Metabolic fingerprinting of bacteria by fluorescence lifetime imaging microscopy

Bacterial populations exhibit a range of metabolic states influenced by their environment, intra- and interspecies interactions. The identification of bacterial metabolic states and transitions between them in their native environment promises to elucidate community behavior and stochastic processes, such as antibiotic resistance acquisition. In this work, we employ two-photon fluorescence lifetime imaging microscopy (FLIM) to create a metabolic fingerprint of individual bacteria and populations. FLIM of autofluorescent reduced nicotinamide adenine dinucleotide (phosphate), NAD(P)H, has been previously exploited for label-free metabolic imaging of mammalian cells. However, NAD(P)H FLIM has not been established as a metabolic proxy in bacteria. Applying the phasor approach, we create FLIM-phasor maps of Escherichia coli, Salmonella enterica serovar Typhimurium, Pseudomonas aeruginosa, Bacillus subtilis, and Staphylococcus epidermidis at the single cell and population levels. The bacterial phasor is sensitive to environmental conditions such as antibiotic exposure and growth phase, suggesting that observed shifts in the phasor are representative of metabolic changes within the cells. The FLIM-phasor approach represents a powerful, non-invasive imaging technique to study bacterial metabolism in situ and could provide unique insights into bacterial community behavior, pathology and antibiotic resistance with sub-cellular resolution.


Results
Fluorescence lifetime phasor fingerprint of different bacterial species. Two-photon FLIM of five clinically relevant species of bacteria, E. coli, S. Typhimurium, P. aeruginosa, B. subtilis and S. epidermidis, was performed on planktonic cells immobilized in agarose. Each species was imaged at a mid-exponential phase of growth to generate a phasor fingerprint ( Fig. 1a and Supplementary S1). A comparative analysis of all the phasor coordinates of each bacterial cell, represented by a single data point, within each species population (n = 10) shows variations along both g and s axes of the phasor plot (Fig. 1b). The bacterial phasor distribution displays greater variation along the g axis (between 0.26 to 0.41) than the s axis on the phasor plot (Fig. 1c). Supplementary  Fig. S2 shows the fluorescence emission spectra of the five bacterial species (n = 10) excited at 740 nm.
Response of E. coli phasor to antibiotic exposure. Bacteria encounter various stresses in their natural environments which elicit specific and highly regulated responses via changes in their metabolism 7,9,46,47 . To determine the relationship between the FLIM phasor and metabolic activity, we probed the phasor response of E. coli to antibiotic treatment. Agarose embedded cells were treated with nalidixic acid or ampicillin, a bacteriostatic antibiotic and bactericidal antibiotic, respectively, then NAD(P)H FLIM was performed. The phasor position of E. coli cells treated with nalidixic acid shifted towards longer fluorescence lifetimes with increasing concentration (light to darker shades of brown dots), as compared to the untreated control population (cyan dots) (Fig. 2a). The g distribution exhibited a sharply increasing mean value at the lowest nalidixic acid concentration, while the mean s values were weakly sensitive to nalidixic acid concentration (Fig. 2b). E. coli cells similarly exposed to increasing concentrations of a bactericidal antibiotic, ampicillin, also exhibited corresponding phasor position shifts to larger g values, while s values remained relatively constant with ampicillin concentration (Fig. 2c and d) when compared with cells in control conditions. These shifts of the phasor position of cells treated with antibiotics corresponds to a concentration-dependent increase in the fraction of dead cells when stained in a live/dead assay ( Supplementary Figs S3 and S4). Exposure of E. coli cells to the same volume of water for the same amount of time did not cause a shift and the position of the phasor was maintained at the same position as the control population ( Supplementary Fig. S9).
To benchmark the FLIM phasor data with an independent metabolic activity measure, we used a resazurin assay to quantify aerobic respiration in planktonic cultures. Resazurin is a redox indicator, sensitive to the presence of oxygen and reducing equivalents, such as NADH. Under reducing conditions, resazurin is reduced to the fluorescent compound, resorufin, and the fluorescence intensity has been linked to bacterial respiration rates 48,49 . Exposure of E. coli to the bacteriostatic nalidixic acid resulted in a significant drop in culture respiration, even at the lowest concentration tested in our study ( Supplementary Fig. S5). Exposure of E. coli to bactericidal ampicillin, however, resulted in a slight increase in metabolic activity for cells exposed to 0.04 mg/mL ampicillin, which decreased with further increases in ampicillin concentration. The resazurin data are consistent with results and trends in the FLIM phasor data, which show a large shift to higher free to bound NAD(P)H ratios with exposure to 0.116 mg/mL nalidixic acid, and little to no concentration dependence thereafter (Fig. 2b). The FLIM data do not show a shift to lower free to bound NAD(P)H at 0.04 mg/mL ampicillin, but the data is within error of the starting culture and the change indicated by the resazurin assay is small relative to that observed for nalidixic acid exposure (Fig. 2d). The resazurin assay and FLIM data, however, show a trend to a lower respiration rate and  higher free to bound NAD(P)H ratio, respectively, with increasing ampicillin concentration, which is distinct from the cell response to nalidixic acid as assessed by both methods.

