A biosensor derived from Listeria innocua LMG2785 was used in this study. This bacterial strain serves as a nonpathogenic surrogate organism for Listeria monocytogenes, a major foodborne human pathogen (Hoffmann et al., 2012; Sarno et al., 2021). Listeria innocua is a Gram-positive bacterium that is susceptible to various bacteriocins, including nisin. The biosensor strain contains the pNZ-pHin2 ^Lm plasmid, which encodes the pHluorin2 gene under the control of the constitutive P_help promoter. Plasmid stability is ensured by chloramphenicol (10 μg mL⁻¹) resistance (Reich et al., 2022).

The pHluorin2 fluorophore has two pH-dependent excitation maxima at approximately 400 and 480 nm (Mahon, 2011; Reich et al., 2022). After excitation at these peaks, the fluorescence intensity at 520 nm changes in a ratiometric, pH-dependent manner. Thus, the extent of the effects of nisin on individual cells after pore formation in the cell wall and the disruption of pH homeostasis in a slightly acidic environment can be measured (Crauwels et al., 2018). The relative fluorescence units (RFU, emission at 520 nm) after excitation with the two excitation maxima were measured to determine changes in the intracellular pH by calculating the ratio of the two values according to the following formula: ratio  RFU 400 / 480 = RFU 400 RFU 480

The complex medium brain-heart-infusion (BHI) (Carl Roth GmbH + Co. KG, Germany) is commonly used for cultivating Listeria species (Jones and D'Orazio, 2013) and was used for precultivation of the biosensor strain. The pH was adjusted to 7.4 with hydrochloric acid or sodium hydroxide if needed, and the BHI base solution was autoclaved. Shortly before inoculation, sterile-filtered chloramphenicol from stock solution was added to the medium to a final concentration of 10 μg mL⁻¹. A chloramphenicol stock solution with a concentration of 34 mg mL⁻¹ was prepared by diluting chloramphenicol sulfate in 100% ethanol.

The optimized Listeria minimal buffer (LMBO: 200 mM MES, 4.82 mM K₂HPO₄, 11.55 mM Na₂HPO₄, 1.7 mM MgSO₄, 0.6 mg mL⁻¹ (NH₄)₂SO₄, and 55 mM glucose) differed slightly in the buffer used (MES instead of MOPS) and the pH (6.2 instead of 6.5) from the previously published LMB (Crauwels et al., 2018), which was based on modified synthetic minimal medium for L. monocytogenes (Tsai and Hodgson, 2003). It was prepared in distilled H₂O, and the pH was adjusted to 6.2 with hydrochloric acid or sodium hydroxide. Afterward, the solution was sterile-filtered and stored at 4 °C. Shortly before inoculation, sterile-filtered chloramphenicol stock solution was added to the buffer to a final concentration of 10 μg mL⁻¹.

A 250 μg mL⁻¹ nisin stock solution was prepared by diluting commercial nisin (2.5% nisin balance sodium chloride, Merck KGaA, Germany) in distilled H₂O. After mixing and sterile filtration, the solution was divided into aliquots and stored at −20 °C.

Listeria innocua LMG2785/pNZ-pHin2 ^Lm was precultivated in 100 mL baffled shaking flasks in 10 mL of BHI medium supplemented with 10 μg mL⁻¹ chloramphenicol. The culture was inoculated with 200 µL of inoculum taken from a glycerol stock and cultivated overnight at 37 °C on a shaker at 120 rpm.

An overnight culture of L. innocua LMG2785/pNZ-pHin2 ^Lm was washed once with phosphate-buffered saline (PBS) and diluted in LMBO buffer to an OD₆₀₀ of 3. Then, 100 µL of cell suspension was mixed with 100 µL of nisin solution in a black microtiter plate. The nisin concentration used in the pHluorin2 assay ranged from 0 to 5 µg mL⁻¹ nisin. After the addition of nisin, the assay mixture was incubated for 30 min in the dark at room temperature (Figure 1A). The emission of pHluorin2 at 520 nm (excitation at 400 and 480 nm) was recorded with a multimode plate reader (Infinite^® M200, Tecan, Switzerland).

