Introduction

Bacteria can survive under an amazing variety of different conditions, ranging from deep sea waters1,2 to aerobic3,4, anaerobic5,6, and high temperature conditions7,8 and can even be revived from frozen or freeze-dried conditions9,10 under which metabolic processes have been fully stopped. While it is clear when bacteria are alive and demonstrate thriving growth in an optimized laboratory setting, it is unclear how to establish their death because in between bacterial life and death, bacteria can exhibit slowed or suspended forms of life11. This makes the outcome of assays to establish bacterial death difficult to interpret. Yet, in environmental, industrial as well as in biomedical applications and regulatory affairs, establishing bacterial death can be of crucial importance with potentially catastrophic consequences if not properly established. Several essays are in use to establish bacterial “death”, but bacteria considered dead by one assay may very well be alive according to another method12. Culturing and enumeration of colony-forming units (CFUs) upon appropriate agars has been the gold standard for many years based on the notion that lack of reproductive ability is synonymous with death13, till the discovery of non-culturable, but yet viable bacterial strains14. Fluorescent probes are available that allow to distinguish between live and dead bacteria15, but technically, are frequently based on the ability of fluorescent probes to enter bacterial cells suffering membrane damage16, which in many cases can be self-repaired17 yielding viable bacteria able to reproduce upon culturing18. Absence of metabolic activity, such as measured by reproduction upon culturing, MTT-conversion19, or bioluminescence imaging20, can be caused by bacterial death but also by temporary arrest21 of metabolic processes. Oppositely, some types of metabolic activity may increase when bacteria are under antimicrobial pressure22, e.g., due to activation of efflux pumps12, while sub-populations of bacteria may continue to grow at a slowed speed (“persisters”).

Given the complexity of establishing bacterial death, this article aims to shed light on the questions of when bacteria can be considered dead and which assay or combination of assays is required to establish that a bacterium is truly dead. To this end, we applied five commonly used assays to determine bacterial death (agar-plating, microscopic counting, SYTO9-propidium iodide staining, MTT-conversion, and bio-luminescence imaging). For killing, bacteria were exposed to 16 different hostile, environmental conditions ranging from exposure to metal ions to antibiotic exposure at different concentrations. Three different Gram-negative and three different Gram-positive bacterial strains were exposed to these hostile conditions and “death” established by each of the five assays applied.

Results

Assay outcomes for establishing bacterial death

Five common assays (see Table 1) to establish bacterial death were used, and their outcomes upon exposure to 16 different antimicrobial treatments (see Table 2) and a control comparison. In Fig. 1, examples of assay outcomes are presented for a Gram-positive and a Gram-negative strain. Negative outcomes point in the direction of bacterial cell death as compared with the control. The 16 different antimicrobial treatments and the control are arranged along the horizontal axes to yield monotonously decreasing assay outcomes for each strain and antimicrobial treatment, i.e., increasing the efficacy of the treatment involved. Based on the outcome of the enumeration of colony-forming units (CFUs) using the agar-plating assay, a clear gap occurs, indicating the difference between culturable and non-culturable bacteria (Fig. 1A). For MTT-conversion, a less clear gap occurs (Fig. 1B), while the order of the different antimicrobial treatments varies in the monotonously decreasing arrangement of the assay outcomes. Note that the outcome of the MTT-conversion assay can be slightly positive, pointing to increased metabolic activity in the bacterial struggle to survive antimicrobial pressures. In the examples presented, increased metabolic activities are likely due to the activation of the Staphylococcus aureus NorA efflux pump extruding ciprofloxacin23 and the Escherichia coli CopA efflux pump extruding copper24. Considering its recognition in applied microbiology as a gold standard, all assay outcomes have been presented as a function of CFU assay outcome in Supplementary Fig. 1.

Table 1 Overview of the different assays used in this paper to establish bacterial “death”, including their working mechanisms
Full size table
Table 2 Antimicrobials used, their potential application, and antimicrobial working mechanisms perceived33,34, together with the concentrations at which they are applied in this study
Full size table
Fig. 1: Examples of averaged assay outcomes for Gram-positive S. aureus ATCC12600 and Gram-negative E. coli ATCC25922 with the antimicrobial treatments arranged along the horizontal axes to yield a monotonous decrease.
figure 1

a Averaged assay outcomes of CFU enumeration. b Averaged assay outcomes of MTT-conversion.

