Overnight E. coli cultures were diluted 1:200 in super optimal broth (SOB) (Sigma Aldrich) and incubated at 37°C until an OD₆₀₀ value of 0.3–0.5 (~3 h). Next, the cultures were spun at 5000 g for 10 min. The supernatant was removed and remaining pellet was concentrated 10 times in transport and storage solution (TSS: LB Miller supplemented with 10% PEG 3350, 20 mM MgSO₄ and 5% dimethyl sulfoxide), placed immediately on ice and left for an hour. 100–200 ng of DNA to be transformed was diluted to 200 μL in chilled TCM (10 mM Tris/HCl at pH 7.5, 10 mM MgCl₂,10 mM CaCl₂) and kept on ice. 200 μL of prepared as above competent cells was then transferred to tubes containing the TCM-diluted DNA and mixed by pipetting, taking care to not warm up the tubes. The mix was left on ice for an hour. Afterwards the samples were transferred to a heating block set to 45°C for 2 min and returned to ice for 5 min. Next the samples were diluted with 0.6 mL of SOC (SOB medium supplemented with 20 mM filter-sterilized glucose) and incubated at 37°C, with shaking at 200 rpm for 1 hour. 100 μL of the sample was then spread on Lysogeny Broth (LB) plates supplemented with 100 μg/mL ampicillin and incubated at for 24 h. PUC19 plasmid was used as positive transformation control and non-transformed competent E. coli cells were used as the negative control.

Glycerol (25%) in LB bacterial stocks were stored at −80°C, streaked on LB (Sigma Aldrich) agar plates or LB agar plates containing 100 μg/mL of ampicillin for pRSETB-Peredox transformed strains. Pates were incubated at 37°C overnight. Next, single colonies were inoculated into 1 mL of LB or LB with 100 μg/mL ampicillin added for plasmid retention. Cultures were incubated at 37°C overnight, with shaking at 200 rpm. Overnight cultures were then transferred to a black, clear-bottom, 96-well plate (ThermoFisher Scientific) (200 μL/well). Total biomass (OD₆₀₀) and fluorescence intensity measurements were performed in the CLARIOstar plate reader (BMG LabTech). Fluorescence intensity was recorded at 575 nm excitation, 610 nm emission for mCherry (plasmid copy number), in precise mode, sequential well scan and automatic gain adjustment. Fluorescence intensity was normalized to OD values and compared between pRSETB-Peredox-transformed and untransformed samples in GraphPad Prism (GraphPad).

Overnight (16–18 h) E. coli cultures were diluted to an OD₆₀₀ of 0.05 in LB and incubated for 2–3 h at 37°C, with shaking at 200 rpm until the cultures reached mid-log phase of growth (OD₆₀₀ of ~1 or CFU/mL of 8^10⁸–1.5^10⁹). OD₆₀₀ measurements in a 1 cm plastic cuvette, as well as a Miles & Misra (MM) dilution series (Miles et al., 1938) on LB agar plates was performed to determine starting CFU/mL values (i.e., time 0, prior to antibiotic addition). MM dilutions were prepared as 200 μL final volume and 10 μL/‘spot’ were plated with 3 ‘spots’ per replicate Next, ciprofloxacin was added at 25XMIC concentration (0.4 μg/mL, determined experimentally) to achieve concentration-independent killing, as outlined by a recent consensus statement (Balaban et al., 2019). Cultures were then returned to the shaking incubator and an MM series was performed at 5 and 24 h following antibiotic addition. At T₅ and T₂₄ antibiotic was removed by washing prior to carrying out the MM series: 100 μL sample was removed from the culture at each timepoint and centrifuged at 5000 g for 5 min. The cell pellet was then resuspended in 100 μL of standard phosphate-buffered saline (PBS), pH 7.4 (Sigma Aldrich) or fresh media and this suspension was used in the MM dilution series. MM plates were incubated at 37°C for 48 h, as it was previously determined that the HipQ strain persisters required 48 h for regrowth on agar plates. For direct comparison, time-kill assays and cytosolic [NAD(H)] (section 2.4) measurements were carried out on the same bacterial cultures simultaneously.

