located in close vicinity to the origin of replication on the chromosome, its number is significantly increased in fast-growing cells compared to slow-growing cells. Thus, the rate of rRNA synthesis will be higher in fast-growing cells due to their increased gene dosage. The promoter activity can therefore be correlated to single cell growth which was also shown in transcriptomics studies and studies of cells from batch and chemostat cultures [19,36,40].
Expression of the alternative sigmafactor 38 rpoS is induced in response to various kind of stress, e.g. starvation and nutrient limitation and influences the expression of around 500 genes to enable the cell to cope with the stress experienced. Thereby it has, among others, influence on intracellular pH regulation, the metabolism and its flexibility and the growth rate [41]. Furthermore, rpoS expression plays an important role when cells change from exponential growth to stationary phase, which is accompanied by a steep increase in rpoS levels [22]. Earlier, rpoS reporter strains could reveal differences between the general stress response of cells in well-mixed fed-batch processes and the equivalent scaledown reactor [12].
The narGHIJ operon was found to be upregulated upon dissolved oxygen depletion [8]. It codes for the alpha subunit of a nitrate reductase, which allows E. coli, in conjugation with NADH-dehydrogenase, to use nitrate as an electron acceptor during anoxic growth while still gaining energy through oxidative phosphorylation [42,43]. In cultivation studies its expression was found to be upregulated when oxygen saturation in the extracellular environment drops below 40% [37].
Escherichia coli BL21(DE3) has several advantageous characteristics like low acetate production at high glucose levels as well as being protease deficient [44]. The second strain used in this study, E. coli T7E2, should be more robust than its ancestor in an industrial cultivation setup [45]. Its modifications are related to removal of major parts of the lambda DE3 prophage and replacement of the cryptic prophage Rac by the pgl gene encoding 6-phosphogluconolactonase which BL21(DE3) has lost and that prevents accumulation of sugar-intermediates.

Triple reporter strains
Two Escherichia coli triple reporter strains, E. coli G7 BL21(DE3) and E. coli G5 T7E2 , expressing three chromosomally integrated fluorescent proteins connected to the expression of three different cellular markers (see below and Additional file 1: Table A1) were characterized in this study. E. coli G7 BL21(DE3) originates from wildtype E. coli BL21(DE3), whereas E. coli G5 T7E2 is based on a modified version of E. coli BL21(DE3), T7E2 [45]. Additionally, during initial characterization experiments the corresponding single-and double reporter strains were employed (see Additional file 1). All strains were synthesized by Gene Bridges (Heidelberg, Germany) and stored as 15% glycerol stocks at − 80 °C. Wildtypes E. coli BL21(DE3) and E. coli T7E2 were used as experimental reference.

Genetic modifications
Wildtype E. coli BL21(DE3) as well as E. coli T7E2 were altered by a series of knock-in recombinations with synthetic cassettes. Each reporter protein was thereby under transcription control of a certain native promoter. The mStrawberry gene was inserted downstream of rpoS in conjunction with its own ribosomal binding site (RBS) without removing any native sequence. The second reporter TagRFP657 is placed behind the narGHIJ gene cluster without removing any native sequence. For the third marker a synthetic copy of the rrnB promoter complex together with EmGFP(A206K) and a consensus RBS in front and a transcriptional terminator behind the open reading frame (ORF) was integrated in the rhamnose operon. The knock-in was performed removing 833 bp of native sequence (5′-terminal 293 bp of rhaS to 5′-terminal 453 bp of rhaB). In all of these transcriptional fusions a strong synthetic RBS (5′-AAA GAG GAG AAA -3′) [46] was used, followed by six intervening nucleotides until the start of the respective ORF, to ensure efficient translation of the fluorescence protein. Each of the three DNA cassettes was integrated into the bacterial genome in connection with bilaterally recombinase flippase (FLP) recognition target sequence [47] flanked antibiotic resistance markers. This allowed for subsequent removal of antibiotic markers by FLP recombination steps, so that none of the strains carries an antibiotic resistance. Correct insertion and orientation were confirmed by amplification of the corresponding loci using locus specific primer via PCR and subsequent Sanger sequencing of the obtained amplicon (data not shown).

Pre-cultures
Single colonies from minimal medium [48] agar plates of the triple-reporter strains and the respective wildtypes were used to inoculate 50 mL minimal medium [48] cultures in 500 mL baffled shake flasks. The flasks were shaken for around 6 h at 150 rpm and 37 °C. The minimal medium was composed of 13.3 g L −1 KH 2 PO 4 , 4 g L −1 (NH 4 ) 2 HPO 4 , 1.859 g L −1 citric acid ·H 2 O, 20 g L −1 glucose, 1. Zn(CH 3 COO) 2 *2H 2 O and 3 g L −1 Fe(III)citrate). To avoid foaming the medium was supplemented with 1 mL −1 Antifoam 204 (Sigma Aldrich, MO, USA). Stirred-tank bioreactors were inoculated with the volume necessary to achieve an optical density at 600 nm (OD 600 ) of 0.02.

Batch processes
The stirred-tank bioreactor employed in this study was a 1.5 L reactor (Labfors 5, Infors AG, Switzerland) with a working volume of 1.2 L. Cells were cultured in minimal medium according to Riesenberg et al. [48] with an initial glucose concentration of 20 g L −1 . Sensors for pH and dissolved oxygen (EasyFerm Bio HB and VisiFerm DO ECS, respectively, Hamilton Bonaduz AG, Switzerland) were calibrated according to standard procedures using a two-and one-point calibration, respectively (pH 4.02 and pH 9.21, calibration for 100% dissolved oxygen). The pH was controlled at pH 6.8 with 1 M ammonia. Temperature and stirring were kept constant at 37 °C and 1000 rpm. Aeration was 2 vvm in the beginning and automatically increased in 0.5 steps until 8 vvm to avoid oxygen limitation. Continuous analysis of off-gas carbon dioxide (CO 2 ) was performed using a gas sensor (Blue-InOne Ferm, BlueSens, Germany). Samples for OD 600 , high performance liquid chromatography (HPLC), dry cell weight (DW) measurements and flow cytometry (FC) analysis were withdrawn frequently following all phases of a typical batch culture of E. coli. Samples for OD 600 , FC analysis and DW were analyzed directly. Samples for HPLC analysis were centrifuged for 5 min at 16.438 g, filtrated (pore size 0.2 µm) and kept at − 20 °C until analysis. All experiments were performed in triplicates and variation is considered using error bars.