Full size image

Rank correlations between assay outcomes for establishing bacterial death

In order to better understand possible relations between two pairs of assays establishing bacterial death, all data were subjected to a Spearman’s non-parametric, rank correlation test followed by a Fisher z-transformation. Figure 2 presents examples of relations between two pairs of assays for an S. aureus strain, to illustrate the meaning of a relatively high and low rank correlation coefficient. Spearman’s rank correlation coefficients and Fisher z-transformed scores between different assay outcomes for each individual strain are presented in Supplementary Table 1. Correlations between assay outcomes appear scattered across individual strains, which does not allow for drawing any generally valid conclusions.

Fig. 2: Examples of the dimensionless assay outcomes for S. aureus ATCC12600 showing a high and low Spearman’s rank correlation.
figure 2

a %Bacteria with an intact cell membrane as a function of total counts in an S. aureus strain (rank correlation coefficient 0.95, Fisher z-transformed score 1.83). b %Bacteria with an intact cell membrane as a function of the %MTT-conversion in an S. aureus strain.

Full size image

Overall, with respect to correlations between assay outcomes averaged for all strains (Table 3), rank correlation coefficients and Fisher z-transformed scores were low with the possible exception of the correlation between total counts and %bacteria with an intact cell membrane, but nevertheless significant (p < 0.005), indicating that the correlations were not due to random chance while each assay measures a different aspect of bacterial death.

Table 3 Spearman’s non-parametric, rank correlation coefficients (lower-left corner of each table section) between different assay outcomes for all strains employed, including Fisher z-transformed scores (upper-right corner of each table section)
Full size table

Interestingly, although both bioluminescence radiance as well as %MTT are based on the presence of active metabolic processes, also their correlation is low. The MTT assay, however, is a colorimetric assay based on the reduction of a tetrazolium salt to formazan crystals by mitochondrial dehydrogenase enzymes19, while bioluminescence involves oxidation of luciferin under the influence of luciferase during which light is produced20. Thus correlation between both metabolic assay outcomes can only be expected in case all metabolic processes in a bacterium are fully arrested.

Principle component analyses of the outcome of different assays to establish bacterial death

A principal component analysis was conducted in order to determine which assay outcomes can be comprised in one principal component and how many principal components indicate bacterial death. Principal component analysis was carried out excluding the bioluminescence radiance assay because this assay can only be carried out on bioluminescent strains, that are mostly genetically engineered. In the principal components analysis, the outcomes of the four remaining assays to establish bacterial “death” are plotted as data points in a 4-dimensional space, and a unit vector along the direction of the largest possible variance is constructed, which is called the first principal component. Each successive principal component is orthogonal to the preceding component and responsible for the largest remaining variance in the data. Accordingly, Table 4 lists the first two principal components, together explaining 89% of the total variance in outcome assays in all strains employed.

Table 4 The two first principal components, PC1 and PC2 calculated on the basis of the variance between different assay outcomes for all strains employed, excluding the bioluminescence assay
Full size table

71% of the variance in the data is expressed by PC1. PC1 is comprised of CFU enumeration, total count, and the %bacteria with intact cell membranes. The three assays included in PC1 reflect different aspects of bacterial killing and death. CFUs purely reflect reproductive ability, but during the process of killing an entire bacterial population, reproduction may still occur in a part of the population, yielding an increase in CFUs due to the growth of some bacteria next to a decrease in CFUs in another part of the population due to bacterial killing. The net effect is reflected in the total counts. The involvement of the possession of an intact cell membrane reflects that membrane damage can be the prelude to bacterial death, but depending on the severity of the damage, not a necessary condition. PC2 accounts for 18% of the variance across the data and is, for the largest part, comprised of %MTT-conversion. Thus PC2 reflects metabolic activity that can be extremely low in dormant bacteria that may revive and reproduce under suitable environmental conditions. Collectively, the principal component analysis indicates that bacterial death is a 2-dimensional phenomenon, characterized by two principal components that can be named “reproductive ability” and “metabolic activity”.