Recovery of E. coli cells after exposure to antibiotics.
To verify whether the bacterial phasor position correlates with global metabolic activity, E. coli cells were allowed to recover after exposure to both antibiotics. Cells were subjected to 0.464 mg/ml nalidixic acid for 30 min and FLIM data was acquired. The cells were then washed and incubated in growth media for 30 min and imaged again. As observed in the previous section, exposure to nalidixic acid shifted the phasor to the right, indicating a higher free to bound NAD(P)H ratio ( Fig. 3a and b). When the cells recovered in media, the phasor distribution of bacteria shifted back towards the left, indicating a lower free to bound NAD(P)H ratio. The phasor of the recovered cells shifted to significantly longer lifetimes than the initial population before exposure to nalidixic acid. Further, the phasor distribution of individual cells within the population along the g axis was much narrower than both the initial control and antibiotic-exposed populations ( Fig. 3a and c).
Ampicillin treated E. coli displayed similar shifts in cell phasors towards shorter lifetimes ( Fig. 3d and e). However, recovery of these cells in fresh media showed a limited shift back to smaller g values compared to cells recovered from nalidixic acid exposure ( Fig. 3c and f). The number of viable cells in the ampicillin treated samples was also lower by 1.8 times compared to the initial and nalidixic acid treated cells ( Supplementary Fig. S6). The phasor distribution of cells in the recovered population was similar to the phasor of the cells in the initial population (Fig. 3f). In both cases, cells exposed to either antibiotic had phasors shifted more significantly along the g axis than along the s axis of the phasor plot in response to treatment ( Fig. 3c and f). By comparing these phasor positions, changes in the ratio of free to bound NAD(P)H within E. coli cells were differentiated between initial and antibiotic-exposed cells. Further, the phasor positions and distribution of the recovered populations upon recovery from bacteriostatic and bactericidal antibiotics represent distinguishing metrics captured by FLIM.

Bacterial phasors as a function of culturing time.
To track the change in bacterial phasor position at different growth phases, we imaged E. coli, S. Typhimurium, P. aeruginosa and B. subtilis cells from shaking cultures at varied growth times ( Fig. 4a and b). The cell density of these bacterial cultures was tracked in parallel using optical density (OD) measurements (Fig. 4c). FLIM data was collected from shaking culture aliquots at regular intervals, starting at 2 h growth time ( Supplementary Fig. S7). No cells were found in the agarose-embedded samples at shorter times to acquire FLIM data. Comparing exponential phase cells, at 4 and 6 h, to the 2 h time point, all cultures except P. aeruginosa exhibited a shift in their mean phasor position to the right, i.e. a larger free to bound NAD(P)H ratio ( Fig. 4a and b). At 8 h and beyond, the mean phasor position of E. coli, S. Typhimurium, and B. subtilis cells shifted back to smaller values of g and began to oscillate, to differing degrees, along the g axis as cultures reached the stationary phase (Fig. 4b). In E. coli and B. subtilis, these oscillations brought the mean phasor positions to even smaller g values than the cultures sampled at 2 h growth. s positions of the mean phasor were not as sensitive as the g position, but tended towards larger values, also non-monotonically, with culture time. As opposed to the other three species, the mean phasor of P. aeruginosa tended to smaller values in both g and s, and no oscillations were observed in g as a function of growth time.
Growth recovery from extended stationary phase cultures. The FLIM phasors of E. coli cells grown for 15 h, 24 h, 48 h and 307 h (13 days) were compared to those grown for the same amount of time and resuspended in fresh media. Figure 5 shows the cell phasor distribution of each of the growth time groups. We observed an increase in the ratio of free to bound NAD(P)H in the cells resuspended in fresh media when compared to their corresponding spent culture media counterparts at all time points except 13 day old cultures. The phasors of recovered cells remained at about the same position for each growth time point, but the phasors of cells taken directly from the spent culture medium shifted closer to the recovered cell phasor position with growth time. The difference in phasor position between the spent media and recovered cultures decreased with longer growth times. E. coli cells grown for 13 days did not show any change in their phasor position after 2 h recovery in fresh media.