Continuous processes
Aerobic, glucose-limited, level-based chemostat cultures were started after an initial batch phase performed as described above. After glucose depletion, detected by a rapid rise in dissolved oxygen in the reactor, cultures were switched to chemostat operation. A constant feed with the same composition as the batch medium was applied to reach respective dilution rates. Steady state was confirmed by constant values for O 2 , exhaust CO 2 and DW after five mean hydraulic residence times. When steady state was verified, cultures were either sampled at different dilution rates or perturbed with additional substrate and oxygen limitation.

Change in dilution rate
Chemostat cultures with the two triple reporter strains were run with different dilution rates. For E. coli G7 BL21(DE3) , chemostats were run at dilution rates of 0.1 h −1 , 0.2 h −1 , 0.25 h −1 , 0.3 h −1 and 0.4 h −1 . For E. coli G5 T7E2 , chemostats were run at dilution rates of 0.14 h −1 , 0.2 h −1 , 0.3 h −1 and 0.4 h −1 . After verification of steady state, three samples for OD 600 , DW, HPLC and FC analysis were taken during three successive residence times. Samples were taken in triplicates and treated as described for batch cultures. Variation between replicates is considered using error bars.

Pulses of glucose and oxygen limitation
Chemostat cultures were run with a dilution rate of 0.2 h −1 . After verification of steady state, cultures were simultaneously perturbed with a final concentration of 2 g L −1 additional glucose in the reactor and reduced oxygen availability. For this purpose, aeration was turned off for 30 min. The perturbation was followed by frequent sampling before (60 min and 30 min prior to perturbation), during shutdown of aeration (0 min, 5 min, 10 min and 20 min) and after perturbation (40 min, 60 min, 90 min and 120 min) for OD 600 , DW, HPLC and FC analysis. Samples were taken in triplicates and treated as described above. Variation between replicates is considered using error bars.

Sample analysis Optical density and dry cell weight
An UV-Vis spectrophotometer (Genesys 10uv-vis, Thermo Scientific, MA, USA) was used for at-line measurement of OD 600 . For biomass determination (DW), 2 mL cell suspension was centrifuged for 5 min at 16.438 g in pre-dried (48 h at 80 °C) and weighted Eppendorf tubes. The pellet was dried for 48 h at 80 °C, cooled down to room temperature (RT) and weighted. DW was determined by weight-difference between empty and filled tube. After measurement of OD 600 and DW in batch cultures, a correlation was established through linear regression (Additional file 2).

Flow cytometry analysis
Samples taken from the reactor were diluted in deionized water adjusted to the dilution used for OD 600 measurements. Diluted samples were centrifuged for 1 min at 16.438 g. The pellet was re-suspended in 1 mL sterile 0.9% saline (NaCl) solution by vortexing. Afterwards cells were kept at RT until analysis of EmGFP, mStrawberry and TagRFP657 fluorescence. For FC analysis a CytoFLEX (Beckman Coulter, CA, USA) equipped with two lasers 488 nm and 638 nm was used. The minimum laser power was 50 mW and the beam spot size 5 µm × 80 µm. The FC has nine detecting channels for FSC, SSC and seven fixed wavelengths (525/40 nm, 585/42 nm, 610/20 nm, 660/20 nm and 2× 780/60 nm) avoiding spectral overlap. To eliminate background noise, the detection thresholds were set in SSC. 100,000 events were recorded with a rate of ∼ 1000 events/s (around 20 µL/s). During measurement, fluorescence distributions were visualized with the FC proprietary software CytExpert. For all experiments, measured fluorescence was put in relation to respective negative controls (see Additional file 1). All obtained results were saved in FCS 3.0 format.

HPLC analysis
Concentrations of glucose, succinate, acetate, citrate, malate, lactate and formate were quantified by HPLC (Finnigan Surveyor, Thermo Fisher Scientific, Waltham, USA). Detection was performed by refractive index (RI) (RID Agilent 1200, Agilent Technologies, CA, USA) with a 300 mm × 7.8 mm Aminex HPX-87H ion exchange column (Bio-Rad, Hercules, CA, USA). 20 µL sample was injected into the mobile phase of 5 mM H 2 SO 4 (aq.), temperature was 60 °C and flow rate 0.35 mL min − 1 . The results from HPLC analysis were used together with DW measurements to calculate yield coefficients. Triplicate datasets are considered using error bars.

Data analysis
Data analysis was performed using MATLAB ® R2017a (The MathWorks, Inc., Natick, MA, USA). Flow cytometry files were loaded using the function fcsread and saved into mat files (by L. Balkay, University of Debrecen, Hungary, available on MATLAB ® central file sharing). Applying the MATLAB ® built-in hist function to the recording channels of the FC, cell count in each channel was saved and enabled histogram plots with relative cell count vs. fluorescence intensity for each detector. To account for variance in fluorescence distributions, which correlates for fluorescence originating from cellular characteristics with noise in gene expression, the coefficient of variance (CV) was determined. The CV is defined as the ratio between standard deviation and mean of a distribution, which were calculated using the MATLAB ® builtin functions mean and std. Consequently, plots for CV against mean fluorescence respectively plots for CV and mean fluorescence intensity were generated for all fluorescent proteins. Normalized mean EmGFP fluorescence was estimated dividing mean EmGFP fluorescence by OD 600 . Linear regression analysis was performed employing the MATLAB ® built-in polyfit function with a degree of one. The result was evaluated using the MATLAB ® built-in polyval function.