Definition of the 2-dimensional space of bacterial death

Considering the 2-dimensional nature of bacterial death, demonstration of a death gap cannot be done anymore on a 1-dimensional line as applied in Fig. 1 but should be done in a 2-dimensional space, as illustrated in Fig. 3. Despite the recognition of agar-plating as the gold-standard, there is no assay that allows to refer to any of the data points in the 2-dimensional space of Fig. 3 as belonging to dead bacteria. Hence, data belonging to a combination of a treatment and bacterial strain that yields data for all four assays used below the dead gap (see Fig. 1), are assumed to reflect dead bacteria.

Fig. 3: Bacterial life and death are represented in two dimensions: PC1 (reproductive ability) and PC2 (metabolic activity).
figure 3

Bacteria indicated as red dots are considered dead by all four assays included in the principal component analysis, while live bacteria are indicated as green dots. The ellipses drawn indicate the 95% confidence limit for bacterial death.

Full size image

Based on this assumption, the ellipses drawn in Fig. 3, define a region in 2-dimensional space within which bacteria can be considered dead within 95% confidence. Thus, it can be concluded that bacterial cell death is a 2-dimensional phenomenon, that represents in one dimension “reproductive ability”, complemented with a second dimension representing “metabolic activity”. Figure 3 is schematically summarized in Fig. 4, presenting different regions in the 2-dimensional space of bacterial life and death.

Fig. 4: Schematic presentation of different regions in the 2-dimensional space of bacterial life and death, as defined by “reproductive ability” PC1 and “metabolic activity” PC2, distinguishing different forms of bacterial life11, i.e., control and thriving growth, slowed and suspended growth from which transition to the death may arise.
figure 4

Border values for PC1 and PC2 are taken from Fig. 3.

Full size image

Accordingly, in the quadrant of the 2-dimensional space that represents bacterial life and death (Fig. 4), where both reproductive ability PC1 and metabolic activity PC2 are zero or positive, growth proceeds as under untreated, control conditions (PC1 = PC2 = 0) or growth is thriving (PC1 > 0 and PC2 > 0). In the quadrant where both PC1 < 0 and PC2 < 0, a reduction in reproductive ability PC1 to smaller than −3.5 combined with a reduction in metabolic activity PC2 to smaller than −1.2 defines the bacterial death region (see also Fig. 4). Whereas this is in line with CFU enumeration as the gold-standard in microbiology, it clearly adds metabolic activity as an important parameter to account for in case of low-metabolically active bacteria. Therewith, we can identify not only the regions of bacterial life and death in 2-dimensional space but also regions of slowed or suspended growth11. As the reproductive ability PC1 becomes more negative, bacteria enter their lag phase in which growth is absent, but metabolic activity is present in preparation for exponential growth. Similarly, the region where reproductive ability is little compromised (PC1 between 0 and −3.5), but the metabolic activity PC2 becomes more negative, can be identified as the region of suspended life, commencing with quiescence and moving into dormancy when PC2 becomes highly negative and the chances to enter the death region may increase.

The importance of the different dimensions of bacterial death and the distinction between death versus slowed or suspended life is illustrated in Fig. 5, showing S. aureus ATCC12600 growing on an agar plate after exposure to treatments yielding PC1 and PC2 values that make them reside in the bacterial death region (PC1 and PC2 values of −5.1 and −2.0, respectively) or the slowed or suspended life region (PC1 and PC2 values of −4.6 and −0.4, respectively). Bacteria that were identified to reside in the death region, did not form any colonies on an agar plate, neither after 24 h nor after 48 h of growth. However, bacteria identified to reside in the slowed or suspended life region displayed a small colony after 24 h of growth on an agar plate that grew into a clear colony upon doubling the time allowed for colony formation as a first step toward entering the life region. For bacteria in the death region, no entry to the life region was observed.

Fig. 5: The distinction between death versus slowed or suspended life.
figure 5

S. aureus ATCC12600 colony formation on agar plates was observed after 24 or 48 h of growth. Bacteria were put on agar plate untreated (“control”, PC1 = PC2 = 0) and after treatments that made them reside in the bacterial death region (PC1 and PC2 values of −5.1 and −2.0, respectively) or in the slowed or suspended life region (PC1 and PC2 values of −4.6 and −0.4, respectively). The dilution rate applied was 103, while images pertaining to the same treatment were taken from the same agar plate.

