Contrary to established approaches, which typically quantify the “average” behaviour of many bacteria with host cells, we used live-cell confocal microscopy to analyse individual host cell pathogen interactions at a single bacterium resolution. This approach provides insight into the individual infection events that underpin the overall infection process.

Monolayers of a murine macrophage cell line (RAW 264.7) and primary murine bone marrow derived macrophages (BMDMs) were infected with L. monocytogenes expressing green fluorescent protein (referred herein as Lm-GFP) from a chromosomally integrated plasmid (see Materials and Methods). We used a membrane impermeant gentamicin protection assay (where start of the imaging experiment in referred as t0) and a low MOI of 0.25 equivalent to 4:1 host cell to pathogen ratio to exclude multiple invasion events per host cell and thus spatially separate individual host-pathogen interactions ( Figure 1A ). In a typical experiment this resulted in 4.1% of host cells (90 per 2200 cells) harbouring exactly 1 bacterium at t0. An additional 4 host cells were infected with multiple bacteria on average per replicate, an event too rare to account statistically thus removed from subsequent analyses.

Upon infection ( Figure 1B ; Video 1 ) we identified three main outcomes at 5 h post infection resulting in: (1) intracellular replication of bacteria; (2) non-replicative invasion and (3) disappearance. We found that upon invasion of RAW 264.7 macrophages, on average only 32% (±6% standard deviation, SD) of individual host-pathogen interactions resulted in a replicative infection ( Figure 1C ). These replication events were typically initiated within the 2 h post infection and resulted in rapid growth (up to 30 bacteria in 5 h) and spread to neighbouring host cells forming characteristic replicative foci (as indicated by white boxes in Figure 1B ). Likewise, 32% (±8%) of single-cell interactions resulted in non-replicative invasion, where individual bacteria remained associated with the original host cell for the duration of the experiment. In some cases, bacteria established a new interaction event at a later time point (see black arrow in Figure 1B ), these bacteria were not included in the analysis. In addition, 36% (±8%) of bacteria disappeared within 5 h post infection from the imaging region. This disappearance is consistent with phagosome killing of bacteria; however, we cannot exclude the possibility that some bacteria escaped to the media. Importantly, the invasion of murine BMDMs at MOI=0.25 resulted in similar infection outcomes; 25% (±3%) of interactions resulted in replicative invasion, which was not statistically different from the macrophage cell line ( Figure 1D ). Overall, these data demonstrate the heterogeneous nature of interactions between L. monocytogenes and host macrophages, with only a fraction of bacteria being able to replicate and spread within the host cell population.

The intracellular life cycle of L. monocytogenes is well characterised, the bacteria must escape the phagosome into the cytoplasm, where they replicate and accumulate actin for intracellular propulsion (20). We therefore wanted to determine the localisation of replicative vs non-replicative bacteria, to determine if non-replicative bacteria were merely those that failed to escape the phagosome. We used phalloidin staining to detect actin ( Figure 1E ) and lysosomal-associated membrane protein 1 (LAMP1) staining ( Figure S1 ) to determine cytoplasmic or phagosomal location, respectively (31–36). We found that 81% (±5%) of replicative bacteria co-localised with actin and thus were present in the host cell cytoplasm ( Figure 1F ), while 17% (±1%) did not colocalise with actin so were predicted to be localised in phagosomes ( Figure 1G ). This is in strong agreement with a previous study using RAW 264.7 cells (36), demonstrating that almost all bacteria present at replicative foci exhibit cytoplasmic localisation. For non-replicative bacteria, we found that 15% (±6%) also showed robust co-localisation with actin staining ( Figure 1F ). This shows that some bacteria are present in the host cytoplasm, but do not replicate. The presence of actin also indicates these bacteria were not destined for autophagy (37). The 49% (±6%) of non-replicative bacteria that exhibited LAMP1 staining ( Figure 1G ) corresponded to intracellular bacteria that failed to escape the phagosome. The remaining 35% of non-replicative bacteria likely represent cells that are not internalised or are cytoplasmic but do not express ActA at the time of measurement, as previously shown (28).

Our live single-cell imaging approach was therefore able to track the heterogeneous outcomes of host and pathogen interactions, with only a third of L. monocytogenes seeding subsequent infections. Our localisation assays demonstrate that the majority of replicative L. monocytogenes are cytoplasmic, consistent with the ability to escape the phagosome being important for replicative success. However, not all non-replicative L. monocytogenes are within phagosomes, indicating phagosomal escape is not the only determinant of replication.

