Genome-Wide Functional Screen for Calcium Transients in Escherichia coli Identifies Increased Membrane Potential Adaptation to Persistent DNA Damage

All eukaryotic cells use calcium as a critical signaling molecule. There is tantalizing evidence that bacteria also use calcium for cellular signaling, but much less is known about the molecular actors and physiological roles.

ABSTRACT

Calcium plays numerous critical roles in signaling and homeostasis in eukaryotic cells. Far less is known about calcium signaling in bacteria than in eukaryotic cells, and few genes controlling influx and efflux have been identified. Previous work in Escherichia coli showed that calcium influx was induced by voltage depolarization, which was enhanced by mechanical stimulation, which suggested a role in bacterial mechanosensation. To identify proteins and pathways affecting calcium handling in bacteria, we designed a live-cell screen to monitor calcium dynamics in single cells across a genome-wide knockout panel in E. coli. The screen measured cells from the Keio collection of knockouts and quantified calcium transients across the population. Overall, we found 143 gene knockouts that decreased levels of calcium transients and 32 gene knockouts that increased levels of transients. Knockouts of proteins involved in energy production and regulation appeared, as expected, as well as knockouts of proteins of a voltage sink, F₁F_o-ATPase. Knockouts of exopolysaccharide and outer membrane synthesis proteins showed reduced transients which refined our model of electrophysiology-mediated mechanosensation. Additionally, knockouts of proteins associated with DNA repair had reduced calcium transients and voltage. However, acute DNA damage did not affect voltage, and the results suggested that only long-term adaptation to DNA damage decreased membrane potential and calcium transients. Our work showed a distinct separation between the acute and long-term DNA damage responses in bacteria, which also has implications for mitochondrial DNA damage in eukaryotes.

IMPORTANCE All eukaryotic cells use calcium as a critical signaling molecule. There is tantalizing evidence that bacteria also use calcium for cellular signaling, but much less is known about the molecular actors and physiological roles. To identify genes regulating cytoplasmic calcium in Escherichia coli, we created a single-cell screen for modulators of calcium dynamics. The genes uncovered in this screen helped refine a model for voltage-mediated bacterial mechanosensation. Additionally, we were able to more carefully dissect the mechanisms of adaptation to long-term DNA damage, which has implications for both bacteria and mitochondria in the face of unrepaired DNA.

INTRODUCTION

Calcium is an important and ubiquitous signaling molecule in eukaryotic cells (1), but much less is known about its role in bacteria (2, 3). High-resolution tools to measure cellular calcium, including fluorescent biosensors, have enabled detailed knowledge of calcium pathways and dynamics at the subcellular level in eukaryotic cells (4, 5) and have recently been applied in bacteria (6). In bacteria, intracellular calcium concentration changes have been associated with infection (7), cell division (8), and motility (9, 10), along with other important cellular processes (3). Bacteria are known to tightly regulate cytoplasmic calcium levels (11, 12), while changes in the chemical or mechanical environment have been shown to induce calcium dynamics (6, 13–15). Given the overwhelming importance of calcium in eukaryotes, as well as the many potential bacterial signals, it is critical to understand exactly how and why bacteria regulate calcium levels, as well as the proteins that bind calcium and exert calcium-responsive cellular functions.

One main challenge in uncovering the spectrum of signals encoded by calcium dynamics in bacteria has been a lack of tools to monitor calcium concentrations in bacteria with high resolution. Luminescent tools have been used to monitor populations of cells with low time resolution (11), but they lack the sensitivity to investigate individual cells. The advent of ultrasensitive genetically encoded calcium sensors (4) enabled their recent application to E. coli, and the results revealed that these bacteria acted as electrically excitable cells, where voltage depolarization led to a transient calcium influx, which arose from mechanical stimulation (6). These new fluorescent sensors can bridge the single-cell gap and potentially reveal new facets of calcium signaling in bacterial cells.

Recently, our group used genetically encoded fluorescent sensors to establish links between mechanosensation, voltage depolarization, and calcium influx (6). Similarly to vertebrate sensory neurons, mechanically stimulated E. coli cells induced voltage depolarization, which in turn caused an increase in calcium influx into the cytoplasm. Thus, there is a direct link between mechanosensation and cytoplasmic calcium concentration. Furthermore, mechanosensing and surface sensing by bacteria are critical steps in processes associated with human health such as host infection (16) and biofilm formation (17, 18).