Full size image

Discussion

Extended evaluation of bacterial death upon exposure to antimicrobials not only yielded low correlations between assay outcomes, but moreover, principal component analyses indicated that bacterial death can only be established on the basis of two principal components, i.e., “reproductive ability” (PC1) and “metabolic activity” (PC2). Plotting of these principal components in 2-dimensional space (Fig. 4) reveals a dead region, with borders given by their PC1 and PC2 values, but at the same time, makes clear that regions exist in 2-dimensional space where either reproductive ability or metabolic activity are reduced that represent slowed or suspended life from which bacteria can enter the life region again. Sensu stricto implies that all assays comprised in PC1 and PC2 calculations must be carried out in order to establish bacterial death.

Whereas Fig. 4 represents a comprehensive, schematic presentation of bacterial life and death, it is important to emphasize its limitations. A clear limitation involves the absence of an assay to identify dead bacteria with 100% certainty, and as a result, the assumption has been used that bacteria are dead when four assays yield a data point below the dead gap (Fig. 1). Another possible limitation is the selection of antimicrobials included and admittedly, there are more antimicrobials around than can ever be included in a study like the present one. However, it is believed that the mechanisms by which the selected antimicrobials operate (Table 2), cover all mechanisms currently prevailing, including those exerted by novel nano-antimicrobials25, with the exception of mechanical killing by magnetically-propelled, sharp-edged magnetic nanoparticles or micro-robots26,27. Here, too, however, the same assays as evaluated in the current work are applied to establish bacterial death, and at the large population level of bacterial occurrence in different applications, Fig. 4 remains valid. Although Fig. 4 is based on a combination of three Gram-positive and three Gram-negative strains and, in principle, is believed to be generally applicable, the exact border values for PC1 and PC2 of the death region may be subject to change upon the inclusion of more strains in the analysis but without affecting the 2-dimensional nature of bacterial death and life.

The implications of the conclusion that bacterial death is 2-dimensional are far-reaching for experimentally establishing bacterial death. Sensu stricto, this conclusion implies that in order to establish bacterial death, CFU enumeration, total counts, and possession of an intact membrane must be experimentally determined in order to calculate the “reproductive ability” according to PC1. In addition, MTT conversion should be measured and expressed as “metabolic activity” according to PC2, after which data should be positioned in a 2-dimensional space according to Fig. 3. In the Supplementary Information, we provide a simple EXCEL program requiring the assay outcomes as defined in this article as input to calculate PC1 and PC2 and position the data in 2-dimensional space with respect to the death region. Whereas this implication is clearly the ideal, it is unlikely to become ever generally applied as it represents an overload of experimental work. In a milder interpretation, the implication of this conclusion would be that in order to establish bacterial death, at least one of the assays required to calculate PC1 and the MTT conversion assay should be carried out for the calculation of PC2. In the mildest, minimal interpretation of this conclusion, researchers should specifically describe which dimension of bacterial death has been assessed and using which assay, when stating in their work that bacteria were killed and considered death. Implementation of the conclusion that bacterial death is 2-dimensional either in its sensu stricto or mildest interpretation, will not only increase the comparability of academic work but will also affect industrial development of new antiseptics and antimicrobials and their regulatory approval that still largely relies on 1-dimensional bacterial killing, solely using the agar-plating gold-standard.

Methods

Bacterial strains and growth conditions

Three Gram-positive strains, including S. aureus ATCC12600, bioluminescent S. aureus Xen36, and Streptococcus mutans ATCC700610, as well as three Gram-negative strains, including E. coli ATCC25922, bioluminescent E. coli Xen16, and Pseudomonas aeruginosa SG81, were used in this study. ATCC strains were obtained from the American Type Culture Collection and the Xen strains from PerkinElmer, Inc., Waltham, MA, USA. P. aeruginosa SG81 is a mucoid environmental isolate28. Each strain was inoculated from a single colony taken from an agar plate, pre-cultured in 10 mL growth medium for 24 h, and diluted 1:20 into fresh growth medium for further growth for 18 h at 37 °C. Staphylococci were grown in Tryptone Soya Broth (TSB, OXOID, Basingstoke, UK), S. mutans in Todd Hewit Broth (THB, OXOID), E. coli in Luria Broth (LB, OXOID) and P. aeruginosa in Lennox broth. All bacteria were cultured in ambient air, except for S. mutans which was cultured in 5% CO2. Bacteria were harvested by centrifugation at 5000g, and resuspended in growth medium, after which total counts were enumerated using a Bürker–Türk counting chamber and phase contrast microscopy.