Given that the PrfA regulon plays such a critical role in L. monocytogenes virulence, and particularly phagosomal escape (20, 23–25), we wanted to understand the relationship between PrfA activity and replicative outcome at the single cell level. We therefore used reporter strains for the PrfA regulated hly and actA genes within our live single cell assay to investigate correlations between PrfA activity and replicative outcomes.

To follow virulence gene expression in individual cells we developed dual reporter Lm strains, in which promoter (Phly or PactA) reporter gene fusions were chromosomally integrated to drive GFP expression, in addition to constitutively expressed tagRFP (see Materials and Methods). In agreement with previous analyses (26), we found that Phly mediated expression of GFP (Phly -GFP) exhibited substantial heterogeneity in L. monocytogenes when cultured in tissue culture cell media for up to 1.5h, with only a subset of cells reaching high expression levels ( Figure S2 ). While a control ΔprfA mutant showed no detectable Phly-GFP expression, the PrfA* strain, in which PrfA is constitutively activated (24), exhibited substantially elevated fluorescence levels and reduced cell-to-cell variability compared to that of the wild type (coefficient of variation 0.93 vs. 0.62, Figure S2 ).

We hypothesised that the ability to establish a replicative invasion was dependent on the level of PrfA activity in individual bacteria. We therefore used confocal microscopy to follow temporal regulation of PrfA activity via Phly-GFP and PactA-GFP expression and the fate of individual bacteria upon infection of RAW 264.7 macrophages ( Figure 2A ; Videos 2 , 3 ). First, we found that following invasion, representative replicating bacteria tracked with a high temporal resolution (every 5 min) exhibited induction of PrfA activity ( Figure 2B , pink arrowheads), consistent with PrfA induction which is known to occur when L. monocytogenes is inside a host cell (38). Specifically, Phly-GFP expression rapidly increased and was maintained within the 2 h time window, notably through multiple division events. Similarly, PactA-GFP was robustly induced with delayed kinetics, as previously indicated at the population level (39) and predicted by differences in PrfA-PrfA box binding specificities between hly and actA promoter regions (40). The robust activation from both promoters was observed through multiple divisions, which suggests ongoing transcription. In contrast, representative bacteria that did not replicate showed lower PrfA activity, however at least one non-replicative tracked bacterium robustly upregulated Phly-GFP expression (depicted in blue in Figures 2A, B , black arrowheads).

To analyse the patterns of PrfA activity more systematically, we quantified Phly-GFP and PactA-GFP at selected times post invasion. At t0, the levels of Phly-GFP and PactA-GFP expression were not statistically different between those bacteria that went on to establish a replicative infection and those that did not. However, at 1 and 2 h post invasion, replicative bacteria induced significantly higher Phly-GFP or PactA-GFP reporter expression, compared to those bacteria that did not replicate ( Figure 2C ). While some non-replicative bacteria exhibited substantial Phly-GFP expression over time, corresponding PactA-GFP fluorescence always remained at low basal levels. This suggested a correlation between the elevated levels and activity of PrfA required for activation of the PactA promoter and the infection outcome. To refine these analyses, we estimated the time to the (first) replication from the time-lapse imaging data, which highlighted a large variability in onset of replication (0 to 90 mins from t0) with some bacteria able to replicate immediately following gentamicin treatment at t0 ( Figure S3 ). Normalisation of temporal PrfA trajectories according to replication time ( Figure 2D ) demonstrated that temporal increases of PactA-expression was a very strong indicator for the replicative invasion, highlighted by a high temporal correlation of expression levels for PactA (R² = 0.36, p-value <0.001) but not Phly (R² = 0.03, p-value= 0.21).

To test this apparent correlation functionally, and to specifically quantify if enhanced PrfA activation increases replication probability, we analysed infections with the PrfA*-strain, in which PrfA shows higher activity compared to a WT strain ( Figure S2 ). We found no statistical difference in the replication probability of wild type or PrfA* strains in RAW 264.7 and BMDMs macrophages (at MOI 0.25), demonstrating that increased PrfA activity per se was not sufficient to induce replication.

As our data indicated that PrfA activation does not determine intracellular replication, we next wanted to address the fundamental question of whether L. monocytogenes ability to replicate is determined by the host or the pathogen, or both. To discriminate between these possibilities, we simultaneously infected macrophages with L. monocytogenes expressing either green (Lm-GFP) or red (Lm-dsRed) fluorescent protein using a combined MOI of 5 (at 1:1 ratio between red and green bacteria). This enabled us to determine outcomes of multiple invasion events per individual host cell. For example, if two bacteria share the same fate upon invasion of the same host cell (e.g., the ability to replicate) it would suggest that the single cell outcome is controlled by the host environment, with some host cells being more permissive to replication than others. Conversely, if fates are statistically independent, the infection outcome is controlled by the bacteria ( Figure 3A ).