Given the importance of calcium in eukaryotic cells along with the implications for human health, we sought to understand more about the proteins involved in regulating bacterial cytoplasmic calcium concentrations. In this study, we performed a genome-wide screen of mechanically induced calcium transients with a library of gene knockouts in E. coli. The screen could detect knockouts that decreased or increased the calcium transients. Hits from our screen corresponded to proteins known to influence calcium concentrations, including genes associated with voltage generation through the electron transport chain and ATP generation through F₁F_o-ATPase. However, we also identified additional gene knockouts in various processes, including outer membrane lipid synthesis and DNA repair. The knockouts of genes involved in outer membrane lipid synthesis did not reduce basal voltage but did reduce calcium flux, which suggested a model of mechanosensation in which the chemical structure of the outer membrane plays a critical role in relaying mechanical signals. Knockouts of proteins associated with DNA damage repair pathways showed increased DNA damage and lower voltage levels. However, chemically induced acute DNA damage did not change the voltage or calcium transients, which showed that long-term adaptation to survive DNA damage included lowering the membrane voltage and growth rates. Though we did not definitively identify a single mechanosensor or voltage-gated calcium channel, several hits could potentially perform elusive functions. Overall, our screening strategy revealed several pathways important in voltage and calcium homeostasis and provided several intriguing hits to follow in future work.

RESULTS

High-throughput screen for calcium effectors in E. coli.

Our previous model of E. coli calcium transients suggested that mechanical stimulation induced voltage depolarization, which led to a transient calcium influx (6) (Fig. 1A). Therefore, a screen for modulators of calcium transients would report conditions that altered mechanosensation, membrane voltage, or calcium handling proteins. To identify genes that modulate mechanically induced calcium transients in E. coli, we developed a live-cell screen (Fig. 1B) using automated 96-well plate imaging of cells expressing a fusion of GCaMP6f (calcium sensor) and mScarlet (insensitive control). The Keio collection of knockouts (19) was transformed with the GCaMP6-mScarlet plasmid and grown on dual-selection plates to maintain the gene knockout (kanamycin [Kan]) and the sensor plasmid (carbenicillin [Carb]). After transformation, ∼97% of the knockouts grew on double-selection plates, which were moved forward for imaging. Cells were spotted onto agarose pads and pressed into glass-bottom well plates (see Fig. S1 in the supplemental material). Thus, within each well were many cells, each of which had a genotype defined by the Keio collection.

FIG 1.

Design for a genome-wide screen for calcium effectors in bacteria. (A) Current model underlying the observed calcium transients in E. coli. Mechanical stimulation induces voltage depolarizations, which induce calcium influx into the cell. Voltage is generated by respiration via the electron transport chain (ETC) and is consumed by F₁F_o-ATPase. (B) Outline of the live-cell screen. The Keio collection is transformed with a constitutive GCaMP-mScarlet-expressing plasmid and imaged on an automated microscope. Custom analyses segment out individual cells and calculate the calcium AUC. All the cells from a given genotype are then combined, giving rise to the AUC score, which is then used to screen for knockouts that increase or decrease calcium transients.

From each well, we acquired a 90-s video of the mechanically induced calcium activity to identify genetic knockouts with altered calcium handling. From each field of view (FOV), individual cells were segmented and data representing the mean levels of GCaMP and mScarlet fluorescence were extracted from each frame (see Materials and Methods; see also Fig. S1). From each cell’s fluorescence, we defined transients and calculated the total calcium area under the curve (AUC). Combining all of the cells representing a given genetic identity, we calculated a population-level AUC representing the median of the sum of the data from all cells within a given field of view. Comparing the median population AUC values from a given genotype, we were able to identify knockouts that reduced or increased the number of Ca²⁺ transients (Fig. 1B).

To test the resolving power of the screen, carbonyl cyanide m-chlorophenylhydrazone (CCCP) and apramycin were used as positive controls that increased or decreased calcium transients, respectively. These chemical treatments had a Z′ factor of 0.32 for CCCP (low transients [6]) and a Z′ factor of 0.28 for apramycin, an aminoglycoside that induces increased transients (20) (Fig. 2A; see also Fig. S2). Though the Z′ score was <0.5, we considered it sufficient for moving forward into a genome-wide screen by accepting the potential presence of false positives during the primary data collection.

FIG 2.