Bacterial exposure to antimicrobials

Antimicrobials used in this study were selected to represent an as wide as possible range of application and working mechanisms (see Table 2). For exposure of bacteria to an antimicrobial, 100 μL bacterial suspension (5 × 108/mL) in a growth medium was added to a 96-well plate (Greiner suspension culture plate, flat bottom), after which 100 μL of an antimicrobial in ultrapure water was added in the concentration ranges mentioned in Table 2 and cultured for 20 h at 37 °C under shaking at 80 rpm.

After the exposure to hostile conditions, bacterial death was determined according to five different assays, described below (see Table 1 for overview). In order to allow comparison of assay outcomes, outcome was defined as the difference between the logarithm of the assay outcome of a treatment and the logarithm of the outcome of an untreated, control culture, i.e., the logarithm of their quotients. Accordingly, unit-less assay outcomes resulted that were zero for the controls in the absence of treatment and effective treatments yielded more negative assay outcomes.

Colony forming units (CFUs)

Bacterial suspensions, after exposure to different antimicrobials, were serially diluted and plated on appropriate agar plates. The agar plates were cultured under the conditions mentioned above, and CFUs were enumerated 24 h after growth. Each experiment was repeated three times with newly prepared samples and separately cultured bacteria.

Total counts

The total numbers of microscopically visible bacteria were manually counted in a Bürker–Türk counting chamber using phase contrast microscopy with a 40× objective, taking five images of randomly chosen fields from each sample. Each experiment was repeated three times with newly prepared samples and separately cultured bacteria.

Percentage of bacteria with intact cell membranes

The percentage of bacteria with intact cell membranes was determined as described above, but after staining with SYTO9 (3.34 mM) and propidium iodide (20 mM) (BacLight™, Molecular probes, Leiden, The Netherlands). SYTO9 can enter bacteria with an intact or damaged cell membrane and interact with intracellular DNA to yield green fluorescent bacteria. Propidium iodide however, can only enter bacteria with a damaged membrane to displace SYTO9 to yield red-fluorescent bacteria. Staining was done by adding 3 μL of SYTO9/propidium iodide (1:1) to 1 mL demineralized water, after which 10 μL was added to 10 μL of bacterial suspension on a microscope slide and incubated for 15 min, while kept in the dark at room temperature. After staining, bacteria were imaged using a fluorescence microscope (DM 4000B, Leica, Wetzlar, Germany) as described above, and the percentage of bacteria with an intact membrane was determined from the ratio of green—over the total number of (green- and red-) fluorescent bacteria, counted with the aid of ImageJ (National Institutes of Health and the Laboratory for Optical and Computational Instrumentation (University of Wisconsin)29,30. Note that this assay is frequently considered erroneously as a live-dead assay for bacteria. Each experiment was repeated three times with newly prepared samples and separately cultured bacteria.

MTT-conversion assay

Metabolic activity was determined using the MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) reduction assay31. MTT, a yellow-soluble tetrazolium salt, is reduced by reductase enzymes into purple formazan and the reduction was measured spectrophotometrically. 100 µL of an exposed bacterial suspension was mixed with 100 µL MTT solution (5 mg MTT dissolved in 10 mL potassium phosphate-buffered saline (PBS) containing 1% glucose, and 100 µL 10 mM menadion) in an Eppendorf tube and incubated for 45 min at 37 °C and 150 rpm in the dark. After incubation, the suspension was centrifuged, and the bacterial pellet was suspended in 250 µL acid isopropanol (5% 1 M HCl in isopropanol). After 15 min, 200 µL was pipetted in a new 96 wells plate, and the absorption of the MTT solution was measured spectrophotometrically at 560 nm (FLUOstar plate reader, Optima, BMG Labtech, Offenburg, Germany). As a control, a growth medium without bacteria was taken. Each experiment was repeated three times with newly prepared samples and separately cultured bacteria.

Bioluminescence radiance

Bioluminescence radiance could only be measured for S. aureus Xen36 and E. coli Xen16. 200 µL of the exposed bacterial suspensions were put in a 96-well plate, and the total bioluminescence was measured over the area of each individual well using an in vivo imaging system (IVIS® Lumina Imaging System, Perkin Elmer) and expressed as photons per second (p/s). Bioluminescence images were obtained using 20 s exposure time, medium binning, 1F/Stop, open emission filter, and corrected automatically for background luminescence. Each experiment was repeated three times with newly prepared samples and separately cultured bacteria.

Evaluation of Spearman’s rank correlation

Spearman’s non-parametric, rank correlation coefficient was used to measure the strength and direction of a monotonic relationship between different assay outcomes that quantify bacterial death. Rank correlation coefficients were calculated by replacing each data value with its rank, sorted from the smallest to the largest values. In the case of tied ranks, the average rank is assigned. The correlation coefficient is then calculated based on these ranks. Spearman’s rank correlation ρ is subsequently calculated as

$${\rm{\rho }}=1-\frac{6\sum {d}^{2}}{n({n}^{2}-1)}$$
(1)

in which d is the difference in rank of corresponding values and n is the number of data points, yielding correlation coefficients ranging from −1 to +1. A value of +1 indicates a perfect positive monotonic relationship, −1 indicates a perfect negative monotonic relationship, and 0 indicates no monotonic relationship. A positive correlation signifies that as one assay increases, the other assay tends to increase as well. Conversely, a negative correlation indicates that as one assay increases, the other tends to decrease. All Spearman’s correlation coefficients were calculated using Origin 8.5 software (OriginLab Corporation, Northampton, MA, USA). For statistical comparison and interpretation, correlation coefficients were Fisher z-transformed into an approximately normal distribution using

$$z=0.5\times \mathrm{ln}\frac{(1+\rho )}{(1-\rho )}$$
(2)

in which z is the transformed score.

Principal component analyses

Principal component analysis was conducted to determine which assay outcomes constitute one principal component and how many principal components indicate bacterial death. First, all assay outcomes Xij were arranged into an original matrix X, where each row i represents one of the 16 treatments and each column j represents one of the four assay outcomes. Next, all Xij outcomes were standardized in a new matrix Z, according to

$${Z}_{{ij}}=\frac{{X}_{{ij}}-{{\rm{\mu }}}_{{ij}}}{{\sigma }_{j}}$$
(3)

in which Zij is the standardized assay outcome with i and j indexing the treatment and the assay respectively, µij is the mean of the jth assay across all treatment outcomes, and σj is the standard deviation over all treatment outcomes obtained using the jth assay. This standardization ensures that each assay has a mean of 0 and a standard deviation of 1. Next, the covariance matrix C for Z was calculated as follows

$$C=\frac{1}{n-1}Z{Z}^{T}$$
(4)

in which ZT denotes the transpose of the matrix Z and n is the total number of observations. Eigenvalues λ and eigenvectors ν indicating the direction of C were defined as

$$C\nu =\lambda \nu$$
(5)

Eigenvalues were ranked in descending order together with their corresponding eigenvalues, after which the eigenvectors ν1 and ν2 were selected, possessing the two largest values of λ. Principal components PC1 and PC2 for each treatment i were subsequently constructed according to

$${PC}{1}_{i}=\mathop{\sum}\limits_{j=1}^{4}{Z}_{{ij}}{\nu }_{1j}$$
(6)
$${PC}{2}_{i}\mathop{\sum}\limits_{j=1}^{4}{Z}_{{ij}}{\nu }_{2j}$$
(7)

All principal components PC1i and PC2i were subsequently plotted in a 2-dimensional scatter plot using GraphPad Prism version 9.0 software (GraphPad Software Inc.; San Diego, CA, USA). The distinction between live and dead bacteria was made on the basis of the death gaps between a monotonous ranking of the outcomes of each assay, denoting bacteria as being dead when all assays yielded data below the death gap of each assay. 95% confidence ellipses encompassing dead bacteria were constructed using R software (Version: 2023.09.1-494).