Fluorescent secreted bacterial effectors reveal an intravacuolar replication compartment for Listeria monocytogenes

Real-time imaging of bacterial virulence factor dynamics is hampered by the limited number of fluorescent tools suitable for tagging secreted effectors. Here, we demonstrated that the fluorogenic reporter FAST could be used to tag secreted proteins, and we implemented it to monitor infection dynamics in epithelial cells exposed to the human pathogen Listeria monocytogenes (Lm). By tracking individual FAST-labelled vacuoles after Lm internalisation into cells, we unveiled the heterogeneity of residence time inside entry vacuoles. Although half of the bacterial population escaped within 13 minutes after entry, 12% of bacteria remained entrapped over an hour inside long term vacuoles, and sometimes much longer, regardless of the secretion of the pore-forming toxin listeriolysin O (LLO). We imaged LLO-FAST in these long-term vacuoles, and showed that LLO enabled Lm to proliferate inside these compartments, reminiscent of what had been previously observed for Spacious Listeria-containing phagosomes (SLAPs). Unexpectedly, inside epithelial SLAP-like vacuoles (eSLAPs), Lm proliferated as fast as in the host cytosol. eSLAPs thus constitute an alternative replication niche in epithelial cells that might promote the colonization of host tissues.


Introduction
Bacterial pathogens harness distinct colonization strategies to take advantage of their host resources. While some remain extracellular, others adopt an intracellular lifestyle. Internalisation into host cells provides invasive bacteria with multiple abilities, such as that of crossing cellular barriers, escaping humoral immune surveillance, or disseminating throughout the organism as cargo of circulating cells. After internalisation, bacteria are entrapped inside primary vacuoles from where they can follow two distinct routes: either subverting endomembrane compartments, or leaving them. For instance Chlamydia trachomatis, Brucella abortus or Legionella pneumophila perturb the maturation and rearrange the properties of vacuoles, thereby creating a compartment prone to their replication [1]. Others, such as Shigella flexneri or Listeria monocytogenes, typically do not grow inside endomembrane compartments, but rather escape from entry vacuoles and gain access to the host cell cytoplasm, where they can replicate as well as exploit the host actin cytoskeleton for intracellular motility and cell-to-cell spread [2].
The foodborne pathogen Listeria monocytogenes (hereafter, Lm) is the causative agent of listeriosis, and has emerged as a model facultative intracellular bacterium [3,4]. This pathogen can cross the protective barriers of its host and colonize tissues and organs by promoting its internalisation into non-phagocytic cells. The classical scheme of Lm intracellular life cycle implies that, both in professional phagocytes and in epithelial cells, Lm rapidly escapes from entry vacuoles due to the combined action of a potent pore-forming toxin, listeriolysin O (LLO), and of two phospholipases C (PlcA and PlcB), before replicating in the cytosol [5]. All three genes (hlyA that encodes LLO, plcA and plcB) are part of Lm LIPI-I virulence gene cluster and are transcriptionally induced by PrfA, the main regulator of Lm virulence gene, in intracellular bacteria [6].
LLO is a cholesterol-dependent pore-forming toxin secreted by Lm via the general secretion system (Sec) [7]. LLO assembles into oligomers on biological membranes, forming arcs and pores of several tens of nm that disrupt membrane integrity [8]. Its activity is optimal at acidic pH representative of the acidification occurring during the maturation of phagosomes (pH = 4.9 to 6.7), which has been proposed to facilitate the escape of bacteria from entry vacuoles while avoiding damages to the host plasma membranes at neutral pH [9,10]. Whereas LLO-deficient Lm cannot gain access to the host cytosol in many cell types, the activity of the phospholipases PlcA and PlcB and the influence of host factors render LLO dispensable for vacuole escape in several human epithelial cell lines [11]. In phagocytes, it has been shown that bacteria secreting reduced amounts of LLO could remain entrapped in long-term compartments named Spacious Listeria-containing Phagosomes (SLAPs) and even replicate extremely slowly therein, with a doubling time in the range of 8 h [12].
The escape dynamics from the entry vacuole has been experimentally addressed using several distinct strategies. One of them consisted in using medium containing a membrane-impermeant fluorescent dye during infection [9]. Upon encapsulation into the internalisation vacuoles together with invading bacteria, the fluorescent dye stained the intravacuolar space until it broke down. However, this method required a washing step after bacterial entry in order to remove the unwanted background due to the high extracellular amount of the dye. This washing step thus prevented the observation of the first 10 minutes of the infection dynamics.
Alternative strategies were based on the assessment of vacuole rupture events and bacterial access to the host cytosol using fluorescent sensors. For instance, galectin-3 was shown to label membrane remnants of damaged vacuoles and thereby allowed the spotting of vacuole lysis [13]. Likewise, actin or the Cell-wall Binding Domain CBD (a domain from the Lm phage endolysin Ply118) are recruited to the bacterial surface only once Lm has escaped the vacuole [5,14]. Cytoplasmic FRET probes that are cleaved by a b-lactamase secreted by invasive bacteria have also been described as efficient reporters of average vacuole rupture time at a cellular scale [15,16]. Although these approaches yielded the order of magnitude of the time lapse between bacterial entry and vacuole escape in various cell types (between 15 min and 1 h), they did not allow a precise recording of the onset of entry events, which prevented their use for establishing the distribution of Lm residence time in entry vacuoles with accuracy. These constraints have limited the possibilities of refined quantitative comparisons of variations in intravacuolar residence times between different conditions. In order to measure the heterogeneity of Lm residence time in entry vacuoles and to assess the role played by LLO in the dynamics of bacterial escape from these compartments, we developed live imaging assays allowing an accurate measurement of the time elapsed between the moment when individual bacteria were internalised into cells and the moment when the integrity of the vacuole membrane was disrupted. We devised a strategy based on tagging proteins secreted by bacteria with the FAST reporter system [17]. FAST is a 14-kDa protein tag which displays fluorescence upon binding with a synthetic fluorogenic probe supplied in the medium. The fluorogen is membrane permeant, non-toxic, and has very little fluorescence by itself. The small size of FAST, its rapid folding kinetics, the reversible binding of fluorogens together with good brightness and photostability made this system an ideal candidate for tagging secreted proteins such as LLO and imaging them in real time.
Using live imaging of FAST-tagged proteins, we quantified the distribution of Lm residence times in primary vacuoles in the LoVo intestinal epithelial cell line. We observed that a fraction of the population of entry vacuoles lasted for several hours and were reminiscent of SLAPs. However, in contrast with SLAPs, the prolonged residence of Lm inside vacuoles was observed in cells infected with wild type (WT) Lm as well as with a hlyA deletion strain. The secretion of LLO allowed Lm to proliferate actively inside these compartments, suggesting that besides its role in vacuole escape, LLO may contribute to setting up an intravacuolar niche permissive for Lm replication in epithelial cells.