To test these possibilities, we first focused on a subset of host cells that at t0 were infected with exactly one green and one red bacteria (167 cells from triplicate experiments, Figure 3B ; Video 4 ). We found that the marginal probabilities that green or red bacteria replicate, although slightly different from each other, p_G= 0.257 (±0.16) vs. p_R=0.148 (±0.07), respectively, were not statistically different (Mann-Whitney test p value 0.4). Assuming the statistical independence, the expected probability that both red and green bacteria replicate in the same host cell is the product of their marginal probabilities, p= p_G x p_R = 0.257 x 0.148 = 0.036 (±0.04), while the expected probability that neither red nor green replicate is p_n=(1-p_G) x (1-p_R)= 0.63 (±0.19). The data show the probability that both green and red bacteria replicated to be 0.029 (±0.025), while the probability that neither replicated was 0.58 (±0.18). These could not be statistically distinguished from the expected probabilities (Fisher exact test p value 0.18). Therefore, the fate of individual bacteria in the same host is independent from each other, suggesting it is the behaviour of individual L. monocytogenes cells that determine the overall outcome of the infection.

The presence of multiple bacteria per host cell raises questions about replication strategy of L. monocytogenes; do multiple bacteria cooperate to increase the likelihood of replication or in contrast, are multiple bacteria cleared more efficiently by host cells. To address this, we tested whether the probability of replication depended on the number of L. monocytogenes associated with cells at t0, based on the live-cell microscopy movies that accurately demonstrate whether the number of bacteria per host cell increases over time or not ( Figure 3C ). We define expected probability that at least one bacterium replicates using the formula p_n=1-(1-p₀)ⁿ where n denotes the number of bacteria per cell at t0 and p₀ is the replication probability for one bacterium per host cell. If there is cooperatively between individual bacteria in the host cell enabling more efficient replication then you would predict probabilities greater than p_n. In turn, lower probabilities would indicate that host cell is sensing multiple bacteria and actively limiting their spread ( Figure 3D ). We found that the observed probabilities exhibited sub-linear increases for up to 7 bacteria per host at t0 ( Figure 3E ). For example, the probability of replication of at least one bacterium if two were present at t0 was p₂ = 0.29 (±0.05), which increased to p₅ = 0.62 (±0.02) when five bacteria were initially present. We found that the statistical independence model accurately recapitulates the data, with the expected p₀ = 0.17 (±0.01). This demonstrates that regardless of the number of bacteria per host cell, each bacterium has the same probability to establish a replicative invasion, thus acts independently and non-cooperatively.

We noted that the distribution of number of bacteria associated with host cells at t0 exhibits substantial heterogeneity; 33% of host cells were not infected, while some macrophages harboured up to 15 bacteria ( Figure S4A ). If the bacterial association was due to a purely random process, the distributions of associated bacteria should follow a one parameter Poisson distribution (41). However, the Poisson fit could not capture the data suggesting that a more complex process is involved. Indeed, we found that a negative binomial distribution accurately captures both the increased number of host cells with a very high pathogen number as well as reduced number of those with few or zero bacteria at t0. This suggests that bacterial association with host cells is not purely random, but rather some host cells appear to be more susceptible. We found that the number of adherent bacteria was negatively correlated with the number of neighbouring cells (correlation coefficient R² = 0.47, p-value <0.01, Figure S4B ), suggesting that adherence is mainly driven by physical accessibility. In the invasion of non-immune cells, specific ligand/receptor interactions between L. monocytogenes and the host are required (20), which may explain heterogeneity of cell adhesions observed in endothelium (29). However, in the case of macrophages, L. monocytogenes is actively taken up through the host mediated phagocytosis (42). This suggested that at least in our infection experiments, isolated host cells are more likely to be infected by L. monocytogenes than those surrounded by neighbours, probably through increased cell surface available for bacteria to bind. While this may be a consequence of our monolayer cell assay, the physical accessibility in vivo to L. monocytogenes may be governed by complex spatial interactions between macrophages and other cell types (8).

Overall, these analyses indicate that fate of individual L. monocytogenes are independent and non-cooperative in the same host, and the ability to replicate is controlled by behaviour of bacteria.