The live-cell screen can identify knockouts with altered calcium flux. (A) Box plot of the calcium AUC from cells treated with DMSO (negative control), CCCP (eliminates transients), or apramycin (increases transients). Each of the box plots has 16 biological replicates. Red lines indicate medians, blue boxes indicate 25/75 limits, black lines indicate 10/90 limits, and red crosses indicate outliers. *, P < 0.001. (B) Calcium AUC from the primary screen. Each gene in the Keio collection is plotted on the y axis, and the log of the calcium AUC is plotted on the y axis. The genes corresponding to the values ranging from 4300 to 5400 on the x axis are not present in the Keio collection. (C) A histogram of the calcium AUC from the entire screen. (D) A random sample of individual cells from wild-type cells, ΔrecG cells (reduced calcium AUC), or ΔbtuR cells (increased calcium AUC). AU, arbitrary units.

We then proceeded to use our assay to screen the Keio collection, a genome-wide knockout library in E. coli (19). The Keio collection consisted of a total of 45 96-well plates. We screened all 45 plates as a primary screen, which consisted of raw video data comprising 7.9 TB. After removal of unconfirmed strains, wells too dim to image, and any other artifacts we could determine from the fluorescence images, we extracted time traces from a total of 4.04 million individual cells from 3,504 genetic knockouts. The top hits from the primary screen were reimaged in biological triplicate to confirm true positive hits. Across the entire screen, we generated criteria to call knockouts with high and low population-wide calcium AUC values (Fig. 2B). Overall, we found 143 knockouts that decreased the calcium AUC values and 32 genes that increased the calcium AUC values (see Tables S1 and S2 in the supplemental material). A random sample of traces from ΔrecG cells (low transients) and ΔbtuR cells (high traces) qualitatively matched the expectations from the median AUC screen measurements (Fig. 2C) compared to wild-type (WT) cells.

Expected knockouts of proteins involved in cell energetics affect calcium transients.

Given our previous work linking membrane voltage with calcium transients, we expected that knockouts known to reduce membrane potential (21, 22) would have a corresponding reduction of observed calcium transients. Indeed, many knockouts of genes involved in the electron transport chain showed reduced calcium AUC compared to the WT results (Table S1). Gene ontology (GO) enrichment analysis (23, 24) of knockouts with reduced levels of calcium transients also showed high enrichment in molecular functions associated with energy production, including cytochromes and the tricarboxylic acid (TCA) cycle. These data provided confirmation that our screen could identify known pathways affecting voltage regulation, which in turn affected voltage transients (6, 25).

In addition to gene knockouts involved with voltage generation, we observed that gene knockouts of several components of F₁F_o-ATPase had higher calcium AUC values than were seen with the WT cells. Considering our current model, these data were also in agreement with identification of F₁F_o-ATPase as a sink of membrane voltage. We expected that inhibiting or eliminating that pathway would result in higher overall membrane potential and increased levels of calcium transients (15). Measurements performed with TMRM (tetramethylrhodamine, methyl ester) confirmed the higher membrane potential (Fig. S3). Though it has long been known that the electron transport chain generates a voltage that is used by F₁F_o-ATPase (26), these hits gave us confidence that our screen could detect proteins that affected both voltage generation and mechanical induction of calcium transients.

Included among the knockouts that both increased and decreased calcium levels of transients were many proteins associated with anaerobic metabolism, including dimethyl sulfoxide (DMSO) reductase (ΔdmsB), oxygen sensing (ΔarcA), and menaquinone biosynthesis (ΔmenB, ΔmenD, ΔmenF) (Tables S1 and S2). This observation is especially interesting because previous work showed that the voltage transients occurred only in the presence of oxygen (25). To confirm that the calcium transients required oxygen also, we imaged cells expressing GCaMP6 before and after the removal of oxygen from the fluidic chamber. Similarly to the voltage data, all induction of calcium transients ceased upon the removal of oxygen but then reinitiated upon its restoration (Fig. S4). Oxygen sensing and metabolism clearly played a critical role in regulating the voltage and calcium transients, but the mechanisms underlying these changes remained unknown.

Outer membrane integrity and exopolysaccharides are critical for mechanosensation.

