Spinach (Spinacia oleracea L.) grown in Italy under conventional farming practices and harvested at baby leaf stage (BBCH stage 13) was used in the experiments (January–May 2020). The spinach was imported by the Swedish company Vidinge Grönt AB (Kävlinge, Sweden) as unwashed material. Leaves that showed no visible symptoms of damage were selected for the experiments.

Different levels of standardized artificial damage in known patterns were inflicted upon the spinach leaves using a Derma stamp (HudRoller Of Sweden; 36 microneedles; 1 mm) or a Derma roller (HudRoller Of Sweden; 1 mm). Both of these generate dot-like lesions. Additionally, a scalpel was used to inflict more severe damage to leaves (Figure 1A). The extent of artificial damage was varied to generate four different levels of damage severity: undamaged (U), low (L), moderate (M), and high (H).

Leaves in the undamaged (U) group were not artificially damaged, and only damage to the leaf petiole caused by harvest was considered during damage quantification. Although this treatment served as a control, it did not preclude presence of microscopic damage, as leaves used in all treatments were only macroscopically observed from the abaxial and adaxial sides.

For low-level (L) damage, a total of 108 lesions of the same size were inflicted using the Derma stamp tool (3 × 36), causing total of about 0.1 cm² damage per leaf. Damage at level L was constant and independent of leaf size, and thus smaller leaves had higher relative damage than large leaves.

In moderate-level (M) damage, a Derma roller tool was used to inflict lesions along the midrib area, with about 0.04 cm² damage per cm² leaf area. The amount of damage was not constant, as the number of individual lesions per leaf varied with leaf area (small leaves had fewer lesions than large leaves), but relative damage per leaf was constant and not affected by leaf size. Individual lesions inflicted at damage level M were of same size, and were more numerous than at damage level L, but smaller in size.

For high-level (H) damage, leaves were first damaged as in group M and then damaged further by making three cuts with a scalpel, giving cut-related damage that can potentially occur during harvest and post-harvest processing (Mulaosmanovic et al., submitted). At damage level H, about 0.05 cm² damage was inflicted per cm² of leaf area. Number of lesions was not constant, but relative damage per leaf was not affected by leaf size.

As both downstream damage quantification and microorganism extraction methods were destructive, one separate batch of artificially damaged leaves was prepared per method. Measured leaf and lesion areas obtained from batch used for damage quantification were further used as a basis for prediction of leaf-scale damaged area (mm², %) and number of lesions on E. coli O157:H7 inoculated leaves.

An E. coli O157:H7 strain, lacking virulence factors verotoxin-1 and -2, but expressing eae, was used for the experimental work. This strain hosts the pGLO plasmid, which encodes for ampicillin resistance and green fluorescent protein (GFP), coupled with an arabinose responsive promoter. The bacteria were routinely grown in lysogeny broth (LB; Sigma-Aldrich, United States) supplemented with ampicillin (100 μg mL^–1; Sigma-Aldrich, Belgium) and arabinose (0.2%; Merck KGaA, Germany) for approximately 18 h at 37°C. Details of the E. coli O157:H7 inoculum preparation protocol are provided by El-Mogy and Alsanius (2012). The final concentration of inoculum solution was set to 10⁶ CFU of E. coli O157:H7 gfp+ × mL^–1 in 0.085% NaCl.

Directly post-damage, spinach leaves used for all damage levels (n = 15) were individually dip-inoculated by immersing them for 15 s in the inoculum suspension using forceps. The leaf petiole was kept outside the inoculum during inoculation, to prevent contact of the inoculant with harvest damage to the basal edge of the petiole. Excess inoculum suspension was removed by holding the leaf tip against a paper towel until all excess suspension had been removed.

E. coli O157:H7 gfp+ was extracted from individual spinach leaves directly post-inoculation for all damage levels (0 days post-inoculation, dpi), in order to determine the amount of bacterial cells added to single leaves. The remainder of the inoculated leaves were incubated in sealed plastic bags at room temperature for 24 h (1 dpi) and 48 h (2 dpi) (Supplementary Figure 1 in Supplementary Material).