Fluorescent tagging with FAST of proteins secreted by Listeria monocytogenes
With the aim of detecting proteins that were secreted by intracellular bacteria into their host cells in livecell microscopy experiments, we explored the possibilities offered by the FAST reporter system for the fluorescent tagging of Lm secreted bacterial effectors. A set of integrative plasmids harbouring gene fusions under control of the PHYPER promoter were designed ( Fig 1A) and introduced in the genome of Lm strain LL195.
These plasmids drove the constitutive production of either FAST or eGFP, either for intrabacterial localisation, or fused in their N-terminus to the secretion signal peptide (SP) of listeriolysin O (LLO) (SP-FAST and SP-eGFP constructs), or to full-length LLO (LLO-FAST, LLO-eGFP and untagged LLO constructs), a classical Sec substrate. A Myc tag in the C-terminus of all constructs allowed detection by immunoblotting. Protein production and secretion by each one of these seven strains was assessed by in-gel colloidal Coomassie staining and immunoblotting against the Myc tag, on bacterial total extracts and culture supernatant fractions from 16h cultures in brain heart infusion (BHI) (S1 Fig). All transgenes were efficiently expressed by Lm, even though in varying amounts. As expected, constructs harbouring either the LLO SP, or full-length LLO, were recovered in bacterial culture supernatants, indicating that the SP of LLO promoted Sec-dependent export of FAST or FAST-tagged proteins, as well as eGFP-fusion proteins albeit to a lesser extent (S1C, D Fig). Constructs devoid of signal peptides were not detected in supernatant fractions, arguing against the release of proteins into the medium due to bacterial lysis. FAST-tagged Sec substrates can thus efficiently undergo secretion through the general secretion pathway.
To assess whether the FAST reporter system remained fluorescent after secretion, we quantified the fluorescence signals in the filtered culture medium of bacteria grown for 6 h in Listeria synthetic medium (LSM) (Fig 1B). In presence of 5 μM of HBR-3,5DM, fluorescence was detected in the culture supernatant of strains secreting SP-FAST or LLO-FAST. By calibrating fluorescence measurements with a standard curve of known FAST concentrations diluted in the same medium, we estimated the concentration of secreted tagged proteins; that of SP-FAST reached 325 ± 55 nM, and that of LLO-FAST was 28 ± 6 nM. In contrast, fluorescence levels in the culture medium of strains producing non-secreted FAST remained low, indicating that the release of fluorescent proteins in the culture medium due to bacterial lysis was minor. We conclude that FAST-labelled proteins retained their fluorescent properties after undergoing secretion through Sec.
Diverse attempts by others in Gram-negative bacteria [18] and our own unpublished observations using tagged Lm virulence factors suggested that the Sec-dependent secretion and subsequent maturation of an eGFP tag into its active, fluorescent fold was inefficient. Surprisingly, the secretion of SP-eGFP-but not that of LLO-eGFP-also gave rise to fluorescent signals in culture supernatants, even though in a range 10-fold lower than that obtained for the secretion of SP-FAST (S2 Fig). A consistent proportion of eGFP undergoing Secdependent secretion was thus able to acquire its mature fold in bacterial culture medium, at least in conditions where it was not fused to a bacterial effector and in LSM.

Fluorescent tagging with FAST of Shigella effectors secreted through the type III secretion system
To evaluate the versatility of FAST as a reporter of bacterial secretion, we next asked if FAST was suitable for fluorescent tagging of effectors secreted through the syringe of the type III secretion system (T3SS) from a Gram-negative pathogen, Shigella flexneri (Sf) strain M90T. As model T3SS substrates, we tagged the Cterminal ends of the effectors OspF and IpaB with FAST-Myc (Fig 1C), which are translocated upon adhesion of Sf to host cells [19]. Bacterial total extracts and culture supernatant fractions were recovered from 16-h cultures in M9 medium, with or without stimulation of type-III dependent secretion by addition of Congo red.
By immunoblotting these fractions against the Myc epitope, we observed that tagged OspF and IpaB were secreted into the bacterial culture medium upon Congo red induction (S3A Fig). The secretion of both tagged effectors was constitutive when using a ∆ipaD mutant strain for which translocation lacks gating controls [20] (S3B Fig). We then assessed whether the fusion proteins secreted by the ∆ipaD strain had retained their fluorescent properties, by measuring fluorescence intensities in the supernatants of 16-h bacterial cultures in M9 medium (Fig 1D). Fluorescence levels were consistently higher with this constitutively secreting strain ∆ipaD than the fluorescence leakage measured for the WT strain when the T3SS was not induced. The concentration of OspF-FAST by the ∆ipaD strain was estimated to be 3.8 ± 0.3 nM, that of IpaB-FAST of 9.4 ± 1.7 nM. Like Sec substrates, FAST-tagged T3SS substrates can thus pass through the needle of the TS33, and keep fluorescent properties after secretion at least when gating controls are lacking.

FAST-tagging of secreted Listeria effectors for live fluorescence microscopy
We next investigated whether the FAST reporter system was suited for detecting proteins secreted into the cytoplasm of host cells by real-time microscopy during infection. To this end, we monitored FAST signals in LoVo cells infected with Lm producing SP-FAST by confocal spinning disk microscopy over an infection time course (Fig 2, S2 Movie). FAST fluorescence labelled uniformly the cytoplasm of infected cells and increased over time (Fig 2A). At 562 nm (the emission wavelength specific for FAST:HBR-3,5DM), fluorescent signals accumulated in cells infected with a strain producing SP-FAST, and not with a control isogenic strain that constitutively expressed mCherry (Fig 2B). In infected cells, measured fluorescence intensities-which reflects the intracellular concentration of SP-FAST-increased exponentially with time ( Fig 2C), likely mirroring the exponential growth of Lm in the host cytosol. After several hours of signal accumulation, the intracellular fluorescence dropped abruptly. This corresponded to a sudden shrinkage of infected cells, probably resulting from their death and from the concomitant permeation of their membranes. For each cell, we fitted the dynamics of fluorescent signals to an exponential curve as shown in Fig 2C (blue curve), measured the rates of fluorescence increase for each exponential curve, calculated fluorescence doubling times (t = !"# $ ), and then plotted their distribution ( Fig 2D). The mean doubling time of FAST signals was 106.7 ± 41.9 min (n = 39). This value, which represents the characteristic time for SP-FAST accumulation in

Residence time of Listeria monocytogenes in internalisation vacuoles
When FAST-tagged proteins were secreted into the large volume of the host cell cytoplasm, fluorescent signals were diluted and therefore only became clearly visible after several hours of infection, once secreted FAST had accumulated sufficiently to be significantly discriminated from non-specific signals. Meanwhile, we reasoned that if Lm was confined inside micron-sized internalisation vacuoles, the higher concentration of secreted FAST molecules in a reduced volume would allow their detection and tracking until the disruption of vacuole membranes, thereby providing an accurate measurement of individual vacuole lifetimes ( Fig 3A).
Indeed, we observed that secreted FAST signals were enhanced in compartments that co-localized with mCherry-expressing bacteria within minutes after bacterial adhesion to cells, until these signals suddenly  (Fig 3D). When using the ∆hlyA::SP-FAST strain, the median residence time was significantly longer (21.1 ± 1.4 min) but remained of the same order of magnitude as for a strain producing LLO, confirming previous observations that Lm gained efficient access to the cytoplasm independently of LLO in epithelial cells [11]. Unexpectedly, a significant proportion of the entry vacuoles lasted for more than one hour (12.0 % for the WT strain; 14.8 % for the ∆hlyA mutant), and a consistent number of intact vacuoles was still observed 3 h p.i. (4.6 % for the WT strain; 6.2 % for the ∆hlyA mutant) (Fig 3C). The fact that the WT strain remained entrapped in vacuoles in proportions nearly identical to that of the ∆hlyA strain could either suggest that a sub-population of WT Lm failed to escape primary vacuoles in spite of LLO secretion, or that LLO was not produced by this sub-population of intravacuolar bacteria. To discriminate between these two hypotheses, we investigated whether LLO fused to a FAST tag was detected in vacuoles from which Lm had failed to escape.

Long-term residence and rapid replication of Listeria inside LLO-decorated vacuoles
To examine whether LLO was produced and secreted by bacteria that remained entrapped in vacuoles, we engineered a Lm strain where LLO was C-terminally fused with FAST at the endogenous hlyA locus (S5A Fig).
In this strain, the fluorescence of FAST reported not only for LLO secretion and localisation, but also for hlyA expression under its natural promoter. In order to be relevant for monitoring the dynamics of Lm intracellular infection, the 15-kDa FAST-Myc tag should not interfere with the function of the protein it reports for. We controlled that the haemolytic properties of the strain expressing hlyA-FAST did not differ from that of the WT strain (S5C Fig); the production, secretion and activity as a cytolysin of LLO are thus quantitatively and qualitatively preserved after C-terminal fusion with FAST.
The strain producing the LLO-FAST fusion also constitutively expressed mCherry, which allowed us to ( Fig 4B). This doubling time was consistently faster than that previously described in SLAPS, which was in the range of 8 h [12].

Role of listeriolysin O in the long-term intravacuolar residence and replication of Listeria
Our above-mentioned results showed that LLO-FAST was secreted by Lm and functional, and also that it was present in eSLAPs, which is likely to permeate their membranes. However, despite the presence of LLO, the integrity of eSLAPs was preserved over several hours, and intravacuolar replication of Lm occurred without vacuole rupture. To investigate whether LLO concentration influenced Lm residence time in eSLAPs, we took advantage of the LLO-FAST reporter strain in order to measure the variability in LLO abundance in vacuoles. Symmetrically, we used a prfA* mutant strain to investigate the outcome of LLO overexpression. The prfA* allele encodes a PrfA variant with a G145S substitution that has been previously described to be constitutively active, and to lead to the strong overexpression of PrfA-dependent virulence genes, including that of hlyA [21]. Accordingly, the in-vitro haemolytic titre of the LL195 prfA* strain was fifty-fold higher than that of the isogenic WT strain, indicative of LLO hyperproduction (S5C Fig). Despite very high levels of LLO secretion, eSLAPs were still detectable for the prfA* strain. indicating that large amounts of the poreforming toxin did not hamper the ability of Lm to reside and multiply inside vacuoles during several hours.

LLO
This feature contrasts with SLAPs, which formed in phagocytes only when the expression of hlyA was moderate [12]. LLO quantity did not influence intravacuolar bacterial growth, since the prfA* strain replicated at a similar rate as the WT strain in eSLAPs ( Fig 4C). Consistently, the growth rate of LLO-FAST-secreting bacteria in eSLAPs was correlated with neither the average nor the maximal level of LLO-FAST fluorescence intensity (S7E, F Fig). Nevertheless, we observed by live-cell imaging that the escape of the prfA* strain from eSLAPs occurred earlier than for the WT strain ( Fig 4C, S9 Fig). In agreement with this observation, the ratio of eSLAPs to the initial number of entry events was lower when cells were infected with the prfA* strain than with the WT strain ( Fig 4D). This higher probability of vacuole escape for the prfA* strain suggests that very a high concentration of LLO exerts a mild destabilising effect on the integrity and duration of eSLAPs, though it does not preclude their formation.
Altogether, our results suggest that the secretion of LLO, but not the precise control of its concentration, is required for proliferation of Lm in eSLAPs, and that LLO exerts only a minor influence upon eSLAP longevity.

Origin and properties of Listeria long residence vacuoles in epithelial cells
The eSLAPs in which Lm replicated (Fig 4, Movies 3, 4) likely originated from internalisation vacuoles from which bacteria had failed to escape (Fig 3C), unless they derived from secondary vacuoles produced by cell-to-cell spread, or by autophagy vacuoles where bacteria would have been entrapped after a first exposure to the host cytoplasm. To assess whether eSLAPs resulted from primary vacuoles, we monitored the intravacuolar stages of mCherry-expressing bacteria in LoVo cells transfected with the YFP-CBD fusion protein reporter [14]. This reporter has been previously described to specifically label the surface of bacteria Because the replication compartments we observed were reminiscent of SLAPs, we hypothesized that they could originate from a process analogous to LC3-associated phagocytosis (LAP), except it would occur in epithelial cells rather than in phagocytes. We thus carried out a molecular characterization of this intravacuolar replication niche in order to analyse whether it had typical features of endosomal, lysosomal and/or noncanonical autophagy-derived compartments. By immunofluorescence staining of LoVo cells infected with mCherry-expressing Lm for 3 hours, we observed that the vacuoles containing several bacteria were negative for the early endosomal marker Rab5 (9% of colocalisation, n = 22), while they were positive for the late endosomal marker Rab7 (88.5%, n = 26), LC3 (100%, n = 63), as well as the lysosomal marker LAMP1 (80.5%, n = 41) (Fig 5B, S10B Fig). These are typical markers of SLAPs, suggesting that, similar to what occurs in phagocytes, LC3 is lipidated and the noncanonical autophagy machinery recruited to entry vacuoles in epithelial cells. Also, as in SLAPs the pH inside eSLAPs remained neutral, which is revealed by their absence of staining when using the acidophilic fluorescent probe LysoTracker Deep Red (0%, n = 17) (Fig 5B,   S10B Fig). Altogether, we conclude that eSLAPs display molecular characteristics highly reminiscent of SLAPs, although they allow a faster replication of Lm, and their maturation and rupture is less sensitive to the concentration of secreted LLO than the compartments observed in phagocytes.

Discussion
Exploring the dynamics of secreted virulence factors at the (sub-)cellular scale constitutes one of the main challenges for real-time microscopy of infectious processes. Here, we bring evidence that FAST offers a versatile, convenient opportunity for tackling this challenge. We took advantage of this system to measure the lifetime of Lm internalisation vacuoles, and to monitor the endomembrane localisation of the secreted Lm virulence factor LLO in live cells. As a result, we uncovered an intravacuolar replication niche for Lm in epithelial cells.

Real-time imaging of LLO during infection
On fixed samples, observing the localisation of LLO in infected cells has often constituted a hurdle, due to the poor quality of the labelling allowed by existing anti-LLO antibodies in immunofluorescence assays [14]. LLO localisation at vacuole membranes, or more recently in bacterial-derived membrane vesicles, was first observed by electron microscopy using immunogold labelling [22,23]. However, the precise dynamics of infectious processes cannot accurately be caught by fixed-cell studies. Besides, the high spatial resolution gained by electron microscopy compromises the observation of events at a cellular scale. As a complementary approach, LLO-eGFP fusions that were ectopically-expressed in host cells have enabled live imaging, yielding precious insight into the dynamics of LLO localisation at membranes and its turnover [24]. Nevertheless, ectopic expression by host cells cannot mimic the concentrations, location, and insertion into membranes from the inside of the vacuole obtained with bacterial secretion. Moreover, host cell signalling pathways and membrane dynamics differ between non-infected and infected cells. Here, we report that (a) the FAST system can be used to tag LLO without loss of function, (b) the LLO-FAST fusion, expressed from its endogenous promoter, is secreted by Lm in infected cells, (c) the vacuoles it decorates can be imaged with accuracy, and (d) some of these vacuoles unexpectedly last for several hours.

FAST, a versatile fluorescent reporter of bacterial secretion
Beyond the live detection of LLO secreted by L. monocytogenes through the general Sec secretion system, FAST opens new perspectives for real-time imaging of bacterial proteins secreted by a broader range of bacterial models and secretion systems. For instance, we provide data supporting that FAST-tagged effectors can also be efficiently secreted through the T3SS of S. flexneri.
In recent years, several strategies have emerged for fluorescent labelling of Sec-or T3SS-dependent substrates [25]. Tagging bacterial effectors with Split-GFP provides a possible solution that has been successfully applied for live detection of Salmonella T3SS-dependent effectors or Listeria Sec-dependent secreted substrates [26,27]; however, the reconstitution process is slow compared with microbial growth, and requires the stable expression of GFP1-10 in recipient cells, which limits its application in most biological systems. Superfolder GFP (sfGFP) or its derivative rsFolder have been successfully used for labelling E. coli periplasmic proteins exported through the Sec pathway [18,28], but to our knowledge has not been applied yet for other bacterial systems or in the context of host-pathogen interactions. Other fluorescent tags such as FlAsH and phiLOV were successfully used for monitoring the secretion of Sf T3SS-dependent effectors [29,30].
Nevertheless, the toxicity in eukaryotic cells of the biarsenite dye used for FlAsH labelling and the rather modest brightness of phiLOV hamper their general use.
FAST compares with previously existing tools, while broadening the possible range of applications, due to (a) its ease of implementation (compared with Split-GFP); (b) its low toxicity (compared with FLASH); (c) its independence to oxygen allowing studies in anaerobes [31,32] as well as (d) its rapid and reversible folding dynamics allowing transport through the T3SS (compared with GFP-derived probes); (e) its reasonable brightness and fast maturation time (compared with PhiLOV). In addition, FAST-labelled proteins can be imaged at different wavelengths between 540 and 600 nm by selecting the appropriate fluorogen [33], thereby providing users with flexibility in the choice of other fluorescent reporters in co-localisation studies. Redshifted fluorogens also limit the toxicity of certain wavelength for bacteria when performing long-term imaging, and membrane-impermeant fluorogens offer the possibility to discriminate between intracellular and extracellular proteins [34], for instance when addressing the localisation of bacterial effectors that anchor to the bacterial cell wall or to membranes.
Hence, FAST expands the panel of fluorescent reporters for monitoring secreted virulence factors and offers a wealth of opportunities to accurately seize the spatiotemporal aspects of infectious mechanisms.

eSLAPs, an alternative replication niche for Listeria monocytogenes in epithelial cells
We document that in LoVo and Caco-2 epithelial cells, a consistent proportion of Lm fails to escape from internalisation vacuoles, but instead replicates efficiently inside epithelial SLAP-like vacuoles (eSLAPs), which are positively labelled by LLO-FAST (Fig 6). After several hours of intravacuolar residence and growth, eSLAPs eventually break open and bacteria resume a canonical cytosolic lifestyle.
The decoration of eSLAPs by LC3, Rab7 and LAMP1 as well as their neutral pH are reminiscent of the SLAPs (Spacious Listeria-containing Phagosomes) previously described in phagocytes [12], and which derive from LAP (LC3-associated phagocytosis) [35]. Upon infection by Lm, we propose a model for the formation of replicative eSLAPs, analogous to the current model of SLAP formation (Fig 6). Whereas the eSLAPs observed in LoVo cells display similarities with SLAPs, they are notably distinct from LisCVs, which are an intravacuolar persistence niche of Lm recently described in human hepatocytes and trophoblast cells [36]. Contrary to SLAPs and eSLAPs, LisCVs do not derive from primary vacuoles. Instead, they form late in the intracellular cycle of Lm by recapture of bacteria that have lost ActA-dependent motility.
Indeed, bacteria found in LisCVs are labelled with YFP-CBD, while the bacteria we observe in eSLAPs are not. Moreover, whereas eSLAPs are lipidated by LC3, LisCVs are not. Last, Lm replicates in eSLAPs, whereas it adopts a viable but non-culturable state in LisCVs and does not grow. Altogether, though occurring in epithelial cells, the features we describe for eSLAPs are consistent with compartments similar to SLAPs, and distinct from LisCVs. However, the replication of Lm inside eSLAPs is significantly faster than the 8 hours of doubling time reported in SLAPs [12], perhaps due to a lower bactericidal capacity of the epithelial niche compared with phagocytes. Membrane permeation by LLO might also attenuate the bactericidal properties of eSLAPs, and/or allow nutrient uptake through the permeated membrane, thereby promoting bacterial replication.

Conclusion
Together with LisCVs and SLAPs, eSLAPs enrich the palette of Lm intravacuolar lifestyles that can establish in various cells types. Apprehending the importance of eSLAPs in the context of in vivo infections prompts future investigation. Indeed, whilst intravacuolar lifestyles impose constraints on motility or nutrient uptake, these compartments might provide shelter from cytosolic surveillance mechanisms such as autophagy and RIG-I-dependent activation of type-I interferon signalling, or favour chronic forms of infections by dampening cell-to-cell spread within tissues. Conversely, delayed residence within vacuoles could promote recognition of Listeria pathogen-associated molecular patterns by endosomal Toll-like receptors and activate NF-kB-dependent inflammatory pathways. Prolonged exposure to the intravacuolar environment could also tune the expression of Lm virulence genes. Deciphering the extent to which these intravacuolar niches influence the balance between bacterial fitness and host defences becomes critical to better appreciate longterm relationships between Lm and its host.

Bacterial strains, plasmids and culture conditions
The bacterial source strains used in this work were Escherichia coli NEB5a (New England Biolabs) for plasmid constructions, Rosetta(DE3)pLysS (Novagen) for recombinant protein production, the clinical isolate of Listeria monocytogenes LL195 (lineage I, ST1) [37] for most of the experiments involving Lm, and Shigella flexneri M90T [38] for experiments on Sf T3SS-dependent secretion. Lm reference strain EGD-e (lineage 2, ST9) [39] (lineage II, ST9) was also used as a control that the observed eSLAPs were not specific to LL195.
All strains were grown at 37°C under shaking at 190 rpm in Luria Bertani (LB) medium for E. coli, in LB or tryptic soy broth (TSB) for Sf, in brain hear infusion (BHI) or Listeria synthetic medium (LSM) [40] for Lm.
In order to favour the expression of transgenes, the DNA coding sequence for FAST, fused with a Myctag, was codon-optimized for Lm (S1A Text) or Sf (S1B Text) using the online Optimizer application (http://genomes.urv.es/OPTIMIZER/) in guided random mode. The optimized sequences were obtained as synthetic Gene Fragments (Eurofins genomics). The Lm-optimized sequence additionally contained the 5'untranslated (5'-UTR) of the hlyA gene, and the sequence encoding the signal peptide (SP) of LLO in its Nterminal part.
For plasmid constructions in the pAD vector derived from the pPL2 backbone [41,42], the 5'-UTRhlyA-SPhlyA-FAST-Myc fusion was amplified with primers oAL543-4, the sequence of the Lm hlyA gene encoding LLO was amplified from the EGD-e genomic DNA with primers oAL549-50b, and the coding sequence for eGFP was amplified from pAD-cGFP (BUG2479) [42] with primers oAL543-7. The UTRhlyA-SP-FAST-Myc amplicon was inserted instead of UTRhlyA-eGFP into the EagI-SalI restriction sites of pAD-cGFP, thus generating pAD-SP-FAST, where FAST is under control of the PHYPER constitutive promoter (Fig 1A). pAD-FAST, pAD-eGFP, pAD-SP-eGFP, pAD-hlyA, pAD-hlyA-FAST and pAD-hlyA-eGFP, all containing the 5'-UTR of hlyA and a Myc tag, were likewise generated by inserting the cognate DNA amplicons into the same restriction sites (Fig 1A). After cloning in E. coli NEB5a, these plasmids were integrated in the genome of L.
monocytogenes strains LL195 at the tRNA Arg locus by electroporation as previously described [41]. The pHpPL3-mCherry plasmid was introduced at the tRNA Arg locus by conjugation [43]. genomes of L. monocytogenes strains LL195 and EGD-e were obtained as previously described [44]. For complementation purposes in haemolysis assays, a simple in-frame deletion mutant of the hlyA gene was also created using the pMAD backbone.
The complete lists of bacterial strains and oligonucleotides used in this work are supplied as Tables S1 and S2, respectively.

Analysis of protein contents in bacterial total extracts or culture media
Bacterial total extracts or culture supernatants were recovered from 1 ml of Lm strains grown to an OD600nm of 2.0 in BHI at 37°C as previously described [46].  [pAD-LLO] or Sf M90T ∆ipaD) and subtracted to each sample. The standard curve linking FAST fluorescence to its concentration was performed by diluting, in control medium corresponding to a culture of the corresponding negative control strain, known amounts of recombinant FAST produced in E. coli Rosetta(DE3)pLysS as previously described [17]. This enabled the calculation of fluorescent FAST-tagged proteins released in each culture.

Fluorescence measurements on culture supernatants
As a negative control that the fluorescence was due to the formation of the FAST-HBR-3,5DM complexes, the fluorescence was also measured on samples mixed with 20 μl of PBS instead of PBS containing HBR-3,5DM. No fluorescence was detected above that of the control culture media.
For normalisation between measurements for FAST-and eGFP-tagged proteins, fluorescence intensities measured in filtered culture media were expressed relatively to the fluorescence measured for a suspension of OD600nm = 1 of Lm constitutively expressing either non-secreted FAST, or eGFP. The fluorescence intensities emitted by each one of these reference suspensions were fixed arbitrarily to 100 A.U.
Each experiment was reproduced three times independently.

Haemolysis assay
The supernatants of 16-h cultures of Lm in BHI were recovered by centrifugation for 1 min at 6,000 ´ g followed by filtration through 0.2-μm pore filters, in order to eliminate bacteria. Serial two-fold dilutions of these supernatants were performed in round-bottom, clear, 96-well plates (100 μl final volume per well) using as a diluent PBS, the pH of which was adjusted to 5.6, and supplemented with 0.1% bovine serum albumin (BSA). Erythrocytes from defibrinated mice blood were washed twice in PBS pH 6.4 and diluted 1/10 th in PBS pH 5.6. 50 μl of this suspension was added to each one of the wells containing diluted culture supernatants.
After 30 min of incubation at 37°C, the plates were centrifuged for 10 min at 430 ´ g and haemolytic titres were calculated as the reciprocal of the dilution for which 50% of haemolysis was observed [47]. Lm strains were grown in BHI until they reached early stationary phase (OD600 of 2 to 3), washed in prewarmed D-MEM, and then diluted in culture medium without serum to achieve a multiplicity of infection (MOI) of 5 (for long-term infections) to 30 (for short-term infections). Except for short-term imaging when bacterial entry was monitored under the microscope, after 1 h of bacteria-cell contact the inoculum was washed away twice with serum-free medium containing 40 μg/ml gentamicin, then the medium was replaced by complete culture medium without phenol red containing 25 μg/ml gentamicin in order to kill extracellular bacteria.

In-situ growth measurements of mCherry-labelled bacteria
We performed for each time point an Otsu-thresholding on the entire z-stack of mCherry images in order to obtain the 3D segmentation of bacteria. Although the segmentation was not able to isolate individual bacteria in dense regions, it yielded the total volume occupied by bacteria in the field of view for each frame. The growth rate of mCherry bacteria was measured by fitting the total volume of segmented objects ( ) to an exponential curve ( ) = % $& , with the growth rate and % the initial volume. The doubling time t was then computed according to the following formula: t = !"# $ . Image computing was performed using MatLab RRID:SCR_001622.

Tracking of primary vacuoles in short term infection assays
The slices of the z-stacks obtained from spinning confocal imaging were projected onto a single plane (maximal projection). Fluorescent vacuoles were tracked using the plugin TrackMate in Fiji. The time at which tracks began during the infection was used to compute the time of Lm entry into LoVo cells. The distribution of residence times in primary vacuoles was computed from the statistics of track lengths.

Tracking of eSLAPs in long-term infection assays
At 2 h p.i., Ibidi μslides were mounted on a confocal spinning disc microscope for time-lapse observations. Pixel size was 250 x 250 nm and step size in z was 1 µm. For each time point taken every 5 or 15 minutes, we recorded a z-stack of fluorescent images in two channels for FAST signals and mCherry labelled bacteria.
Given the good signal-to-noise ratio of mCherry images, we applied for each time point Otsu's thresholding algorithm on the entire z-stack in order to obtain a 3D segmentation of bacteria. The Otsu segmentation did not allow to isolate bacteria when they were too dense, for instance in eSLAPs, or after prolonged infection when cells were full of bacteria. Hence, we detected objects that could either be single bacteria or clusters of bacteria. We tracked each segmented object from frame to frame based on their similarities in size and location.
Our method took benefit from the fact that cytosolic bacteria were moving very fast compared to those encapsulated in eSLAPs. The growth rates of bacteria inside eSLAPs were computed by fitting the dynamics of segmented mCherry volumes to an exponential function ( ) = % $& , with the growth rate and % the initial volume of the vacuole. The doubling time t was then computed according to the following formula: t = !"# $ . No growth was observed for the ∆hlyA strain in long-term vacuoles tracked by the co-occurrence of FAST and mCherry signals (based on the spherical shape of the object). For the WT and the prfA* strains, an object was classified as a growing eSLAP if it met the two following criteria: (a) the initial size was equal to or larger than the size a of single bacterium (32 voxels) (S11 Fig), and  to the initial number of objects bigger than 32 voxels on the first frame of the movie (S11 Fig), which is a good proxy for the initial number of entry events. To quantify LLO-FAST signals in a given vacuole, we used the binary mask of mCherry labelled objects to measure the average intensity in the corresponding region of the FAST image. Image computing was performed using MatLab.

Quantification of YFP-CBD signals
Images were first z-projected by maximum intensity. For both fluorescent channels (mCherry bacteria and YFP-CBD), we measured the average fluorescence in a rectangular ROI with an area of 20 pixels at each time point. We tracked each vacuole manually for 5 frames before it ruptured. Then, for each vacuole, we tracked manually two bacteria that were issued from a given vacuole until we could not follow them anymore (2 to 6 frames after rupture). To quantify the number of vacuoles that associated with the indicated markers, 9 to 14 microscopic fields were examined from coverslips (or from live infection for the LysoTracker staining), and processed with Fiji.

Immunofluorescence or LysoTracker staining of infected cells
A vacuole was defined as a round aggregate of at least 4 bacteria (labelled with either mCherry in immunofluorescence experiments, or with eGFP in live experiments) that did not colocalize with actin staining.
The presence of each marker was assessed by seeking for a corresponding fluorescent signal in the vicinity of vacuoles, and displaying a shape similar to the edge of the vacuole.
equipment that was instrumental to this work, and providing expert support whenever needed. We are indebted to the IBENS animal facility for kindly supplying the mice blood used in haemolysis assays.     (Fig 2). To get an estimate of the number of bacteria in each field, the total volume occupied by bacteria (the number of voxels that were labelled with mCherry) was divided by the average size of bacteria (32 voxels). Each colour represents an independent biological replicate (in blue, two wells were recorded in the same experiment). The exponential fit associated to each experiment is displayed as orange dashed lines.

S4 Movie. Observation of the decoration of vacuoles by LLO-FAST in Listeria cells infected by
Lm EGD-e.