In addition to components regulating voltage, we also expected to find knockouts of proteins that affect bacterial mechanosensation. Knockouts of several inner membrane (IM) proteins with domains of unknown function were hits with reduced calcium transients, which may represent the channels or sensors involved in relaying the mechanical signal across the plasma membrane (Table S1). However, no single knockout of any IM protein eliminated all calcium transients, which might have been due to functional redundancy or to the presence of a nonproteinaceous calcium channel (27). However, we did see a number of knockouts of proteins associated with outer membrane and polysaccharide synthesis that had reduced calcium AUC values (Fig. 3A). We hypothesized that a fully intact outer membrane and the presence of polysaccharides are critical components of the mechanical relay. Earlier work showed that these knockouts had no reduction in growth rate compared to WT cells (28). Furthermore, these knockouts showed no reduction in basal voltage as measured by the use of TMRM compared to WT cells (Fig. 3B), which suggested that the proteins were a part of the mechanosensitive relay.

FIG 3.

Knockouts of outer membrane synthesis and extracellular polysaccharide secretion refine the model for mechanosensation. (A) Box plot of knockouts involved in outer membrane (OM) synthesis and exopolysaccharide (EPS) secretion showed reduced calcium AUC compared to WT cells. Each box plot represents results from 5 biological replicates. (B) Knockouts corresponding to OM synthesis and EPS secretion did not show reduced membrane potential compared to WT cells. Each mark represents the median of the population and corresponds to one biological replicate. Each genotype was measured in biological triplicate. (C) Box plot of calcium AUC of wild-type cells adhered with poly-l-lysine or an agarose pad compared to a Δ₄pol knockout maintained under the same conditions. Each box plot corresponds to 4 biological replicates. *, P < 0.01. (D) Box plot of calcium AUC for E. coli adhered under agarose similarly to the method used for the other measurements or encased in agarose by adding the cells before solidification of the gel. Each box plot corresponds to 5 biological replicates. *, P < 0.01. (E) Proposed model for E. coli mechanosensation. A force is applied by the agarose pad which is opposed by an equal and opposite force from the coverslip. The coverslip force arises from adherence with secreted EPS, which then interacts with lipopolysaccharide (LPS) on the outer membrane of the cell. This force is relayed to a mechanosensor in the inner membrane which remains to be identified. For all box plots, red lines indicate medians, blue boxes indicate 25/75 limits, black lines indicate 10/90 limits, and a red cross indicates an outlier.

On the basis of these data and the previous work showing mechanical induction of calcium transients, we proposed a model of mechanosensation in which E. coli cells use the outer membrane and polysaccharide adhesion to stick to the surface of glass. Upon mechanical deformation by the use of an agarose pad, the outer membrane undergoes a shear stress which induces the activation of the mechanical sensor, followed by voltage and calcium transients. To test this model, we measured an E. coli strain with 4 components of exopolysaccharide synthesis removed (29) (Δ₄pol) which showed very low levels of calcium transients (Fig. 3C). A second experiment measured calcium transients of bacteria either sandwiched between the glass surfaces, which was used as the standard in this assay, or encased in agarose, which solidified around the cells and exerted no shear force. The cells encased in agarose showed significantly lower levels of calcium transients than the glass-sandwiched cells (Fig. 3D) but continued to grow. Combined, these data are consistent with mechanosensation arising from a shear force relayed through the outer membrane to induce voltage and calcium transients (Fig. 3E).

Knockouts involved in DNA repair reduce voltage and calcium transients.

One unexpected cellular process that showed dramatically decreased calcium AUC was knockouts of genes involved in recombination and DNA repair (Fig. 4A). These knockouts had extremely low calcium transients compared to WT cells. These striking data suggested a link between persistent DNA damage and membrane voltage in E. coli. DNA damage is known to reduce respiration in bacteria via the SOS response, but respiration is restored upon repair of the damage (30). Furthermore, DNA damage is associated with depolarized membrane potential in eukaryotic mitochondria (31–33), and any changes to the metabolic state could affect membrane voltage and calcium currents.

FIG 4.

Knockouts of genes associated with DNA repair have reduced calcium AUC. (A) Box plot of knockouts involved in DNA repair compared to WT cells. Each box plot corresponds to 5 biological replicates. (B) TUNEL fluorescence measured by a flow cytometer for WT, ΔrecA, and ΔrecG cells. *, P < 0.05. (C) TMRM fluorescence for comparisons of WT cells to rec knockouts. Each marker represents the median of the population for one biological replicate, and 3 biological replicates were measured for each condition. Addition of 50 μM CCCP is used to show fluorescence from zero voltage. (D) Growth curves measured by OD₆₀₀ for comparisons of WT cells to rec knockout cells. Each dark line represents the mean and shaded areas represent standard deviations of results from 3 biological replicates. For all box plots, red lines indicate medians, blue boxes indicate 25/75 limits, and black lines indicate 10/90 limits.