Batch cultivation of triple reporter strains in comparison to the wildtype
After preliminary characterization in shake flask cultures and confirmation that all fluorescence proteins are inducible under relevant bioprocess conditions and the expression of the fluorescent proteins is not a burden to the cells respectively their fluorescence is pH insensitive (see Additional file 1), biological triplicate batch cultivations in stirred-tank bioreactors with minimal medium were performed for the triple reporter strains E. coli G5 T7E2 and E. coli G7 BL21(DE3) as well as for the wildtypes E. coli BL21(DE3) and E. coli T7E2. Population level growth was followed by OD 600 and biomass concentration. Furthermore, consumed respectively produced metabolites including the off-gas composition were followed. On single cell level, growth (expression of rrnB-EmGFP), general stress response (expression of rpoS-mStrawberry) and oxygen limitation (expression of nar-TagRFP657) of the triple reporter strains E. coli G5 T7E2 and E. coli G7 BL21(DE3) were measured with flow cytometry.

Population level physiology
Generally, the wildtypes E. coli BL21(DE3) and E. coli T7E2 as well as the triple reporter strains, E. coli G7 BL21(DE3) and E. coli G5 T7E2 , showed a typical growth behaviour. After inoculation, a lag phase of around five hours for the wildtype E. coli BL21(DE3) and E. coli G7 BL21(DE3) was characterized by constant values for glucose concentration and dissolved oxygen as well as no significant rise in carbon dioxide or biomass concentration ( Fig. 1a, b, substrates and products). Furthermore, no production of organic acids could be detected (Fig. 1a, b, organic acids). Also E. coli T7E2 and E. coli G5 T7E2 exhibited a lag-phase, which however was around 1 h shorter than for the other strains (Fig. 1c, d, substrates and products). Following the lag phase, the cells grew exponentially consuming the glucose in the medium and producing carbon dioxide, biomass, acetate and small amounts of malate and formate (see Fig. 1a-d, exponential phase indicated as grey area). While the wildtype E. coli BL21(DE3) and the triple reporter strain E. coli G7 BL21(DE3) did not produce any further metabolites, Fig. 1 Biomass, metabolites (glucose, acetate (blues squares), malate (green left-pointing triangle) and formate (red right-pointing triangle)) concentrations, off-gas carbon dioxide and dissolved oxygen following batch cultures in stirred-tank bioreactors with the triple reporter strains E. coli G7 BL21(DE3) (b) and E. coli G5 T7E2 (c) in comparison to wildtypes E. coli BL21(DE3) (a) and E. coli T7E2 (d) with minimal medium (T = 37 °C, pH = 6.8, initial glucose concentration 20 g L −1 ). Error bars indicate triplicate datasets. Grey areas mark the exponential growth phases of the different strains E. coli T7E2 and E. coli G5 T7E2 additionally produced minor amounts of citrate, lactate and succinate (total 2% respectively 2.5% of overall carbon, data not shown). The maximum specific growth rate was similar for all strains .004 h −1 ) and in accordance with earlier studies for E. coli BL21(DE3) [49,50].
From mid-exponential growth phase onwards, E. coli G7 BL21(DE3) and its wildtype re-assimilated the prior produced acetate, malate and formate simultaneously with consumption of residual glucose (Fig. 1a, b, organic acids). In contrast, E. coli G5 T7E2 and its wildtype started consuming secondary carbon sources first towards the end of the exponential growth phase (Fig. 1c, d, organic acids). This short second growth phase is visible as small second rise in carbon dioxide. Afterwards, depletion of all metabolites in the cultures is characterized by a steep rise in air saturation in the reactor accompanied by a decrease in carbon dioxide levels for all cultured strains ( Fig. 1a-d, substrate and products). Consequently, the cells reach stationary phase, where no further increase in biomass concentration is found (Fig. 1a-d, products). All strains reached a final biomass concentration of around 12 g L −1 at the end of the exponential growth phase, which is comparable to published data for E. coli BL21(DE3) [51,52].
Due to the shorter lag-phase, the batch processes with E. coli T7E2 and E. coli G5 T7E2 are slightly shorter than for the two other strains. Moreover, E. coli G5 T7E2 and its wildtype produced lower amounts of acetate compared to E. coli G7 BL21(DE3) and the wildtype E. coli BL21(DE3) ( Table 1). Apart from that, there is no significant difference comparing yield coefficients for the four strains and carbon balances closed with at least 98%.
Expectedly, comparing E. coli G7 BL21(DE3) and its wildtype, the strains exhibit strong resemblance in their growth behaviour in batch cultures, which further confirms, that expression of the fluorescent proteins is not a burden to the cells. E. coli G5 T7E2, and its wildtype however, behave similar but slightly different from the other two strains (see also Additional file 2). One reason could be, that E. coli T7E2 carries a knock-in of pgl that encodes for 6-phosphogluconolactonase and which E. coli BL21(DE3) lacks [45]. A loss of this gene leads to a blockage of the pentose-phosphate-pathway, so that intermediates of the remaining part of the central carbon metabolism can be stronger accumulated.