The time-kill assay method described above was followed, with the exception that at T₀ (prior to antibiotic addition) and T₂₄ (24 h post antibiotic addition) sample aliquots were diluted 1:500 in sterile PBS (to 10⁶ total cells/mL). Next, cells were live/dead stained with Zombie near-infrared (near-IR) fixable dye (BioLegend) (1:1000 or 1 μL/10⁶ cells) and incubated for 20 min in the dark. For heat-killed control samples, undiluted aliquots (1 mL) were incubated at 75°C for 30 min and diluted to 10⁶ total cells/mL in PBS prior to live/dead staining. Next, cells were loaded onto the LSR Fortessa Cell Analyser (BD Biosciences) and forward scatter (FSC), side scatter (SSC) and zombie near-IR fluorescence (650 nm excitation, 746 nm emission) was recorded. Additionally, T-sapphire fluorescence intensity ([NADH:NAD+]) was recorded at 405 nm excitation and 520 nm emission and mCherry fluorescence intensity (Peredox plasmid copy number) was recorded at 561 nm excitation and 610 nm emission wavelengths. Laser power was kept consistent throughout the independent experiments. Subsequent data analysis was performed in FlowJo v.10.8 (BD Biosciences). First, doublets were excluded based on FSC-A/FSC-H ratio (Cosma, 2020) and then cell population was gated manually based on FSC & SSC values. Live/dead discrimination was performed by applying a bifur gate on histogram plot of zombie near-IR fluorescence, using live cells (mid-exponential cultures, no antibiotic) and heat-killed cells as negative and positive staining controls, respectively. The singlet subpopulation was further split into NADH biosensor expressing (mCherry positive) and non-expressing subpopulations (mCherry negative) with a bifur gate, using untransformed cells as the negative control. Untransformed cells were also used to quantify autofluorescence for both aforementioned fluorophores. To quantify relative NAD(H) of individual cells, T-sapphire fluorescence intensity was plotted against mCherry fluorescence intensity of all events was done in FlowJo v.10.8 (BD Biosciences). For statistical analysis, fluorescence intensity values of T-sapphire and mCherry of individual events were exported into Microsoft excel, T-sapphire fluorescence was normalized to mCherry fluorescence and values were averaged within biological replicates. All statistical analyses (see respective figure legends) were performed in GraphPad Prism v.8. The antibiotic tolerant subpopulation (persisters & VBNCs) was classified as cells which stained as ‘live’ following antibiotic exposure was estimated by subtracting the % of persister cells in a given biological replicate from the % of live cells at T₂₄.

pRSETB-Peredox (also referred to as NADH biosensor) was successfully transformed into the highly persistent E. coli HipQ and its parental strain, confirmed by growth on ampicillin (pRSETB selection marker) agar plates. The biosensor consists of T-Sapphire-Rex fusion protein whose fluorescence intensity correlates with intracellular [NADH:NAD+], as well as constitutively expressed mCherry which can be used to normalize T-sapphire fluorescence to plasmid copy number (Hung et al., 2011; see Supplementary Figure S1 for pRSETB-Peredox plasmid map).

Since NADH biosensor expression is driven by the viral T7 promoter (Supplementary Figure S1), initially no expression was expected without a subsequent transformation of E. coli strains with a T7 polymerase-expressing vector. However, pRSETB-Peredox transformants formed red-tinted colonies on agar plates, and it was subsequently confirmed by fluorescence measurements, relative to the untransformed E. coli controls, that those cells expressed mCherry, confirming expression from the T7 operon (Figure 1). Consequently, it was confirmed by PCR for the T7 polymerase gene that neither of the E. coli strains was infected with the T7 phage (Supplementary Figure S2). It is therefore reasonable to assume that the mCherry expression is a result of ‘leaky’ expression from the T7 promoter and therefore this biosensor can be used in T7 polymerase negative E. coli strains without further genetic modification.

Once expression of the NADH biosensor was confirmed in E. coli, the next step was to investigate whether it affects antibiotic tolerance, possibly by slowing down cell growth and/or inducing the stress response due to the presence of protein aggregates (Schramm et al., 2020). Although growth rates, measured as optical density change over 24h, showed a significant difference between NADH biosensor-expressing cultures and their respective untransformed controls (Figure 2A), no statistically significant difference in CFU/mL was observed after at 3 h of growth (Figure 2B), the starting time point of the time-kill assays. Since CFU/mL measurements are considered the ‘gold-standard’ for growth assays, as they are not influenced by cell size and/or the presence of dead cells, both biosensor-expressing and untransformed cultures were assumed to be in the same growth phase at 3 h.

Next, ciprofloxacin persister frequencies of biosensor-expressing and untransformed cultures were compared. Ciprofloxacin was chosen as it is non-lytic (LeBel, 1988) so the biosensor would be retained in the cells following antibiotic exposure. Also, the HipQ strain is known to be highly persistent to fluoroquinolones (Wolfson et al., 1990). As can be seen in Figure 2C, there was no statistically significant difference in the % survival (persister subpopulation) of biosensor-expressing and untransformed cultures following 24 h of antibiotic exposure. There was also no statistically significant difference in the frequency of ciprofloxacin persisters between E. coli HipQ and its parental strain, however there was a high degree of variability between biological replicates. The difference in the frequency of persisters between HipQ and its parental strain was in line with previous observations, with E. coli HipQ generating 10–1,000 fold higher number of persister cells (Wolfson et al., 1990). It is possible that the non-lytic mechanism of action of ciprofloxacin could have resulted in a high frequency of damaged, VBNC-like cells whose regrowth was extremely sensitive to environmental conditions, resulting in increased variance in recorded frequencies of persister cells. This hypothesis was further validated by viability staining of antibiotic-exposed cultures (Figure 3), which demonstrated that a significant proportion of cells were alive following antibiotic exposure, although only a small minority of these cells regrew during our experimental timeframe.