A flow cytometer (FC) (Amnis^® CellStream^®, Luminex, United States) was used to analyze nisin-treated L. innocua LMG2785/pNZ-pHin2 ^Lm as described previously (Reich et al., 2022). Biosensor cells were diluted 1:50 in PBS, and 50 µL of cell suspension was analyzed for pHluorin2 fluorescence emission at 528 nm using 405 and 488 nm laser irradiation. The laser powers used were 10% (forward scatter and side scatter), 35% (405 nm) and 40% (488 nm). The flow speed was set to “slow” for maximal sensitivity. Gating was performed as described by Reich et al., 2022. The flow cytometry data were analyzed using FlowJo (v10) software.

A polydimethylsiloxane (PDMS)-based microfluidic system was used for cell trapping and time lapse imaging (Grünberger et al., 2015). The chip design features inlet and outlet structures for loading cells and injecting nisin to maintain a stable nisin concentration and defined conditions during measurement. Cell trapping was achieved by several chambers (80 × 90 µm) with a height of approximately 650 nm arranged in multiple arrays, as shown in Figure 1B. A silicon wafer containing the negative structure was used as a mold for fabrication. Two-component PDMS was mixed at a base:linker ratio of 10:1 (Sylgard 184 Silicone Elastomer, Dow Corning Corporation, United States) and used to cast the structure of the wafer. After the air was removed in a desiccator and the mold was baked at 80 °C for 2 h, the mold was cut out and prepared for plasma bonding. This included cleaning the glass substrate (D 263 T eco, 39.5 × 34.5 × 0.175 mm, Schott, Germany) and the mold itself with analytical-grade isopropanol as well as punching the inlet and outlet structure with a 0.75 mm biopsy puncher (Reusable Biopsy Punch, 0.75 mm, WPI, United States). After surface activation in a plasma generator (Diener Femto Model 1 B2, Diener electronic, Germany), the PDMS mold was placed on a glass substrate for covalent bonding, creating closed structures and a functional microfluidic chip. For a detailed fabrication protocol the reader is referred to Gruenberger et al. (Gruenberger et al., 2013).

Time lapse imaging was performed with an inverse microscope (Nikon Eclipse Ti2 Series, Nikon, Germany) (Figure 1C). After fabrication, the microfluidic chip was fixed on an in-house manufactured sample plate and loaded with stationary cells from the preculture as described above (Figure 1A).

Data from the MTP assay heterogeneity comparison experiments (chapter 3.1) were collected using a sample from the preculture that was prepared by a procedure similar to that described for the microtiter plate assay. The sample was loaded on the chip shortly before the end of nisin incubation and measured after 30 min of nisin application. For temporally resolved experimental data (chapters 3.2 and 3.3), the loaded sample was obtained immediately after mixing the stationary culture with nisin. During the measurement, the chip was perfused with media with the tested nisin concentrations using a pressurized pump system (Microfluidic Flow Control System™, Fluigent, Germany) to eliminate changes in concentration due to solvent evaporation or nisin degradation. After inoculation of the microfluidic chip, the chamber area was homogenously covered with scattered single cells to counteract limitation-like effects and decrease the possibility of nisin deficiency in supply channels to a negligible level. Furthermore, cell-to-cell fluorescence heterogeneity effects can be distinguished from effects that can be attributed to cell proliferation.

Phase contrast microscopy was used to identify chambers suitable for measurement; the region of interest (ROI) was marked, and the possible movement of cells over time was checked to exclude false values due to deviations in the cell position and chosen ROI. Quantitative fluorescence signal measurements were conducted using two different fluorescence filter combinations with the same emission filter to represent the two pHluorin2 excitation maxima at 400 nm and 480 nm. The illumination intensity was regulated using a light engine (SOLA Light Engine, Lumencor, United States) and was set to 10% to reduce photobleaching.