Single cell level physiology Single cell growth
The growth of single cells could be followed by expression of EmGFP together with the ribosomal promoter rrnB (Fig. 2). During batch cultivations, in agreement with the course of biomass concentration (Fig. 1b, products) and the growth rate on population level (Fig. 2a), EmGFP-expression of E. coli G7 BL21(DE3) remained low during the first 5 h (Fig. 2a). Thereafter, E. coli G7 BL21(DE3) started to grow exponentially (Fig. 1, grey area), which led to a steep increase in growth rate on population level and a rise in fluorescence intensity of EmGFP (Fig. 2a). Then, during exponential growth, the growth rate on population level remained about constant, while EmGFPexpression and thereby its mean constantly increased further (Fig. 2a, grey area). With glucose depletion after around 10 h of batch culture (Fig. 1b, substrates), EmGFP-expression intensity endured at constant values until the end of the batch, while the growth rate on population level decreased (Fig. 2a).
EmGFP-expression of E. coli G5 T7E2 showed a similar trend as for E. coli G7 BL21(DE3) , however the different growth phases of the batch culture were clearer reflected by EmGFP-expression of E. coli G5 T7E2 (Fig. 2a,  b). Consistent with the shorter lag-phase on population level (Fig. 1d, substrates and products), the growth rate on population level of E. coli G5 T7E2 as well as the expression of EmGFP started to rise earlier than for E. coli G7 BL21(DE3) , after around 4 h of cultivation (Fig. 2b). During exponential growth (Fig. 2b, grey area), EmGFPexpression and thereby its mean increased further until the beginning of stationary phase after around 8 h of batch cultivation, where it stagnated (Fig. 2b). As expected and similar to E. coli G7 BL21(DE3) , the population level growth rate remained about constant during exponential growth and slowly decreased with glucose depletion (Fig. 2b). Generally, no subpopulation division was found in any phase of the batch culture, which is consistent with earlier studies [53][54][55]. However, the fluorescence distributions varied in shape and fluorescence intensity. These changes can be described by the coefficient of variance (CV) of a distribution that represents the ratio between its standard deviation and its mean fluorescence intensity and can also be correlated to noise in gene expression [56]. The CV of EmGFP fluorescence distributions of E. coli G7 BL21(DE3) (Fig. 5a) was high during lag-phase, but decreased with increasing mean fluorescence until the beginning of the exponential growth phase (compare with Fig. 1b, substrates). During exponential growth, mean fluorescence intensity further increased while the CV only gradually decreased until the end of the process. This is in agreement with studies that revealed low noise levels for stronger expressed genes whereas the opposite is observed when expression levels are low [15,57]. Compared to E. coli G7 BL21(DE3) , E. coli G5 T7E2 exhibited a similar mean fluorescence intensity and CV at the end of the process (Fig. 5a). Strikingly, fluorescence distributions broadened right before the onset of exponential growth phase, then, during exponential growth their CV decreased significantly, which was also found earlier [15,58]. Also for E. coli G5 T7E2 the CV revealed an inverse correlation between noise in EmGFP-expression and EmGFP-expression (Fig. 5a).

General stress response of single cells
General stress response of single cells was tracked by the expression of rpoS in combination with mStrawberry (Fig. 3). For E. coli G7 BL21(DE3) , irrespective of a moderate upshift at the beginning of the lag-phase (compare with Fig. 1b), expression of mStrawberry remained low during glucose consumption until mid-exponential growth phase (around 9 h of cultivation, grey area, Fig. 3a). Then, when the glucose concentration fell below 5 g L −1 and during transition to stationary phase, mStrawberryexpression rose until the end of the batch process and depletion of the glucose in the reactor (Fig. 3a).
Again, E. coli G5 T7E2 showed a similar trend in mStrawberry-expression than E. coli G7 BL21(DE3) . At the beginning of the process, expression levels were low and remained about constant until mid-exponential growth phase (compare with Figs. 1d, 3b, grey area). Then, mStrawberry-expression significantly increased until the end of the process and depletion of the glucose in the reactor (Fig. 3b). Compared to E. coli G7 BL21(DE3) , mean fluorescence intensity reached slightly higher values at the end of the cultivation and additionally the rise in fluorescence intensity around mid-exponential growth phase was detected earlier for E. coli G5 T7E2 than for E. coli G7 BL21(DE3) .
As for EmGFP-expression, the development of the CV was for both strains, E. coli G5 T7E2 and for E. coli G7 BL21(DE3) , inversely correlated to that of mean fluorescence intensity with highest levels of noise in mStrawberry-expression at the beginning of the process (Fig. 5b). Then, the CV decreased until mid-exponential phase and thereafter further, however with a lower rate (Fig. 5b).

Oxygen limitation of single cells
The expression of the nar-operon in combination with TagRFP657 was used to track oxygen limitation of single cells (Fig. 4). Generally, expression levels of this marker were lower than for the other markers, nevertheless a distinct fluorescence upshift under oxygen limited conditions could be detected. For both triple reporter strains, E. coli G7 BL21(DE3) and E. coli G5 T7E2 , Fig. 3 Single cell level physiology for the triple reporter strains E. coli G7 BL21(DE3) (a) and E. coli G5 T7E2 (b) following batch cultures in stirred-tank bioreactors with minimal medium (T = 37 °C, pH = 6.8, initial glucose concentration 20 g L −1 ). Fluorescence distributions (lower subplots) and mean fluorescence characteristics (upper subplots) for the general stress response of single cells related to rpoS-mStrawberry expression is depicted together with substrate availability. All values are shown as average of triplicate measurements following the batch. Grey areas mark the exponential growth phases of the different strains TagRFP657-expression only slightly rose during the transition from exponential growth phase to stationary phase where oxygen saturation in the reactor dropped to close to 40% (Fig. 4a, b). Mean expression levels were similar for E. coli G7 BL21(DE3) and E. coli G5 T7E2 . Interestingly, TagRFP657-expression was found to be induced earlier than in previous studies where nar-expression first rose when oxygen saturation in the reactor fell below 40% [37]. Afterwards, expression levels remained elevated until the end of the cultivation (Fig. 4a, b), probably related to a higher half-life of TagRFP657, while the air saturation returned to 100% with the depletion of the substrate in the reactor (Fig. 1b, d). Considering the correlation between mean fluorescence intensity and CV (Fig. 5c), as for the other markers, noise in gene expression seem to decrease with increasing mean fluorescence intensity, but the trend is less clear than for expression of the other fluorescence markers. Moreover, the CV seemed to fluctuate for E. coli G7 BL21(DE3) , especially towards the end of the process (Fig. 5c).

Population heterogeneity in chemostat cultures with the triple reporter strains
Since in batch cultures, all fluorescent proteins could be detected and showed the fluorescence pattern throughout different process phases that were expected due to their characteristics, the triple reporter strains E. coli G7 BL21(DE3) and E. coli G5 T7E2 were applied for investigation of population heterogeneity in continuous cultures in two case studies. As for batch cultures, all experiments were followed with frequent sampling for population and single cell level physiology.