Non-washed spinach leaves retrieved from a commercial processing plant for leafy vegetables were used and the leaves were not decontaminated before starting the experiment, so they were inhabited by an autochthonous (resident) microbiome. Leaf weight and area were correlated, with leaf weight explaining 71% of leaf area variations (p < 0.001) (Figure 1B). Damaged leaf area varied widely between the damage treatments (Figure 1C) and separated the treatments into three distinct groups, i.e., with no (U), low (L), and moderate (M)/high (H) artificial damage. Thus, saturation in damaged leaf area was observed in the moderately damaged leaves, and additional leaf cuts did not significantly alter the damage level (Figure 1C). This might be because the proportion of large leaves was higher for group M than for the high damage group (Supplementary Figure 2), thus leading to more lesions per leaf (Figure 1D).

The design of this study considered pathogen invasion with double perturbation dimensions caused by leaf damage, namely (i) nutrient pulse and (ii) landscape modification (Figure 2). Damaged leaves not only exhibit a richer pool of readily available nutrients, but also, at the damage site, a larger surface to be colonized, free from barriers existing in intact leaves. Different studies have investigated bacterial invasion on leaves (Melotto et al., 2006; Underwood et al., 2007; Barak et al., 2011; Van Elsas et al., 2012). With respect to plants, invasion as a concept has mainly been studied with respect to the microbial community structure of soils and roots (Kinnunen et al., 2016). Several studies have compared altered resource (nutrient) availability (Mallon, 2015; Mallon et al., 2015; Berg and Koskella, 2018), revealing contrasting invasion success. Although there is no such thing as an intact leaf (Mulaosmanovic et al., 2020) in natural or crop production and processing settings [Mulaosmanovic et al., submitted], the impact of modifications in leaf landscape and nutritional conditions can be assessed. The present study examined interactions between leaf damage size and invasion success of the target organism E. coli O157:H7, for which plants are not considered to be a primary habitat (Brandl, 2006), but did not quantify the strength of the nutrient pulse caused by wounding. Changes in the microbial community structure assisted by damage have previously been characterized by Mulaosmanovic et al. [submitted], and were not included in the present assessment of the biological and food safety relevance of leaf damage.

Although the inoculum density was standardized to log 6.74 ± 0.03 CFU × mL^–1, the sum of culturable E. coli O157:H7 gfp+ × cm^–2 detached from spinach leaves directly after inoculation varied between the four damage levels, with the highest numbers on moderately and severely damaged leaves, and the lowest on low damaged and undamaged leaves (M, H > U, L; Supplementary Table 1). This general trend was found throughout the experiment, and at first sight might appear to be a systematic experimental error. However, the inoculated cells were subjected to nutrient stress when transferred from the nutrient-rich propagation conditions to deprived circumstances in the inoculum suspension, and then to leaves with no or low damage. The difference in sum of E. coli attached to the leaves might then be a consequence of physiological changes due to starvation, leading to cell death or viable but not culturable (VBNC) cells (Nyström, 2001) and incapacity to overcome the abiotic and biotic hurdles of the leaf habitat. A limnological invasion study has found that inoculum density is a key factor for successful invasion (Acosta et al., 2015). The considerable load of E. coli O157:H7 does not compare with natural contamination conditions, but with the presence of a severe E. coli shedder (Omisakin et al., 2003). Comparable contamination rates were used in previous studies (Wright et al., 2017) and may occur when storing leafy vegetables at abuse temperature (Söderqvist et al., 2017). Furthermore, studies utilizing different inoculum densities show that inoculum size is not in itself the most decisive parameter for leaf colonization; populations with low densities at the start produce similar outcomes to E. coli populations with high initial densities (Wright et al., 2017). Crop type appears to be more crucial for the fate of E. coli on plants (Hartmann et al., 2017; Darlison et al., 2019).