Furthermore, there was no statistically significant difference in the average of total antibiotic-tolerant cells (persisters and VBNCs, or cells which stained as ‘live) following 24 h of ciprofloxacin exposure between biosensor-expressing and untransformed E. coli (Figure 3), or between HipQ and its parental strain. The frequency of live cells (persisters and VBNCs) was also several orders of magnitude higher than the frequency of persisters, demonstrating that following ciprofloxacin exposure the vast majority of the antibiotic-tolerant subpopulation is in the VBNC state. It is noted that the live/dead discrimination method was based on cellular membrane integrity and therefore could have led to an overestimation of ‘live’ cells, given the fact that ciprofloxacin is a non-lytic antibiotic. However, it was subsequently demonstrated, that ‘live’ cells undergo respiration at higher rates than ‘dead’ cells (Figure 4), investigated further in section 3.1.1 providing evidence that this subpopulation consists of metabolically active (i.e., living) cells.

Interestingly, the frequency of persisters in a biological replicate did not correlate with the frequency of VBNCs, i.e., a higher % of live cells did not consistently equal to a higher number of persisters (Figure 5). This suggests that ciprofloxacin-related persister and VBNC state is potentially influenced by different mechanisms, or in other words that likelihood of regrowth is independent of likelihood of antibiotic exposure survival. This explanation is further supported by the finding that the proportion of total antibiotic-tolerant cells (persisters and VBNCs) is not significantly different between the highly-persistent HipQ strain and its parental counterpart after treatment with ciprofloxacin (Figure 3).

In this assay exponentially growing cultures were exposed to 25XMIC of ciprofloxacin and cytosolic [NADH:NAD+], as well as viability of individual cells was measured with a flow cytometer. As mentioned previously (section 3.2), ciprofloxacin is non-lytic and therefore it was reasoned that the majority of ciprofloxacin-exposed cells will retain their protein content (including the Peredox biosensor), regardless of their viability status.

Previous research has demonstrated that ciprofloxacin exposure results in a high number of VBNC cells (up to 90% of the population) (Mason et al., 1995) which is consistent with observations presented in this article. This is likely a consequence of mechanism of action of this antibiotic since ciprofloxacin inhibits DNA gyrase, impairing DNA replication and repair without causing cell lysis (LeBel, 1988). However, VBNC cells have also been observed in response to lytic antibiotics, such as ampicillin, although at a much lower frequency (Mohiuddin et al., 2020b; Goode et al., 2021).

There has been an ongoing debate in the field whether the VBNC state is separate phenomenon to persistence, with some research groups reporting a distinct phenotypic state associated with VBNC cells (Bamford et al., 2017; Goode et al., 2021), and others suggesting that VBNCs are simply ‘aged’ persisters, which consequently display an extremely long lag phase (Ayrapetyan and Williams, 2018; Kim et al., 2018). Regardless of classification, VBNCs are a clinically relevant subpopulation and pathogen’s such as enteropathogenic E. coli, Vibrio cholerae (Senoh et al., 2012) and Mycobacterium tuberculosis (McCune et al., 1966) VBNC cells have been shown to be resuscitated in in vivo (or in vivo mimicking) conditions, which could contribute to infection recurrence. Therefore, better understanding of the VBNC state would aid in design of improved antibiotic treatment regimens, effective against heterogenous antibiotic-tolerant subpopulations which include both persisters and VBNCs.

It is important to note that the definition of ‘viability’ regarding the viable but non culturable state differs and can be based either on transcriptional/translational activity and membrane integrity (e.g., Goode et al., 2021) or on membrane integrity alone (e.g., Ayrapetyan and Williams, 2018; Kim et al., 2018). However, recent findings suggest that membrane integrity is not a sufficient determinant of viability and that there are cell-like particles, filled with protein aggregates or lacking cytosolic contents altogether, which maintain intact membranes (Song and Wood, 2021). The authors of the aforementioned findings suggest that these ‘shells’ could have been mistakenly classified as VBNCs in previous research. Therefore, in this article we based our live/dead cell discrimination on both membrane integrity and metabolic activity, demonstrating that ciprofloxacin VBNCs are not dead ‘shells’ as they actively undergo respiration. However, it is possible that the ‘zombie’ subpopulation we observed, which consisted of ‘live’ cells (i.e., cells with intact membranes) with a very low level of biosensor fluorescence were these cell-like particles described by Song and Wood (2021). Moreover, these observations further justify our reasoning of excluding the ‘zombie’ subpopulation from our analysis.

In this work, we sought to investigate the intracellular redox homeostasis of ciprofloxacin-tolerant cells. Previous work by Orman and colleagues has demonstrated, with the innovative use of flow cytometry combined with redox sensor green (RSG) staining, that in exponentially growing E. coli cultures, ofloxacin persisters were more likely to have a lower reductase activity than the bulk population (Amato et al., 2013). We validated these results with a closely related quinolone, ciprofloxacin. However, we note that reductase activity measured by Orman et al. is not specific to [NAD(H)], as there are NAD(H) independent redox reactions occurring within cells, such as NADP(H) or ATP mediated reactions (Gowda, 2018). To the best of our knowledge the enzyme(s) involved in RSG dye redox reaction, used by the authors of the aforementioned study, are not specified. Here we demonstrated that subpopulation tolerance to ciprofloxacin in E. coli is related specifically to a lower intracellular [NADH] and subsequently, downregulated cellular respiration (Pinzon et al., 2011; Ricciardi et al., 2014; Braissant et al., 2020). Respiration-inducing drugs have been previously reported to be active against ciprofloxacin persisters (Marques et al., 2014), and our work suggests that those could also be effective against VBNCs. However, we also found that cells that survive prolonged antibiotic exposure consisted of two subpopulations: those whose [NADH:NAD+] was lower than the bulk population prior to antibiotic exposure and a smaller subpopulation with a higher NADH:NAD+ than the bulk population. This is further evidence that the antibiotic-tolerant subpopulation is itself heterogenous and suggests a combination therapy approach might be necessary for its eradication. It is noted that coenzyme flavin adenine nucleotide (FAD) has a similar fluorescence spectrum to T-Sapphire. Cytosolic FAD has previously been shown to increase in stressful environmental conditions, such as antibiotic exposure (Surre et al., 2018). Although in this work we did not find a significant increase of autofluorescence between the bulk population and antibiotic-tolerant cells on batch level, it is possible that on a single-cell level antibiotic-tolerant cells which displayed a high level of fluorescence produced a higher amount of FAD, rather than NADH.

It has been previously shown that E. coli HipQ generates 10–1000 fold more ofloxacin persisters than its parental strain (Wolfson et al., 1990). This agrees with ciprofloxacin persister cell frequencies presented in this work, since ofloxacin is a closely related quinolone. Moreover, we demonstrated that E. coli HipQ antibiotic-tolerant cells have a lower [NADH:NAD+] than those of its parental strain. This suggested that respiration occurs at a lower rate in these cells. E. coli HipQ has previously been shown to generate an increased number of stochastic persisters, i.e., cells at the ‘tail-end’ of growth/division rate distribution (Balaban et al., 2004; Hingley-Wilson et al., 2020). Our observations point towards a link between stochastic persistence and cellular respiration, with cells at the ‘tail-end’ of the respiration rate distribution being more likely to survive antibiotic exposure, however this hypothesis requires further experimental validation.

This study presents a framework for the use of NADH biosensor Peredox in investigations of NAD(H) homeostasis of antibiotic-tolerant and antibiotic-susceptible bacterial cells.

In this work we demonstrated that the majority of E. coli ciprofloxacin-tolerant cells are in the VBNC state. These observations could be easily expanded on in the future work to other non-lytic antibiotics and/or bacterial species, following the method we established and validated here. Furthermore, we evaluated dysregulated NADH homeostasis as a potential marker for antibiotic tolerance and showed that the antibiotic tolerant subpopulation differs significantly from the bulk population in their cytosolic [NADH:NAD+]. We note that the biggest limitation of our study is that we were not able to determine whether this dysregulation occurs as cause or effect of antibiotic tolerance (in other words, whether antibiotic-tolerant cells showed this NAD(H) homeostasis dysregulation prior to antibiotic exposure). To expand on the observations presented here, we suggest employing fluorescence-activated cell sorting (FACS), in order to sort populations based on [NADH:NAD+] prior to antibiotic exposure, which would allow to answer the aforementioned research question, and subsequently validate whether [NADH:NAD+] can be used as antibiotic tolerance marker. Furthermore, to increase sensitivity of this system, higher level of Peredox expression could be obtained by inducing T7 promoter-dependent expression. However, the potential effect of high amounts of recombinant protein on cellular metabolic state (e.g., through ribosomal sequestration or formation of protein aggregates), and subsequently antibiotic tolerance, would need to be evaluated first.