Case study 1-population heterogeneity at different growth rates
First, aerobic glucose-limited chemostat processes with different dilution rates were performed to evaluate the influence of growth rate on physiology on population and single cell level. Especially, the relation between the population level growth rate and the expression of EmGFP in combination with rrnB should be evaluated. With E. coli G7 BL21(DE3) , chemostats were run at dilution rates of 0.1 h −1 , 0.2 h −1 , 0.25 h −1 , 0.3 h −1 and 0.4 h −1 . Chemostats at the two latter mentioned dilution rates were also run with E. coli G5 T7E2 and additionally at dilution rates of 0.14 h −1 and 0.2 h −1 (Additional file 3).

Population level physiology
Comparing physiology on population level at different dilution rates, the biomass concentration gradually increased by about 45% in cultures with 0.1 h −1 until 0.3 h −1 for E. coli G7 BL21(DE3) (Fig. 6a). With further increase of the dilution rate to 0.4 h −1 , the biomass concentration remained constant. For E. coli G5 T7E2 , the trend was similar, but with a lower rise in biomass concentration by about 39% with increasing dilution rate (Fig. 6a). Generally, biomass concentration was slightly lower compared to steady state biomass concentrations found under equivalent conditions in continuous cultures with E. coli BL21(DE3) carrying the plasmid pET28a-PAmCherry [59].
For none of the triple reporter strains residual glucose could be detected in the reactor at any of the dilution rates (detection limit of glucose in the HPLC around 0.01 g L −1 ), except at 0.2 h −1 for E. coli G7 BL21(DE3) where minor amounts around 0.01 g L −1 were observed (data not shown). Acetate formation as product of overflow metabolism, was only found at dilution rates above 0.3 h −1 with both triple reporter strains (Fig. 6b). At 0.3 h −1 and beneath only minor amounts of acetate (less than 0.02 g L −1 ) could be measured. At 0.4 h −1 , acetate started to accumulate in low amounts of around 0.03 g L −1 . According to published studies (e.g. [59,60]) the onset of overflow metabolism for E. coli lies around 0.45 h −1 which is consistent with these results.

Single cell level physiology
Consistent with population level physiology, on single cell level no expression of nar-TagRFP657 respectively oxygen limitation of single cells was detected for any of the dilution rates and none of the two triple reporter strains (Additional file 3). Furthermore, no significant induction of the general stress response of single cells was seen in chemostats at dilution rates of up to and including 0.3 h −1 for E. coli G7 BL21(DE3) and E. coli G5 T7E2 . Solely at 0.4 h −1 expression of rpoS-mStrawberry was slightly increased (10% rise in mean fluorescence intensity compared to negative control, Additional file 3) at the beginning of the continuous phase with E. coli G7 BL21(DE3) , but not with E. coli G5 T7E2 . In steady state at 0.4 h −1 , no significant rpoS-mStrawberry expression could be detected with E. coli G7 BL21(DE3) .
Expression of rrnB-EmGFP correlated to single cell growth increased as expected with dilution rate for E. coli G7 BL21(DE3) and E. coli G5 T7E2 . When normalizing mean fluorescence intensity of EmGFP with OD 600 , a linear correlation between normalized mean EmGFP fluorescence and dilution rate was found for both strains (Fig. 7). However, the resulting regression line reveals a 5% better linear fit for the normalized mean fluorescence data of E. coli G5 T7E2 than of E. coli G7 BL21(DE3) (coefficient of determination R 2 = 0.97 vs. R 2 = 0.92). These findings are in agreement with published transcriptomics data and studies at different dilution rates [36,40].

Case study 2-population heterogeneity in response to combined oxygen and glucose perturbation
After characterisation of steady state of chemostat cultures at different dilution rates (growth rates), continuous cultures at a dilution rate of 0.2 h −1 of both triple reporter strains were perturbed with a 2 g L −1 concentrated glucose pulse and simultaneous interruption of aeration for 30 min. In this way, the appearance of  substrate gradients in industrial scale bioreactors could be simulated. These lead to formation of zones in the reactor with low oxygen saturation and high substrate availability close to the point of feeding, whereas distant to that, generally opposite conditions are found [61]. Perturbations were followed with frequent sampling for population and single cell level physiology. Thereby, a special focus was put on inducibility of nar-TagRFP657 under oxygen limited conditions.

Population level physiology
Following perturbation on population level, the triple reporter strains exhibited a similar, batch-like behaviour, however produced different amounts of intermediates of the central carbon metabolism.
For E. coli G7 BL21(DE3) , the response to perturbation can be divided into four phases. Until around 15 min after the perturbation, only minor amounts of the pulsed glucose were consumed (Fig. 8a, substrates). Nevertheless, cells produced formate, lactate and acetate (Fig. 8a,  organic acids). Malate, already present in the culture Fig. 8 Concentrations of biomass and metabolites [glucose, acetate (blue squares), malate (green left-pointing triangles), formate (red right-pointing triangles) and lactate (grey circles)], off-gas carbon dioxide and dissolved oxygen of the triple reporter strains E. coli G7 BL21(DE3) (a) and E. coli G5 T7E2 (b) following perturbation with 2 g L −1 concentrated glucose and simultaneous interruption of air supply for 30 min in aerobic glucose-limited chemostat cultures with minimal medium in stirred-tank bioreactors at 0.2 h −1 (T = 37 °C, pH = 6.8, initial glucose concentration 20 g L −1 ). Grey areas illustrate the period without air supply, whereas dashed lines indicate when the pulsed glucose was depleted (around 50 min after perturbation with E. coli G7 BL21(DE3) and around 45 min after perturbation with E. coli G5 T7E2 ). Error bars indicate triplicate datasets during steady state, was consumed. Simultaneously, the air saturation in the reactor decreased to around 2%, whereas only minor amounts of carbon dioxide were produced (Fig. 8a, substrates and products). Thereafter, until 30 min after perturbation, the pulsed glucose was actively consumed accompanied by a rise in carbon dioxide (Fig. 8a, substrates and products). During this phase, formate, acetate, lactate and minor amounts of malate were also accumulated (Fig. 8a, organic acids). When aeration was restarted, the residual glucose was cometabolized with the prior produced acetate, malate and lactate. Formate was not re-assimilated and instead gradually washed-out of the reactor (Fig. 8a, organic acids). The dissolved oxygen content in the liquid phase slowly recovered concurrently with decreasing level of carbon dioxide (Fig. 8a, substrates and products). With depletion of the pulsed glucose around 50 min after perturbation, all values returned to steady state from before perturbation, except of formate, which did not return to steady state values within the measurement period (Fig. 8a, organic acids).
In contrast to E. coli G7 BL21(DE3) , E. coli G5 T7E2 started to consume the pulsed glucose directly after perturbation (Fig. 8b, substrates). Simultaneously, acetate and formate were accumulated as well as minor amounts of malate, which was, opposite to the culture with E. coli G7 BL21(DE3) , not present in steady state prior to perturbation ( Fig. 8b organic acids). Furthermore, E. coli G5 T7E2 did not acquire any lactate during absence of aeration. While the final concentration of formate was similar to that of E. coli G7 BL21(DE3) (about 1 g L −1 ), the acetate concentration was around 70% higher. With perturbation, the dissolved oxygen concentration in the liquid phase rapidly decreased to around 1% (Fig. 8b, substrates). During consumption of the pulsed glucose, the off-gas carbon dioxide rose to around 1% (Fig. 8b, products). When aeration was restarted, the prior produced acetate was re-assimilated together with consumption of residual glucose and malate, while low amounts of lactate were produced and shortly after abolished again (Fig. 8b, organic  acids). About 45 min after perturbation, the pulsed glucose was depleted, which was faster than for E. coli G7 BL21(DE3) (Fig. 8a, b, substrates). Afterwards, residual acetate, which was first depleted around 15 min after the pulsed glucose, was re-assimilated, while formate levels remained constant (Fig. 8b, organic acids). Accordingly, carbon dioxide and dissolved oxygen levels in the gas phase only returned slowly to steady state values (Fig. 8b,  substrates and products). With depletion of acetate, all values returned to steady state prior to perturbation. Solely formate was slowly washed-out of the reactor and did not return to steady state values within the measurement period.