Success of leaf surface attachment was monitored indirectly by analyzing the prevalence of E. coli O157:H7 gfp+ in the supernatant suspension after the repeated centrifugation (detachment, washing). The treatments used in this study to remove E. coli O157:H7 gfp+ loosely attached to spinach leaves directly after inoculation showed substantially higher removal rates than approaches using rinsing only (Uhlig et al., 2017) or immersion of leafy vegetables at various water temperatures and storage temperatures using Listeria monocytogenes (Natsou et al., 2012). Of the attached E. coli cells, 97.1–99.3% could be removed by the detachment and washing steps (M > U > L > H; % CFU × cm^–2: 99.27A, 98.86A, 97.95A, 97.05A) directly after inoculation (Figure 3). This aligns with results presented by Jensen et al. (2015) and Pezzuto et al. (2016) when using high inoculum density of Salmonella spp. However, the proportion of cells removed from the total population through the two centrifugation (detachment, washing) steps decreased gradually over time in all treatments involving artificial leaf damage, leaving 2–15% of the total E. coli O157:H7 gfp+ associated with the leaves at 2 dpi (Figure 3). With respect to log values in the washed off fraction, the CFU of E. coli × cm^–2 increased over time, indicating a greater total population (Supplementary Figure 3 and Supplementary Table 1). Proliferation and loose attachment of the inoculant, or resuscitation from VBNC, might serve as alternative explanations for this observation. The conflicting trends in removal rates of E. coli are puzzling. The lower removal rate of E. coli O157:H7 gfp+ from no and low damaged leaves might indicate better retention of E. coli O157:H7 gfp+ than on leaves with moderate and high damage. However, this was not supported by the results for the latter treatments. An alternative explanation could be that organic nutrients in the rinsate from moderately to highly damaged leaves might have supported reproduction of the inoculant. Interestingly, this effect persisted over time, whereas the initial nutrient influx due to damage should have been exhausted by 1 and 2 dpi. Rinsate nutrients cannot account for the increased number of E. coli O157:H7 gfp+ detached from moderately and highly damaged leaves directly after inoculation. Alternatively, wounding might impact the physiological status of E. coli O157:H7 gfp+ cells, retaining cells on highly wounded tissue in a viable and culturable stage, whereas cells on nutrient-depleted leaves progress to VBNC or die. Furthermore, wounding per se might create conditions more conducive to E. coli O157:H7 gfp+ cells passively adhering to the leaf surface, e.g., increased surface area, altered physical and chemical conditions, or altered net charge of leaf surface. The initial higher recovery is most likely a result of additional inoculation solution adhering the leaves due to change in the hydrophobicity of the leaf surface in damaged area. The total population of E. coli O157:H7 gfp+ undoubtedly grew more on damaged leaves. The numbers of introduced E. coli removed from moderately to severely damaged leaves at 1 and 2 dpi were higher, compared with leaves with no, or low levels of damage. We could speculate that leaves with damage nearing saturation partly supported greater population of E. coli over 24 and 48 h. This is based on the assumption that cells detach and wash off at similar rates, regardless of the level of damage inflicted on leaves. This, however, does not fully explain the magnitudes higher recovery after 1 and 2 dpi, and the explanations provided herein are not mutually exclusive.

For surface sterilization, the commercially administered concentration of sodium hypochlorite was used. However, verification of the surface decontamination using abaxial and adaxial leaf printing on agar plates showed that the inoculated strain survived the treatment to a great extent as >90% of the leaf imprints were positive for growth of the artificially introduced strain. In fact, limited efficacy of standard sanitation practices for removal of E. coli has been reported in different studies (Allende et al., 2008, 2009; López-Gálvez et al., 2009; Ölmez, 2010; Keskinen and Annous, 2011; Davidson et al., 2013; Wright et al., 2017). A limitation of surface sterilization followed by maceration and plating is that bacteria which are not internalized, but somehow shielded from surface sanitation, may be classified as internalized. We therefore avoid referring to macerate-associated E. coli as “internalized,” as we could not fully exclude presence of shielded non-internalized bacterial cells on the leaf surface.