Our data demonstrate that approximately a third of individual bacteria associated with host cells (referred to here as the overall replication) were able to establish a replicative infection at MOI 0.25 ( Figure 1C ). Surprisingly, when the infection was performed at MOI 5 the overall replication probability was reduced by 2-fold ( Figure 4A ). Specifically, the probability of replication when one bacterium interacted with a host cell (p₁) was reduced from 0.32 (±0.06) for MOI 0.25 to 0.17 (±0.01) for MOI 5, while the expected probabilities as a function of number of interacting bacteria exhibited statistically non-overlapping trends. These data demonstrate that changing the MOI affects the overall replication probability.

In order to replicate in macrophages bacteria must enter the host cell through phagocytosis and escape to the cytoplasm (21). To test how MOI affects replication probability, we used anti-L. monocytogenes (anti-Lm) antibody staining to distinguish bacteria that were internalised from those that were adhering to the cell surface in cells fixed at t0 ( Figure 4B ). The proportion of internalized bacteria at t0 at MOI 5 was 28.12 ±4.4% (based on 749 cells that had associated bacteria at t0), which is in good agreement with the previously published 25% internalisation rate in RAW 264.7 cells at MOI=10 (43). In comparison, 47.4 ± 4.3% of bacteria were internalised at MOI 0.25 (based on 952 cells) demonstrating that higher MOI reduced the rate of phagocytosis ( Figure 4C ).

Subsequently, we developed a live cell version of the internalisation assay, where we followed Lm-GFP after live-cell anti-Lm extracellular staining at t0, allowing us to determine the fates of internalised vs. extracellular bacteria ( Figure 4D ; Video 5 ). We found that for MOI 0.25 (for which individual bacteria can be tracked over time) 52% (±14%) of bacteria that were internalized at t0 replicated, while only 3.5% (±1.9%) of the bacteria that were extracellular at t0 replicated by 5h ( Figure 4E ). Due to gentamicin-mediated killing of extracellular bacteria (44), we predict the latter likely correspond to bacteria that are in the process of internalisation at t0. We then introduced conditional probabilities to simplistically characterise the two-step internalisation/replication process, such that the conditional probability of replication given that a bacterium is internalised (probability of replication of intracellular bacteria) P(R/I)=P(R)/P(I) is the ratio of the overall replication probability P(R) and the probability of internalisation P(I) ( Figure 4F ). According to this model for MOI 0.25, the expected conditional probability of replication given that one bacterium is internalised was P(R/I)=0.62, based on the measured overall replication probability P(R)=0.29 ( Figure 4E ) and the probability of internalisation P(I)=0.47 ( Figure 4C ). Therefore, the expected probabilities obtained using fixed cell internalisation assay and observed conditional proportions based on dual live-cell staining assay are in the good agreement.

Having confirmed the conditional probability model, we next examined whether the change of the internalisation rate might explain apparent differences in probability of replication observed at different MOIs. The conditional probabilities that at least one bacterium replicates calculated for MOI 0.25 and MOI 5 based on the associated internalisation rates ( Figure 4C , see Materials and Methods for derivations) followed very close trends ( Figure 4G , black solid line vs. blue line). While initially we assumed the overall internalisation rate for MOI 5, we additionally examined the proportion of internalised bacteria as a function of number of bacteria at t0 ( Figure S5 ). We found a small but significant linear increase from 26.4 (±1.3%) when one bacterium is present up to 41.6 (±5.7%) when seven bacteria were present (R² = 0.27, p-value 0.01), suggesting that increased number of bacteria may increase rate of phagocytosis or, according to our previous finding, isolated cells exhibit higher rate of phagocytosis ( Figure S4B ). Nevertheless, the conditional probabilities that at least one bacterium replicates calculated for MOI 0.5 based on the internalisation rates calculated as function of bacteria per cell ( Figure 4G , blue circles) were also in a good agreement with conditional probabilities for MOI 0.25. Therefore, these analyses demonstrate that the changes of the overall replication probability in response to MOI is controlled through the rate of the phagocytosis, while the replication probability of intracellular bacteria was unaffected (52% vs 59% for MOI 0.25 vs 5, respectively, according to the fitted model).