Single cell level physiology Single cell growth
The growth of single cells of E. coli G7 BL21(DE3) was only partly consistent with population level physiology (Figs. 8, 9). Mean EmGFP-fluorescence levels of E. coli G7 BL21(DE3) slightly fluctuated with perturbation, where after, during consumption of the pulsed glucose (Fig. 8a,  substrates), a constant increase in mean EmGFP-fluorescence until around 25 min after perturbation was found (Fig. 9a, Additional file 4). Then, EmGFP-fluorescence levels remained constant apart from a slight deflection when aeration was restarted (Fig. 9a), even though the residual pulsed glucose was further consumed on population level (Fig. 8a, substrates). With depletion of the pulsed glucose around 50 min after perturbation (Fig. 8a,  substrates), mean fluorescence intensity returned to steady state levels. Noise in gene expression, depicted as CV of EmGFP-expression (Fig. 9a), slighty changed with perturbation. During glucose consumption, noise in gene expression remained constant, but exhibited a slight deflection when aeration was restarted. Afterwards, variation levels gradually returned to steady state values.
In E. coli G5 T7E2 cultures, mean fluorescence intensity levels for EmGFP-expression (Fig. 9a, Additional file 4) rose following glucose consumption (Fig. 8b, substrates). A constant growth in mean EmGFP fluorescence intensity was seen until aeration was restarted (Fig. 9a). Then, fluorescence intensity increased further until the pulsed glucose was depleted (around 45 min after perturbation, Fig. 8b, substrates). Afterwards, mean EmGFP fluorescence intensity slowly returned to steady state levels prior to perturbation that were reached within around 35 min after depletion of the pulsed glucose (Fig. 9a). Consequently, in contrast to E. coli G7 BL21(DE3) , EmGFP-expression of E. coli G5 T7E2 could mirror growth on population level. Moreover, the CV, which was for E. coli G5 T7E2 generally lower than for E. coli G7 BL21(DE3) , remained almost constant throughout the perturbation (Fig. 9a).
Generally, E. coli G5 T7E2 exhibited higher mean fluorescence levels in conjunction with lower CVs for steady and throughout perturbation than E. coli G7 BL21(DE3) and could more clearly reflect single cell level growth characteristics described on population level. The prolonged rise in mean fluorescence intensity in cultures with E. coli G5 T7E2 compared to E. coli G7 BL21(DE3) might be connected to subsequent rather than co-consumption of glucose and acetate, that is accumulated in response to perturbation (Fig. 8a, b, organic acids).

General stress response of single cells
General stress response of single cells exhibited the strongest response immediately after perturbation. E. coli G7 BL21(DE3) showed a non-distinctive response to interruption of aeration and simultaneous introduction of the glucose pulse. With perturbation, a slight fluctuation in mean fluorescence levels could be suspected, that reverted, when the cells started to consume the pulsed glucose (Figs. 8a, 9b, Additional file 4). Thereafter, mean mStrawberry-expression levels remained about constant until aeration was restarted, when mean mStrawberry-expression decreased. With glucose depletion around 50 min after perturbation (8a, substrates), mean fluorescence intensity started to gradually rise towards steady state until the end of the measurement period (Fig. 9b). The CV solely showed light deflections during the perturbation period (Fig. 9b). Other than that, no significant changes were found.
Cells from E. coli G5 T7E2 cultures generally exhibited higher mean fluorescence intensities than cells from cultures with E. coli G7 BL21(DE3) . Furthermore, E. coli G5 T7E2 cultures showed a distinct and steep increase in Fig. 9 Mean fluorescence (left) and coefficient of variance (right) of the triple reporter strains E. coli G7 BL21(DE3) (grey squares) and E. coli G5 T7E2 (black circles) following perturbation with 2 g L −1 concentrated glucose and simultaneous interruption of air supply for 30 min in aerobic glucose-limited chemostat cultures with minimal medium in stirred-tank bioreactors at 0.2 h −1 (T = 37 °C, pH = 6.8, initial glucose concentration 20 g L −1 ). Values are shown for single cell growth related to rrnB-EmGFP expression (a), general stress response of single cells related to rpoS-mStrawberry expression (b) and oxygen limitation of single cells related to nar-TagRFP657 expression (c) as average of triplicate measurements. Grey areas illustrate the period without air supply whereas dashed lines indicate when the pulsed glucose was depleted (around 50 min after perturbation with E. coli G7 BL21(DE3) (grey line) and around 45 min after perturbation with E. coli G5 T7E2 (black line)). Error bars indicate triplicate datasets mean mStrawberry-expression intensity with perturbation (Fig. 9b). However, fluorescence intensity reverted until around 15 min after perturbation to levels prior to perturbation. Afterwards, mStrawberry-fluorescence intensity levels remained about constant until the end of the measurement period. The CV showed the opposite behaviour as mean fluorescence intensity. CV-values decreased compared to steady levels prior to perturbation, but also reverted to steady state values within around 15 min after perturbation (Fig. 9b). Similar to the single cell growth rate, noise levels are generally lower for E. coli G5 T7E2 than for E. coli G7 BL21(DE3) .