Determination of E. coli O157:H7 gfp+ concentrations in the macerate during incubation showed that overall log values varied significantly with respect to dpi (p < 0.001; Figure 3 and Supplementary Figure 3). Leaf damage had a significant impact on log CFU E. coli O157:H7 gfp+ re-isolated from the macerate (Figure 4A). Leaf weight, but not leaf area in all cases, was negatively correlated with the prevalence of E. coli counts in the macerate, and explained between 21 and 80% of the variation in the introduced organism. When exposed to the same level of damage, lighter leaves accumulated higher counts per gram tissue in the macerate. This may be due to leaf age rather than leaf weight (Supplementary Figure 4). By definition for commercially grown baby spinach leaves, younger baby spinach leaves are lighter. For iceberg lettuce, substantially richer nutritional conditions, and hence higher E. coli colonization, has been detected on younger than older leaves (Brandl and Amundson, 2008). This may indicate that young undamaged leaves pose a similar risk of causing food-borne disease outbreaks as young leaves with low or moderate damage.

Interestingly, already directly after inoculation (0 dpi), the inoculated strain was significantly more prevalent in macerate of severely damaged leaves (H) than in macerate from the other three treatments (Figure 4B). This is particularly surprising as the removal rate of various washing and detachment steps was highest in the treatments with moderate and severe damage (Supplementary Figure 3 and Supplementary Table 1) and might indicate changes in leaf landscape (Figure 5) and environmental and nutritional conditions due to wounding (Figure 2). A gradual change in counts of the introduced strain occurred during the subsequent days. The introduced strain increased in all artificially damaged leaves, but the pace was slower in leaves with a lower damage level. The strongest increase in numbers of macerate-inhabiting E. coli was found within the first day post-inoculation, with significantly higher presence of E. coli in macerate of leaves with higher damage levels than lower levels (U, L). Nevertheless, E. coli counts were more than 36- and 20-fold higher in leaf macerate at damage level L and M, respectively. The increase in culturable E. coli O157:H7 gfp+ in macerate from non-damaged and severely damaged leaves was less dramatic (U: 4.3-fold increase; H: 5.3-fold increase). This might be a consequence of the nutritional limitations between the treatments, where E. coli O157:H7 gfp+ regained its culturability (Cuny et al., 2005), potentially following access to unexploited nutrient pulses in the L and M leaves, but not in undamaged leaves. After 48 h of incubation (2 dpi), no difference in occurrence of the inoculated E. coli strain (log CFU × cm^–2) was found between leaves exposed to the three different levels of artificial damage. The increase in E. coli between dpi 1 and 2 was similar in macerate from damaged leaves (1.55- to 1.59-fold). The values all deviated significantly from log values of introduced E. coli in the non-damaged leaves, for which E. coli counts in leaf macerate increased 1.13-fold between dpi 1 and 2.

As the damaged area did not differ significantly between M and H leaves, a possible explanation for the higher amount of E. coli O157:H7 gfp+ associated with the macerate of highly damaged leaves directly after inoculation could be the type of damage. High damage introduces cuts to edges, changing a considerable part of the leaf to be hydrophilic rather than hydrophobic. This results in higher changes to the physicochemistry of the leaf, hence, more inoculum will adhere to the surface. Indeed, Takeuchi and Frank (2000) showed that E. coli O157:H7 preferably colonizes cut edges as opposed to the intact leaf surface, which is supported by the verification results obtained in the present study using confocal laser scanning microscopy (see section “Verification of Leaf Damage and E. coli O157:H7 Interactions Using Imaging”). By definition, leaf wounds are three-dimensional. In addition to the previously mentioned alteration in leaf landscape and abiotic conditions, the picture might be biased by the fact that quantification of damage area in this study was based on a two-dimensional assessment.

Although plants are not its primary habitat, E. coli O157:H7 is able to colonize leaves of leafy vegetables as both an epiphyte and endophyte (Deering et al., 2012; Hartmann et al., 2017; Wright et al., 2017). Microscopy-based techniques, such as scanning electron microscopy (SEM) and confocal laser scanning microscopy (CLSM) have been used to visualize plant-microbe interactions. Bacterial invasion into stomata or damaged tissue has been studied with SEM (Keskinen et al., 2009; López-Gálvez et al., 2010; Wang et al., 2010). Although SEM micrographs reveal presence of bacteria on leaf surface, with this approach it is not possible to distinguish between artificially introduced and residential bacteria. CLSM is the most common approach for corroborating presence of the particular fluorescent protein-tagged microorganism, and identifying internalized bacteria, where images can be taken at different microscopic depths. It has previously been used for visualization of internalized E. coli cells and colonies, and to determine depth of E. coli penetration into tissue of leafy vegetables (Takeuchi and Frank, 2000; Solomon et al., 2002; Brandl, 2008; Mitra et al., 2009; Hartmann et al., 2017; Wright et al., 2017).

