Differences in internalization and growth of Escherichia coli O157:H7 within the apoplast of edible plants, spinach and lettuce, compared with the model species Nicotiana benthamiana

Summary Internalization of food‐borne bacteria into edible parts of fresh produce plants represents a serious health risk. Therefore, internalization of verocytotoxigenic E. coli O157:H7 isolate Sakai was assessed in two species associated with outbreaks, spinach (Spinacia oleracea) and lettuce (Lactuca sativa) and compared to the model species Nicotiana benthamiana. Internalization occurred in the leaves and roots of spinach and lettuce throughout a 10 day time‐course. The plant species, tissue type and inoculum dose all impacted the outcome. A combination of low inoculum dose (~102 CFU) together with light microscopy imaging highlighted marked differences in the fate of endophytic E. coli O157:H7 Sakai. In the fresh produce species, bacterial growth was restricted but viable cells persisted over 20 days, whereas there was > 400‐fold (~2.5 Log10) increase in growth in N. benthamiana. Colony formation occurred adjacent to epidermal cells and mesophyll cells or close to vascular bundles of N. benthamiana and contained components of a biofilm matrix, including curli expression and elicitation, extracellular DNA and a limited presence of cellulose. Together the data show that internalization is a relevant issue in crop production and that crop species and tissue need to be considered as food safety risk parameters.


Introduction
Verocytotoxigenic Escherichia coli (VTEC) is a foodborne pathogen that can cause serious disease ranging from haemorrhagic colitis to life-threatening haemolytic uraemic syndrome (HUS) and central nervous system damage (Kaper et al., 2004). Although most cases were previously associated with contamination of meat and milk products, in recent years fresh fruit and vegetables have been increasingly implicated as the sources of infection and on an international scale,~20% of foodborne VTEC outbreaks are thought to have arisen from fresh produce (Greig and Ravel, 2009). One of largest outbreaks of VTEC occurred in Japan in 1996, with almost 10 000 cases and 12 deaths, as a result of contamination of white radish sprouts with E. coli O157:H7 (Michino et al., 1999;Watanabe et al., 1999). The isolate responsible, E. coli O157:H7 Sakai, forms the basis of the current study. Whilst transient surface contamination of fresh produce has been demonstrated, it is now recognized that VTEC are able to colonize plants as secondary hosts in a manner that has some key differences from their primary host, cattle (Holden et al., 2009). Indeed, specific E. coli isolates appear to have adapted to their hosts with isolates in phylogenetic group B1 more commonly associated with plants, whereas phylogroups A and B2 are linked to an animal-associated lifestyle (M eric et al., 2012).
VTEC and other food-borne pathogens, such as Salmonella enterica, can exist on both external and internal tissues of plants (Deering et al., 2012;Erickson, 2012;Hirneisen et al., 2012;Hou et al., 2013;Wright et al., 2013;Martinez et al., 2015). This capability presents a food safety threat in crop production, as internalized bacteria cannot be removed with standard sanitation practices, although treatments such as irradiation, ultrasound and cold plasma can be effective (Gomes et al., 2009;Bilek and Turantas ß, 2013;Ziuzina et al., 2015). Internalization into plants is defined as the ability of bacteria to penetrate into the internal tissues, where the bacteria normally reside in the extracellular spaces of the apoplast (Godfrey et al., 2010). Some well-characterized phytopathogens also reside close to the leaf surface in the stomatal pores (Yu et al., 2013), and other plant-associated bacteria are known to internalize (Turner et al., 2013). Internalization of E. coli has been demonstrated by recovery of bacteria following surface sterilization of the plant tissue (Solomon et al., 2002;Warriner et al., 2003b;Gomes et al., 2009;Deering et al., 2011;Hirneisen et al., 2012;Wright et al., 2013;Erickson et al., 2014).
We have previously demonstrated that a large proportion (81-91%) of spinach plants maintained in hydroponic culture support an internal population of bacteria, which accounted for~0.5% of the total population (Wright et al., 2013). Within the roots, these bacteria were located inside the cell walls of epidermal and cortical cells and within the apoplast between the plant cells. Similarly, colonization and invasion of fiveto 7-day-old Arabidopsis thaliana roots by E. coli O157:H7 strain Odwalla were observed along with proliferation of bacteria on the roots and shoots over a 3 day period (Cooley et al., 2003). Young lettuce seedlings grown in soil containing E. coli O157:H7-spiked manure were colonized on the surface, with some bacteria also located within the leaves (Solomon et al., 2002). As a second route of contamination, spiked irrigation water applied to the roots yielded bacteria from the mature leaves (Solomon et al., 2002). However, the route by which bacteria migrate from roots to shoots remains unclear, particularly in relation to whether bacteria can move through the vasculature.