Oxygen limitation of single cells
Oxygen limitation of single cells showed for both strains distinct changes in fluorescence pattern and -intensity in response to perturbation, which were consistent with population level physiology. Shortly after perturbation, mean TagRFP657-fluorescence levels of both triple reporter strains rose steeply coincident with dissolved oxygen concentration in the liquid phase dropping below 40% air saturation (Figs. 8a, b, substrates, 9c and Additional file 4). Fluorescence levels continued to increase until aeration was restarted, where after they decreased with glucose depletion and relaxation of dissolved oxygen concentration. For E. coli G7 BL21(DE3) , former steady state levels were reached with glucose depletion whereas for E. coli G5 T7E2 steady state levels were first resumed around 60 min after perturbation. The later reversion to steady state values for E. coli G5 T7E2 was probably connected to re-assimilation of residual acetate after glucose depletion.
As the distribution shape seemingly remained constant, the CV is slightly decreasing with increasing TagRFP657-fluorescence levels for E. coli G7 BL21(DE3) around 15 min after perturbation (Fig. 9c). Lowest noise in gene expression levels were found close to when aeration was restarted, where after the CV returned to steady state values prior to perturbation. Apart from a slight irruption around 15 min after perturbation, CV values for E. coli G5 T7E2 remained constant until aeration was restarted (Fig. 9c). Then, noise in gene expression levels gradually returned to steady state values. Strikingly, in comparison to the other two markers of the E. coli triple reporter strains, the CV for TagRFP657 was generally 30-50% lower.

Discussion
In the present study, two E. coli triple reporter strains based on wildtype E. coli BL21(DE3) and on a modified version of E. coli BL21(DE3), T7E2, have been characterized and evaluated for their suitability for investigation of population heterogeneity in laboratory scale bioprocesses simulating industrial scale process conditions. Both strains could monitor single cell growth through the expression of EmGFP together with the ribosomal promoter rrnB, the general stress response of single cells with the expression of rpoS-mStrawberry and oxygen limitation of single cells by expression of nar-TagRFP657 in different growth phases as well as at different growth rates and in response to perturbations of oxygen and glucose. Furthermore, comparing fluorescence characteristics of the different markers of the triple reporter strains could reveal physiological correlations. Though, the responses of the strains deviated and were not always consistent with population level physiology.

Differences between the two triple reporter strains Population level physiology
As expected, E. coli G7 BL21(DE3) exhibited stronger resemblance with E. coli BL21(DE3) than with E. coli G5 T7E2 in growth behaviour in batch cultures. However, major characteristics as growth rate, yields and final biomass concentration were similar for the two triple reporter strains. Generally, E. coli G5 T7E2 needed less time to adjust to new environmental conditions, as seen by the shorter lag-phase in batch processes and almost immediate consumption of pulsed glucose in chemostat cultures. Furthermore, the E. coli T7E2 based triple reporter strain showed a higher diversity in by-product formation in different phases of the batch as well as in response to the pulse, nevertheless this also resulted in lower steady state biomass concentrations in chemostat cultures at different dilution rates. As mentioned above, the reason could be a higher flexibility in the central carbon metabolism of E. coli G5 T7E2 [45]. Surprisingly, while E. coli G7 BL21(DE3) cometabolized acetate and glucose in batch and continuous culture, E. coli G5 T7E2 rather consumed acetate subsequently to glucose. Possibly, the higher metabolic diversity results in lower priority for acetate re-assimilation, as energy can be restored through other parts of the metabolism [62,63].

Single cell level physiology
Also on single cell level a higher flexibility and responsiveness of E. coli G5 T7E2 was visible as this strain was able to reflect more clearly and sensitively even small changes in population variation and fluorescence intensity in different growth phases in batch cultures, in response to perturbation as well as distinct general stress responses in batch and continuous cultures. Moreover, linearity of normalized mean fluorescence values in continuous culture at different growth rates was higher for E. coli G5 T7E2 than for E. coli G7 BL21(DE3) . Apart from the marker for oxygen limitation in batch cultures, E. coli G5 T7E2 generally exhibited higher mean fluorescence values for all markers under all conditions investigated. At the same time, population variation respectively noise in gene expression was predominantly lower than for E. coli G7 BL21(DE3) which points towards higher alignment and robustness of E. coli G5 T7E2 cultures as also suspected earlier [45].
Consequently, due to its advantageous physiological characteristics E. coli G5 T7E2 seems to be better suited when population variation in a dynamic bioprocess environment should be monitored. However, E. coli G7 BL21(DE3) can still be applied to study general trends and moderate changes with slight loss in resolution of physiological changes.