Scanning electron micrographs of intact and artificially damaged and inoculated plant material at 2 dpi (Figure 5) revealed that bacteria on leaf surface preferred junctions between epidermal cells and stomatal openings on undamaged tissue. Edges of artificial lesions were highly colonized compared with undamaged leaf areas, and greater colonization was observed on larger wounds (Supplementary Video 2). CLSM assessment revealed green fluorescence along stomatal openings and in substomatal cavities on both intact and damaged leaves, even after surface sanitation with 0.25% sodium hypochlorite, but surface sanitation may have affected the density of E. coli O157:H7 gfp+ cells. CLSM scans of leaf lesions and adjacent areas at 2 dpi revealed increased surface colonization in the form of fluorescent matrix on the edges of artificial lesions and cells surrounding lesion (Figure 5 and Supplementary Figures 5–7). This indicates that, in addition to stomata, lesion edges may be preferred attachment sites, irrespective of their size, which corroborates previous findings (Takeuchi and Frank, 2000; Brandl, 2008).

Invasion of artificial wounds by E. coli O157:H7 gfp+ was confirmed in our study, and the severity of invasion was affected by the size of individual lesions (Figure 5, Supplementary Figures 5–7, and Supplementary Videos 1, 2), and relative leaf damage (Figure 4A). On leaves without artificial damage (U), E. coli O157:H7 gfp+ was mainly located in junctions between epidermal cells and along stomatal openings (Supplementary Figure 5B), or in sub-stomatal cavities at depth down to 5 μm (Supplementary Figure 5C and Supplementary Video 1). In artificially damaged leaves, invasion of the inoculant was observed in deeper tissue layers, 15–40 μm below the epidermal surface, with invasion depth depending on damage level (Supplementary Figures 5D–L). The greatest accumulation of E. coli O157:H7 gfp+ cells and penetration depth was observed with the high (H) damage level, next to the cut area, where E. coli O157:H7 gfp+ formed 8 μm × 8 μm sized colonies at a depth between 20 and 40 μm in the mesophyll tissue layer, possibly as a result of side internalization via artificial cut in the leaf tissue (Supplementary Figures 5J–L, 6J–L and Supplementary Video 2).

In lesions and adjacent areas in M and H damage leaves, aggregates of E. coli O157:H7 gfp+ were detected in the apoplast (Supplementary Figures 5, 7), a nutrient-rich habitat (Pignocchi and Foyer, 2003; Merget et al., 2019). E. coli O157:H7 gfp+ was found internalized in spaces between cells adjacent to lesions within the mesophyll tissue. This was characteristic for all damage levels, but particularly pronounced for leaves with high damage, where E. coli O157:H7 gfp+ was found at the similar depth in areas adjacent to lesions as in lesions (Supplementary Figures 7G–I). This indicates that, once internalized, E. coli O157:H7 gfp+ can multiply and disperse in apoplastic spaces, and not just persist as previously demonstrated (Deering et al., 2012; Wright et al., 2013). Bacterial invasion into the plant cells was not observed, only into the apoplastic spaces of the leaf and attached to mesophyll cells. This is in agreement with findings for E. coli Sakai (Wright et al., 2017) and for the same E. coli O157:H7 gfp+ strain used in the present study (Hartmann et al., 2017).