Occurrence of internalization under conditions that are relevant to horticulture and at ecologically appropriate inoculum levels has been demonstrated, although it appears to occur at a low frequency (Erickson et al., 2010(Erickson et al., , 2013. This raises questions relating to the internalization ability of food-borne pathogens and its relevance in food safety, i.e. do sufficiently high enough numbers of bacteria get to and stay within the plant material to pose a food safety threat to the consumer at the end of the food chain? More specifically, how well do VTEC internalize; how do internalized VTEC fare over prolonged time periods; does any specificity exists in the plant-microbe interactions or is the interaction generic among hosts? As such, the aim here was to determine the ability and subsequent fate of the most common VTEC serotype, O157:H7 (isolate Sakai), to internalize into edible species that have been associated with food-borne VTEC outbreaks, as well as a wellcharacterized member of the Solanaceae, Nicotiana benthamiana, grown under commercially relevant conditions and to harvestable age. Use of N. benthamiana allowed utilization of genetic resources that are not available for edible species. Finally, the growth rates of internalized E. coli O157:H7 were measured using a very low density of starting inocula (tens of CFU) and the molecular nature of apoplastic colonies described.

Variation in the ability of E. coli Sakai to internalize into different hosts and tissues
In our previous work, we found that E. coli O157:H7 Sakai internalized into the roots and leaves of spinach and lettuce grown under hydroponics conditions (Wright et al., 2013), but the rate at which this occurred for plants grown under conditions more relevant to commercial production is unknown. Thus, the ability of E. coli Sakai to internalize was measured in the roots and leaves of spinach and lettuce grown in compost, in a glasshouse. Two different inoculum levels were used: a high dose at 10 7 CFU ml À1 that is appropriate for laboratory-scale assessments of bacteria-plant interactions and microscopy, and at a lower dose, 10 3 CFU ml À1 that is more likely to be found in nature from contaminated soil or irrigation water (Matthews et al., 2006). Internalization was defined as the number of bacteria recovered from plant tissue following surface sterilization with 200 ppm calcium hypochlorite for 15 min. These levels, which are~2.5 times higher than that used in industrial production for maintaining a degree of cleanliness of produce washing water (Ramos et al., 2013), were used to keep the work relevant to vegetable production. A large level of variation in the efficiency of surface sterilization occurred, ranging from 38% for spinach leaves inoculated with a high dose, to 100% for lettuce roots or leaves inoculated with a low dose of bacteria (Table 1). Incomplete surface sterilization meant that it was not always possible to obtain a full data set, e.g. for lettuce leaves inoculated with a high dose of bacteria, and indicated that higher concentrations of hypochlorite are required for complete removal of surface-associated E. coli Sakai.
Recovery of internalized E. coli Sakai varied widely between the plant species, tissue type and starting inoculum, although internalized bacteria were recovered in every case and for each time point (Table 1). Inoculation with the high density (10 7 CFU ml À1 ) resulted in~5 log 10 CFU total population on roots or leaves for the starting time point, whereas inoculation at the low density (10 3 CFU ml À1 ) resulted in~2 log 10 CFU total population for the same point. Inoculation at the higher density resulted in the highest number of internalized E. coli Sakai, with the maximum numbers found in spinach roots 10 dpi (2.37 log 10 CFU), followed by lettuce roots (1.87 log 10 CFU) and spinach leaves (1.81 log 10 CFU), at the same time point. However, there were no significant differences between any of the groups at this dosage, based on presence/absence analysis. The lowest numbers recovered occurred for the lower inoculum at 5 dpi in spinach roots (0.08 log 10 CFU) and spinach leaves (0.27 log 10 CFU), although in both cases, the numbers increased by 10 dpi (to 0.21 and 1.50 log 10 CFU respectively). The numbers of internal E. coli Sakai were significantly higher in lettuce roots compared with spinach roots, determined by an unbalanced ANOVA on presence/absence counts (P < 0.05) for the low dosage, and there was a significant interaction between the mean responses for plant species and tissue type (P < 0.01). The highest proportion of internalization occurred at 10 dpi in the leaves of Nicotiana benthamiana inoculated with a low dose, with 36% of the total population (i.e. external + internal) internalized (1.34 log 10 CFU). At the low dose, regression analysis showed that there were significant differences for the total populations of bacteria present, between both plant species (P < 0.01) and for a plant species by tissue interaction (P < 0.05). The increase in the total population on the leaves of lettuce and the roots of both lettuce and spinach inoculated at a low dose was indicative of proliferation of the bacteria on external tissue.