The standard workflow for data analysis is shown in Figure 1D. Image processing and analysis were performed using the Fiji distribution of the open-source software ImageJ (Schindelin et al., 2012). The cells and regions for the endpoint measurements 30 min after nisin exposure (Figure 3) were selected and measured manually, whereas images for dynamic heterogeneity analysis (Figure 6; Figure 7) were obtained with automatic built-in tools in Fiji. As a first step, the cells were identified in the phase contrast image using the “Analyze Particles” feature, and watershed transformation was applied to separate multiple single cells that were initially identified as one cell. Fine-tuning of selected regions was carried out manually. Image stabilization was performed prior to selecting the regions, if necessary. After checking and applying the ROI for every timepoint and fluorescence channel, the fluorescence values were measured as the grayscale value of the light intensity at the image sensor. The ratiometric signal of each cell at each timepoint was then calculated from the maximum values of the respective cell in both fluorescence channels after subtracting the mean background signal value from the fluorescence value. While maximum values generally might not be as robust as the mean value for the assignment of a determined value to a specific cell, the negative impact of manual analysis and differing parameters for ROI identification on the obtained data is greater for the mean, which is thus more prone to error. The determination of the blank value of each image was the opposite. Single pixels could be overexposed due to impurities, refraction or reflection phenomena, leading to high impacts on the determined signal when calculating fluorescence ratios. The mean values over an extended area of the chamber were more reliable and descriptive of the background signal. Cells that did not show a notable signal higher than the background were not taken into account and were excluded from the measurement to prevent false distortion of the measuring principle and signal heterogeneity.

Signal heterogeneity can be quantified and displayed as the standard deviation of all measured single-cell values, whereas the mean value of all measured cells corresponds to the mean sensor signal of the population at the respective nisin concentration. This statistical evaluation method also addresses outlier values, which must be considered when analyzing signal heterogeneity at the single-cell level, as the signals of individual cells can differ immensely from the mean value of a given population.

Similar to the gating method used in flow cytometry analysis (see Supplementary Material, Supplementary Figure S1), the cutoff of the distributions can be used to divide the cultures into two populations and to define signal ranges in which the cells can be classified as “dead” or “alive”. In the microfluidic setup, the signal range for living cells was greater than approximately 0.5 (Figure 3, 0 μg mL⁻¹ nisin), and the signal range for dead cells was a ratiometric fluorescence signal less than approximately 0.4 (Figure 3, 5 μg mL⁻¹ nisin).

In the next step, microfluidic experiments were performed to determine the temporal dynamics of single cells. As observed in the results presented thus far, at concentrations of 0 and 5 μg mL⁻¹ nisin, cells behaved quite uniformly, and individual single-cell values were nearly indistinguishable from one another (Supplementary Figure S4).

However, at intermediate nisin concentrations (0.3125–1.25 μg mL⁻¹), considerable single-cell variability in the temporal response of the sensor bacteria was observed (Figure 7). At a nisin concentration of 0.3125 μg mL⁻¹, the population still behaved quite uniformly, but individual cells were susceptible to nisin, causing a decrease in the ratio RFU over time. These cells were outliers, as the majority of cells were only slightly affected by low nisin concentrations. With increasing concentrations, an increasing proportion of susceptible cells and a greater impact of time on cell-to-cell heterogeneity were observed. At a nisin concentration of 0.625 μg mL⁻¹, individual cells showed an initial effect after nisin exposure, but the ratio RFU of each cell, and therefore the cell-to-cell heterogeneity, remained almost constant. In contrast, at a nisin concentration of 1.25 μg mL^−1, the population showed increasing cell-to-cell heterogeneity over time. This increase was due to the disruption of single cells over time, which increased the proportion of dead cells at subsequent time points. When viewing the temporal course of individual cells, different transition times and dynamics from intact to disturbed pH homeostasis can be observed. Changes in cell-to-cell heterogeneity over time can easily be observed as signal changes of single cells in ratiometric images (Figure 7). Earlier studies investigated E. coli after exposure to the antimicrobial peptides LL-37 and Cecoprin A with high spatio-temporal resolved analysis (Sochacki et al., 2011; Rangarajan et al., 2013). The studies revealed concentration and time dependent cell-to-cell heterogeneity as well as heterogeneity in degree of binding and spatial distribution of LL-37 (Sochacki et al., 2011; Choi et al., 2016). Similar underlying effects that caused the heterogeneity in the time lag between peptide addition and permeabilization of the outer membrane in E. coli (Choi et al., 2016), could cause changes in cell-to-cell heterogeneity of the L. innocua biosensor over time.