To mechanistically understand how the overall replication probability depends on the different number of bacteria in the environment we devised a dual colour experiment where Lm-GFP (green Lm) equivalent of MOI 0.25 was supplemented with Lm-dsRed such that the overall MOI was maintained at 5 ( Figure 5A ). We found that supplementation with live Lm-dsRed significantly reduced the proportion replicating bacteria, while supplementation with PFA fixed (dead) Lm-dsRed had no significant effect in comparison to the control Lm-GFP at MOI 0.25 ( Figure 5B ). This suggests a role for a secreted factor produced by bacteria, which is consistent with the soluble pore-forming toxin LLO (45). Indeed, we found when WT Lm-GFP were supplemented with Δhly Lm-dsRed, unable to produce LLO, the replication probability was not affected at an MOI of 5 ( Figure 5C ). Replication probability was similarly affected by live WT but not Δhly L. monocytogenes upon invasion of BMDMs ( Figure 5D ).

LLO is known to play multiple roles during invasion, including activation of host immune responses (45–48), we therefore tested whether recombinant LLO alone can inhibit L. monocytogenes replication and internalisation. Indeed, we found that incubation with previously used non-cytolytic doses of recombinant LLO (49) significantly reduced replication at MOI=0.25 ( Figure 5E ), without affecting cell membrane permeability and viability ( Figure S6 ). In BMDMs LLO treatment almost completely prevented replication, with <1% bacteria able to establish replicative invasions ( Figure 5F ). Consistent with a role for phagocytosis, we observed limited changes of internalisation of Δhly strain at MOI 0.25 vs 5, in contrast to the WT bacteria ( Figure 5G ). Finally, we showed that treatment with recombinant LLO also significantly reduced the internalisation of bacteria ( Figure 5H ). Overall, these data demonstrate that the amount of LLO in the environment surrounding the host macrophages regulates the overall L. monocytogenes replication by controlling bacterial uptake.

L. monocytogenes EGDe : InlA^m (77) was used as the wild type (WT), with all mutations generated in this background. L. monocytogenes was grown in tryptone soya broth (TSB) unless otherwise stated, when needed antibiotics were added at final concentrations of: chloramphenicol (Cm) 7 μg ml^-1 and erythromycin (Em) 5 μg ml^-1. Escherichia coli DH5α was used for cloning and grown in Luria-Bertani broth (LB), when needed antibiotics were added at final concentrations of: chloramphenicol (Cm) 35 μg ml^-1 and erythromycin (Em) 150 μg ml^-1. Viability of stored bacteria was confirmed by routine measurements of colony forming units of the inoculum.

Plasmids ( Table 1 ) were electroporated into L. monocytogenes to generate fluorescently tagged and fluorescent reporter strains of L. monocytogenes described in the same table. Chromosomal integration of integrative plasmids was confirmed by PCR as described previously (81). Correct fluorescence of strains was confirmed by microscopy.

L. monocytogenes PrfA* and ΔprfA mutants were constructed using the temperature sensitive shuttle plasmid pAUL-A as described previously (22).

RAW 264.7 macrophages were maintained in Dulbecco’s modified Eagle medium (DMEM) supplemented 10% (v/v) fetal bovine serum (FBS) and 1% (v/v) 100x MEM non-essential amino acid solution (NEAA) at 37 °C 5% (v/v) CO₂. Bone marrow derived macrophages (BMDMs) were generated from C57BL/6 female mice using L929-conditioned media, once differentiated BMDMs were maintained for up to 3 days in DMEM supplemented with 10% (v/v) FBS at 37 °C 5% (v/v) CO₂. All animal procedures were performed with appropriate personal and project licenses in place, in accordance with the Home Office (Animals) Scientific Procedures Act (1986), and approved by the Home Office and the local Animal Ethical Review Group, University of Manchester. Cell membrane permeability and viable cell count was measured with haemocytometer (Merck) using 0.4% (w/v) Trypan Blue (ThermoFisher) staining as described (82).

RAW 264.7 macrophages or BMDMs were seeded in 35 mm TC treated imaging dishes (Cellview Greiner) at 3.5x10⁵ and 7x10⁶ cells ml^-1 respectively and incubated overnight. L. monocytogenes mid-log (OD₆₀₀ 0.45-0.6) aliquots stored at -80 °C in PBS glycerol (15% v/v) were used for infections. Cells were infected with L. monocytogenes at a MOI of 0.25 in pre-warmed media for 45 min and washed three times prior to the addition of 10 μg ml^-1 gentamicin media ( Figure 1A ). For assays with recombinant listeriolysin (rLLO, Abcam), rLLO was added to cells with the L.monocytogenes at a final concentration of 0.05-2 nM. Infections were immediately imaged by live-cell time-lapse microscopy using a Zeiss LSM710, Zeiss LSM780 or Zeiss LSM880 microscope. Data was visualised using the Zeiss Zen Black software.

For internalisation and actin staining assay infections were performed as described above but fixed at 0 h and 5 h respectively with 4% (w/v) PFA PBS for 30 min at room temperature. For the internalisation assay, extracellular L. monocytogenes were stained with polyclonal rabbit anti-listeria (anti-Lm) antibody (Abcam, ab35132), and washed 3 times before secondary antibody staining with anti-rabbit IgG 594 (Sigma-Aldrich). For the live-cell internalisation and infection outcome assay, anti-Lm was added to live cells in pre-warmed media for 30 s, washed 3 times before secondary antibody staining with Brilliant violet 421 donkey anti-rabbit antibody (Biolegend) in pre-warmed media for 30 s, and washed 3 times before imaging.

For the actin staining assay fixed cells were permeabilised 0.1% triton X-100 (v/v) PBS for 4 min and washed 3 times with PBS. As permeabilization sometimes affected Lm-GFP signal intensity, anti-Lm staining was then performed as described above, but with anti-rabbit IgG 488 (Biolegend) secondary antibody. Alexa fluor 594 Phalliodin (ThermoFisher) was used to stain actin.

For lysosomal staining, cells were fixed using 4% (w/v) PFA for 15 minutes before washing with PBS. Cells were then permeabilised with 0.1% Triton X-100 (v/v) in PBS with 1% (w/v) BSA and incubated for an hour in the presence of anti-LAMP1 (Biolegend, cat no. 121622) conjugated with Alexa Fluor 594, followed by a final wash with PBS and 1% (w/v) BSA.

To identify infection outcomes in time-lapse microscopy data, individual L. monocytogenes visible at t0 were visually tracked in Zen Black software and recorded as bacterial replication, no-replication or bacteria disappear ( Figure 1B ). Correlations between actin or anti-Lm staining were manually assessed in Zen Black. Co-localisation of anti-LAM and LAMP1 staining was assessed in Cell Profiler (83). Evidence of single/multiple bacteria per host cells was used to stratify non-replicating/replicating bacteria, respectively.

For tracked L. monocytogenes GFP reporter expression (Phly or PactA) during infection, individual L. monocytogenes were highlighted as regions of interest for selected time points in FIJI (84), and relative fluorescence intensities (RFU) exported for downstream analysis.

For L. monocytogenes Phly-GFP reporter expression in media at selected timepoints, automated analysis of exported tif images was performed in CellProfiler (83). Brightfield images were used to segment images and identify bacterial cell outlines, relative fluorescence intensities for individual bacteria were then exported for downstream analysis.

In general, the conditional probability that bacteria replicates (R) given that it is internalised (I); P(R/I), can be expressed as the ratio of the overall replication probability P(R) and the internalisation probability P(I); such that P(I)=P(R)/P(I). Then P_n(I) = P_n(R)/P_n(I) denotes respective probabilities for n bacteria adhering to the host cell at t0. The expected conditional probability that at least one bacteria replicates is given by p_n(R/I)=1-(1- P_n(R/I))ⁿ, where in general P_n(R/I) may depend on n. However, under the statistical independence model ( Figure 2D ), these relationships are equivalent to p_n(R/I)=1-(1- p₀(R/I))ⁿ where n is the number of adherent bacteria and p₀(R/I)=I(R/I)= P₁(R)/P₁(I) is the probability of replication if one bacteria is present. For MOI 0.25, p₀(R/I) was measured directly using live-cell microscopy with additional staining ( Figure 3D ), and subsequently used to calculate expected probabilities for n>1 ( Figure 3G , blue curve). For MOI 5, we used a previously fitted p_{0 =} 0.164 ( Figure 2E ) such that p₀=P_n(R) and p_n(R/I)=1-(1- p₀/P_n(I))ⁿ. Then the probability of internalisation P_n(I) was either measured for each n ( Figure S3 ) or a single average rate was used (as in Figure 3C ).

Statistical analysis was performed using GraphPad Prism 8 software (version 8.4.2). The D’Agostino-Pearson test was applied to test for normal (Gaussian) distribution of acquired data. Two-sample comparison was conducted using non-parametric Mann Whitney test, for analyses of variance Kruskal-Wallis ANOVA with Dunn’s multiple comparisons test was performed. Simple linear regression and Pearson’s correlation coefficient R² was used to test association between two selected variables. Schematic Displays were created with BioRender.com.