Discussion
In this work, we employ label-free NAD(P)H FLIM to characterize fluorescence lifetime fingerprint of bacteria with cellular resolution. The phasor approach to FLIM proves to be a powerful tool to differentiate metabolic states of bacteria via the relative quantification of free to bound NAD(P)H ratios. This ratio is expected to be related to metabolic activity, as NAD(P)+ reducing enzymes are central to catabolic pathways such as glycolysis, the citric acid cycle and the pentose phosphate pathway, which are each active in bacteria under different growth conditions. Similarly, enzymes such as NADH dehydrogenases and NAD(P)H oxidases play a central role in various anabolic pathways and show a high level of specificity in binding to NADPH or NADH [50][51][52] . NAD(P)H FLIM of E. coli, P. aeruginosa, B. subtilis, S. Typhimurium and S. epidermidis was used to generate the bacterial species fingerprint phasors as well as those for the antibiotic experiments. We observed small variations between phasors of control populations of E. coli even when grown under the same conditions and harvested at the same time and OD, perhaps due to normal metabolic variability ( Supplementary Fig. S8). The phasor fingerprint exhibits a distribution of values of individual cells within a population of same species (Fig. 1b and c). These differences suggest a distribution of metabolic activity within and between species populations at the same time of growth. The observed diversity of phasor positions may be a result of varying rates of NAD(P)+/NAD(P) H reduction/oxidation and shifts in NAD(P)H utilization due to catabolic and anabolic adaptations 32,33,53,54 .
The shift of the NAD(P)H phasor position towards shorter lifetimes in response to antibiotic exposure suggests a decrease in enzyme binding of NAD(P)H, indicated by a larger free to bound ratio. In case of bacteriostatic antibiotics this shift tracks with lower respiration activity indicated by the resazurin assay ( Supplementary  Fig. S5). These results are consistent with previously reported data 46,55 which suggest that bacteriostatic antibiotics induce lower oxygen consumption rates (OCR) in cultures by arresting respiration in cells. The increase in respiration activity at low ampicillin concentrations, on the other hand, is consistent with findings that OCR increases upon exposure to bactericidal antibiotics 46,55 . While the effects of higher concentrations of ampicillin on respiration have not been studied, to our knowledge, the resazurin and NAD(P)H FLIM data show a shift to a more reducing environment and an increase in the free to bound NAD(P)H ratio in the phasor plots, respectively.
In the event of cell death, the metabolic state is irreversible; however bacterial cells are known to recover from a state of static growth and dormancy induced by bacteriostatic compounds 9,46,47,56 . The observed return of the mean phasor position to the left of the plot when nalidixic acid was washed away from the E. coli cells ( Fig. 3a and b) is consistent with these previous results. In addition to the shift in the mean phasor, the distribution of single-cell phasors about the mean in the recovered population exposed to nalidixic acid was much narrower than the control and treated populations (Fig. 3c). In the context of the above results, the decrease in phasor variance indicates a recovery to a relatively homogeneous metabolic state throughout the population, as compared to the initial and nalidixic acid-treated populations, which exhibit a greater spread in NAD(P)H lifetimes.
In contrast to the recovery results from nalidixic acid, the shift of the phasor position of cells treated with ampicillin and recovered in fresh medium was less pronounced and their distribution was much broader (Fig. 3d). This phasor data was acquired from a population of cells that did not lyse as a result of the ampicillin treatment, so the extent of phasor recovery is of a biased population and characteristic only of cells that survived the treatment. Indeed, the density of cells that survived this treatment was 19.1 times lower than the nalidixic acid treatment (Supplementary Fig. S6). In addition, the phasor distribution of the recovered cell population was much broader than that of cells recovered from nalidixic acid treatment, indicating more metabolic heterogeneity. The response of bacteria to high-stress environments depends on a range of factors, such as the rate of cell division and initial metabolic states 9, 57-60 . Such factors can direct the fate of part of the population towards persister or dormant states and the rest towards death. Exposure to bactericidal antibiotics also results in generation of reactive oxygen species 61, 62 and broad, and ultimately toxic, changes to the activity of central metabolic pathways in E. coli 55,62 . Consequently, the bactericidal effects of ampicillin, in contrast to nalidixic acid, may result in a population with the diversity of metabolic states observed in the recovered populations.
The metabolism of bacterial populations vary drastically over time, resulting in a collection of different metabolic states that cannot be captured by optical density measurements alone [63][64][65] . In the case of aerobically grown bacteria, these variations depend on the type of nutrient source, temperature of growth, and rate of aeration, among others. Moreover, the bacterial growth curve extends beyond the conventional lag, exponential and stationary phases of growth [66][67][68] . Many interesting physiological phenomena occur during late stationary phase growth, such as the emergence of persister cells and growth advantage in stationary phase (GASP) phenotypes, changes in gene expression and cell morphology, and programmed cell death 60,[68][69][70][71] . Most importantly each bacterium within the population may experience different metabolic and growth rates compared to neighboring cells, yet this information cannot be obtained by cell density measurement or CFU counts.
We employed NAD(P)H FLIM phasor to fingerprint the metabolic states of 4 bacterial species at different stages of growth (Fig. 4a). The earliest time point, 2 h, is already near the end of the lag phase for these cultures. The mean phasor positions of E. coli, S. Typhimurium, and B. subtilis all begin at relatively small g values. In comparison with observed phasors from the antibiotic exposure and recover experiments, these smaller g values suggest a greater metabolic activity. As cells enter their exponential phase, their phasors shift right to larger g values, consistent with a ramp down of metabolic rates as previously determined via indirect methods 64,65 . At 8-10 h growth, this phasor trend reverses for these three cultures, suggesting an additional metabolic transition commensurate with the change in growth rate as cultures enter the stationary phase. In the stationary phase, their phasor positions shift back and forth in g in a manner seemingly unrelated to the cell density (Figs 4b and Supplementary S10). Bacterial cultures display an oscillatory behavior of cell density in the culture with rise and fall during the stationary phase due to GASP phenotypes 69,71,72 , but these typically occur at late stationary phase growth, and the observed phasor shifts are not directly correlated with cell density. Instead, these results indicate that the NAD(P)H phasor is sensitive to metabolic changes within the cells not captured by cell density or CFU analyses.
To look at metabolic adaptations in the stationary phase cultures, FLIM data was obtained from E. coli cultures grown for 15 h, 24 h, 48 h, and 13 d (Fig. 5). Interestingly, cells grown for an additional 2 h in the spent culture medium from 15, 24, and 48 h cultures had phasors at longer lifetimes than the early and mid-exponential phase cells in Figures 1, 2, 3 and 4 (g ~ 0.4). The NAD(P)H lifetimes in these populations were also longer than those of cells from those same cultures that recovered for an additional 2 h in fresh growth medium, which were themselves consistent with early exponential phase E. coli cells (Fig. 4), suggesting a growth recovery. The difference in mean phasor position between cells recovered in fresh and spent medium, however, shrinks with increasing culture time at 48 h and 13 d, indicating a diminished capacity of these bacteria to recover ( Supplementary  Fig. S11). Indeed, cells from the 13 d culture did not show any recovery, suggesting that most cells are dead or in a phenotypic state with an extended lag time to exponential growth. Future experiments will focus on FLIM and independent metabolic rate measurement in late stationary phase cultures to draw quantitative correlations between the two.
In conclusion, we have demonstrated a label-free, two-photon FLIM method for tracking changes in bacterial metabolism. Our results represent the first NAD(P)H FLIM phasor fingerprinting of various bacterial species, and they indicate that the NAD(P)H fluorescence lifetime, as captured in the phasor position, is sensitive to changes in metabolic states within cell populations. The NAD(P)H phasor of cells exposed to antibiotics exhibited the expected shifts due to diminished metabolic capacity, as well as detecting the recovery of cells resistant or tolerant to ampicillin exposure. The distribution of cell phasors within a population also changes as a function of treatment conditions, providing unique insight into single-cell and community behavior. Lastly, NAD(P)H FLIM of bacteria from planktonic cultures exhibited changes in the phasors that did not track with cell density, indicating that this technique can provide additional information about cell metabolism beyond what is inferred from conventional growth rate measurements. This work demonstrates the power of NAD(P)H FLIM to track metabolic states of individual bacteria in situ, and that the bacterial phasor represents a unique and complimentary set of data to conventional metabolism and growth characterization results.

Material and Methods
Bacterial strains and growth conditions. The bacterial strains used in this study are listed in Table S1.