An especially interesting finding was obtained at a nisin concentration of 2.5 μg mL⁻¹. All cells showed a direct effect and susceptibility to nisin, as expected, which resulted in a ratio RFU of approximately 0.25–0.3 from the start of the measurement. Overall, the majority of cells behaved quite homogenously and shared a similar ratio RFU throughout the measurement time. However, the temporally resolved single-cell data revealed persistent cells that appeared in rare instances and partially reestablished pH homeostasis (verification in ratiometric pictures–Figure 7). The possibility of reestablishing pH homeostasis, which is suggested by the data, indicates the adaptation processes and strategies of the whole-cell biosensor strain. Thus, the ratiometric fluorescence signal of single cells should be carefully analyzed and critically interpreted because even cells with disrupted cell membranes and an unusually low intracellular pH might show persistent behavior in response to their cytotoxic environment and regain pH homeostasis. Nisin-resistant strains of L. innocua were reported a long time ago (Maisnier-Patin and Richard, 1996). Abnormal cell wall synthesis and thickened cell walls were assumed to be resistance mechanisms that enabled the resistant strains to withstand high nisin concentrations. In addition, many other mechanisms of nisin resistance have been described for both L. monocytogenes and L. innocua and might explain these cells that adapted to nisin treatment (Kaur et al., 2011; Bergholz et al., 2013; Zhou et al., 2014; Pang et al., 2022). However, because multiple strategies to cope with nisin treatment have been identified for L. monocytogenes, including resistance and tolerance (Zhou et al., 2020) as well as persister cell formation (Wu et al., 2017), further experiments are needed to distinguish the underlying principles and mechanisms involved. More studies investigated persistence and resistance of pathogenic bacteria after treatment with antimicrobial compounds. Differences in the survival rate of single cells and frequency of persister cells might be growth phase dependent. This was observed for Pseudomonas syringae pv. phaseolicola 1448A after streptomycin and tailocin treatment at different growth phases, i.e., log, early stationary and late stationary phase (Patel et al., 2021). Monitoring of intracellular pH of Mycobacterium avium subsp. paratuberculosis after exposure to nisin and neutralized cell-free supernatants of Lactobacillus plantarum PCA 236 over a time span of 24 h revealed the formation of a probably resistant subpopulation (Gaggìa et al., 2010). Single cell analysis of antibiotic-resistant isolates of Salmonella enterica revealed heterogenous gene expression levels and efflux pump activity between cells, suggesting that functional activity and gene expression fluctuations can contribute to adaptive resistances (Sánchez-Romero et al., 2014).

Overall, the data revealed two important findings. First, the mean signal value of the assay arises from the formation of subpopulations and cells with bistable behavior between the signal ratios of intact and disrupted cells. Second, individual persistent or resistant cells were observed that might adapt to nisin-containing environments and regain pH homeostasis. Possible adaptation processes must be investigated and characterized in further experiments.

The methods presented are applicable for characterization and investigation of other bacteriocins and antimicrobial compounds with the same or similar modes of action, for instance pore forming, membrane disrupting, and pH homeostasis influencing compounds. This has already been showed for bacteriocins from class IIa, i.e., Pediocin PA-1 (Chikindas et al., 1993; Khorshidian et al., 2021; Reich et al., 2022) and class IId, i.e., Garvicin Q (Tosukhowong et al., 2012; Desiderato et al., 2022). However, bacteriocins with different modes of action (Perez et al., 2018; Pérez-Ramos et al., 2021; Darbandi et al., 2022), such as DNA replication or protein synthesis inhibiting microcins (Yang et al., 2014), as well as bacteriocins originating from the species of the model organism itself (Darji et al., 1995; Lee, 2020; Meza-Torres et al., 2021) cannot be characterized using the presented biosensor strain. Modified or novel fluorescence-based biosensor strains might be used in combination with the presented methodology for analogous characterization of other antimicrobial compounds and their effects on the respective model organism.