Stability of the reporter plasmid in internalized E. coli O157:H7 Sakai The extent of variation in internalization ability showed, to some extent, a plant species and tissue type influence. Therefore, we used microscopy to examine the fate in internalized E. coli Sakai in more detail. The bacteria were transformed with a plasmid-borne GFP reporter, under the control of the E. coli gyrA promoter (termed E. coli Sakai-GFP), to aid in detection. The long-term maintenance of the plasmid-borne GFP reporter was found to be stable for at least 10 days following examination of E. coli Sakai-GFP drop inoculated onto the leaves of spinach or N. benthamiana leaves. There were no significant differences in the total population (i.e. external and internal) of E. coli Sakai-GFP recovered on agar either lacking or containing the selective antibiotic from N. benthamiana (average of 6.09 or 6.13 log 10 CFU, respectively) or spinach leaves (average of 4.47 or 4.56 log 10 CFU respectively).  Colony development by internalized E. coli Sakai-GFP in N. benthamiana leaves Once it was established that the GFP reporter could be used to detect E. coli Sakai, it was possible to characterize the fate of internalized bacteria. E. coli Sakai was infiltrated into plant leaves to investigate its fate in isolation of epiphytic bacteria. Bacteria for plant inoculations were cultured at 18°C to avoid any temperature-shifts, as temperature is a key environmental factor known to affect E. coli gene expression (Crozier et al., 2016) that could in turn, affect the fate of internalized bacteria. E. coli Sakai-GFP were infiltrated into N. benthamiana leaves at the high dose (10 7 CFU ml À1 ) to ensure that there were sufficient fluorescent cells visible for examination by microscopy. N. benthamiana expressing a fluorescent plasma membrane reporter (termed mOrg-LTI-benth) was used to aid with in planta localization of E. coli Sakai-GFP. Four days of postinfiltration, the bacteria were observed in colonies frequently associated with the internal face of the abaxial epidermal cells in the channels between adjoining cells (Fig. 1A). At a later time point, 23 dpi, increased sized colonies were located associated with the epidermis (Fig. 1B-C). Colonies were also visible at a much greater depth into the tissue, associated with the bundle sheath cells ( Fig. 1D-G). All E. coli Sakai-GFP were observed to be in the apoplastic space of the leaf, with no indication of penetration by the bacteria into the plant cells. Some individual bacterial cells within the colonies could be distinguished (Fig. 1H) along with a fluorescent matrix, which in some cases could also be detected using transmitted light (Fig. 1I).
To determine whether development of colonies of infiltrated E. coli Sakai-GFP had arisen as a result of bacterial proliferation, wild-type (i.e. non-labelled) N. benthamiana were infiltrated with a mixture of E. coli Sakai transformed with either the GFP reporter or a RFP reporter. After 19 days, separate GFP-and RFP-expressing colonies were detected at the internal boundary of the epidermal cells ( Fig. 1J-L), with larger sized colonies associated with the mesophyll cells (Fig. 1M). The lack of mixing of differently labelled bacterial cells indicates that colonies developed from proliferation of a single GFP-or RFP-expressing bacterium.
Internalized E. coli Sakai-GFP do not form colonies in spinach, lettuce or tomato leaves To compare the fate of internalized E. coli Sakai in edible crop species associated with outbreaks of food-borne pathogens, infiltrated bacteria were examined in the leaves of spinach, lettuce or tomato. Infiltration of a combination of E. coli Sakai-GFP or Sakai-RFP resulted in a different pattern of colonization to that seen in N. benthamiana. Even after prolonged incubation of 14-21 days, only a few bacteria were detected within spinach ( Fig. 2A). On frequent occasions, a single bacterium was observed attached to mesophyll cells adjacent to a region of autofluorescence within the cell wall (Fig. 2B). Such regions were not observed in tissue infiltrated with MS buffer-only negative control (Fig. 2D). Small chains of E. coli Sakai cells, grouped by the GFP or RFP reporter (Figs 2C, F and H-I), were indicative of limited proliferation in all three species.
The lack of E. coli Sakai colony development in the apoplast of spinach, lettuce or tomato compared with N. benthamiana indicated large differences in the ability of internalized bacteria to grow. Therefore, we determined the in planta growth rate for E. coli Sakai-GFP. Very low inoculum densities of~20 to 50 cells were used to minimize background 'noise' from cell turnover and thus allow growth to be definitively measured. This necessitated the use of MPN to estimate numbers below the limit of detection by direct plating. The numbers of E. coli Sakai-GFP infiltrated into the leaves of N. benthamiana significantly increased during the first four sampling times, up to 15-21 dpi (Fig. 3) although there was variation between individual plant replicates ( Table 2). Growth between d1 and d14 was in the order of 10.5 generations, which represents a generation time of~1.3 days. A similar growth rate was measured with a higher starting inoculum, suggesting that growth was not limited by nutrient availability within the apoplast (Table 2). In contrast, infiltration of E. coli Sakai-GFP into spinach or tomato leaves showed either minimal or no increase in the bacterial population ( Fig. 3B-C). Infiltration into lettuce leaves generated large variation in the numbers of bacteria recovered (data not shown), possibly due to technical variation in the success of infiltration compared with the other species, and thus in our hands it was not possible to obtain a sufficiently robust growth rate for this species. Therefore, tomato leaves were included as a second fresh produce crop: although tomato leaves are not eaten, VTEC has been recovered from retail tomato fruits (Gomez-Aldapa et al., 2013) and the related food-borne pathogen S. enterica is often linked to their consumption (Bennett et al., 2015).

Characterization of E. coli Sakai apoplastic colonies in N. benthamiana leaves
Development of E. coli Sakai-GFP colonies within the apoplast of N. benthamiana were often accompanied by an unknown matrix (Figs 1I and 4A). As colonization of plants by enteric pathogens has been reported to be, in part, dependent on the ability to produce the biofilm components curli (Macarisin et al., 2012;Hung et al., 2013) and cellulose (Barak et al., 2007;Carter et al., 2011), it was possible that the VTEC matrix in planta also contained these components. Therefore, indicator dyes, calcofluor white for cellulose and Congo red for curli, together with a plasmid-borne curli reporter fusion (csgBA-gfp), were used for their detection on well-developed E. coli Sakai-GFP apoplastic colonies in N. benthamiana or mOrg-LTI-benth. Treatment with either calcofluor white or Congo red did not result in staining of E. coli Sakai prior to infiltration indicating that neither component was expressed to detectable levels during in vitro growth (Fig. S1). However, Congo red staining occurred on the periphery and some internal structures of the apoplastic E. coli Sakai-GFP colonies Fig. 4B and C). Expression of the curli structural gene promoter, csgBA, was also observed in apoplastic E. coli Sakai, (Fig. 4D and E). Prior to infiltration, E. coli Sakai was grown in in vitro conditions (RD-MOPS glucose at 18°C) that repressed csgBA expression, with only background levels of fluorescence, 67.5 AE 63 RFU, detected from a mid-log phase culture (OD 600 of 0.6). In comparison, in vitro induction occurred when E. coli Sakai was cultured in RD-MOPS glycerol (287 AE 36 RFU, 95% CI), and to a marginal, but not significant extent in spinach leaf lysates (167 AE 39, 95% CI). Elicitation of curli fibres from E. coli O157:H7 Sakai, from Congo red staining, was observed in vitro but only when cultured on curli induction medium at 37°C (red colonies) and not at 18°C, or on indicator plates made with M9 glucose (white colonies). Together, the data indicate a specific signal derived from live plants was responsible for curli production in the apoplast biofilms. Calcofluor white binds cellulose from plants and bacteria non-discriminatorily, and as expected the majority of staining occurred in plant cell walls, external to the mOrange-labelled plasma membrane (Fig. 4A). There was only limited evidence of bacteria-derived cellulose in the biofilm matrix, where the majority of bacterial colonies observed did not stain with calcofluor white (Fig. 4A and B), although staining of internal colony structures that did not coincide with Congo red was observed occasionally (Fig. 4C). Extracellular DNA (eDNA) has also been associated with E. coli biofilms (Tetz et al., 2009)  and staining with DAPI indicated discrete structures within a colony ( Fig. 4F and G). Whilst DAPI stained E. coli Sakai cells in vitro prior to infiltration (Fig. S1C), the apoplastic stained structures did not always coincide with intact bacterial cells within the colony (Fig. 4G). Together, these data demonstrate that curli, eDNA, and to a lesser extent cellulose are components of the biofilm matrix that are expressed in apoplastic E. coli Sakai colonies and not under relevant in vitro growth conditions.

Discussion
VTEC are able to colonize plants as secondary hosts, and although the vast majority of the colonizing population exists as epiphytes, a subpopulation can penetrate the internal tissues where they are located in the apoplast. As endophytic behaviour of human pathogens represents a public health threat, there is a need to better understand the ability of the bacteria to internalize and determine their fate over time. The presence of E. coli bacteria in the substomatal cavity and in association with the spongy mesophyll has only been demonstrated convincingly in seedlings germinated from inoculated seed (Itoh et al., 1998;Warriner et al., 2003a;Deering et al., 2011). Here, we show some of the dynamics of E. coli O157:H7 movement to internal tissue of mature plants of spinach, lettuce and N. benthamiana by examining bacterial location over time. One of the key findings was that internalization occurred in every case and at each time point tested, underscoring the propensity of E. coli Sakai to invade plant tissue. Both the plant species and tissue type significantly impacted internalization for the low inoculum dose, with the highest level observed in lettuce roots and the lowest in spinach leaves. Others have reported on the presence of endophytic VTEC under a variety of conditions, reviewed in Deering et al. (2012) and Hirneisen et al. (2012); however, it has not been possible to determine the fate of the bacteria over time. Using a combination of a very low-cell-density inoculation and taking advantage of advances in detector sensitivity for deep imaging to vascular tissue, species-specific differences that affect internalized E. coli Sakai have come to light.
The fate of internalized E. coli O157:H7 Sakai in spinach or lettuce was distinct from those cells on the external tissue. Higher levels of total colonization (both internal & external) tended to occur on the roots for all species tested. Although roots of these species are not consumed, mechanized harvesters can introduce root tissue, which subsequently requires removal during trimming and packaging, thus running the risk of microbial cross-contamination to edible tissue (Reed, 2011). In  Table 2 for others). spinach, lettuce or tomato. The occurrence of chains of two to four bacteria indicated limited proliferation, which equated to~threefold increase in the population in tomato over a 20 day period. A similar pattern of bacterial persistence without multiplication in spinach was reported for E. coli O157:H7 14 dpi, although in this case a higher inoculum density was used (Mitra et al., 2009). The lack of growth in spinach leaves contrasted with that seen in spinach roots as we previously found endophytic colonization of roots by E. coli Sakai-GFP bacteria, with large numbers of bacteria accumulating within epidermal cells, presumably as a result of proliferation (Wright et al., 2013).
In contrast to the situation in edible crop species, infiltration of E. coli Sakai into the apoplastic spaces of N. benthamiana leaves resulted in establishment of large colonies with an increase of > 400-fold in the numbers of bacteria after 20 days. Formation of essentially distinct colonies that were labelled with either GFP or RFP precluded the possibility that the colonies arose as a result of bacterial aggregation. The molecular basis to the differences in the ability of E. coli Sakai to grow, either in different tissues or in different species, is unknown, although it is likely to be a factor of the plant defence system (Melotto et al., 2014).
The colonies formed within N. benthamiana exhibited some of the characteristics of biofilm extracellular matrix, curli and in some instances cellulose, which have previously been associated with E. coli and S. enterica colonization of plants (Eberl et al., 2007;Lapidot and Yaron, 2009;Macarisin et al., 2012;Yaron and Romling, 2014;Carter et al., 2016), as well as eDNA. Curli fibres are adhesive structures that play a role in biofilm formation (Barnhart and Chapman, 2006) via adhesion to both biotic and abiotic surfaces and autoaggregation (Goulter et al., 2010). Cellulose production is widespread in bacteria (R€ omling and Galperin, 2015) and frequently coexpressed with curli (Hufnagel et al., 2015). Curli production was evident from both Congo red staining and reporter expression of the csgBA promoter, whilst calcofluor white, which stains b-polysaccharides with high affinity for cellulose (Adav et al., 2010), indicated bacterial-derived cellulose over and above the strong plantderived signals. Production of curli was more frequent than cellulose for E. coli O157:H7 Sakai in planta, indicating differences in expression profiles. In E. coli and Salmonella enterica, the expression of curli (csgBA) and cellulose (bcsA) is co-activated by the transcriptional activator CsgD (R€ omling et al., 1998;Weber et al., 2006). However, intra-and interstrain variability shows that cellulose is not an essential biofilm component (Uhlich et al., 2006) and for some E. coli isolates, cellulose regulation occurs through an alternative mechanism, independent of curli (Da Re and Ghigo, 2006). Neither curli nor cellulose formation was detected in the E. coli O157:H7 Sakai cultures immediately prior to infiltration and gene expression of csgBA only occurred during colony formation and establishment in planta, indicating specific signals for induction that could be plant-and/or nutrient-associated. This supports published findings for E. coli (Sakai) (Lim et al., 2010) and a comparison of E. coli isolates from plants or humans and other mammals, where the plant-associated strains were significantly more likely to display the rdar morphotype, indicative of a biofilm comprising curli and cellulose (M eric et al., 2012). A distinction was also found within the E. coli O157:H7 serogroup, where plant-derived isolates were more likely to express curli under nutrient limitation than animal-derived isolates (Carter et al., 2011). A potential trigger for apoplastic curli production in live plants supports data that show curli-independent biofilm formation on abiotic surfaces for E. coli O157:H7 cultured in spinach leaf lysates (Carter et al., 2016). The presence of eDNA has been reported previously in E. coli biofilms (Tetz et al., 2009) and is an essential component for others such as Pseudomonas aeruginosa (Whitchurch et al., 2002). eDNA in the E. coli O157:H7 Sakai biofilm matrix in planta may have occurred via an active secretion mechanism or as a result of cell lysis (Whitchurch et al., 2002;Wu and Xi, 2009).
Confocal laser scanning microscopy combined with fluorescent fluorochromes has developed as a valuable method to examine biofilms formed on slides (Lawrence et al., 2007;Adav et al., 2010;Neu et al., 2010), or on leaf surfaces (Rossez et al., 2014) and for bacteria derived from leaf surfaces (Carter et al., 2016). It has also been used to illustrate colony development of phytopathogens in planta, e.g. Misas-Villamil et al. (2013), and alternative approaches have shown a role for biofilm development in planta for related members of the Enterobacteriaceae (Koczan et al., 2011). Here, we were able to exploit technical advances for deep tissue imaging to characterize biofilm components of endophytic E. coli Sakai in planta. Biofilm formation on the leaf epidermal surface is suggested to protect the bacteria from extreme variations in hydration (Morris and Monier, 2003;Eberl et al., 2007) but whether they perform a similar function within the moist environment of the leaf interior, offer protection from components of the plant defence or have identical composition is as yet unknown. It has been suggested that the a key difference between plant (phyto-) and human enteric pathogens is that the latter do not show significant multiplication on leaf surfaces of mature plants (Yaron and Romling, 2014). However, the fact that growth can occur under specific conditions such as the germination of sprouts (Cooley et al., 2003), on cut surfaces (Wachtel et al., 2002), or as demonstrated here in a tissue-and speciesspecific manner indicates a more complex and dynamic interaction. Plant age is another factor that affects the colonization ability of enteric pathogens (Brandl and Amundson, 2008). Here, relatively mature plants of at least 4 weeks old at the point of inoculation were used, to maintain relevance to the harvestable age for leafy vegetables. To determine the risk of internalization to public health, additional considerations including the dose response need to be considered, which for VTEC is variable but estimated to be below a dose of~1000 bacteria (Strachan et al., 2005). As the bacteria do not proliferate within spinach or lettuce leaves, internalization of at least the minimum infectious dose would be required to cause a threat to public health, notwithstanding the contribution of the epiphytic population. However, in species where growth is not restricted, such as N. benthamiana, the occurrence of only a few individual bacteria to penetrate into and proliferate within the leaf apoplast would provide sufficient inoculum to exceed the minimum infectious dose. Whilst the lack of bacterial proliferation in the apoplast of spinach or lettuce leaves is reassuring for public health, it is important to identify the factors involved in the multiplication of bacteria within the N. benthamiana apoplast and identify other plant species, particularly those used as uncooked food sources, where similar growth can occur.

Experimental procedures
Bacterial strains and growth conditions Escherichia coli O157:H7 Stx-negative strain Sakai (kanamycin resistant) (Dahan et al., 2004) was used for the experimental work and E. coli strain AAEC189A (Teunis et al., 2004) was used for cloning. Bacteria were routinely cultured~18 h in lysogeny broth (LB) supplemented with chloramphenicol (25 mg ml À1 ) or kanamycin (25 mg ml À1 ), at 37°C with aeration. Prior to plant inoculations, E. coli Sakai was subcultured at 1:100 dilution into MOPS medium supplemented with amino acids and 0.2% glucose, termed rich-defined RD-MOPS glucose (Neidhardt et al., 1974) with antibiotics as required, and incubated with aeration for~20 h at 18°C to pre-adapt the bacteria to plant-relevant growth temperatures. Cultures were diluted to OD 600 of 0.02 (equivalent to 10 7 CFU ml À1 ) in 0.5 9 MS medium (Murashige and Skoog) without sucrose and diluted further as required. Long-term stocks of bacteria were stored in 20% glycerol at À80°C. MacConkey agar was used for selection of E. coli, otherwise LB agar was used.
The plasmid-borne GFP fluorescent reporter was generated previously with the gyrA promoter was fused to gfp+, termed pgyrA-gfp (Holden et al., 2006). An equivalent red fluorescent reporter was generated by fusing the same promoter to a far-red RFP, mKate (Shcherbo et al., 2007) derived from pNW725 (Marlow et al., 2014) and subcloned into pACYC184 using BamHI and HindIII restriction enzyme sites. The gyrA promoter from E. coli was then cloned upstream of mKate using primers gyrA_H3 (5 0 -CCCAAGCTTCAATATAGCCCAGACGCA) and gyrA-BH1 (5 0 -CGCGGATCCGCTATCCCTCTACTG TATCC) and termed pgyrA-rfp. E. coli Sakai was transformed with the plasmids to produce two strains referred to as E. coli Sakai-GFP and Sakai-RFP. Stability of the GFP plasmid was tested by inoculating 3 9 2 ll drops of GFP-Sakai diluted in 0.5 9 MS to OD 600 of 0.02 onto the abaxial epidermis of mOrg-LTI-benth or spinach leaves, seven replicated plants each, and maintained in the growth chamber for 12 days. The inoculated area was cut from the leaf with a cork borer, extracted in PBS following maceration using a pestle and mortar, the cells serially diluted 10-fold in PBS and recovered on MacConkey agar plates either with or without chloramphenicol. A paired ttest for means was used to compare the results AE antibiotic selection, at the 95% confidence level.

Curli expression and production
The csgBA promoter from E. coli O157:H7 (Sakai) was cloned into the XbaI site in pKC026 using primers csgB.XbaI.F (5 0 -CTCTAGATATTTACGTGGGTTTTAATA CTTTGG) and csgB.XbaI.R (5 0 -GGTCTAGAGTTGTC ACCCTGGACCTGG), to generate a csgBA-gfp+ transcriptional fusion plasmid, termed pAH004. E. coli Sakai transformed with pAH004 or pKC026 was incubated in LB at 37°C, 200 r.p.m. for~18 h, and then subinoculated at 1:100 dilution into RD-MOPS with glucose or glycerol (0.2%), or in MOPS supplemented with 40% (v/v) spinach leaf lysate and incubated statically at 18°C for 72 h. Triplicate samples of 200 ll and 1 ml were taken for fluorescence measurement in a black 96-well microplate (GloMax-Multi Detection System with blue optical kit, Promega), or for cell density measurement at OD 600 respectively . Relative fluorescence units (RFU) were corrected for background fluorescence from the vector-only control (pKC026) and plotted against the cell density measurements to obtain a nonlinear line of best fit, second-order polynomial within 95% confidence intervals (PRISM5, GRAPHPAD). Elicitation of curli fibres was assessed on indicator plates, containing no-salt lysogeny broth or M9 glucose medium supplemented with 40 lg ml À1 Congo red and 20 lg ml À1 Coomassie brilliant blue, solidified with 15 g L À1 agar, from three independent experimental replicates, with Salmonella enterica serovar Senftenberg (isolate 20070885) as a positive control. Colony colour was scored dark red (curli positive, e.g. for S. enterica), red, pink or white (negative control) as per Zhou et al. (2013).

Plant material
Nicotiana benthamiana Domin was grown from seed stocks maintained at the James Hutton Institute. Seeds of spinach (Spinacia oleracea L. var. Amazon), lettuce (Lactuca sativa L. var. All Year Round, termed AYR) and tomato (Solanum lycopersicum L. var. Moneymaker) were obtained from Sutton Seeds, UK. Seeds were sown in commercial compost and maintained under glasshouse conditions with 16 h daylight, daytime temperature 26°C maximum, night-time 22°C before transfer to an environmental cabinet with 16 h daylight, at a constant temperature of 18°C and 65-70% humidity prior to inoculation. Plants were used for experiments at around 4 weeks after sowing. Spinach leaf lysates were generated as described previously (Crozier et al., 2016), essentially by grinding fresh leaves from~4 week old plants in liquid nitrogen, and clarifying by centrifugation and heat treatment at 50°C, prior to filtration. The lysates were used at 40% (v/v) final concentration.
mOrange-LTI6b plasmids were generated by modification of the EGFP-LTI6b marker of Kurup and coworkers (Kurup et al., 2005), were mobilized into Agrobacterium strain LBA4404 and used to transform N. benthamiana leaf segments as described by Horsch and associates (Horsch et al., 1985), with spectinomycin-resistant plants (termed mOrg-LTI-benth) being regenerated and screened for expression by fluorescence microscopy. As the mOrange protein labels the plant cell plasma membrane, these plants were utilized to identify the position of intact plant cells in the absence of other staining methods.