To first determine if defects in recombination could lead to increased DNA damage, we assayed damage by measuring terminal deoxynucleotidyltransferase-mediated dUTP-biotin nick end labeling (TUNEL). As expected, the knockouts tested showed significantly more DNA damage than the WT cells (Fig. 4B). Removal of proteins associated with DNA repair is reported to reduce expression of components of the electron transport chain in mitochondria, so we hypothesized that the cells with unrepaired DNA would have lower membrane potential. TMRM measurements showed that the recombination knockouts indeed had lower voltage (Fig. 4C) than the WT cells. The exception was the ΔrecG knockout, which had significantly higher TMRM fluorescence than the WT cells. We suspected this might have been due to cytoplasmic pH differences. For all gene knockouts of DNA repair, there was a corresponding reduction in the growth rate of these cells compared to WT (Fig. 4D). Despite these signatures, the knockout cells observed were not elongated, which suggested that the SOS response and sulA expression levels had been reduced (34). Taking the data together, we observed that cells with incomplete repair machinery and persistent DNA damage had reduced voltage and calcium transients, and a reduced growth rate, compared to wild-type cells.

Persistent adaptation, not acute adaptation, to DNA damage lowers membrane voltage.

DNA damage in eukaryotic mitochondria is known to be associated with depolarized membrane potential, and the depolarization has been attributed to ROS buildup and damage to the membrane (35). In bacteria, it has been suggested that recA filaments interact with the membrane, which might provide a mechanism for reduced voltage (36, 37). The high temporal and spatial resolution of the fluorescent measurements offered the opportunity to investigate the dynamic interplay between DNA damage and depolarization. Genetic knockouts from the Keio collection had had many generations of growth and adaptation before imaging was possible, so we instead turned to genotoxic chemicals to induce damage and monitor the acute voltage response.

To control the onset of DNA damage, we used mitomycin C (MMC), a DNA cross-linker purified from Streptomyces caespitosus (38, 39). Addition of 5 μM MMC showed bactericidal activity within 2 h as measured by CFU levels (Fig. 5A), confirming high DNA damage within a short time frame. The SOS response was upregulated as measured by filamenting of cells 2 h after treatment (Fig. S5). Unexpectedly, calcium transients in cells treated with 5 μM MMC were identical to those in untreated cells (Fig. 5B), and even after 8 h of treatment, the treated cells showed no difference from untreated cells in calcium AUC (Fig. 5C), suggesting that the voltage levels were unchanged. E. coli cells were then grown for 36 h in liquid culture in the presence of 5 μM MMC, and the cells reached stationary phase, which indicated that some fraction of the population had adapted to the new environment. These cells, which were continuously treated with the drug MMC, showed severely reduced transients, similarly to knockouts of DNA repair machinery (Fig. 5D). Taken together, our data supported a model whereby cells undergo a long-term adaptation to persistent DNA damage by lowering their membrane potential followed by a reduction in the SOS pathway level. The adaptations might arise through mutation or toxin/antitoxin systems or by other means, but they occur at long time scales compared to ROS generation (seconds to minutes).

FIG 5.

Persistent DNA damage induces an adaption via voltage reduction. (A) CFU assay of cells treated with 5 μM MMC compared to untreated cells. Gentamicin (gent)-treated cells were used as a positive control for a bactericidal compound. (B) Time course of calcium AUC taken from 90-s movies comparing untreated cells to cells treated with 5 μM MMC. Each point was taken from a unique well, alternating between treated and untreated conditions. (C) Time course of calcium AUC from a titration of MMC concentrations over 8 h. AUC was calculated from 90-s movies, and data were normalized to results from untreated cells. (D) Calcium AUC comparing persistent DNA damage (ΔrecA cells, 36 h of 5 μM MMC) to acute DNA damage. *, P < 0.01. For box plots, red lines indicate medians, blue boxes indicate 25/75 limits, and black lines indicate 10/90 limits.