From the perspective of biological relevance, presence of leaf lesions and their size were found to affect the prevalence and invasion of E. coli O157:H7 gfp+ into leaf tissue of baby spinach. However, due to rapid proliferation after successfully breaching abiotic and biotic hurdles, the initial differences in CFU leveled out over the observation period at lower damage levels (L, M). As expected, driven by the suitable nutritional conditions, E. coli O157:H7 gfp+ preferentially colonized the edges of fresh wounds. Damage level played a role in depth of invasion and bacterial dispersal, with higher damage levels leading to deeper invasion and dispersal to adjacent areas. The lesion sizes tested were relevant for commercial settings. With respect to biological relevance of damage, new questions arise concerning e.g., the interactions between the persistently lower counts of E. coli O157:H7 gfp+ loosely associated with plant tissue and in macerate, and the presence and fate of VBNC cells. This study provided possible explanations for conflicting results, but we strongly recommend examination of this issue in forthcoming studies. From a consumer perspective, leaves displaying U-M damage levels would be macroscopically judged as immaculate in a shop display, while individual leaves exhibiting the most severe damage level (H) can occur in convenience leafy vegetable packages [Mulaosmanovic et al., submitted]. Mixing injured with macroscopically intact leaves inside the same bag enhances deterioration of intact leaves (Ariffin et al., 2017), which, in light of our results, may be of food safety relevance.

The starting point for the food safety risk assessment was a scenario with one contaminated leaf of spinach within a retail pack. In this study, contamination with the inoculation O157:H7 was considered. Although the leaves were washed and decontaminated, the inoculant internalized in some leaves. As a result of sodium hypochlorite treatment limitations, inoculated strain survived the sanitation treatment to a great extent (>90% positive leaf imprints). Therefore we acknowledge the presence of shielded non-internalized E. coli O157:H7 gfp+ cells on the leaf surface, retrieved from the macerate together with internalized cells. We considered the risk presented by a bag of spinach leaves (30 g or 70 g) with one or more internally contaminated leaves; whether the risk changed (increased) if the leaves were heavily damaged (macerate measured as undamaged, low, medium, and high); and whether the risk changed if leaves were included in the bag on the day of contamination or 1 or 2 days later.

Based on the geometrical means, the occurrence of culturable E. coli O157:H7 gfp+ in the macerate increased by one order of magnitude within 2 dpi. The risk of consumers’ infection from consumption of 1 g of contaminated spinach is shown in Table 1. The high likelihood of infection indicated by ingesting 1 g of leaves implies that even one contaminated leaf in a bag of 30 or 70 gram would present a high risk to the consumer.

Macroscopically intact leaves appeared to represent an infection risk if contaminated at high inoculum densities. This complements the post-EHEC 2011 outbreak debate (European Food Safety Authority [EFSA], 2011), and supports findings in earlier studies of Brandl and Mandrell (2002) for macroscopically intact leaf areas after co-inoculation of Salmonella Thompson and Pantoea agglomerans. It was demonstrated in this study for E. coli O157:H7 (Figures 3–5, Supplementary Figures 3, 5–7, Supplementary Video 1, and Supplementary Table 1).

Based on the risk assessment, the presence of one contaminated leaf in a 70 g pack appears to result in an infection risk of almost 100%, irrespective of leaf damage level. The infection risk remained high even 1 and 2 days after introduction of the inoculant. However, the inoculum density used in this study was very high. The inoculation parameters were chosen in order to retrieve the inoculant in the macerate even of undamaged leaves. It should be recalled that the inoculum density is comparable to the impact of exposure to fecal material from cattle, which are super-shedders (MacCabe et al., 2019). Moreover, the 15 s of submersion in inoculant was shorter than the usual exposure time in the primary washing step (Grudén et al., 2016).

The risk assessment findings extracted from this study are of interest for the establishment of hurdles and measures preventing transmission within the horticultural value network:

The incubation period used in the present experiment can be viewed as the interval between onset of contamination and consumption. To translate these results into prevention measures, two hurdles can be identified, namely:

Each of these hurdles appear to represent around a one log reduction in bacterial numbers under high inoculum density conditions. Future studies on whether these hurdles represent the same log reduction at lower levels of contamination are needed.

The following food safety insights could be inferred from these results:

This study showed that leaf damage has a significant impact on numbers of E. coli O157:H7 gfp+ retrieved from macerate. In order to follow the introduced strain in the different segments of plant tissue, we performed dip inoculation of entire single leaves. Due to the destructive nature of the analyses, the results indicate changes over time, but cannot be translated into growth and survival of the introduced strain. The results showed that: