Dissecting antibiotic effects on the cell envelope using bacterial cytological profiling: a phenotypic analysis starter kit

ABSTRACT Phenotypic analysis assays such as bacterial cytological profiling (BCP) have become increasingly popular for antibiotic mode of action analysis. A plethora of dyes, protein fusions, and reporter strains are available and have been used for this purpose, enabling both rapid mode of action categorization and in-depth analysis of antibiotic mechanisms. However, non-expert researchers may struggle choosing suitable assays and interpreting results. This is a particular problem for antibiotics that have multiple or complex targets, such as the bacterial cell envelope. Here, we set out to curate a minimal set of accessible and affordable phenotypic assays that allow distinction between membrane and cell wall targets, can identify dual-action inhibitors, and can be implemented in most research environments. To this end, we employed BCP, membrane potential, fluidity, and cell wall synthesis assays. To assess specificity and ease of interpretation, we tested three well-characterized and commercially available reference antibiotics: the potassium ionophore valinomycin, the lipid II-binding glycopeptide vancomycin, and the dual-action lantibiotic nisin, which binds lipid II and forms a membrane pore. Based on our experiments, we suggest a minimal set of BCP, a membrane-potentiometric probe, and fluorescent protein fusions to MinD and MreB as basic assay set and recommend complementing these assays with Laurdan-based fluidity measurements and a PliaI reporter fusion, where indicated. We believe that our results can provide guidance for researchers who wish to use phenotypic analysis for mode of action studies but do not possess the specialized equipment or expert knowledge to employ the full breadth of possible techniques. IMPORTANCE Phenotypic analysis assays using specialized fluorescence fusions and dyes have become increasingly popular in antibiotic mode of action analysis. However, it can be difficult to implement these methods due to the need for specialized equipment and/or the complexity of bacterial cell biology and physiology, making the interpretation of results difficult for non-experts. This is especially problematic for compounds that have multiple or pleiotropic effects, such as inhibitors of the bacterial cell envelope. In order to make phenotypic analysis assays accessible to labs, whose primary expertise is not bacterial cell biology, or with limited equipment and resources, a set of simple and broadly accessible assays is needed that is easy to implement, execute, and interpret. Here, we have curated a set of assays and strains that does not need highly specialized equipment, can be performed in most labs, and is straightforward to interpret without knowing the intricacies of bacterial cell biology.

TEXT S1 Selecting antibiotic concentration, growth conditions, and fluorescence dyes for phenotypic analysis.

Selection of antibiotic concentrations
Choosing the right antibiotic concentration and treatment time is crucial for successful phenotypic analysis.Too low concentrations will not lead to visible effects, while too high concentrations can kill or lyse the cells, leading to pleiotropic effects due to cell death and disintegration.Even for bacteriostatic antibiotics an overdose can lead to off-target activities that would not be observed at inhibitory concentrations.Likewise, treatment time can be detrimental.In order to observe the direct effects of an antibiotic, short treatment times are indicated.In our experience, 10 to 15 minutes tend to be suitable for most antibiotics.However, some compounds require longer treatment times to show an effect e.g., beta-lactams, fluoroquinolones, or cell division inhibitors, whose effects only become apparent after one or more rounds of cell division have (or should have) occurred (1).
Simply choosing the MIC is not sufficient in many cases, as the conditions differ between MIC incubation and phenotypic experiments with respect to cell density, growth phase, culture volume, and aeration, to name a few.Similarly, killing kinetics are not suitable to determine the optimal treatment time as dead cells should be avoided.Therefore, acute shock experiments to determine the growth behavior after antibiotic addition in log phase are the method of choice (see Figure S1).Usually, a concentration leading to a 50-70% reduction in growth rate is suitable for phenotypic experiments (referred to as optimal stressor concentration, OSC) (2).However, for some compound classes, in particular membrane-active molecules, this window can be hard to near-impossible to find as they tend to result in an 'allor-nothing' response.In such a case, it can be necessary to choose multiple concentrations and monitor the growth rate during each experiment to ensure that growth inhibition but no lysis occurred (3).Some compounds have a two-staged mechanism of action e.g., nisin which binds to lipid II and after reaching a critical concentration forms a transmembrane pore.In this case, in order to capture both activities of the compound, slightly higher, near-inhibitory concentrations have to be used.Accordingly, in this study we used near-inhibitory concentrations of all antibiotics to have comparable stress levels under the different conditions (Figure S1).

Selection of growth conditions
It is pivotal to perform acute shock experiments under the same conditions phenotypic analysis is carried out.Likewise, conditions should be the same for all phenotypic assays in a study, as far as the specific experimental setups allow.Like antibiotic concentration and treatment time, media composition, growth temperature, culture volume, and aeration, to name a few, can influence how cells grow, how they present phenotypically, and how well an antibiotic compound works.
Media composition is a crucial and well-known factor that influences antibiotic activity (4,5).Even for very similar media like MHB and LB, which are both full media commonly used for microbiology, significant differences can be found in MIC (e.g., 8-fold for nisin, Table S1).Likewise, the OSCs differ considerably (e.g., 5-fold for valinomycin, Table S1, Figure S1-2).Thus, conditions must be kept as similar as possible and if an antibiotic requires a specific supplement (e.g., valinomycin requiring KCl media), a separate untreated control culture in the modified medium is essential.

Selection and compatibility of dyes
Fluorescence dyes are crucial for BCP and many other phenotypic assays, but not all dyes are suitable for all setups e.g., many membrane dyes display phototoxicity and are therefore not suitable for timelapse microscopy (6).Importantly, some membrane dyes are not compatible with certain culture media.Thus, both Nile red and mitotracker green deliver good membrane stains in full media like MHB and LB.Yet, mitotracker green leads to pronounced membrane stress in minimal media such as BMM and SMM (Figure S3), making it unusable under these conditions.Similarly, dye concentration is crucial.For example, too low concentrations of Nile red and mitotracker green lead to unclear and blurry membrane stains while too high concentrations display toxicity that becomes apparent as clear membrane stress phenotypes.
Likewise, an overdose of DAPI leads to membrane stress and cell lysis (Figure S4).Staining times can be crucial as well.Too short incubation leads to heterogenous stains while too long incubation may lead to phenotypic effects due to toxicity.For BCP dyes, 5 min has been proven adequate in our hands.DiIC12, which stains fluid membrane domains is strongly affected by a number of conditions as RIFs behave differently at different growth phases, temperatures, and media conditions (Figure S5).We have made similar observations for RIF-associated proteins like MurG and PlsX (data not shown).
When working with fluorescent protein fusions, inducer concentrations and induction times can be critical as well.For example, we typically grow strains expressing GFP fusions to MurG, MraY, and MreB in constant presence of inducer.However, TB35 expressing GFP-MinD requires shorter induction times and lower inducer concentrations as overexpression of MinD leads to inhibition of cell division.Hence, elongated cells with disturbed localization patterns are observed (Figure S6).

Selection of model organism
In this study, which was aimed at evaluating assays that are accessible for most labs/researchers, we have chosen B. subtilis as model because it is non-pathogenic and can be grown in most labs.It is sensitive to most antibiotics and has been used extensively for antibiotic mode of action studies in the past, including phenotypic analysis studies, making it a very good reference organism.Further, the availability of a range of different strains and libraries offers near-endless possibilities for deeper cell biological characterization of antibiotic mechanisms.However, it is of course possible to employ BCP and other phenotypic assays to other species including pathogens.Thus, adaptation to other Gram-positive bacteria is usually seamless (7,8).
Gram-negative bacteria, such as E. coli can pose more of a challenge as their outer membrane does not permit the passage of many dyes.However, outer membranepermeable strains e.g., overexpressing outer membrane porins (9,10), or selective outer membrane-permeabilizing agents e.g., polymyxin B nonapeptide (11), can solve this issue.Yet, for BCP specifically another problem is the specificity of membrane dyes as the vast majority of dyes stain both the inner and outer membrane, even when the outer membrane is permeabilized.However, mitotracker green appears to be inner membrane-selective, yet outer membrane-impermeable, enabling specific inner membrane visualization in outer membranepermeabilized cells (Figure S7).An alternative strategy to visualize membrane defects is using an ubiquitous membrane protein fused to GFP e.g., GlpT (1,(12)(13)(14), as proxy.
TEXT S2 Quantitative image analysis of microscopic phenotypic analysis assays.
While the experienced eye can easily interpret microscopy images from appearance alone, quantification can aid interpretation by less experienced researchers, increase comparability between labs, and visualize effects of population heterogeneity.Licensed microscope software such as NIS Elements (Nikon) or Zen (Zeiss) have inbuilt functions for image analysis that are suitable for most basic needs.However, freely available software is more popular as it allows easy transfer of image analysis workflows between labs without the obstacle of licensing and operating different software interfaces.ImageJ is typically the program of choice, mostly due to its plethora of functions and easy customization through plugins and macros.Programs based on ImageJ that are specifically tailored to the needs of analyzing bacterial cells are ObjectJ (developed for Escherichia coli and later amended with plugins aiding single-cell analysis of B. subtilis cells growing in chains) (15)(16)(17) and MicrobeJ (developed for use on various microbial cells) (18).These tools enable automated detection and quantification of bacterial cells based on phase contrast or fluorescence intensity.Details on these programs and their capacities can be found in the aforementioned references.Here, we want to briefly mention workflows for quantification of the phenotypic analysis assays used in this study and address typical limitations and pitfalls of automated or semi-automated image analysis.
One major challenge when working with B. subtilis is that it forms a cell division septum before cells separate, resulting in chains of cells that are indistinguishable by phase contrast.This poses a challenge for programs like ObjectJ that are optimized for E. coli, in which septation and separation occur simultaneously, allowing easy single cell detection by phase contrast alone (17).One solution to this problem are the plugins Chain Tracer and Nuc Tracer, which were developed to automatically detect single cells based on membrane and nucleoid stains, respectively (17).An alternative is MicrobeJ, which can also detect cells based on membrane stains but also allows manual separation of cells (18).Once single cell detection is established, quantitative analysis of phenotypes is feasible.

Quantification of BCP
Since BCP is a combination of methods, several image analysis workflows must be employed to quantify each parameter.Phase contrast can deliver information on cell morphology.
MicrobeJ can easily detect cell length and width, parameters that are of importance for cell division inhibitors and certain antibiotics targeting cell wall synthesis (e.g., fosfomycin, which leads to shape deformations).
Likewise, the program can automatically quantify fluorescence intensity per cell, which can be used to quantify GFP (or dye) leakage and identify pore-forming compounds (Figure S10).While the image analysis is straight-forward in this case, pore assays themselves are sensitive to the conditions and need to be carefully controlled.Thus, cell lysis will lead to the leakage of intracellular content as well as the uptake of pore dyes such as propidium iodide and Sytox Green, leading to false-positive results, especially in species like B. subtilis, which is prone to undergoing autolysis under unfavorable conditions.Therefore, careful selection of antibiotic concentrations and treatment times as well as lysis controls are paramount.Since phase contrast can serve as internal control for cell lysis, it is possible to exclude single lysed cells from the analysis or include them and state their abundance.This is not possible when using spectroscopic batch measurements or even flow cytometry and is a clear advantage of microscopic assessment of pore formation.
Nucleoid compaction can be quantified in a manner similar to whole-cell fluorescence.
To this end, the DAPI signal is detected and its area compared to the area of the whole cell based on phase contrast (Figure 9A).One limitation of this analysis is the fluorescence intensity.Some compounds diminish or eliminate the DAPI stain e.g., nitrofurantoin and peroxide, which at higher concentrations lead to DNA disintegration (9,19).In such cases, the analysis will not result in sensical data.For compounds with such clear phenotypes, this is not a problem in praxis as the effects are usually abundantly clear without the need to quantify.However, in cases with pronounced population heterogeneity, this can be a major problem.This was the case for valinomycin in this study.While nucleoid relaxation was clearly visible by eye (Figure 2), image analysis showed no clear difference in nucleoid compaction values (Figure 9A).This is due to the reduced or absent DAPI signal in the portion of cells that show nucleoid relaxation (Figure 9B).While this phenotype makes sense in light of the proposed mechanism of DNA fragmentation (20), it will be excluded from DNA compaction analysis as the software will not detect the fluorescence signal.Such examples illustrate that, while image analysis can be very useful, it must always be double-checked by manual assessment and compared to the phenotypes observed by eye.
Membrane foci in the Nile red (or any other membrane) stain can in principle be analyzed by foci detection.However, in practice, we have made the experience that it is near-impossible to accurately capture all foci due to the variety in foci size and fluorescence intensity as well as the inability of the software to distinguish an aberrant membrane focus from a freshly forming cell division septum.While this limitation could be overcome in the future e.g., with the implementation of machine learning plugins, the currently most reliable quantification method is simple counting of phenotypes.This can be facilitated by the cell straightening function in MicrobeJ, which displays all detected cells in an organized and numbered manner.Thus, while the analysis does rely on visual inspection and is therefore not unbiased, the datasets can easily be made available for independent assessment and reevaluation.

Quantification of GFP localization
In principle, GFP localization can be quantified similarly to BCP.However, the method must be adapted for each fusion depending on the localization pattern.Thus, intracellular proteins can be quantified in intensity in the same manner as GFP and nucleoid-associated proteins may be detected analogously to nucleoid compaction.Longitudinal line scans, either manual in ImageJ or automated in ObjectJ, are suitable for most cell division proteins that localize at midcell, cell poles, or both (21) e.g., MinD (Figure 3).Loss of membrane binding of evenly distributed membrane proteins, such as MraY, can be analyzed with line scans across the cell.
Membrane proteins that localize in foci suffer from the same foci detection problems as membrane stains, making automated analysis difficult, and often require manual counting.
Quantification of MreB movement can be done in two ways, kymographs and colocalization analysis.Kymographs are static representations of a series of timelapse images and visualize movement.For labs equipped for extended live cell imaging, kymographs constitute a powerful visualization tool.Yet, the sensitivity of MreB to oxygen limitation (22) makes timelapse imaging difficult or impossible, when the required conditions are not met, and is thus not possible in every lab.
In principle, it is possible to quantify the co-localization between MreB foci from only two individual images as displayed in Figure 6 and S14, using the co-localization function in ImageJ.However, it must be noted that the fast movement of MreB results in fully mobile foci overlapping by chance.Additionally, the propensity of MreB to partially lose its membrane binding under stress creates a general cellular fluorescence background, which may result in a false-positive analysis.Thus, simple overlay images or timelapse videos, looping the two individual images, often provide a clearer readout than co-localization analysis.

Conclusion
While many of the issues addressed here can in principle be solved by careful adjustment of image analysis parameters, in particular detection thresholding, the diversity of phenotypes elicited by antibiotics often makes it necessary to adjust parameters individually for each sample, sometimes for each image, or even for different cell populations within the same image, undermining the purpose of unbiased analysis and requiring rather extensive knowledge and experience in image analysis.To avoid analysis artefacts, researchers should always manually check their automated analyses, investigate suspicious values, and carefully crossevaluate their quantitative data with visual inspection.

TEXTS3
Step-by-step protocol for Nile red, DAPI, and GFP microscopy.

Day 1
 inoculate the respective strain in 2 mL MHB in a 50 mL culture tube and grow overnight at 30 °C1 .
o if a GFP fusion strain is used, the medium is supplemented with appropriate concentrations of inducer (see Table S4).

Day 2
 1:100 dilute the overnight culture in 2 mL fresh medium supplemented with appropriate inducer concentrations, where applicable, and grow at 30°C.
 allow cultures to grow to an OD600 of 0.3.
 transfer 100 µL of culture to 2 mL microtubes in a pre-warmed thermoshaker 2 .
 add antibiotics of interest to the samples and incubate for the desired time (usually, short treatment times of 5-10 min give best results)  if fluorescence staining is desired, add 0.5 µg/mL Nile red and 1 µg/mL DAPI 5 min prior to imaging 3 .

Notes:
1 Media and growth temperatures may be adjusted as necessary.B. subtilis is sensitive to oxygen availability and the ratio of liquid to tube/flask volume is critical.Furthermore, culturing B. subtilis in reusable glassware may result in pre-stressed cells due to remnants of cleaning agents.In our hands, 2 mL medium in 50 mL single-use plastic culture tubes have proven optimal for small-scale experiments.
2 Maintenance of constant temperature and shaking is crucial as both temperature shifts and oxygen depletion elicit effects similar to membrane-acting compounds and drastically affect phenotypic analyses.

TEXT S4
Step-by-step protocol for DiSC35 spectroscopy.
Detailed descriptions and protocol variations including alternative dyes are described in detail in te Winkel et al. (23) and Buttress et al. (11).

Day1:
 inoculate the respective strain (here B. subtilis 168CA) in 2 mL MHB supplemented with 50 µg/mL BSA 1 in a 50 mL culture tube and grow overnight at 30 °C2 .

Day2:
 1:100 dilute the overnight culture in 2 mL fresh medium supplemented with 50 µg/mL BSA and grow at 30 °C.
 while the cells are growing, prepare the following solutions: o dilute 100 µM DiSC35 stock solution to 15 µM DiSC35 working solution in MHB containing 50 µg/mL BSA.
o dilute antibiotics 3 to 100x of the test concentration in a minimum volume of 10 µL.Gramicidin can be used as a positive control at a concentration of 1 µg/mL.
 when the culture reaches an OD600 of 0.3 4 , add 138.5 µL to the wells of a black polystyrene microtiter plate.
 after three measurements (medium-only baseline), pause the run and add 10 µL of 15 µM DiSC35 working solution to each well and restart the run.
 continue the measurements until the fluorescence baseline is stable.
 pause the run and add 1.5 µL of 100x concentrated antibiotics to each sample.
 restart the run and continue readings for 30 minutes 6 .

Notes:
1 BSA helps keeping the baseline stable in the spectroscopic measurements.In order to avoid any effects of media changes, it is added to the cultures from the start of the experiment.
2 Media and growth temperatures may be adjusted as necessary. 3When testing a new antibiotic, interference with DiSC35 should be tested by measuring the fluorescence of the dye in the absence of cells.To this end, follow the same protocol but add sterile medium containing 50 µg/mL BSA to the wells instead of bacterial culture. 4Exponentially growing cultures can be diluted down to an OD600 of 0.3, but it should be avoided diluting from transition/stationary phase cultures as those display different membrane potential levels and higher population heterogeneity, leading to less reproducible results. 5For the same reasons mentioned in the previous protocol, the instrument should be temperature-controlled and capable of continuous shaking. 6Longer measurements are affected by photobleaching and endpoint measurements may be a better choice.Please refer to Schäfer et al. (9) for experimental procedures.

TEXT S5
Step-by-step protocol for Laurdan spectroscopy.
Detailed and varied protocols for Laurdan measurements including alternative dyes are described in detail in Wenzel et al., Scheinpflug et al.,and Humphrey et al. (8,24,25).
This assay is particularly temperature-sensitive.Temperature shifts must be avoided.All consumables, including tubes, microplates, pipette tips, etc., as well as media and buffers must be pre-warmed to the desired temperature (here, 30 °C).

Day 2:
 1:100 dilute the overnight culture in 2 mL fresh medium supplemented with 0.2% glucose and grow at 30 °C.
 while the cultures are growing, prepare the following:  after cultures reach an OD600 of 0.6, add 10 µM Laurdan (from 1 mM stock) and incubate for 5 min.
 wash cells four times in pre-warmed Laurdan buffer using short centrifugation steps of 30 s at 16 000 × g.
 after the last washing step, adjust the OD600 to OD 0.8.
 after three measurements, pause the run and add 100 µL of the prepared buffer antibiotic mix to the samples.
 restart the run and continue readings for 30 min.

Notes:
1 Measurements are conducted in buffer due to the autofluorescence of yellow-tinted culture media such as MHB.Glucose is added to maintain cell energization and membrane potential during washing and measurements.To avoid additional effects of media changes, glucose is added to the cultures from the start of the experiment.If experiments are conducted in colorless medium, measurements can be conducted in the same medium (i.e., washing and resuspending in medium instead of Laurdan buffer) and the addition of glucose can be omitted.
2 Media and growth temperatures may be adjusted as necessary. 3When testing a new antibiotic, interference with Laurdan should be tested by measuring the fluorescence of the dye in the absence of cells.To this end, follow the same protocol but add sterile Laurdan buffer containing 10 µM Laurdan to the plate instead of the stained and washed bacteria.

TEXT S6
Step-by-step protocol for DiIC12 staining.
Detailed protocols for DiIC12 measurements are described in detail in Wenzel et al. and Humphrey et al. (8,24).
 after reaching an OD600 of 0.3 4 , wash cultures four times with pre-warmed MHB containing 1% DMSO using short centrifugation steps of 30 s at 16 000 × g.
 resuspend cell pellets in MHB containing 1% DMSO and re-adjust to an OD600 of 0.3.
 transfer 100 µL of culture to 2 mL microtubes in a pre-warmed thermoshaker 5 .
 add antibiotics of interest to the samples and incubate for the desired time (usually, short treatment times of 5-10 min give best results6 )  spot 0.5 µL of each sample on agarose-coated (1.2%) microscopy slides 7 .

Notes:
1 Media and growth temperatures may be adjusted as necessary. 2The OD600 of the overnight culture should not exceed 4 as overly stationary cultures may display long lag phases and slower growth, which affects the distribution of membrane microdomains resulting in aberrant or smooth patterns. 3Maintenance of a final DMSO concentration of 1% is crucial to keep the dye in solution. 4Natural RIFs are only visible in fast-growing exponential cultures (26). 5Maintenance of constant temperature and shaking is crucial as both temperature shifts and oxygen depletion elicit effects similar to membrane-acting compounds and drastically affect phenotypic analyses.

TEXT S7
Step-by-step protocol for PliaI disk diffusion assay.
Antibiotic activity can vary drastically between agar and liquid medium assays.To ensure that an appropriate inhibition zone (approximately 2 cm) is reached, initial concentration tests should be conducted on MHB agar plates before proceeding to PliaI activation assays.
 prepare MHB agar plates containing 100 µg/mL X-gal and store in the dark at room temperature 1 .

Day2:
 1:100 dilute the overnight culture in MHB without erythromycin and grow at 30 °C.
 while the cells are growing, prepare antibiotic dilutions so that a maximum of 10 µL have to be added to the disk to reach the desired final concentration.As a positive control, 5 µg nisin can be used.
 when the cells reach an OD600 of 0.3, plate 100 µL of culture per agar plate 3 .
 add the appropriate amounts of antibiotic solution to sterile 6 mm filter paper disks.
 place disks on the agar plate and gently press down 4 .
 incubate agar plates over night at 30 °C in the dark.

Day 3:
 after 16 h of incubation, inspect agar plates for blue color around the inhibitions zones and take images with an appropriate scanner or camera.

Notes:
1 Plates should be freshly prepared on the day of or before the experiment is conducted.
2 Media and growth temperatures may be adjusted as necessary. 3If necessary, adjust volume so that a confluent layer of bacteria is achieved after overnight incubation. 4Mind the expected diameter of the inhibition zone and place discs accordingly.

FIG
FIG S3Media compatibility of Nile red and mitotracker green.All cells were grown at 30 °C and stained in mid-exponential growth phase.Scale bars 2 µm.

TABLE S1
Minimal inhibitory and optimal stressor concentrations for valinomycin, vancomycin, and nisin.

TABLE S2
Optimal stressor concentrations and reported mechanisms of action of comparator compounds.

TABLE S3
(1,12,13)f effects of comparator compounds in each assay.Clear effects on both DNA and membrane are observed after 1 h of treatment(1,12,13).Smooth localization, may be observed when growth is slowed or halted(1,12,13).(+): Small and/or transient effect.na: Not applicable due to interference of the compound with the assay.

TABLE S4
Strains used in this study.
FIG S1 Growth inhibition of B. subtilis 168CA by valinomycin, vancomycin, and nisin in MHB at 30 °C.Arrow marks timepoint of antibiotic addition.FIG S2 Growth inhibition of B. subtilis 168CA by valinomycin, vancomycin, and nisin in LB at 37 °C.Arrow marks timepoint of antibiotic addition.