All strains were revived from frozen stocks by streaking on lysogeny broth (LB) agar (1.5%) plates. Shaking cultures of bacterial strains were grown in 2 ml LB at 37 °C for 5 h unless stated otherwise. For imaging, the samples were prepared by mixing LB shaking cultures with 1% agarose in a 3:7 ratio. 100 μl of the resulting solution was cast on a glass coverslip by spin coating at 500 rpm for 10 seconds. Growth curve sample preparation. For each growth curve, 20 ml LB media was inoculated with bacteria to an optical density at 600 nm (OD) of 0.02 and aliquoted into 1 ml cultures in 20 ml glass culture tubes with loose-fitting metal caps, which were kept shaking at 200 rpm and 37 °C. One culture aliquot was taken out for each bacterium every 2 h, and samples for FLIM measurements were prepared as described above.
Growth curve recovery sample preparation. For growth curve recovery 1 ml cultures were grown for 15 h, 24 h, 48 h and 13 d in duplicates. 100 μl from one of the duplicate samples was diluted into 900 μl of fresh LB media and shaken for 2 h at 200 rpm and 37 °C. The other set of cultures from each time point was allowed to grow in the same spent culture media for 2 more hours. After 2 h both cultures were taken out of the incubator and samples were prepared for FLIM imaging as described above.
FLIM data acquisition. FLIM was performed on a Zeiss LSM 710 microscope (Carl Zeiss, Jena, Germany) coupled to an 80 MHz multiphoton excitation laser source, Titanium:Sapphire MaiTai laser (Spectra-Physics, Mountain View, CA) with excitation at 740 nm using a 60X, 1.2 N.A. oil immersion objective, (Carl Zeiss, Oberkochen, Germany). The image scan speed was 25.21 µs/pixel with an image size of 256 × 256 pixels. Excitation from emission signal were separated at 690 nm followed by bandpass emission filter 460/80 nm (Semrock, Rochester, NY). Photomultiplier tube (H7422P-40, Hamamatsu, Japan) was used as the microscope external detector port photo-sensor unit. A320 FastFLIM FLIMbox (ISS, Champaign, IL) was employed to acquire frequency domain FLIM data. SimFCS software (LFD, Irvine) was used for FLIM data acquisition. For calibrating the FLIM system, Rhodamine 110 with known lifetime of 4 ns was measured for every experiment. The (g, s) coordinate system used to indicate phasor cursor coordinates in this article used the first harmonic phasor plots at 80 MHz (repetition rate of the laser).
Spectral data acquisition. NAD(P)H in the bacterial samples was excited with two-photon excitation at 740 nm. Fluorescence spectra were collected on a Zeiss LSM 710 microscope spectral detector consisting of 32 channels, between 416-728 nm, each with 9.7 nm bandwidth. For image acquisition, the pixel frame size was set to 256 × 256 and the pixel dwell time was 177 μs/pixel. Spectral data were acquired from the same sample on which FLIM was performed. Data analysis. FLIM data was analyzed using the SimFCS software developed at the Laboratory for Fluorescence Dynamics (LFD, UC Irvine). To create the bacterial phasors, the average g and s value of phasor distribution from individual bacterial cell were calculated and plotted as a scatter plot. Thus, each point of the bacterial phasor scatter plot represents a single bacterium. Image segmentation, scatter plot and box plots calculations were performed on MATLAB. For Supplementary Fig. S2, the acquired fluorescence spectra were smoothed and normalized to maximum intensity.
Live/dead assay. Agarose-embedded samples were exposed to different antibiotic concentrations for 30 min and stained with Syto 9 and propidium iodide for 20 min. The samples were then washed with ultrapure water and imaged using a Zeiss LSM780 inverted confocal microscope (Carl Zeiss, Jena, Germany) with 63X, NA 1.2 water immersion objective. Dichroic beam splitters were used to reflect laser lines at 488 and 561 nm. Both color channels were imaged using alternating line averaging to avoid blurring of the image due to drifting. These adjustments reduced fluorescence cross-talk to undetectable level in two-color imaging experiments. Confocal images were obtained with a pinhole size of 1 airy unit.
The fraction of dead cells was quantified using the "select objects" from the intensity images from each channel using the Volocity software package after deconvoluting the images by applying the point spread function to each channel.
Colony forming unit (CFU) analysis. 1 ml shaking cultures of E. coli were grown for 5 h and exposed to 0.464 mg/ml nalidixic acid or 0.4 mg/ml ampicillin for 30 min. The cells were centrifuged at 2000 × g, resuspended in LB media and incubated at 37 °C and shaking at 250 rpm for 30 min. The control was not exposed to antibiotics and each condition was done in triplicate. Serial dilutions of the shaking cultures were plated to count CFUs to measure recovery from antibiotic exposure.