Suitability of the marker positions and the integrated fluorescence proteins
In general, all utilized markers could reflect the targeted cellular characteristics through their fluorescence distributions collected with flow cytometry in both strains, even though they exhibit slightly different characteristics. While all applied fluorescence proteins were shown to be pH insensitive and are generally monomers, the maturation time of TagRFP657 is higher than for EmGFP and mStrawberry [28,30].
Probably, therefore rrnB-EmGFP and rpoS-mStrawberry exhibited, especially considering E. coli G5 T7E2 , responses even to small environmental changes as well as an immediate response to perturbation. However, the decay in mean fluorescence when the trigger signal vanishes, was, especially during batch cultures, slow for the signal of both fluorescent proteins. In stationary phase, fluorescence levels remained high until the end of the batch process. Maybe, fluorescence of the reporter proteins is preserved by the conditions that predominate stationary phase cultures of the triple reporter strains, however this assumption would need to be further investigated. Therefore, for futures studies it might be advantageous to add, for instance, a tag to destabilize these two fluorescent proteins to ensure a faster decline in fluorescence signal after its trigger vanishes.
Expression of rrnB-EmGFP could be shown to correlate with the growth rate of single cells when normalizing with optical density, whereas just taking raw fluorescence intensity data into account, fluorescence rather correlated to growth on population level. This is possibly connected to the fact that E. coli cells increase slightly in length and width in nutrient-rich conditions [64]. In response to glucose perturbation and oxygen limitation only a slight upshift in fluorescence levels was found, which fits with the pulse intensity and the corresponding upshift in growth. However, future studies with a higher pulsed substrate concentration at different growth rates would probably reveal a clearer and stronger response of this marker.
Concerning the general stress response followed with expression of rpoS-mStrawberry, it has to be mentioned, that this marker allows to gain a general picture of induction of the stress response. When following batch cultures, the marker exhibited the expected response in mid-/late exponential growth phase. However, during the perturbation with glucose and oxygen limitation the response was rather unspecific, probably due to the fact that cells used specialized stress responses at this point of the process as they are already primed for potential stress respectively encounter generally stable conditions in the chemostat. The latter case would then probably even lead to downregulation of the expression of rpoS [41]. This however would need to be addressed in future studies where the targeted induction of the general stress response is investigated in response to stress of different kind. Consequently, for a deeper investigation of the mechanism behind population heterogeneity, the marker probably has to be placed differently in the genome. For instance, a more specific response to a certain kind of stress that is relevant for E. coli to cope with environmental changes in industrial scale bioprocesses could be followed.
The expression of TagRFP657 seemed to be slightly sluggish, especially following batch cultures and also expression levels were lower than for the other markers. Moreover, the distribution shape, e.g. in response to oxygen perturbations in continuous cultures seemed to fluctuate, which could not easily be correlated to physiological changes. The reason might be that TagRFP657 has a longer half-life than the other markers [28,65]. Possibly in the future TagRFP657 could be replaced by a fluorescent protein that exhibits more similar maturation characteristics to mStrawberry and EmGFP.
But also the expression of the nar-operon itself might be less stable, as during the present study upshifts in TagRFP657-fluorescence in batch cultures were observed earlier than what was expected from previous studies. There, expression of the nar-operon was first induced when oxygen saturation was below 40%, while in the present study expression levels were already increased between 50 and 40% oxygen saturation [37]. However, when oxygen saturation decreased rapidly in response to perturbation, the mean fluorescence response was clear and reverted fast when oxygen saturation in the bioreactor rose again.

Evaluation of physiological findings correlating the response of the three markers
Throughout this study, single cell physiology was in agreement with or could further supplement population level physiology. In batch cultures, single cell growth could be reflected by EmGFP-expression and exhibited an increase in population variation before the onset of exponential growth phase. Probably, E. coli uses noise in gene expression of rrnB during lag-phase to prepare for growth, whereas it only plays a minor role during transition from exponential growth to stationary phase, as no such increase was found. Population heterogeneity as a strategy to prepare for growth was also found elsewhere [15,58]. During exponential growth phase, cells seem to be more aligned than during all other batch phases exhibiting less heterogeneity.
Interestingly, general stress response was induced in mid-exponential growth phase which is consistent with earlier studies [22,41], but did not lead to a significant increase in noise in gene expression or population heterogeneity. Presumably, the general stress response is a rather conserved function in E. coli and therefore not steered by noise in gene expression.
In general for growth rate and general stress response of single cells an inverse correlation between expression and noise in gene expression was observed, which has also been suggested earlier [8]. For oxygen limitation of single cells this correlation was not as clearly visible.
When continuous cultures are simultaneously perturbed with glucose and oxygen, E. coli apparently only uses the general stress response to cope with the environmental change immediately after perturbation whereas later, e.g. when the aeration is restarted, cells possibly employ specialized stress responses and re-adjustment mechanisms of the metabolism, respectively, that are not directly correlated to the general stress response. Furthermore, once growing, E. coli seems to be prepared to encounter oxygen limited conditions, e.g. when nutrients get scarce as no induction of the general stress response was found simultaneously to expression of the marker for oxygen limitation. However, also specialized stress responses that are automatically primed during growth could be employed.
Expectedly, for single cell growth neither during batch cultures or at different growth rates nor in response to perturbation with oxygen and glucose a subpopulation division in slower and faster growing cells was found for any of the triple reporter strains under the well-mixed conditions studied. Similarly, no diversification appeared during higher growth rates in chemostats after onset of overflow metabolism. This finding is in accordance with previous studies [53][54][55] and points towards reasonable robustness of E. coli BL21(DE3) and E. coli T7E2 in response to environmental changes.
Indeed, for future experiments with the triple reporter strains more advanced methods for data treatment for instance algorithms applying flow cytometric fingerprinting [66][67][68][69] that have been developed in the past years should be applied. They could potentially reveal further correlations between the changes in expression of the different markers as well as physiological peculiarities under different environmental conditions in bioprocesses. Especially, small changes that are overlooked when just using averaged data or subjective interpretation of fluorescence histograms could reveal new physiological details [70].

Conclusion
In conclusion, growth rate, general stress response as well as oxygen limitation of single cells of the two triple reporter strains could be detected and the respective markers exhibit the expected fluorescence pattern throughout standard batch cultivations and in response to changes in growth rate and perturbations in glucose and oxygen levels. Consequently, the triple reporter strains represent valuable tools to study population heterogeneity in bioprocesses and supplement population level physiology with less complexity than omicsmethods. Correlating the responses of the different markers to each other significantly increases the level of information compared to conventional experiments investigating single cell physiology with single reporter strains when studying population heterogeneity in bioprocesses.