DISCUSSION

In this study, we performed a single-cell, genome-wide screen for modulators of cytoplasmic calcium in E. coli. Using this screen, we identified both known and novel pathways that modulate the level of cytoplasmic calcium. The imaging assay had a throughput of 384 genotypes per day, each consisting of dozens to hundreds of single cells, with 450 time points for each measurement. The high signal-to-noise ratio from the fluorescent signal, combined with the automated analytical scripts, enabled distinguishing positive and negative chemical controls with high confidence. This setup could be easily adapted to other conditions, including a chemical screen on calcium AUC, additional bacterial species, or other fluorescent sensors. The current screen is limited by throughput, due to the serial nature of the assay and the relatively long time per condition (90 s), and still requires reasonably high levels of operator time and skill. Furthermore, the Keio collection of knockouts is fraught with well-documented cases of mutation and adaptation, so an inducible CRISPRi screen may provide complementary information and allow investigation of essential genes. Bearing in mind these limitations, we believe that single-cell, physiological dynamics have the potential to uncover new information and biological processes not available from static imaging.

One of the most interesting and significant hits in our screen was the identification of knockouts of proteins associated with DNA repair, which had reduced voltage and calcium transients compared to the results obtained with the WT cells. Our data showed that the DNA damage clearly resulted in impaired voltage but also that the decrease in voltage occurred only over the course of adaption over ∼48 h. Acute treatment had no impact on the calcium transients, even at concentrations high enough to prevent cells from future cell division. These data suggest a model whereby DNA damage does not immediately lower voltage through ROS production (30) or through interaction with the plasma membrane (36, 37) but rather occurs only as a mutation or transcriptional adaptation that allows a fraction of the cells to continue to divide in spite of persistent DNA damage. Furthermore, the cells turn off the SOS response despite this persistent DNA damage. Understanding the exact mechanisms by which cells accommodate DNA damage via voltage suppression is an important path forward that we plan to study.

DNA damage induced by the use of MMC also revealed an interesting relationship between cell division as measured by CFU and cell metabolism as measured by cytoplasmic calcium maintenance. Addition of 5 μg/ml MMC was enough to prevent >99% of cells from forming colonies within 2 h. Yet the cells were metabolically active for 8 h after treatment. This intermediate time period between treatment and cell death is similar to previous results seen with aminoglycosides (20) and suggests that this could be an important consideration during antibiotic treatment if those dying cells can still signal to neighbors and can still influence the population-level activity.

The mechanisms that tie DNA damage to decreased membrane potential could have an impact in the mitochondria of eukaryotic cells as well. Depolarized mitochondria are marked for mitophagy via the PINK-1 pathway, and one trigger for depolarization is mitochondrial DNA damage (40, 41). An existing model suggests that DNA damage directly reduces membrane potential through increased ROS (40, 42), but that would be the opposite of what we observed in bacteria. One interesting avenue will be to perform similar experiments on eukaryotic cells, monitoring the mitochondrial potential upon mitochondrial DNA damage.

MATERIALS AND METHODS

Strains and growth.

E. coli BW25113 was acquired from the Yale Coli Genetic Stock Center and was used as the wild-type strain for all experiments except for the comparisons to the Δ₄pol strain, which used strain MG1655. The Keio collection was purchased from Dharmacon (OEC4988). Keio strains were grown in 10 μg/ml kanamycin (Kan) to ensure maintenance of the genetic insertion. Cells were grown overnight in LB at 37°C while shaking at 190 rpm. All strains that carried the GCaMP-mScarlet calcium indicator plasmid were grown in 100 μg/ml carbenicillin (Carb) in addition to other antibiotics necessary to maintain those strains. The Δ₄pol strain was maintained in media with 10 μg/ml chloramphenicol, 30 μg/ml kanamycin, and 25 μg/ml zeocin.

Plasmids and transformation.

The constitutive GCaMP-mScarlet plasmid was generated as described in a previous publication (20) and is available from Addgene (catalog no. 158979). The plasmid was transfected into E. coli knockouts by using transformation and storage solution (TSS) buffer in a 96-well plate. Briefly, cells were grown overnight in LB and Kan in a plate with 96 deep wells. Cells were then spun down at 1,900 × g for 7 min, and the supernatant was removed by pouring. TSS buffer was added, and the reaction mixture was incubated with a 1-μl volume per well from a miniprep of the plasmid. Upon incubation for 1 h, a 5-μl volume of the cells was placed onto a double-selection plate (Carb, 100 μg/ml; Kan, 50 μg/ml) using a multichannel pipette, and the plates were incubated overnight at 37°C.