Measurement of internalization ability
Escherichia coli Sakai, grown as described above, was suspended in sterile distilled water at an OD 600 of 0.02 (~10 7 CFU ml À1 ) and then further diluted 1:100 000 to achieve~10 2 CFU ml À1 , at a volume of 1 L per six plants. To inoculate leaves, the plants (still in their pots) were inverted and the aerial tissue submerged gently into the bacterial suspension for 30 s, and the plants placed upright and left for 1 h before sampling for the first time point. To inoculate the roots, the plant pots were submerged to~1/3 height into the bacterial suspension, left for 1 h and removed. In both cases, sterile distilled water was used as a negative control. Samples were taken at time 0 (i.e. 1 hpi), day 5 and day 10 by aseptic removal of leaves or roots. The roots were washed in sterile distilled water to remove compost particles. The fresh weight of the samples was determined and they were stored briefly in 20 ml PBS prior to processing. To measure the total counts, five replicate samples were homogenized using a mortar and pestle in PBS, plated onto MacConkey agar with kanamycin in serial dilutions, incubated at 37°C and colonies enumerated the next day. To measure internalized counts, a second batch of five replicate samples was submerged in 20 ml 0.03% (w.v) calcium hypochlorite. The samples were incubated for 5 min at room temperature, with gentle shaking at 80 r.p.m. and then washed five times in PBS. The presence of any remaining surface-associated E. coli Sakai was determined by placing each sample onto an agar dish containing MacConkey medium with kanamycin for~10 s, termed 'imprint plates'. The samples were then processed as for the total counts. Any samples with colonies on the imprint plates were discarded from subsequent analysis as incompletely surface-sterilized, whereas those with no surface-associated colonies were deemed to only contain internalized bacteria. Data analyses were conducted separately for the two inoculum levels (Genstat v16; VSNI Ltd., Hemel Hempstead, UK). As the data were highly skewed by the number of internalized samples where no bacteria were detected, two-phase analyses were carried out. First, the data were rescored as presence/absence of bacteria and analysed using logistic regression. Then, the non-responding samples were removed and unbalanced ANOVA was used to test for differences between treatments in samples where bacteria were detected. Differences were considered significant at the 95% confidence level.

Confocal microscopy
For imaging purposes, leaves were infiltrated with bacteria suspended in 0.5 9 MS buffer at 10 7 CFU ml À1 by pressure injection using a 1 ml syringe (without a needle) into the abaxial epidermis and the plants maintained in an environmental cabinet until observed. Leaf segments were infiltrated with sterile distilled water, to displace air from the apoplastic spaces between the spongy mesophyll cells, prior to mounting on microscope slides using double-sided tape. The abaxial surface of the leaf was observed using a Nikon A1R confocal laser scanning microscope mounted on a NiE upright microscope fitted with an NIR Apo 409 0.8W water dipping lens and GaAsP detectors. Images represent falsecoloured single sections, maximum intensity, 3D or orthogonal projections as indicated, produced using NISelements AR software. GFP (green) and chlorophyll (blue) were excited at 488 nm with the emissions at 500-530 nm and 663-737 nm, respectively, and mOrange or mKate (RFP) were excited at 561 nm with emission at 570-620 nm (magenta). Where appropriate, leaves were infiltrated with an aqueous solution of calcofluor white M2R (Fluorescent Brightener 28; Sigma, St. Louis, Missouri, USA) at 0.1 mg ml À1 that was excited sequentially at 405 nm and the emission detected at 425-475 nm (cyan), and/or 5 mM Congo red (5 mM) imaged using the same settings as mOrange. DAPI (4 0 , 6-diamidino-2-phenylindole), 1 lg ml À1 , was imaged using the same settings as calcofluor. On occasion, laser light was directed through the tissue to the transmitted light detector to obtain a transmission image (grey). Isolated bacteria from an 18°C overnight culture were stained in vitro with dyes as above before being mounted under a coverslip on a cushion of 2% Oxoid No. 1 agar and observed using a CFI Plan Apochromat VC 609 WI lens.

Determination of growth rates in planta
Escherichia coli Sakai were diluted to 10 3 or 10 5 CFU ml À1 in 0.5 9 MS and 0.1 ml infiltrated into individual leaves, one per plant, by pressure injection as described above and the plants maintained in an environmental cabinet. Leaves were harvested at random from six plants at the designated time points, ground in 1 ml PBS and added to 5 ml buffered peptone water (BPW) with chloramphenicol. This was serially diluted, 10-fold, to extinction with 3 9 1 ml replicates for each dilution. Visible growth was assessed following overnight incubation at 37°C and 10 ll spots were plated onto MacConkey agar with chloramphenicol and incubated overnight for confirmation of E. coli Sakai-GFP. Colony growth was scored (+/À) and the MPN per ml of extract determined (Cochran, 1950), and this was multiplied by 6 to determine the number in the whole leaf. Using the six replicate measurements for N. benthamiana (four experiments), spinach and tomato (two experiments each), the estimates for most probable numbers of E. coli Sakai-GFP bacteria, after log 10 transformation, were analysed separately using a linear regression to estimate changes in population size during the first 20 days of the experiment (Genstat v16; VSNI Ltd.).