Colonies of the transformed cells were then picked using a pin transfer tool and grown overnight in LB (Carb, 100 μg/ml; Kan, 50 μg/ml), and glycerol stocks were made from this suspension. To grow cells for imaging, the glycerol stock in the 96-well plates was placed into a 96-well plate with LB (Carb, 100 μg/ml; Kan, 10 μg/ml) using a pin transfer tool and cells were grown overnight at 37°C while shaking.

TMRM assays.

A 1-ml Falcon polystyrene round-bottom tube containing 940 μl of M9 minimal medium (MM) with 0.4% glucose and 1× nonessential (NE) amino acids was seeded with 50 μl of an overnight-growth suspension of the strain to be tested (in biological triplicate), and the strain was grown at 28°C with 200 rpm shaking. When the cells reached an OD of ∼0.4, 10 μl of TMRM (20 μM) was added to the suspension to reach a final concentration of 0.2 μM TMRM. Thirty minutes later, cells were quantified with respect to their TMRM incorporation by counting 100,000 events per condition using a BD FACSCellesta flow cytometer with the following voltage settings: forward scatter (FSC) at 700, side scatter (SSC) at 350, with 561-nm-wavelength laser D585/15 at 500, C610/20 at 500, and B670/30 at 481. Emission for each event was collected at the 585/15-nm wavelengths.

TUNEL assay.

DNA damage was measured using a TUNEL Apo-Direct kit (BD, catalog no. 556381) and the manufacturer’s suggested protocol. Cells were fixed with ethanol followed by washing and staining.

Keio imaging.

Imaging took place on an inverted microscope setup using 96-well glass-bottom plates (Brooks, catalog no. MGB096-1-2-LG-L), similarly to earlier work (20). Cells were grown in overnight cultures in 96 wells at 37°C in LB with carbenicillin to maintain the plasmid. Agarose pads in a 96-well format were created using a custom three-dimensional (3D) printed mold (Kraljlab). The mold was designed so that each well held a 200-μl agarose pad. Agarose was dissolved at a final concentration of 2% in partial minimal medium (PMM; 1× M9 salts, 0.2% glucose, 200 μM MgSO₄, 10 μM CaCl₂, 1× minimal essential medium [MEM] amino acids) and added into the mold, which was covered by a piece of glass (McMaster Carr, catalog no. 8476K43). A second piece of glass was added to create flat surfaces on both sides. After the agarose gel solidified (typically >30 min), 2 μl of the cell suspension was added onto each pad and the solution was allowed to fully diffuse into the gel. The pads were then pressed into the glass-bottom plate using a second 3D printed piece. Inversion of the plate (A1 from cell suspension → A12 in the glass-bottom plate) was performed and was accounted for in the analysis software.

Imaging took place on an automated inverted microscope (Nikon Ti2) using a 40× numerical-aperture (NA) 0.95 lens objective for imaging performed with two scientific complementary metal oxide semiconductor (sCMOS) cameras (Hamamatsu, Flash 4 v2) and a custom dichroic splitter in the emission path. The emission filters used were a 525/50 filter (GCaMP) and a 568 long-pass (LP) filter (mScarlet), and we used a 561 LP dichroic to separate the two colors. Cells were illuminated using an LED source (Spectra X; Lumencor) and simultaneous excitation with 470-nm-wavelength and 550-nm-wavelength light. Images were captured continuously with an exposure time of 200 ms (5 Hz) for a total of 90 s.

To enable automated imaging of all 96 wells without additional user input, a single point was selected in each well by the user such that cells were present but not in excessive density (20 < number of cells < 2,000). Selecting all 96 points took on average ∼20 min. Enabling the Perfect Focus setting on the Ti2 microscope kept the plate in focus across the entire well plate. Capture of the entire movie took ∼2.5 h per plate, so we were typically able to image 2 to 3 plates per day. Image data were stored with the metadata in the .nd2 format for downstream processing.

Oxygen removal experiments were conducted using a custom flow cell setup similar to that used in our previous experiments (20). Two syringes containing PMM or PMM plus 5% oxyrase (oxyrase for broth; Oxyrase Inc.) were connected to a single channel via a T valve. Medium was exchanged by flowing through ∼1 ml by manually depressing the syringes. Upon medium exchange, a delay of 10 min was used before initiating imaging to allow full depletion of environmental oxygen.

Image processing.

Image processing from the movie files was performed similarly to our previous reports (20), with a few changes due to the nature of the movie collection. Image analysis was performed in Matlab using custom .m scripts. Image processing followed the following general scheme: (i) estimating the illumination profile for all experiments on a given day; (ii) correcting the uneven illumination for each movie; (iii) registering drift and jitter in XY; (iv) subtracting an estimated background; (v) segmenting cells using a Hessian algorithm; (vi) extracting time traces for individual cells; and (vii) processing each time trace to calculate calcium area under the curve (AUC).

(i) Estimating the illumination profile. For a given day, the images in every movie were averaged across time, opened using a morphological operator, and blurred using a 2D Gaussian filter. These experimental images were then averaged together to give an estimate of the uneven illumination. These images were smooth across the entire field of view and varied in illumination by ∼50% across the entire image.

(ii) Correcting uneven illumination. The individual movie images were then loaded into memory sequentially. Each frame of each movie was converted to a double image and then divided in accordance with the uneven illumination. The value representing this image was then multiplied by the average value determined for the movie, and the resulting image was converted back into a uint16 image (Matlab terminology) to maintain consistent intensity values. Each frame was then reassembled into an illumination-corrected movie.

(iii) Registering drift and jitter in XY. Each frame was aligned to the previous frame using a convolution of the 2D Fourier transform (2DFT). Each sequential image was first estimated by applying the XY warping from the previous frame. Then, the 2DFT was taken for each image and multiplied with respect to the previous frame. The optimal updated XY position was then calculated and applied. Due to the short time of these movies (90 s), little drift was observed for most measurements.

(iv) Subtracting the estimated background. The background was estimated for each frame individually using a morphological operator. A disk structured element with radius of 9 μm was blurred with a Gaussian filter. This background estimation was then subtracted from the original image. To protect against potential negative values, the minimum for the entire movie was set to 50 counts.

(v) Segmenting cells using a Hessian algorithm. To segment cells, first, the foreground was estimated by subtraction of the background image using Otsu’s method. The Hessian value was then calculated using the background subtracted image and was then multiplied elementwise to produce a logical image of the foreground. Otsu’s method was again used on this modified Hessian image to identify individual cells. Hard limits were set to remove potential noise that did not fit given criteria for size or minimum intensity. We found that first increasing the size of the image using a spline interpolation gave superior segmentation results. Not all cells were identified within a microcolony by the use of this method, though we estimate that it can identify ∼96% of the cells accurately.

(vi) Extracting time traces for individual cells. From a given identified cell, for each time point in the movie, we extracted the mean intensity using the Matlab command regionprops. The mean intensity values representing both the GCaMP6f and the mScarlet, or any other fluorophore that the cells expressed, were extracted using this method.

(vii) Processing each time trace to calculate the calcium AUC. To identify calcium transients in a given cell, we first removed any low-frequency photobleaching using a moving median filter, followed by a wavelet denoising (wdencmp.m). To identify the transient starts, any point of the differential that rose above zero (rising slope) was used to ensure that the peak would have an area under the curve value of >3σ. For any given cell, all the transients that met these criteria were summed to give a cellular AUC over the duration of the movie. To calculate the genotype AUC, the median AUC of the population was calculated to minimize the effect of the presence of any outliers from nonstationary cells.

CFU.

CFU were measured by plating treated cells onto LB-agarose without antibiotic and counting growing colonies. CFU measurements were conducted by trying to mimic the experiments performed via microscopy. Briefly, cells were grown overnight in LB and diluted 1:20 in 5 ml PMM. These cultures were grown at room temperature with shaking for 2 h (t = 0) followed by the addition of antibiotic. At each time point, the culture was removed from the shaker, and 100 μl was removed. A 10× series dilution was then conducted by removing 20 μl and adding that volume to 180 μl LB alone in a 96-well plate. The 10-fold-dilution step was performed 7 times, resulting in a change from the original concentration to a dilution of 10⁷. From each of the 10× dilution series, 3 μl was plated onto an LB agar pad and left to dry (1 colony = 333 cells/ml, representing the lower end of our dynamic range). After an entire experiment had been completed (typically 5 h), the agar was placed into an incubator and grown overnight. Colonies were then manually counted the next morning.

Growth curves.

Growth curves were determined using a 96-well plate and a Tecan Spark plate reader. The plate reader was set to 37°C, and data were collected from each well every 5 min over 10 h.